]> RIFT: Routing in Fat Trees Juniper Networks
1137 Innovation Way Sunnyvale CA 94089 United States of America prz@juniper.net
Juniper Networks
1137 Innovation Way Sunnyvale CA 94089 United States of America jhead@juniper.net
Hudson River Trading
United States of America as3957@gmail.com
Individual
France pascal.thubert@gmail.com
Individual
brunorijsman@gmail.com
Yandex
fl0w@yandex-team.ru
RTG rift Routing protocol IGP Anisotropic routing Clos Fat Tree Cloud Distance vector Link state Optimized flooding Zeroconf disaggregation fabric ZTP LIE TIE TIDE TIRE KV-TIE multi-plane ToF This document defines a specialized, dynamic routing protocol for Clos, fat tree, and variants thereof. These topologies were initially used within crossbar interconnects and consequently router and switch backplanes, but their characteristics make them ideal for constructing IP fabrics as well. The protocol specified by this document is optimized towards the minimization of control plane state to support very large substrates as well as the minimization of configuration and operational complexity to allow for a simplified deployment of said topologies.
Introduction Clos topologies have gained prominence in today's networking, primarily as a result of the paradigm shift towards a centralized data center architecture that is poised to deliver a majority of computation and storage services in the future. Such networks are commonly called a fat tree / network in modern IP fabric considerations as a similar term for the original definition of the term Fat Tree. In most generic terms, and disregarding exceptions like horizontal shortcuts, those networks are all variations of a structured design isomorphic to a ranked lattice where the least upper bound is the "top of the fabric" and links closer to the top may be "fatter" to guarantee non-blocking bisectional capacity. Many builders of such IP fabrics desire a protocol that autoconfigures itself and deals with failures and misconfigurations with a minimum amount of human intervention. Such a solution would allow local IP fabric bandwidth to be consumed in a "standard component" fashion, i.e., provision it much faster and operate it at much lower costs than today, similar to how compute or storage is consumed already. In looking at the problem through the lens of such IP fabric requirements, Routing in Fat Trees (RIFT) addresses those challenges not through an incremental modification of either a link-state (distributed computation) or distance-vector (diffused computation) technique but rather a mixture of both, briefly described as "link-state towards the spines" and "distance vector towards the leaves". In other words, "bottom" levels are flooding their link-state information in the "northern" direction while each node generates under normal conditions a "default route" and floods it in the "southern" direction. This type of protocol naturally supports highly desirable address aggregation. Alas, such aggregation could drop traffic in cases of misconfiguration or while failures are being resolved. It could also cause persistent network partitioning, which has to be addressed by some adequate mechanism. The approach RIFT takes is described in and is based on automatic, sufficient disaggregation of prefixes in case of link and node failures. The protocol further provides:
  • optional fully automated construction of fat tree topologies based on detection of links without any configuration () while allowing for conventional configuration methods or an arbitrary mix of both,
  • the minimum amount of routing state held by nodes,
  • automatic pruning and load balancing of topology flooding exchanges over a sufficient subset of links (),
  • automatic address aggregation () and consequently automatic disaggregation () of prefixes on link and node failures to prevent traffic loss and suboptimal routing,
  • loop-free non-ECMP forwarding due to its inherent valley-free nature,
  • fast mobility (),
  • rebalancing of traffic towards the spines based on bandwidth available (), and finally
  • mechanisms to synchronize a limited key-value datastore () that can be used after protocol convergence to, e.g., bootstrap higher levels of functionality on nodes.
illustrates a simplified, conceptual view of a RIFT fabric with its routing tables and topology databases using IPv4 as the address family. The top of the fabric's link-state database holds information about the nodes below it and the routes to them. When referring to , /32 notation corresponds to each node's IPv4 loopback address (e.g., A/32 is node A's loopback, etc.) and 0/0 indicates a default IPv4 route. The first row of database information represents the nodes for which full topology information is available. The second row of database information indicates that partial information of other nodes in the same level is also available. Such information will be needed to perform certain algorithms necessary for correct protocol operation. When the "bottom" (or in other words leaves) of the fabric is considered, the topology is basically empty and, under normal conditions, the leaves hold a load-balanced default route to the next level. The remainder of this document fills in the protocol specification details.
RIFT Information Distribution [A,B,C,D] [E] +---------+ +---------+ A/32 @ [C,D] | E | | F | B/32 @ [C,D] +-+-----+-+ +-+-----+-+ C/32 @ C | | | | D/32 @ D | | | | | | +--+ | | | | | | +---------)--+ | | | | | | | | | | +---------+ | | | | | | [A,B] +-+-----+-+ +-+-----+-+ [A,B] [D] | C | | D | [C] +-+-----+-+ +-+-----+-+ 0/0 @ [E,F] | | | | 0/0 @ [E,F] A/32 @ [A] | | | | A/32 @ A B/32 @ [B] | | +--+ | B/32 @ B | | | | | +---------)--+ | | | | | | +---------+ | | | | | | +-+-----+-+ +-+-----+-+ 0/0 @ [C,D] | A | | B | 0/0 @ [C,D] +---------+ +---------+
Requirements Language The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.
A Reader's Digest This section is an initial guided tour through the document in order to convey the necessary information for different readers, depending on their level of interest. The authors recommend reading the HTML or PDF versions of this document due to the inherent limitation of text version to represent complex figures. The "Terminology" section should be used as a supporting reference as the document is read. The indications of direction (i.e., "top", "bottom", etc.) referenced in are of paramount importance. RIFT requires a topology with a sense of top and bottom in order to properly achieve a sorted topology. Clos, fat tree, and other similarly structured networks are conducive to such requirements. Where RIFT allows for further relaxation of these constraints will be mentioned later in this section. Several of the images in this document are annotated with "northern view" or "southern view" to indicate perspective to the reader. A "northern view" should be interpreted as "from the top of the fabric looking down", whereas "southern view" should be interpreted as "from the bottom looking up". Operators and implementors alike must decide whether multi-plane IP fabrics are of interest for them. illustrates an example of both single-plane in and multi-plane fabric in . Multi-plane fabrics require understanding of additional RIFT concepts (e.g., negative disaggregation in ) that are unnecessary in the context of fabrics consisting of a single-plane only. "Overview" and "Generalized Topology View" aim to provide enough context to determine if multi-plane fabrics are of interest to the reader. "Fallen Leaf Problem" and additionally Sections and describe further considerations that are specific to multi-plane fabrics. The fundamental protocol concepts are described starting in "Specification", but some subsections are less relevant unless the protocol is being implemented. The protocol transport is of particular importance for two reasons. First, it introduces RIFT's packet format content in the form of a normative Thrift model given in , which is carried in an according security envelope as described in . Second, the Thrift model component is a prerequisite to understanding the RIFT's inherent security features as defined in both "Security" and "Security Considerations". The normative schema defining the Thrift model can be found in Sections and . Furthermore, while a detailed understanding of Thrift and the model is not required unless implementing RIFT, they may provide additional useful information for other readers. If implementing RIFT to support multi-plane topologies, should be reviewed in its entirety in conjunction with the previously mentioned Thrift schemas. Sections not relevant to single-plane implementations will be noted later in this section. All readers dealing with implementation of the protocol should pay special attention to the Link Information Element (LIE) definitions as it not only outlines basic neighbor discovery and adjacency formation but also provides necessary context for RIFT's optional Zero Touch Provisioning (ZTP) and miscabling detection capabilities that allow it to automatically detect and build the underlay topology with basically no configuration. These specific capabilities are detailed in . For other readers, the following sections provide a more detailed understanding of the fundamental properties and highlight some additional benefits of RIFT, such as link-state packet formats, efficient flooding, synchronization, loop-free path computation, and link-state database maintenance (see Sections , , , , , , , , , , , and ). RIFT's ability to perform weighted unequal-cost load balancing of traffic across all available links is outlined in with an accompanying example. is the place where the single-plane vs. multi-plane requirement is explained in more detail. For those interested in single-plane fabrics, only is required. For the multi-plane-interested reader, Sections , , , and are also mandatory. is especially important for any multi-plane-interested reader as it outlines how the Routing Information Base (RIB) and Forwarding Information Base (FIB) are built via the disaggregation mechanisms but also illustrates how they prevent defective routing decisions that cause traffic loss in both single-plane or multi-plane topologies. contains a set of comprehensive examples that show how RIFT contains the impact of failures to only the required set of nodes. It should also help cement some of RIFT's core concepts in the reader's mind. Last but not least, RIFT has other optional capabilities. One example is the key-value datastore, which enables RIFT to advertise data post-convergence in order to bootstrap higher levels of functionality (e.g., operational telemetry). Those are covered in . More information related to RIFT can be found in the "RIFT Applicability" document, which discusses alternate topologies upon which RIFT may be deployed, describes use cases where it is applicable, and presents operational considerations that complement this document. "RIFT Day One" covers some practical details of existing RIFT implementations and deployment details.
Reference Frame
Terminology This section presents the terminology used in this document.
Bandwidth Adjusted Distance (BAD):
Each RIFT node can calculate the amount of northbound bandwidth available towards a node compared to other nodes at the same level and can modify the route distance accordingly to allow for the lower level to adjust their load balancing towards spines.
Bidirectional Adjacency:
Bidirectional adjacency is an adjacency where nodes of both sides of the adjacency advertised it in the Node TIEs with the correct levels and System IDs. Bidirectionality is used to check in different algorithms whether the link should be included.
Bow-tying:
Traffic patterns in fully converged IP fabrics normally traverse the shortest route based on hop count towards their destination (e.g., leaf, spine, leaf). Some failure scenarios with partial routing information cause nodes to lose the required downstream reachability to a destination and force traffic to utilize routes that traverse higher levels in the fabric in order to turn south again using a different route to resolve reachability (e.g., leaf, spine-1, superspine, spine-2, leaf).
Clos / fat tree:
This document uses the terms "Clos" and "fat tree" interchangeably where it always refers to a folded spine-and-leaf topology with possibly multiple Points of Delivery (PoDs) and one or multiple Top of Fabric (ToF) planes. Several modifications such as L2L shortcuts and multi-level shortcuts are possible and described further in the document.
Cost:
A natural number without a unit associated with a single entity. The cost is a monoid under addition. A cost may be associated with either a single link or prefix, or it may represent the sum of costs (distance) of links in the path between two nodes.
Crossbar:
Physical arrangement of ports in a switching matrix without implying any further scheduling or buffering disciplines.
Directed Acyclic Graph (DAG):
A finite directed graph with no directed cycles (loops). If links in a Clos are considered as either being all directed towards the top or vice versa, each of two such graphs is a DAG.
Disaggregation:
The process in which a node decides to advertise more specific prefixes southwards, either positively to attract the corresponding traffic or negatively to repel it. Disaggregation is performed to prevent traffic loss and suboptimal routing to the more specific prefixes.
Distance:
The sum of costs (bound by the infinite cost constant) between two nodes. A distance is primarily used to express separation between two entities and can be used again as cost in another context.
East-West (E-W) Link:
A link between two nodes at the same level. East-West links are normally not part of Clos or fat tree topologies.
Flood Repeater (FR):
A node can designate one or more northbound neighbor nodes to be flood repeaters. The flood repeaters are responsible for flooding northbound TIEs further north. The document sometimes calls them flood leaders as well.
Folded Spine-and-Leaf:
In case the Clos fabric input and output stages are equivalent, the fabric can be "folded" to build a "superspine" or top, which is called the ToF in this document.
Interface:
A layer 3 entity over which RIFT control packets are exchanged.
Key Value (KV) TIE:
A TIE that is carrying a set of key value pairs . It can be used to distribute non-topology-related information within the protocol.
Leaf-to-Leaf (L2L) Shortcuts:
East-West links at leaf level will need to be differentiated from East-West links at other levels.
Leaf:
A node without southbound adjacencies. Level 0 implies a leaf in RIFT, but a leaf does not have to be level 0.
Level:
Clos and fat tree networks are topologically partially ordered graphs, and "level" denotes the set of nodes at the same height in such a network. Nodes at the top level (i.e., ToF) are at the level with the highest value and count down to the nodes at the bottom level (i.e., leaf) with the lowest value. A node will have links to nodes one level down and/or one level up. In some circumstances, a node may have links to other nodes at the same level. A leaf node may also have links to nodes multiple levels higher. In RIFT, level 0 always indicates that a node is a leaf but does not have to be level 0. Level values can be configured manually or automatically as described in .
LIE:
This is an acronym for a "Link Information Element" exchanged on all the system's links running RIFT to form ThreeWay adjacencies and carry information used to perform RIFT Zero Touch Provisioning (ZTP) of levels.
Metric:
Used interchangeably with "cost".
Neighbor:
Once a ThreeWay adjacency has been formed, a neighborship relationship contains the neighbor's properties. Multiple adjacencies can be formed to a remote node via parallel point-to-point interfaces, but such adjacencies are not sharing a neighbor structure. Saying "neighbor" is thus equivalent to saying "a ThreeWay adjacency".
Node TIE:
This is an acronym for a "Node Topology Information Element", which contains all adjacencies the node discovered and information about the node itself. Node TIE should not be confused with a North TIE since "node" defines the type of TIE rather than its direction. Consequently, North Node TIEs and South Node TIEs exist.
North SPF (N-SPF):
A reachability calculation that is progressing northbound, for example, SPF that is using South Node TIEs only. Normally it progresses by only a single hop and installs default routes.
Northbound Link:
A link to a node one level up or, in other words, one level further north.
Northbound Representation:
The subset of topology information flooded towards higher levels of the fabric.
Overloaded:
Applies to a node advertising the overload attribute as set. The overload attribute is carried in the NodeFlags object of the encoding schema.
Point of Delivery (PoD):
A self-contained vertical slice or subset of a Clos or fat tree network normally containing only level 0 and level 1 nodes. A node in a PoD communicates with nodes in other PoDs via the ToF nodes. PoDs are numbered to distinguish them, and PoD value 0 (defined later in the encoding schema as common.default_pod) is used to denote "undefined" or "any" PoD.
Prefix TIE:
This is an acronym for a "Prefix Topology Information Element", and it contains all prefixes directly attached to this node in case of a North TIE and the necessary default routes the node advertises southbound in case of a South TIE.
Radix:
A radix of a switch is the number of switching ports it provides. It's sometimes called "fanout" as well.
Routing on the Host (RotH):
A modern data center architecture variant where servers/leaves are multihomed and consequently participate in routing.
Security Envelope:
RIFT packets are flooded within an authenticated security envelope that optionally enables protection of the integrity of information a node accepts if any of the mechanisms in are used. This is further described in .
Shortest Path First (SPF):
A well-known graph algorithm attributed to Dijkstra that establishes a tree of shortest paths from a source to destinations on the graph. The SPF acronym is used due to its familiarity as a general term for the node reachability calculations RIFT can employ to ultimately calculate routes, of which Dijkstra's algorithm is a possible one.
South Reflection:
Often abbreviated just as "reflection", it defines a mechanism where South Node TIEs are "reflected" from the level south back up north to allow nodes in the same level without E-W links to be aware of each other's node Topology Information Elements (TIEs).
South SPF (S-SPF):
A reachability calculation that is progressing southbound, for example, SPF that is using North Node TIEs only.
South/Southbound and North/Northbound (Direction):
When describing protocol elements and procedures, in different situations, the directionality of the compass is used, i.e., "lower", "south", and "southbound" mean moving towards the bottom of the Clos or fat tree network and "higher", "north", and "northbound" mean moving towards the top of the Clos or fat tree network.
Southbound Link:
A link to a node one level down or, in other words, one level further south.
Southbound Representation:
The subset of topology information sent towards a lower level.
Spine:
Any nodes north of leaves and south of ToF nodes. Multiple layers of spines in a PoD are possible.
Superspine, Aggregation/Spine, and Edge/Leaf Switches:
Typical level names in 5 stages folded Clos for levels 2, 1, and 0, respectively (counting up from the bottom). We normalize this language to talk about ToF, Top-of-Pod (ToP), and leaves.
System ID:
RIFT nodes identify themselves with a unique network-wide number when trying to build adjacencies or describe their topology. RIFT System IDs can be auto-derived or configured.
ThreeWay Adjacency:
RIFT tries to form a unique adjacency between two nodes over a point-to-point interface and exchange local configuration and necessary RIFT ZTP information. An adjacency is only advertised in Node TIEs and used for computations after it achieved ThreeWay state, i.e., both routers reflected each other in LIEs, including relevant security information. Nevertheless, LIEs before ThreeWay state is reached may already carry information related to RIFT ZTP.
TIDE:
The Topology Information Description Element carries descriptors of the TIEs stored in the node.
TIE:
This is an acronym for a "Topology Information Element". TIEs are exchanged between RIFT nodes to describe parts of a network such as links and address prefixes. A TIE always has a direction and a type. North TIEs (sometimes abbreviated as N-TIEs) are used when dealing with TIEs in the northbound representation, and South-TIEs are used (sometimes abbreviated as S-TIEs) for the southbound equivalent. TIEs have different types, such as node and prefix TIEs.
TIEDB:
The database holding the newest versions of all TIE headers (and the corresponding TIE content if it is available).
TIRE:
The Topology Information Request Element carries a set of TIDE descriptors. It can both confirm received and request missing TIEs.
Top of Fabric (ToF):
The set of nodes that provide inter-PoD communication and have no northbound adjacencies, i.e., are at the "very top" of the fabric. ToF nodes do not belong to any PoD and are assigned the common.default_pod PoD value to indicate the equivalent of "any" PoD.
Top of PoD (ToP):
The set of nodes that provide intra-PoD communication and have northbound adjacencies outside of the PoD, i.e., are at the "top" of the PoD.
ToF Plane or Partition:
In large fabrics, ToF switches may not have enough ports to aggregate all switches south of them, and with that, the ToF is "split" into multiple independent planes. explains the concept in more detail. A plane is a subset of ToF nodes that are aware of each other through south reflection or E-W links.
Valid LIE:
LIEs undergo different checks to determine their validity. The term "valid LIE" is used to describe a LIE that can be used to form or maintain an adjacency. The amount of checking itself depends on the Finite State Machine (FSM) involved and its state. A "minimally valid LIE" is a LIE that passes checks necessary on any FSM in any state. A "ThreeWay valid LIE" is a LIE that successfully underwent further checks with a LIE FSM in ThreeWay state. A minimally valid LIE is a subcategory of a ThreeWay valid LIE.
RIFT Zero Touch Provisioning (abbreviated as RIFT ZTP or just ZTP):
An optional RIFT mechanism that allows the automatic derivation of node levels based on minimum configuration, as detailed in . Such a minimum configuration consists solely of ToFs being configured as such. RIFT ZTP contains a recommendation for automatic collision-free derivation of the System ID as well.
Additionally, when the specification refers to elements of packet encoding or the constants provided in , a special emphasis is used, e.g., invalid_distance. The same convention is used when referring to finite state machine states or events outside the context of the machine itself, e.g., OneWay.
Topology
A Three-Level Spine-and-Leaf Topology ^ N +--------+ +--------+ Level 2 | |ToF 21| |ToF 22| W <-*-> E ++-+--+-++ ++-+--+-++ | | | | | | | | | S v P111/2 P121/2 | | | | MHP MHP | | | | ^ ^ ^ ^ | | | | | | | | | | | | +--------------+ | +-----------+ | | | +---------------+ | | | | | | | | | +-----------------------------+ | | ^ | | | | | | | All 0/0 0/0 0/0 +-----------------------------+ TIEs v v v | | | | | | | +-+ +<-0/0----------+ | | | | | | | | | | +-+----++ Optional +-+----++ ++----+-+ ++-----++ Level 1 | | E/W Link | | | | | | |Spin111+----------+Spin112| |Spin121| |Spin122| +-+---+-+ ++----+-+ +-+---+-+ ++---+--+ | | | | | | | | | +---0/0--->-----+ 0/0 | +----------------+ | 0/0 | | | | | | | | +---<-0/0-----+ | v | +--------------+ | | v | | | | | | | +-+---+-+ +--+--+-+ +-+---+-+ +---+-+-+ Level 0 | | L2L | | | | | | |Leaf111+~~~~~~~~~~+Leaf112| |Leaf121| |Leaf122| +-+-----+ +-+---+-+ +--+--+-+ +-+-----+ + + \ / + + Prefix111 Prefix112 \ / Prefix121 Prefix122 Multi-homed Prefix +---------- PoD 1 ---------+ +---------- PoD 2 ---------+
The topology in is referred to in all further considerations. This figure depicts a generic "single-plane fat tree" and the concepts explained using three levels apply by induction to further levels and higher degrees of connectivity.
Topology with Multiple Planes
Further, this document will also deal with designs that provide only sparser connectivity and "partitioned spines", as shown in and explained further in .
RIFT: Routing in Fat Trees The remainder of this document presents the detailed specification of the RIFT protocol, which in the most abstract terms has many properties of a modified link-state protocol when distributing information northbound and a distance-vector protocol when distributing information southbound. While this is an unusual combination, it does quite naturally exhibit desired properties.
Overview
Properties The most singular property of RIFT is that it only floods link-state information northbound so that each level obtains the full topology of levels south of it. Link-State information is, with some exceptions, not flooded East-West nor back south again. Exceptions like south reflection is explained in detail in , and east-west flooding at the ToF level in multi-plane fabrics is outlined in . In the southbound direction, the necessary routing information required (normally just a default route as per ) only propagates one hop south. Those nodes then generate their own routing information and flood it south to avoid the overhead of building an update per adjacency. The East-West direction is described later in the document. Those information flow constraints create not only an anisotropic protocol (i.e., the information is not distributed "evenly" or "clumped" but summarized along the north-south gradient) but also a "smooth" information propagation where nodes do not receive the same information from multiple directions at the same time. Normally, accepting the same reachability on any link, without understanding its topological significance, forces tie-breaking on some kind of distance function. And such tie-breaking ultimately leads to hop-by-hop forwarding by shortest paths only. In contrast to that, RIFT, under normal conditions, does not need to tie-break the same reachability information from multiple directions. Its computation principles (south forwarding direction is always preferred) lead to valley-free forwarding behavior. In the shortest terms, valley-free paths allow reversal of direction from a packet heading northbound to southbound while permitting traversal of horizontal links in the northbound phase at most once. Those principles guarantee loop-free forwarding and with that can take advantage of all such feasible paths on a fabric. This is another highly desirable property if available bandwidth should be utilized to the maximum extent possible. To account for the "northern" and the "southern" information split, the link state database (LSDB) is partitioned accordingly into "north representation" and "south representation" Topology Information Elements (TIEs). In the simplest terms, the North TIEs contain a link-state topology description of lower levels and South TIEs simply carry a node description of the level above and default routes pointing north. This oversimplified view will be refined gradually in the following sections while introducing protocol procedures and state machines at the same time.
Generalized Topology View This section and are dedicated to multi-plane fabrics, in contrast with the single-plane designs where all ToF nodes are topologically equal and initially connected to all the switches at the level below them. The multi-plane design is effectively a multidimensional switching matrix. To make that easier to visualize, this document introduces a methodology depicting the connectivity in two-dimensional pictures. Further, it can be leveraged that what is under consideration here is basically stacked crossbar fabrics where ports align "on top of each other" in a regular fashion. A word of caution to the reader: At this point, it should be observed that the language used to describe Clos variations, especially in multi-plane designs, varies widely between sources. This description follows the terminology introduced in . This terminology is needed to follow the rest of this section correctly.
Terminology and Glossary This section describes the terminology and abbreviations used in the rest of the text. Though the glossary may not be clear on a first read, the following sections will introduce the terms in their proper context.
P:
Denotes the number of PoDs in a topology.
S:
Denotes the number of ToF nodes in a topology.
K:
To simplify the visual aids, notations, and further considerations, the assumption is made that the switches are symmetrical, i.e., they have an equal number of ports pointing northbound and southbound. With that simplification, K denotes half of the radix of a symmetrical switch, meaning that the switch has K ports pointing north and K ports pointing south. K_LEAF (K of a leaf) thus represents both the number of access ports in a leaf node and the maximum number of planes in the fabric, whereas K_TOP (K of a ToP) represents the number of leaves in the PoD and the number of ports pointing north in a ToP Node towards a higher spine level and thus the number of ToF nodes in a plane.
ToF Plane:
Set of ToFs that are aware of each other by means of south reflection. Planes are designated by capital letters, e.g., plane A.
N:
Denotes the number of independent ToF planes in a topology.
R:
Denotes a redundancy factor, i.e., the number of ToP nodes in a PoD that are connected to a ToF plane. In a single-plane design, R is equal to K_LEAF.
Fallen Leaf:
A fallen leaf in a plane Z is a switch that lost all connectivity northbound to Z.
Clos as Crossed, Stacked Crossbars The typical topology for which RIFT is defined is built of P number of PoDs and connected together by S number of ToF nodes. A PoD node has 2K number of ports. From here on, half of them (K=Radix/2) are assumed to connect host devices from the south, and the other half is assumed to connect to interleaved PoD top-level switches to the north. The K ratio can be chosen differently without loss of generality when port speeds differ or the fabric is oversubscribed, but K=Radix/2 allows for more readable representation whereby there are as many ports facing north as south on any intermediate node. A node is hence represented in a schematic fashion with ports "sticking out" to its north and south, rather than by the usual real-world front faceplate designs of the day. provides a view of a leaf node as seen from the north, i.e., showing ports that connect northbound. For lack of a better symbol, the document chooses to use the "o" as ASCII visualization of a single port. In this example, K_LEAF has 6 ports. Observe that the number of PoDs is not related to the Radix unless the ToF nodes are constrained to be the same as the PoD nodes in a particular deployment.
A Leaf Node, K_LEAF=6 Top View +---+ | | | o | e.g., Radix = 12, K_LEAF = 6 | | | o | | | ------------------------- | o <------ Physical Port (Ethernet) ----+ | | ------------------------- | | o | | | | | | o | | | | | | o | | | | | +---+ v || || || || || || || +----+ +------------------------------------------------+ | | | | +----+ +------------------------------------------------+ || || || || || || || Side Views
The Radix of a PoD's top node may be different than that of the leaf node. Though, more often than not, a same type of node is used for both, effectively forming a square (K*K). In the general case, switches at the top of the PoD with K_TOP southern ports not necessarily equal to K_LEAF could be considered. For instance, in the representations below, we pick a 6-port K_LEAF and an 8-port K_TOP. In order to form a crossbar, K_TOP leaf nodes are necessary as illustrated in .
Southern View of Leaf Nodes of a PoD, K_TOP=8 +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | | | | | | | | | | | | | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+
As further visualized in , the K_TOP leaf nodes are fully interconnected with the K_LEAF ToP nodes, providing connectivity that can be represented as a crossbar when "looked at" from the north. The result is that, in the absence of a failure, a packet entering the PoD from the north on any port can be routed to any port in the south of the PoD and vice versa. And that is precisely why it makes sense to talk about a "switching matrix".
Northern View of a PoD's Spines, K_TOP=8 W <---*---> E +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | | | | | | | | | | | | | | | +--------------------------------------------------------+ | o o o o o o o o | +--------------------------------------------------------+ +--------------------------------------------------------+ | o o o o o o o o | +--------------------------------------------------------+ +--------------------------------------------------------+ | o o o o o o o o | +--------------------------------------------------------+ +--------------------------------------------------------+ | o o o o o o o o | +--------------------------------------------------------+ +--------------------------------------------------------+ | o o o o o o o o |<--+ +--------------------------------------------------------+ | +--------------------------------------------------------+ | | o o o o o o o o | | +--------------------------------------------------------+ | | | | | | | | | | | | | | | | | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | ^ | | | | ---------- ----------------------- | +----- Leaf Node Top of PoD Node (Spine) --+ ---------- -----------------------
Side views of this PoD is illustrated in Figures and .
Side View of a PoD, K_TOP=8, K_LEAF=6 Connecting to ToP Nodes Connecting to Client Nodes Connecting to ToP Nodes || || || || || || || || +----------------------------------------------------------------+ N | Top-of-PoD Node (Sideways) | ^ +----------------------------------------------------------------+ | || || || || || || || || * +----+ +----+ +----+ +----+ +----+ +----+ +----+ +----+ | |Leaf| |Leaf| |Leaf| |Leaf| |Leaf| |Leaf| |Leaf| |Leaf| v |Node| |Node| |Node| |Node| |Node| |Node| |Node| |Node| S +----+ +----+ +----+ +----+ +----+ +----+ +----+ +----+ || || || || || || || || Connecting to Client Nodes
Other Side View of a PoD, K_TOP=8, K_LEAF=6, 90-Degree Turn in E-W Plane from the Previous Figure Connecting to ToP Nodes Connecting to Client Nodes Connecting to ToP Nodes || || || || || || +----+ +----+ +----+ +----+ +----+ +----+ N |ToP | |ToP | |ToP | |ToP | |ToP | |ToP | ^ |Node| |Node| |Node| |Node| |Node| |Node| | +----+ +----+ +----+ +----+ +----+ +----+ * || || || || || || | +------------------------------------------------+ v | Leaf Node (Sideways) | S +------------------------------------------------+ Connecting to Client Nodes
As a next step, observe that a resulting PoD can be abstracted as a bigger node with a number K of K_POD = K_TOP * K_LEAF, and the design can recurse. It will be critical at this point that, before progressing further, the concept and the picture of "crossed crossbars" is understood. Else, the following considerations might be difficult to comprehend. To continue, the PoDs are interconnected with each other through a ToF node at the very top or the north edge of the fabric. The resulting ToF is not partitioned if and only if (IIF) every ToP node is connected to every ToF node. This topology is also referred to as a single-plane configuration and is quite popular due to its simplicity. There are K_TOP ToF nodes and K_LEAF ToP nodes because each port of a ToP node connects to a different ToF node. Consequently, it will take at least P * K_LEAF ports on a ToF node to connect to each of the K_LEAF ToP nodes of the P PoDs. illustrates this, looking at P=3 PoDs from above and 2 sides. The large view is the one from above, with the 8 ToF of 3 * 6 ports each interconnecting the PoDs and every ToP Node being connected to every ToF node.
Fabric Spines and ToFs in Single-Plane Design, 3 PoDs [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] <-----+ | | | | | | | | | [=================================] | -------------- | | | | | | | | +----- ToF [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] +----- Node ---+ | -------------- | | v +-+ +-+ +-+ +-+ +-+ +-+ +-+ +-+ <-----+ +-+ | | | | | | | | | | | | | | | | | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] ------------------------- | | [ |o| |o| |o| |o| |o| |o| |o| |o<--- Physical Port (Ethernet) | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] ------------------------- | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | | | | | | | | | | | | | | | | | | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] ---------- | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] <--- ToP Node --------+ | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] (Spine) | | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] ---------- | | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | | | | | | | | | | | | | | | | | | -+ +- +-+ v | | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | --| |--[ ]--| | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | ----- | --| |--[ ]--| | [ |o| |o| |o| |o| |o| |o| |o| |o| ] +--- PoD ---+ --| |--[ ]--| | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | ----- | --| |--[ ]--| | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | --| |--[ ]--| | [ |o| |o| |o| |o| |o| |o| |o| |o| ] | | --| |--[ ]--| | | | | | | | | | | | | | | | | | -+ +- +-+ | | +-+ +-+ +-+ +-+ +-+ +-+ +-+ +-+ +-+
The top view can be collapsed into a third dimension where the hidden depth index is representing the PoD number. One PoD can be shown then as a class of PoDs and hence save one dimension in the representation. The ToF node expands in the depth and the vertical dimensions, whereas the ToP nodes are constrained in the horizontal dimension. A port in the 2-D representation effectively represents the class of all the ports at the same position in all the PoDs that are projected in its position along the depth axis. This is shown in .
Collapsed Northern View of a Fabric for Any Number of PoDs / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ] +-+ +-+ +-+ +-+ +-+ +-+ +-+ +-+ ]] | | | | | | | | | | | | | | | | ] ----------------------- [ |o| |o| |o| |o| |o| |o| |o| |o| ] <-- Top of PoD Node (Spine) [ |o| |o| |o| |o| |o| |o| |o| |o| ] ----------------------- [ |o| |o| |o| |o| |o| |o| |o| |o| ]]]] [ |o| |o| |o| |o| |o| |o| |o| |o| ]]] ^^ [ |o| |o| |o| |o| |o| |o| |o| |o| ]] // PoDs [ |o| |o| |o| |o| |o| |o| |o| |o| ] // (depth) | |/| |/| |/| |/| |/| |/| |/| |/ // +-+ +-+ +-+/+-+/+-+ +-+ +-+ +-+ // ^ | -------- +----- ToF Node --------
As simple as a single-plane deployment is, it introduces a limit due to the bound on the available radix of the ToF nodes that has to be at least P * K_LEAF. Nevertheless, it will become clear that a distinct advantage of a connected or non-partitioned ToF is that all failures can be resolved by simple, non-transitive, positive disaggregation (i.e., nodes advertising more specific prefixes with the default to the level below them that is not propagated further down the fabric) as described in . In other words, non-partitioned ToF nodes can always reach nodes below or withdraw the routes from PoDs they cannot reach unambiguously. And with this, positive disaggregation can heal all failures and still allow all the ToF nodes to be aware of each other via south reflection. Disaggregation will be explained in further detail in . In order to scale beyond the "single-plane limit", the ToF can be partitioned into N number of identically wired planes where N is an integer divider of K_LEAF. The 1:1 ratio and the desired symmetry are still served, this time with (K_TOP*N) ToF nodes, each of (P*K_LEAF/N) ports. N=1 represents a non-partitioned ToF (superspine), and N=K_LEAF is a maximally partitioned ToF. Further, if R is any integer divisor of K_LEAF, then N=K_LEAF/R is a feasible number of planes and R is a redundancy factor that denotes the number of independent paths between 2 leaves within a plane. It proves convenient for deployments to use a radix for the leaf nodes that is a power of 2 so they can pick a number of planes that is a lower power of 2. The example in splits the ToF in 2 planes with a redundancy factor of R=3, meaning that there are 3 non-intersecting paths between any leaf node and any ToF node. A ToF node must have, in this case, at least 3*P ports and be directly connected to 3 of the 6 ToP nodes (spines) in each PoD. The ToP nodes are represented horizontally with K_TOP=8 ports northwards each.
Northern View of a Multi-Plane ToF Level, K_LEAF=6, N=2 +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ Plane 1 ----------- . ------------ . ------------ . ------------ . -------- Plane 2 +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | O | | O | | O | | O | | O | | O | | O | | O | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ ^ | | --------------------- +----- ToF Node Across Depth ---------------------
At the extreme end of the spectrum, it is even possible to fully partition the ToF with N=K_LEAF and R=1 while maintaining connectivity between each leaf node and each ToF node. In that case, the ToF node connects to a single port per PoD, so it appears as a single port in the projected view represented in . The number of ports required on the ToF node is more than or equal to P, i.e., the number of PoDs.
Northern View of a Maximally Partitioned ToF Level, R=1 Plane 1 +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ -+ +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | O | | O | | O | | O | | O | | O | | O | | O | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | ----------- . ------------------- . ------------ . ------- | Plane 2 | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | O | | O | | O | | O | | O | | O | | O | | O | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | ----------- . ------------ . ---- . ------------ . ------- | Plane 3 | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | O | | O | | O | | O | | O | | O | | O | | O | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | ----------- . ------------ . ------------------- . --------+<-+ Plane 4 | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | ----------- . ------------ . ------------ . ---- . ------- | | Plane 5 | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | ----------- . ------------ . ------------ . -------------- | | Plane 6 | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | | | O | | O | | O | | O | | O | | O | | O | | O | | | | +-| |--| |--| |--| |--| |--| |--| |--| |-+ | | +---+ +---+ +---+ +---+ +---+ +---+ +---+ +---+ -+ | ^ | | | | ---------------- ------------- | +----- ToF Node Class of PoDs ---+ ---------------- -------------
Fallen Leaf Problem As mentioned earlier, RIFT exhibits an anisotropic behavior tailored for fabrics with a north-south orientation and a high level of interleaving paths. A non-partitioned fabric makes a total loss of connectivity between a ToF node at the north and a leaf node at the south a very rare but possible occasion that is fully healed by positive disaggregation as described in . In large fabrics or fabrics built from switches with a low radix, the ToF may often become partitioned in planes, which makes it more likely that a given leaf is only reachable from a subset of the ToF nodes. This makes some further considerations necessary. A "fallen leaf" is a leaf that can be reached by only a subset of ToF nodes due to missing connectivity. If R is the redundancy factor, then it takes at least R breakages to reach a "fallen leaf" situation. In a maximally partitioned fabric, the redundancy factor is R=1, so any breakage in the fabric will cause one or more fallen leaves in the affected plane. R=2 guarantees that a single breakage will not cause a fallen leaf. However, not all cases require disaggregation. The following cases do not require particular action:
  • If a southern link on a node goes down, then connectivity through that node is lost for all nodes south of that link. There is no need to disaggregate since the connectivity to this node is lost for all spine nodes in the same fashion.
  • If a ToF node goes down, then northern traffic towards it is routed via alternate ToF nodes in the same plane and there is no need to disaggregate routes.
In a general manner, the mechanism of non-transitive, positive disaggregation is sufficient when the disaggregating ToF nodes collectively connect to all the ToP nodes in the broken plane. This happens in the following case:
  • If the breakage is the last northern link from a ToP node to a ToF node going down, then the fallen leaf problem affects only that ToF node, and the connectivity to all the nodes in the PoD is lost from that ToF node. This can be observed by other ToF nodes within the plane where the ToP node is located and positively disaggregated within that plane.
On the other hand, there is a need to disaggregate the routes to Fallen Leaves within the plane in a transitive fashion, that is, all the way to the other leaves, in the following cases:
  • If the breakage is the last northern link from a leaf node within a plane (there is only one such link in a maximally partitioned fabric) that goes down, then connectivity to all unicast prefixes attached to the leaf node is lost within the plane where the link is located. Southern Reflection by a leaf node, e.g., between ToP nodes, if the PoD has only 2 levels, happens in between planes, allowing the ToP nodes to detect the problem within the PoD where it occurs and positively disaggregate. The breakage can be observed by the ToF nodes in the same plane through the north flooding of TIEs from the ToP nodes. However, the ToF nodes need to be aware of all the affected prefixes for the negative, possibly transitive, disaggregation to be fully effective (i.e., a node advertising in the control plane that it cannot reach a certain more specific prefix than the default prefix, whereas such disaggregation in the extreme condition must be propagated further down southbound). The problem can also be observed by the ToF nodes in the other planes through the flooding of North TIEs from the affected leaf nodes, together with non-node North TIEs, which indicate the affected prefixes. To be effective in that case, the positive disaggregation must reach down to the nodes that make the plane selection, which are typically the ingress leaf nodes. The information is not useful for routing in the intermediate levels.
  • If the breakage is a ToP node in a maximally partitioned fabric (in which case it is the only ToP node serving the plane in that PoD that goes down), then the connectivity to all the nodes in the PoD is lost within the plane where the ToP node is located. Consequently, all leaves of the PoD fall in this plane. Since the Southern Reflection between the ToF nodes happens only within a plane, ToF nodes in other planes cannot discover fallen leaves in a different plane. They also cannot determine beyond their local plane whether a leaf node that was initially reachable has become unreachable. As the breakage can be observed by the ToF nodes in the plane where the breakage happened, the ToF nodes in the plane need to be aware of all the affected prefixes for the negative disaggregation to be fully effective. The problem can also be observed by the ToF nodes in the other planes through the flooding of North TIEs from the affected leaf nodes if the failing ToP node is directly connected to its leaf nodes, which can detect the link going down. Then again, the knowledge of the failure at the ToF level can only be useful if it is propagated transitively to all the leaves; it is useless above that level since the decision of placing a packet in a plane happens at the leaf that injects the packet in the fabric.
These abstractions are rolled back into a simplified example that shows that in the loss of the link between spine node 3 and leaf node 3 will make leaf node 3 a fallen leaf for ToF nodes in plane C. Worse, if the cabling was never present in the first place, plane C will not even be able to know that such a fallen leaf exists. Hence, partitioning without further treatment results in two grave problems:
  1. Leaf node 1 trying to route to leaf node 3 must not choose spine node 3 in plane C as its next hop since it will inevitably drop the packet when forwarding using default routes or do excessive bow-tying. This information must be in its routing table.
  2. A path computation trying to deal with the problem by distributing host routes may only form paths through leaves. The flooding of information about leaf node 3 would have to go up to ToF nodes in planes A, B, and D and then "loopback" over other leaves to ToF C, leading in extreme cases to traffic for leaf node 3 when presented to plane C taking an "inverted fabric" path where leaves start to serve as ToFs, at least for the duration of a protocol's convergence.
Discovering Fallen Leaves When aggregation is used, RIFT deals with fallen leaves by ensuring that all the ToF nodes share the same north topology database. This happens naturally in single-plane design by the means of northbound flooding and south reflection but needs additional considerations in multi-plane fabrics. To enable routing to fallen leaves in multi-plane designs, RIFT requires additional interconnection across planes between the ToF nodes, e.g., using rings as illustrated in . Other solutions are possible, but they either need more cabling or end up having much longer flooding paths and/or single points of failure. In detail, by reserving at least two ports on each ToF node, it is possible to connect them together by interplane bidirectional rings as illustrated in . The rings will be used to exchange full north topology information between planes. All ToFs having the same north topology allows, by the means of transitive, negative disaggregation described in , to efficiently fix any possible fallen leaf scenario. Somewhat as a side effect, the exchange of information fulfills the requirement for a full view of the fabric topology at the ToF level without the need to collate it from multiple points.
Using Rings to Bring All Planes and Bind Them at the ToF
Addressing the Fallen Leaves Problem One consequence of the "fallen leaf" problem is that some prefixes attached to the fallen leaf become unreachable from some of the ToF nodes. RIFT defines two methods to address this issue, denoted as positive disaggregation and negative disaggregation. Both methods flood corresponding types of South TIEs to advertise the impacted prefix(es). When used for the operation of disaggregation, a positive South TIE, as usual, indicates reachability to a prefix of given length and all addresses subsumed by it. In contrast, a negative route advertisement indicates that the origin cannot route to the advertised prefix. The positive disaggregation is originated by a router that can still reach the advertised prefix, and the operation is not transitive. In other words, the receiver does not generate its own TIEs or flood them south as a consequence of receiving positive disaggregation advertisements from a higher-level node. The effect of a positive disaggregation is that the traffic to the impacted prefix will follow the longest match and will be limited to the northbound routers that advertised the more specific route. In contrast, the negative disaggregation can be transitive and is propagated south when all the possible routes have been advertised as negative exceptions. A negative route advertisement is only actionable when the negative prefix is aggregated by a positive route advertisement for a shorter prefix. In such case, the negative advertisement "punches out a hole" in the positive route in the routing table, making the positive prefix reachable through the originator with the special consideration of the negative prefix removing certain next-hop neighbors. The specific procedures are explained in detail in . When the ToF switches are not partitioned into multiple planes, the resulting southbound flooding of the positive disaggregation by the ToF nodes that can still reach the impacted prefix is generally enough to cover all the switches at the next level south, typically the ToP nodes. If all those switches are aware of the disaggregation, they collectively create a ceiling that intercepts all the traffic north and forwards it to the ToF nodes that advertised the more specific route. In that case, the positive disaggregation alone is sufficient to solve the fallen leaf problem. On the other hand, when the fabric is partitioned in planes, the positive disaggregation from ToF nodes in different planes do not reach the ToP switches in the affected plane and cannot solve the fallen leaves problem. In other words, a breakage in a plane can only be solved in that plane. Also, the selection of the plane for a packet typically occurs at the leaf level and the disaggregation must be transitive and reach all the leaves. In that case, the negative disaggregation is necessary. The details on the RIFT approach to deal with fallen leaves in an optimal way are specified in .
Specification This section specifies the protocol in a normative fashion by either prescriptive procedures or behavior defined by Finite State Machines (FSMs). The FSMs, as usual, are presented as states a neighbor can assume, events that can occur, and the corresponding actions performed when transitioning between states on event processing. Actions are performed before the end state is assumed. The FSMs can queue events against themselves to chain actions or against other FSMs in the specification. Events are always processed in the sequence they have been queued. Consequently, "On Entry" actions for an FSM state are performed every time and right before the corresponding state is entered, i.e., after any transitions from previous state. "On Exit" actions are performed every time and immediately when a state is exited, i.e., before any transitions towards the target state are performed. Any attempt to transition from a state towards another on reception of an event where no action is specified MUST be considered an unrecoverable error, and the protocol MUST reset all adjacencies and discard all the states (i.e., force the FSM back to OneWay and flush all of the queues holding flooding information). The data structures and FSMs described in this document are conceptual and do not have to be implemented precisely as described here, i.e., an implementation is considered conforming as long as it supports the described functionality and exhibits externally observable behavior equivalent to the behavior of the standardized FSMs. The FSMs can use "timers" for different situations. Those timers are started through actions, and their expiration leads to queuing of corresponding events to be processed. The term "holdtime" is used often as shorthand for "holddown timer" and signifies either the length of the holding down period or the timer used to expire after such period. Such timers are used to "holddown" the state within an FSM that is cleaned if the machine triggers a HoldtimeExpired event.
Transport All normative RIFT packet structures and their contents are defined in the Thrift models in . The packet structure itself is defined in ProtocolPacket, which contains the packet header in PacketHeader and the packet contents in PacketContent. PacketContent is a union of the LIE, TIE, TIDE, and TIRE packets, which are subsequently defined in LIEPacket, TIEPacket, TIDEPacket, and TIREPacket, respectively. Further, in terms of bits on the wire, it is the ProtocolPacket that is serialized and carried in an envelope defined in within a UDP frame that provides security and allows validation/modification of several important fields without Thrift deserialization for performance and security reasons. Security models and procedures are further explained in .
Link (Neighbor) Discovery (LIE Exchange) RIFT LIE exchange auto-discovers neighbors, negotiates RIFT ZTP parameters, and discovers miscablings. The formation progresses under normal conditions from OneWay to TwoWay and then ThreeWay state, at which point it is ready to exchange TIEs as described in . The adjacency exchanges RIFT ZTP information in any of the states, i.e., it is not necessary to reach ThreeWay for ZTP to operate. RIFT supports any combination of IPv4 and IPv6 addressing, including link-local scope, on the fabric to form adjacencies with the additional capability for forwarding paths that are capable of forwarding IPv4 packets in the presence of IPv6 addressing only. IPv4 LIE exchange happens by default over a well-known IPv4 multicast address that may also be administratively configured (e.g., with a local scope). For IPv6 , exchange is performed over the link-local multicast scope address, which is configured or otherwise well-known. In both cases, a destination UDP port defined in the schema () is used unless configured otherwise. LIEs MUST be sent with an IPv4 Time to Live (TTL) or an IPv6 Hop Limit (HL) of either 1 or 255 to prevent RIFT information reaching beyond a single Layer 3 (L3) next hop in the topology. Observe that, for the allocated link-local scope IP multicast address, the TTL value of 1 is a more logical choice since the TTL value of 255 may, in some environments, lead to an early drop due to the suspicious TTL value for a packet addressed to such a destination. LIEs SHOULD be sent with network control precedence unless an implementation is prevented from doing so . Any LIE packet received on an address that is neither the well-known nor configured multicast or a broadcast address MUST be discarded. The originating port of the LIE has no further significance, other than identifying the origination point. LIEs are exchanged over all links running RIFT. An implementation may listen and send LIEs on IPv4 and/or IPv6 multicast addresses. A node MUST NOT originate LIEs on an address family if it does not process received LIEs on that family. LIEs on the same link are considered part of the same LIE FSM independent of the address family they arrive on. The LIE source address may not identify the peer uniquely in unnumbered or link-local address cases so the response transmission MUST occur over the same interface the LIEs have been received on. A node may use any of the adjacency's source addresses it saw in LIEs on the specific interface during adjacency formation to send TIEs (). That implies that an implementation MUST be ready to accept TIEs on all addresses it used as sources of LIE frames. A simplified version MAY be implemented on platforms with limited multicast support (e.g., Internet of Things (IoT) devices) by sending and receiving LIE frames on IPv4 subnet broadcast addresses or IPv6 all-routers multicast addresses. However, this technique is less optimal and presents a wider attack surface from a security perspective and should hence be used only as a last resort. A ThreeWay adjacency (as defined in the glossary) over any address family implies support for IPv4 forwarding if the ipv4_forwarding_capable flag in LinkCapabilities is set to true. In the absence of IPv4 LIEs with ipv4_forwarding_capable set to true, a node MUST forward IPv4 packets using gateways discovered on IPv6-only links advertising this capability. The mechanism to discover the corresponding IPv6 gateway is out of scope for this specification and may be implementation-specific. It is expected that the whole fabric supports the same type of forwarding of address families on all the links; any other combination is outside the scope of this specification. If IPv4 forwarding is supported on an interface, ipv4_forwarding_capable MUST be set to true for all LIEs advertised from that interface. If IPv4 and IPv6 LIEs indicate contradicting information, protocol behavior is unspecified. A node sending IPv4 LIEs MUST set the ipv4_forwarding_capable flag to true on all LIEs advertised from that interface. Operation of a fabric where only some of the links are supporting forwarding on an address family or have an address in a family and others do not is outside the scope of this specification. Any attempt to construct IPv6 forwarding over IPv4-only adjacencies is outside the scope of this specification. outlines protocol behavior pertaining to LIE exchange over different address family combinations. outlines the way in which neighbors forward traffic as it pertains to the ipv4_forwarding_capable flag setting across the same address family combinations. The table is symmetric, i.e., the local and remote columns can be exchanged to construct the remaining combinations. The specific forwarding implementation to support the described behavior is out of scope for this document. Control Plane Behavior for Neighbor Address Family Combinations
Local Neighbor Address Family Remote Neighbor Address Family LIE Exchange Behavior
IPv4 IPv4 LIEs and TIEs are exchanged over IPv4 only. The local neighbor receives TIEs from remote neighbors on any of the LIE source addresses.
IPv6 IPv6 LIEs and TIEs are exchanged over IPv6 only. The local neighbor receives TIEs from remote neighbors on any of the LIE source addresses.
IPv4, IPv6 IPv6 The local neighbor sends LIEs for both IPv4 and IPv6, while the remote neighbor only sends LIEs for IPv6. The resulting adjacency will exchange TIEs over IPv6 on any of the IPv6 LIE source addresses.
IPv4, IPv6 IPv4, IPv6 LIEs and TIEs are exchanged over IPv6 and IPv4. TIEs are received on any of the IPv4 or IPv6 LIE source addresses. The local neighbor receives TIEs from the remote neighbors on any of the IPv4 or IPv6 LIE source addresses.
IPv4, IPv6 IPv4 The local neighbor sends LIEs for both IPv4 and IPv6, while the remote neighbor only sends LIEs for IPv4. The resulting adjacency will exchange TIEs over IPv4 on any of the IPv4 LIE source addresses.
Forwarding Behavior for Neighbor Address Family Combinations
Local Neighbor Address Family Remote Neighbor Address Family Forwarding Behavior
IPv4 IPv4 Only IPv4 traffic can be forwarded.
IPv6 IPv6 If either neighbor sets ipv4_forwarding_capable to false, only IPv6 traffic can be forwarded. If both neighbors set ipv4_forwarding_capable to true, IPv4 traffic is also forwarded via IPv6 gateways.
IPv4, IPv6 IPv6 If the remote neighbor sets ipv4_forwarding_capable to false, only IPv6 traffic can be forwarded. If both neighbors set ipv4_forwarding_capable to true, IPv4 traffic is also forwarded via IPv6 gateways.
IPv4, IPv6 IPv4, IPv6 IPv4 and IPv6 traffic can be forwarded. If IPv4 and IPv6 LIEs advertise conflicting ipv4_forwarding_capable flags, the behavior is unspecified.
IPv4, IPv6 IPv4 IPv4 traffic can be forwarded.
The protocol does not support selective disabling of address families after adjacency formation, disabling IPv4 forwarding capability, or any local address changes in ThreeWay state, i.e., if a link has entered ThreeWay IPv4 and/or IPv6 with a neighbor on an adjacency and it wants to stop supporting one of the families, change any of its local addresses, or stop IPv4 forwarding, it MUST tear down and rebuild the adjacency. It MUST also remove any state it stored about the remote side of the adjacency such as associated LIE source addresses. Unless RIFT ZTP is used as described in , each node is provisioned with the level at which it is operating and advertises it in the level of the PacketHeader schema element. It MAY also be provisioned with its PoD. If the level is not provisioned, it is not present in the optional PacketHeader schema element and established by ZTP procedures, if feasible. If PoD is not provisioned, it is governed by the LIEPacket schema element assuming the common.default_pod value. This means that switches except ToF do not need to be configured at all. Necessary information to configure all values is exchanged in the LIEPacket and PacketHeader or derived by the node automatically. Further leaf flag definitions are found in as they have implications in terms of level and adjacency formation. Leaf flags are carried in HierarchyIndications. A node MUST form a ThreeWay adjacency if, at a minimum, the following first order logic conditions are satisfied on a LIE packet, as specified by the LIEPacket schema element and received on a link (such a LIE is considered a "minimally valid" LIE). Observe that, depending on the FSM involved and its state, further conditions may be checked, and even a minimally valid LIE can be considered ultimately invalid if any of the additional conditions fail:
  1. the neighboring node is running the same major schema version as indicated in the major_version element in PacketHeader and
  2. the neighboring node uses a valid System ID (i.e., a value different from IllegalSystemID) in the sender element in PacketHeader and
  3. the neighboring node uses a different System ID than the node itself and
  4. (the advertised MTU values in the LiePacket element match on both sides, while a missing MTU in the LiePacket element is interpreted as default_mtu_size) and
  5. both nodes advertise defined level values in the level element in PacketHeader and
  6. [
    1. the node is at the leaf_level value and does not already have any ThreeWay adjacencies to nodes that are at the Highest Adjacency ThreeWay (HAT), as defined in , with a level that is different than the adjacent node or
    2. the node is not at the leaf_level value and the neighboring node is at the leaf_level value or
    3. both nodes are at the leaf_level value and both indicate support for that described in or
    4. neither node is at the leaf_level value and the neighboring node is, at most, one level away.
    ]
LIEs arriving with IPv4 Time to Live (TTL) or an IPv6 Hop Limit (HL) different than 1 or 255 MUST be ignored.
LIE Finite State Machine This section specifies the precise, normative LIE FSM, which is also shown in . Additionally, some sets of actions often repeat and are hence summarized into well-known procedures. Events generated are fairly fine grained, especially when indicating problems in adjacency-forming conditions to simplify tracking of problems in deployment. The initial state is OneWay. The machine sends LIEs proactively on several transitions to accelerate adjacency bring-up without waiting for the corresponding timer tic.
LIE FSM | HALChanged | | HALSChanged | | LieRcvd | | FloodLeadersChanged | | SendLie | | MultipleNeighbors | | LevelChanged | | NeighborChangedLevel | | NeighborChangedAddress | | UnacceptableHeader | | MTUMismatch | | HoldtimeExpired | | NeighborDroppedReflection | | HALChanged | | HALSChanged | | LieRcvd | | NeighborDroppedReflection | | NeighborChangedBFDCapability | | NeighborChangedAddress | | UnacceptableHeader | | MTUMismatch | | HoldtimeExpired | | MultipleNeighbors | | FloodLeadersChanged | | SendLie | | LevelChanged | | MultipleNeighborsDone | Enter | MultipleNeighbors | | LevelChanged | | HALChanged | | HALSChanged | | LieRcvd | | NeighborDroppedReflection | | NeighborChangedLevel | | NeighborChangedAddress | | UnacceptableHeader | | MTUMismatch | | NeighborChangedMinorFields | | HoldtimeExpired | | FloodLeadersChanged | | SendLie | | NewNeighbor | | NewNeighbor | | MultipleNeighbors | | NeighborChangedLevel | | NeighborChangedAddress | | UnacceptableHeader | | MTUMismatch | | HoldtimeExpired | | LevelChanged | | HALChanged | | HALSChanged | | LieRcvd | | FloodLeadersChanged | | SendLie | | HATChanged | | ValidReflection | | UpdateZTPOffer | | TimerTick | | TimerTick | | HATChanged | | UpdateZTPOffer | | UpdateZTPOffer | | HATChanged | | TimerTick | | ValidReflection | | ValidReflection | ThreeWay MultipleNeighborsWait OneWay TwoWay | TimerTick | | HATChanged | | ValidReflection | | UpdateZTPOffer | Enter | V +-----------+ | OneWay |<----+ | | | TimerTick | | | LevelChanged | | | HALChanged | | | HATChanged | | | HALSChanged | | | LieRcvd | | | NeighborDroppedReflection | | | NeighborChangedLevel | | | NeighborChangedAddress | | | UnacceptableHeader | | | MTUMismatch | | | NeighborChangedMinorFields | | | HoldtimeExpired | | | FloodLeadersChanged | | | SendLie | | | UpdateZTPOffer | |-----+ | | | |<--------------------- (ThreeWay) | |---------------------> | | ValidReflection | | | |---------------------> (Multiple | | MultipleNeighbors Neighbors +-----------+ Wait) ^ | | | | | NewNeighbor | V (TwoWay) (OneWay) | ^ | | NeighborChangedLevel | | NeighborChangedAddress | | UnacceptableHeader | | MTUMismatch | | HoldtimeExpired | | V | +-----------+ | TwoWay |<----+ | | | TimerTick | | | LevelChanged | | | HALChanged | | | HATChanged | | | HALSChanged | | | LieRcvd | | | FloodLeadersChanged | | | SendLie | | | UpdateZTPOffer | |-----+ | | | |<---------------------- | |----------------------> (Multiple | | NewNeighbor Neighbors | | Wait) | | MultipleNeighbors +-----------+ ^ | | | ValidReflection | V (ThreeWay) (TwoWay) (OneWay) ^ | ^ | | | LevelChanged | | | NeighborChangedLevel | | | NeighborChangedAddress | | | UnacceptableHeader | | | MTUMismatch | | | HoldtimeExpired NeighborDropped- | | | Reflection | | | | V | +-----------+ | | ThreeWay |-----+ | | | |<----+ | | | TimerTick | | | HALChanged | | | HATChanged | | | HALSChanged | | | LieRcvd | | | ValidReflection | | | FloodLeadersChanged | | | SendLie | | | UpdateZTPOffer | |-----+ | |----------------------> (Multiple | | MultipleNeighbors Neighbors +-----------+ Wait) (TwoWay) (ThreeWay) | | V V +------------+ | Multiple |<----+ | Neighbors | | TimerTick | Wait | | HALChanged | | | HATChanged | | | HALSChanged | | | LieRcvd | | | ValidReflection | | | NeighborDroppedReflection | | | NeighborChangedBFDCapability | | | NeighborChangedAddress | | | UnacceptableHeader | | | MTUMismatch | | | HoldtimeExpired | | | MultipleNeighbors | | | FloodLeadersChanged | | | SendLie | | | UpdateZTPOffer | |-----+ | | | |<--------------------------- | |---------------------------> (OneWay) | | LevelChanged +------------+ MultipleNeighborsDone
The following words are used for well-known procedures:
  • PUSH Event: queues an event to be executed by the FSM upon exit of this action
  • CLEANUP: The FSM conceptually holds a "current neighbor" variable that contains information received in the remote node's LIE that is processed against LIE validation rules. In the event that the LIE is considered to be invalid, the existing state held by a "current neighbor" MUST be deleted.
  • SEND_LIE: create and send a new LIE packet
    1. reflecting the neighbor element as described in ValidReflection,
    2. setting the necessary not_a_ztp_offer variable if the level was derived from the last-known neighbor on this interface, and
    3. setting the you_are_flood_repeater variable to the computed value.
  • PROCESS_LIE:
    1. if LIE has a major version not equal to this node's major version or System ID equal to this node's System ID or IllegalSystemID, then CLEANUP, else
    2. if both sides advertise Layer 2 MTU values and the MTU in the received LIE does not match the MTU advertised by the local system or at least one of the nodes does not advertise an MTU value and the advertising node's LIE does not match the default_mtu_size of the system not advertising an MTU, then CLEANUP, then PUSH UpdateZTPOffer, then PUSH MTUMismatch, else
    3. if the LIE has an undefined level or this node's level is undefined or this node is a leaf and the remote level is lower than HAT or the LIE's level is not leaf and its difference is more than one from this node's level, then CLEANUP, then PUSH UpdateZTPOffer, then PUSH UnacceptableHeader, else
    4. PUSH UpdateZTPOffer, construct a temporary new neighbor structure with values from LIE, if no current neighbor exists, then set current neighbor to new neighbor, PUSH NewNeighbor event, CHECK_THREE_WAY, else
      1. if the current neighbor System ID differs from LIE's System ID, then PUSH MultipleNeighbors, else
      2. if the current neighbor stored level differs from LIE's level, then PUSH NeighborChangedLevel, else
      3. if the current neighbor stored IPv4/v6 address differs from LIE's address, then PUSH NeighborChangedAddress, else
      4. if any of the neighbor's flood address port, name, or local LinkID changed, then PUSH NeighborChangedMinorFields
      5. CHECK_THREE_WAY
  • CHECK_THREE_WAY: if the current state is OneWay, do nothing, else
    1. if LIE packet does not contain a neighbor then if the current state is ThreeWay, then PUSH NeighborDroppedReflection, else
    2. if the packet reflects this System ID and local port and the state is ThreeWay, then PUSH the ValidReflection event, else PUSH the MultipleNeighbors event.
States:
  • OneWay: The initial state the FSM is starting from. In this state, the router did not receive any valid LIEs from a neighbor.
  • TwoWay: This state is entered when a node has received a minimally valid LIE from a neighbor but not a ThreeWay valid LIE.
  • ThreeWay: This state signifies that ThreeWay valid LIEs from a neighbor have been received. On achieving this state, the link can be advertised in the neighbors element in NodeTIEElement.
  • MultipleNeighborsWait: Occurs normally when more than two nodes become aware of each other on the same link or a remote node is quickly reconfigured or rebooted without regressing to OneWay first. Each occurrence of the event SHOULD generate a notification to help operational deployments.
Events:
  • TimerTick: One-second timer tick, i.e., the event is provided to the FSM once a second by an implementation-specific mechanism that is outside the scope of this specification. This event is quietly ignored if the relevant transition does not exist.
  • LevelChanged: Node's level has been changed by ZTP or configuration. This is provided by the ZTP FSM.
  • HALChanged: Best HAL computed by ZTP has changed. This is provided by the ZTP FSM.
  • HATChanged: HAT computed by ZTP has changed. This is provided by the ZTP FSM.
  • HALSChanged: Set of HAL offering systems computed by ZTP has changed. This is provided by the ZTP FSM.
  • LieRcvd: Received LIE on the interface.
  • NewNeighbor: New neighbor is present in the received LIE.
  • ValidReflection: Received valid reflection of this node from the neighbor, i.e., all elements in the neighbor element in LiePacket have values corresponding to this link.
  • NeighborDroppedReflection: Lost previously held reflection from the neighbor, i.e., the neighbor element in LiePacket does not correspond to this node or is not present.
  • NeighborChangedLevel: Neighbor changed the advertised level from the previously held one.
  • NeighborChangedAddress: Neighbor changed the IP address, i.e., the LIE has been received from an address different from previous LIEs. Those changes will influence the sockets used to listen to TIEs, TIREs, and TIDEs.
  • UnacceptableHeader: Unacceptable header received.
  • MTUMismatch: MTU mismatched.
  • NeighborChangedMinorFields: Minor fields changed in the neighbor's LIE.
  • HoldtimeExpired: Adjacency holddown timer expired.
  • MultipleNeighbors: More than one neighbor is present on the interface.
  • MultipleNeighborsDone: Multiple neighbors timer expired.
  • FloodLeadersChanged: Node's election algorithm determined new set of flood leaders.
  • SendLie: Send a LIE out.
  • UpdateZTPOffer: Update this node's ZTP offer. This is sent to the ZTP FSM.
Actions:
  • on HATChanged in OneWay finishes in OneWay: store HAT
  • on FloodLeadersChanged in OneWay finishes in OneWay: update you_are_flood_repeater LIE elements based on the flood leader election results
  • on UnacceptableHeader in OneWay finishes in OneWay: no action
  • on NeighborChangedMinorFields in OneWay finishes in OneWay: no action
  • on SendLie in OneWay finishes in OneWay: SEND_LIE
  • on HALSChanged in OneWay finishes in OneWay: store the HALS
  • on MultipleNeighbors in OneWay finishes in MultipleNeighborsWait: start multiple neighbors timer with the interval multiple_neighbors_lie_holdtime_multiplier * default_lie_holdtime
  • on NeighborChangedLevel in OneWay finishes in OneWay: no action
  • on LieRcvd in OneWay finishes in OneWay: PROCESS_LIE
  • on MTUMismatch in OneWay finishes in OneWay: no action
  • on ValidReflection in OneWay finishes in ThreeWay: no action
  • on LevelChanged in OneWay finishes in OneWay: update the level with the event value, PUSH the SendLie event
  • on HALChanged in OneWay finishes in OneWay: store the new HAL
  • on HoldtimeExpired in OneWay finishes in OneWay: no action
  • on NeighborChangedAddress in OneWay finishes in OneWay: no action
  • on NewNeighbor in OneWay finishes in TwoWay: PUSH the SendLie event
  • on UpdateZTPOffer in OneWay finishes in OneWay: send the offer to the ZTP FSM
  • on NeighborDroppedReflection in OneWay finishes in OneWay: no action
  • on TimerTick in OneWay finishes in OneWay: PUSH SendLie event
  • on FloodLeadersChanged in TwoWay finishes in TwoWay: update you_are_flood_repeater LIE elements based on the flood leader election results
  • on UpdateZTPOffer in TwoWay finishes in TwoWay: send the offer to the ZTP FSM
  • on NewNeighbor in TwoWay finishes in MultipleNeighborsWait: PUSH the SendLie event
  • on ValidReflection in TwoWay finishes in ThreeWay: no action
  • on LieRcvd in TwoWay finishes in TwoWay: PROCESS_LIE
  • on UnacceptableHeader in TwoWay finishes in OneWay: no action
  • on HALChanged in TwoWay finishes in TwoWay: store the new HAL
  • on HoldtimeExpired in TwoWay finishes in OneWay: no action
  • on LevelChanged in TwoWay finishes in TwoWay: update the level with the event value
  • on TimerTick in TwoWay finishes in TwoWay: PUSH SendLie event, if last valid LIE was received more than holdtime ago as advertised by the neighbor, then PUSH the HoldtimeExpired event
  • on HATChanged in TwoWay finishes in TwoWay: store HAT
  • on NeighborChangedLevel in TwoWay finishes in OneWay: no action
  • on HALSChanged in TwoWay finishes in TwoWay: store the HALS
  • on MTUMismatch in TwoWay finishes in OneWay: no action
  • on NeighborChangedAddress in TwoWay finishes in OneWay: no action
  • on SendLie in TwoWay finishes in TwoWay: SEND_LIE
  • on MultipleNeighbors in TwoWay finishes in MultipleNeighborsWait: start multiple neighbors timer with the interval multiple_neighbors_lie_holdtime_multiplier * default_lie_holdtime
  • on TimerTick in ThreeWay finishes in ThreeWay: PUSH the SendLie event, if the last valid LIE was received more than holdtime ago as advertised by the neighbor, then PUSH the HoldtimeExpired event
  • on LevelChanged in ThreeWay finishes in OneWay: update the level with the event value
  • on HATChanged in ThreeWay finishes in ThreeWay: store HAT
  • on MTUMismatch in ThreeWay finishes in OneWay: no action
  • on UnacceptableHeader in ThreeWay finishes in OneWay: no action
  • on MultipleNeighbors in ThreeWay finishes in MultipleNeighborsWait: start multiple neighbors timer with the interval multiple_neighbors_lie_holdtime_multiplier * default_lie_holdtime
  • on NeighborChangedLevel in ThreeWay finishes in OneWay: no action
  • on HALSChanged in ThreeWay finishes in ThreeWay: store the HALS
  • on LieRcvd in ThreeWay finishes in ThreeWay: PROCESS_LIE
  • on FloodLeadersChanged in ThreeWay finishes in ThreeWay: update you_are_flood_repeater LIE elements based on the flood leader election results, PUSH the SendLie event
  • on NeighborDroppedReflection in ThreeWay finishes in TwoWay: no action
  • on HoldtimeExpired in ThreeWay finishes in OneWay: no action
  • on ValidReflection in ThreeWay finishes in ThreeWay: no action
  • on UpdateZTPOffer in ThreeWay finishes in ThreeWay: send the offer to the ZTP FSM
  • on NeighborChangedAddress in ThreeWay finishes in OneWay: no action
  • on HALChanged in ThreeWay finishes in ThreeWay: store the new HAL
  • on SendLie in ThreeWay finishes in ThreeWay: SEND_LIE
  • on MultipleNeighbors in MultipleNeighborsWait finishes in MultipleNeighborsWait: start multiple neighbors timer with the interval multiple_neighbors_lie_holdtime_multiplier * default_lie_holdtime
  • on FloodLeadersChanged in MultipleNeighborsWait finishes in MultipleNeighborsWait: update you_are_flood_repeater LIE elements based on the flood leader election results
  • on TimerTick in MultipleNeighborsWait finishes in MultipleNeighborsWait: check MultipleNeighbors timer, if the timer expired, PUSH MultipleNeighborsDone
  • on ValidReflection in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on UpdateZTPOffer in MultipleNeighborsWait finishes in MultipleNeighborsWait: send the offer to the ZTP FSM
  • on NeighborDroppedReflection in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on LieRcvd in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on UnacceptableHeader in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on NeighborChangedAddress in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on LevelChanged in MultipleNeighborsWait finishes in OneWay: update the level with the event value
  • on HATChanged in MultipleNeighborsWait finishes in MultipleNeighborsWait: store HAT
  • on MTUMismatch in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on HALSChanged in MultipleNeighborsWait finishes in MultipleNeighborsWait: store the HALS
  • on HALChanged in MultipleNeighborsWait finishes in MultipleNeighborsWait: store the new HAL
  • on HoldtimeExpired in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on SendLie in MultipleNeighborsWait finishes in MultipleNeighborsWait: no action
  • on MultipleNeighborsDone in MultipleNeighborsWait finishes in OneWay: no action
  • on Entry into OneWay: CLEANUP
Topology Exchange (TIE Exchange)
Topology Information Elements Topology and reachability information in RIFT is conveyed by TIEs. The TIE exchange mechanism uses the port indicated by each node in the LIE exchange as flood_port in LIEPacket and the interface on which the adjacency has been formed as the destination. TIEs MUST be sent with an IPv4 Time to Live (TTL) or an IPv6 Hop Limit (HL) of either 1 or 255 and also MUST be ignored if received with values different than 1 or 255. This helps to protect RIFT information from being accepted beyond a single L3 next hop in the topology. TIEs SHOULD be sent with network control precedence unless an implementation is prevented from doing so . TIEs contain sequence numbers, lifetimes, and a type. Each type has ample identifying number space, and information is spread across multiple TIEs with the same TIEElement type (this is true for all TIE types). More information about the TIE structure can be found in the schema in , starting with TIEPacket root.
Southbound and Northbound TIE Representation A central concept of RIFT is that each node represents itself differently, depending on the direction in which it is advertising information. More precisely, a spine node represents two different databases over its adjacencies, depending on whether it advertises TIEs to the north or to the south/east-west. Those differing TIE databases are called either southbound or northbound (South TIEs and North TIEs), depending on the direction of distribution. The North TIEs hold all of the node's adjacencies and local prefixes, while the South TIEs hold all of the node's adjacencies, the default prefix with necessary disaggregated prefixes, and local prefixes. explains further details. All TIE types are mostly symmetrical in both directions. defines the TIE types (i.e., the TIETypeType element) and their directionality (i.e., direction within the TIEID element). As an example illustrating a database holding both representations, the topology in with the optional link between spine 111 and spine 112 (so that the flooding on an East-West link can be shown) is shown below. Unnumbered interfaces are implicitly assumed and, for simplicity, the key value elements, which may be included in their South TIEs or North TIEs, are not shown. First, shows the TIEs generated by some nodes.
Example TIEs Generated in a 2-Level Spine-and-Leaf Topology
It may not be obvious here as to why the South Node TIEs contain all the adjacencies of the corresponding node. This will be necessary for algorithms further elaborated on in Sections and . For Node TIEs to carry more adjacencies than fit into an MTU-sized packet, the neighbors element may contain a different set of neighbors in each TIE. Those disjointed sets of neighbors MUST be joined during corresponding computation. However, if the following occurs across multiple Node TIEs:
  1. capabilities do not match or
  2. flags values do not match or
  3. the same neighbor repeats in multiple TIEs with different values.
The implementation is expected to use the value of any of the valid TIEs it received, as it cannot control the arrival order of those TIEs. The miscabled_links element SHOULD be included in every Node TIE; otherwise, the behavior is undefined. A ToF node MUST include information on all other ToFs it is aware of through reflection. The same_plane_tofs element is used to carry this information. To prevent MTU overrun problems, multiple Node TIEs can carry disjointed sets of ToFs, which MUST be joined to form a single set. Different TIE types are carried in TIEElement. Schema enum 'common.TIETypeType' in TIEID indicates which elements MUST be present in TIEElement. In case of a mismatch between TIETypeType in the TIEID and the present element, the unexpected elements MUST be ignored. In case of the lack of an expected element in the TIE, an error MUST be reported and the TIE MUST be ignored. The positive_disaggregation_prefixes and positive_external_disaggregation_prefixes elements MUST be advertised southbound only and ignored in North TIEs. The negative_disaggregation_prefixes element MUST be propagated, according to , southwards towards lower levels to heal pathological upper-level partitioning; otherwise, traffic loss may occur in multi-plane fabrics. It MUST NOT be advertised within a North TIE and MUST be ignored otherwise.
Flooding As described before, TIEs themselves are transported over UDP with the ports indicated in the LIE exchanges and use the destination address on which the LIE adjacency has been formed. TIEs are uniquely identified by the TIEID schema element. TIEID induces a total order achieved by comparing the elements in sequence defined in the element and comparing each value as an unsigned integer of corresponding length. The TIEHeader element contains a seq_nr element to distinguish newer versions of the same TIE. TIEHeader can also carry an origination_time schema element (for fabrics that utilize precision timing) that contains the absolute timestamp of when the TIE was generated and an origination_lifetime to indicate the original lifetime when the TIE was generated. When carried, they can be used for debugging or security purposes (e.g., to prevent lifetime modification attacks). Clock synchronization is considered in more detail in . remaining_lifetime counts down to 0 from origination_lifetime. TIEs with lifetimes differing by less than lifetime_diff2ignore MUST be considered EQUAL (if all other fields are equal). This constant MUST be larger than purge_lifetime to avoid retransmissions. This normative ordering methodology is described in and MUST be used by all implementations.
TIEHeader Comparison Function Y.direction: return X is larger else if X.direction < Y.direction: return Y is larger else if X.originator > Y.originator: return X is larger else if X.originator < Y.originator: return Y is larger else: if X.tietype == Y.tietype: if X.tie_nr == Y.tie_nr: if X.seq_nr == Y.seq_nr: X.lifetime_left = X.remaining_lifetime - time since TIE was received Y.lifetime_left = Y.remaining_lifetime - time since TIE was received if absolute_value_of(X.lifetime_left - Y.lifetime_left) <= common.lifetime_diff2ignore: return Both are Equal else: return TIEHeader with larger lifetime_left is larger else: return TIEHeader with larger seq_nr is larger else: return TIEHeader with larger tie_nr is larger else: return TIEHeader with larger TIEType is larger]]>
All valid TIE types are defined in TIETypeType. This enum indicates what TIE type the TIE is carrying. In case the value is not known to the receiver, the TIE MUST be reflooded with the scope identical to the scope of a prefix TIE. This allows for future extensions of the protocol that are within the same major schema and that have types that are opaque to some nodes; some restrictions are defined in .
Normative Flooding Procedures On reception of a TIE with an undefined level value in the packet header, the node MUST issue a warning and discard the packet. This section specifies the precise, normative flooding mechanism and can be omitted unless the reader is pursuing an implementation of the protocol or looks for a deep understanding of underlying information distribution mechanism. Flooding procedures are described in terms of the flooding state of an adjacency, and resulting operations on it are driven by packet arrivals. Implementations MUST implement a behavior that is externally indistinguishable from the FSMs and normative procedures given here. RIFT does not specify any kind of flood rate limiting. To help with adjustment of flooding speeds, the encoded packets provide hints to react accordingly to losses or overruns via you_are_sending_too_quickly in the LIEPacket and "Packet Number" in the security envelope described in . Flooding of all corresponding topology exchange elements SHOULD be performed at the highest feasible rate, but the rate of transmission MUST be throttled by reacting to packet elements and features of the system, such as queue lengths or congestion indications in the protocol packets. A node SHOULD NOT send out any topology information elements if the adjacency is not in a ThreeWay state. No further tightening of this rule is possible. For example, link buffering may cause both LIEs and TIEs/TIDEs/TIREs to be reordered. A node MUST drop any received TIEs/TIDEs/TIREs unless it is in the ThreeWay state. TIEs generated by other nodes MUST be reflooded. TIDEs and TIREs MUST NOT be reflooded.
FloodState Structure per Adjacency For each adjacency, the structure conceptually contains the following elements. The word "collection" or "queue" indicates a set of elements that can be iterated over the following:
TIES_TX:
Collection containing all the TIEs to transmit on the adjacency.
TIES_ACK:
Collection containing all the TIEs that have to be acknowledged on the adjacency.
TIES_REQ:
Collection containing all the TIE headers that have to be requested on the adjacency.
TIES_RTX:
Collection containing all TIEs that need retransmission with the corresponding time to retransmit.
FILTERED_TIEDB:
A filtered view of TIEDB, which retains for consideration only those headers permitted by is_tide_entry_filtered and which either have a lifetime left > 0 or have no content.
The following words are used for well-known elements and procedures operating on this structure:
TIE:
describes either a full RIFT TIE or just the TIEHeader or TIEID equivalent, as defined in . The corresponding meaning is unambiguously contained in the context of each algorithm.
is_flood_reduced(TIE):
returns whether a TIE can be flood-reduced or not.
is_tide_entry_filtered(TIE):
returns whether a header should be propagated in TIDE according to flooding scopes.
is_request_filtered(TIE):
returns whether a TIE request should be propagated to the neighbor or not, according to flooding scopes.
is_flood_filtered(TIE):
returns whether a TIE requested be flooded to the neighbor or not, according to flooding scopes.
try_to_transmit_tie(TIE):
if not is_flood_filtered(TIE), then
  1. remove the TIE from TIES_RTX if present
  2. if the TIE with same key is found on TIES_ACK, then
    1. if the TIE is the same as or newer than TIE, do nothing, else
    2. remove the TIE from TIES_ACK and add TIE to TIES_TX
  3. else insert the TIE into TIES_TX.
ack_tie(TIE):
remove the TIE from all collections and then insert the TIE into TIES_ACK.
tie_been_acked(TIE):
remove the TIE from all collections.
remove_from_all_queues(TIE):
same as tie_been_acked.
request_tie(TIE):
if not is_request_filtered(TIE), then remove_from_all_queues(TIE) and add to TIES_REQ.
move_to_rtx_list(TIE):
remove the TIE from TIES_TX and then add to TIES_RTX, using the TIE retransmission interval.
clear_requests(TIEs):
remove all TIEs from TIES_REQ.
bump_own_tie(TIE):
for a self-originated TIE, originate an empty or regenerate with the version number higher than the one in the TIE.
The collection SHOULD be served with the following priorities if the system cannot process all the collections in real time:
  1. Elements on TIES_ACK should be processed with highest priority
  2. TIES_TX
  3. TIES_REQ and TIES_RTX should be processed with lowest priority
TIDEs TIEID and TIEHeader spaces form a strict total order (modulo incomparable sequence numbers (found in "TIEHeader.seq_nr"), as explained in , in the very unlikely event that a TIE is "stuck" in a part of a network while the originator reboots and reissues TIEs many times to the point its sequence number rolls over and forms an incomparable distance to the "stuck" copy), which implies that a comparison relation is possible between two elements. With that, it is implicitly possible to compare TIEs, TIEHeaders, and TIEIDs to each other, whereas the shortest viable key is always implied.
TIDE Generation
NEXT_TIDE_ID:
ID of the next TIE to be sent in the TIDE.
As given by the timer constant, periodically generate TIDEs by:
  1. NEXT_TIDE_ID = MIN_TIEID
  2. while NEXT_TIDE_ID is not equal to MAX_TIEID do:
    1. HEADERS = Exactly TIRES_PER_TIDE_PKT headers from FILTERED_TIEDB starting at NEXT_TIDE_ID, unless fewer than TIRES_PER_TIDE_PKT remain, in which case all remaining headers.
    2. if HEADERS is empty, then START = MIN_TIEID, else START = first element in HEADERS
    3. if HEADERS size is less than TIRES_PER_TIDE_PKT, then END = MAX_TIEID, else END = last element in HEADERS
    4. send sorted HEADERS as TIDE, setting START and END as its range
    5. NEXT_TIDE_ID = END
The constant TIRES_PER_TIDE_PKT SHOULD be computed per interface and used by the implementation to limit the amount of TIE headers per TIDE so the sent TIDE PDU does not exceed the MTU of the interface. TIDE PDUs SHOULD be transmitted at a rate that does not lead to packet drops. The algorithm will intentionally enter the loop once and send a single TIDE, even when the database is empty; otherwise, no TIDEs would be sent for in case of an empty database and break the intended synchronization.
TIDE Processing
TXKEYS:
Collection of TIE headers to be sent after processing of the packet
REQKEYS:
Collection of TIEIDs to be requested after processing of the packet
CLEARKEYS:
Collection of TIEIDs to be removed from flood state queues
LASTPROCESSED:
Last processed TIEID in the TIDE
DBTIE:
TIE in the LSDB, if found
On reception of TIDEs, the following processing is performed:
  1. LASTPROCESSED = TIDE.start_range
  2. For every HEADER in the TIDE do:
    1. DBTIE = find HEADER in the current LSDB
    2. if HEADER < LASTPROCESSED, then report an error and reset the adjacency and return
    3. put all TIEs in LSDB, where (TIE.HEADER > LASTPROCESSED and TIE.HEADER < HEADER) into TXKEYS
    4. LASTPROCESSED = HEADER
    5. if DBTIE is not found, then
      1. if originator is this node, then bump_own_tie
      2. else put HEADER into REQKEYS
    6. if DBTIE.HEADER < HEADER then
      1. if the originator is this node, then bump_own_tie, else
        1. if this is a North TIE header from a northbound neighbor, then override DBTIE in LSDB with HEADER
        2. else put HEADER into REQKEYS
    7. if DBTIE.HEADER > HEADER, then put DBTIE.HEADER into TXKEYS
    8. if DBTIE.HEADER = HEADER, then
      1. if DBTIE has content already, then put DBTIE.HEADER into CLEARKEYS, else
      2. put HEADER into REQKEYS
  3. put all TIEs in LSDB, where (TIE.HEADER > LASTPROCESSED and TIE.HEADER <= TIDE.end_range) into TXKEYS
  4. for all TIEs in TXKEYS, try_to_transmit_tie(TIE)
  5. for all TIEs in REQKEYS, request_tie(TIE)
  6. for all TIEs in CLEARKEYS, remove_from_all_queues(TIE)
TIREs
TIRE Generation Elements from both TIES_REQ and TIES_ACK MUST be collected and sent out as fast as feasible as TIREs. When sending TIREs with elements from TIES_REQ, the remaining_lifetime field in TIEHeaderWithLifeTime MUST be set to 0 to force reflooding from the neighbor even if the TIEs seem to be the same.
TIRE Processing
TXKEYS:
Collection of TIE headers to be sent after processing of the packet
REQKEYS:
Collection of TIEIDs to be requested after processing of the packet
ACKKEYS:
Collection of TIEIDs that have been acknowledged
DBTIE:
TIE in the LSDB, if found
On reception of TIREs, the following processing is performed:
  1. for every HEADER in TIRE do:
    1. DBTIE = find HEADER in the current LSDB
    2. if DBTIE is not found, then do nothing
    3. if DBTIE.HEADER < HEADER, then put HEADER into REQKEYS
    4. if DBTIE.HEADER > HEADER, then put DBTIE.HEADER into TXKEYS
    5. if DBTIE.HEADER = HEADER, then put DBTIE.HEADER into ACKKEYS
  2. for all TIEs in TXKEYS, try_to_transmit_tie(TIE)
  3. for all TIEs in REQKEYS, request_tie(TIE)
  4. for all TIEs in ACKKEYS, tie_been_acked(TIE)
TIEs Processing on Flood State Adjacency On reception of TIEs, the following processing is performed:
ACKTIE:
TIE to acknowledge
TXTIE:
TIE to transmit
DBTIE:
TIE in the LSDB, if found
  1. DBTIE = find TIE in the current LSDB
  2. if DBTIE is not found, then
    1. if the originator is this node, then bump_own_tie with a short remaining lifetime
    2. else insert TIE into LSDB and ACKTIE = TIE
    else
    1. if DBTIE.HEADER = TIE.HEADER, then
      1. if DBTIE has content already, then ACKTIE = TIE
      2. else process like the "DBTIE.HEADER < TIE.HEADER" case
    2. if DBTIE.HEADER < TIE.HEADER, then
      1. if the originator is this node, then bump_own_tie
      2. else insert TIE into LSDB and ACKTIE = TIE
    3. if DBTIE.HEADER > TIE.HEADER, then
      1. if DBTIE has content already, then TXTIE = DBTIE
      2. else ACKTIE = DBTIE
  3. if TXTIE is set, then try_to_transmit_tie(TXTIE)
  4. if ACKTIE is set, then ack_tie(TIE)
Sending TIEs On a periodic basis, all TIEs with a lifetime of > 0 left MUST be sent out on the adjacency, removed from the TIES_TX list, and requeued onto TIES_RTX list. The specific period is out of scope for this document.
TIEs Processing in LSDB The LSDB holds the most recent copy of TIEs received via flooding from according peers. Consecutively, after version tie-breaking by LSDB, a peer receives from the LSDB the newest versions of TIEs received by other peers and processes them (without any filtering) just like receiving TIEs from its remote peer. Such a publisher model can be implemented in several ways, either in a single thread of execution or in multiple parallel threads. LSDB can be logically considered as the entity aging out TIEs, i.e., being responsible to discard TIEs that are stored longer than remaining_lifetime on their reception. LSDB is also expected to periodically reoriginate the node's own TIEs. Originating at an interval significantly shorter than default_lifetime is RECOMMENDED to prevent TIE expiration by other nodes in the network, which can lead to instabilities.
TIE Flooding Scopes In a somewhat analogous fashion to link-local, area, and domain flooding scopes, RIFT defines several complex "flooding scopes", depending on the direction and type of TIE propagated. Every North TIE is flooded northbound, providing a node at a given level with the complete topology of the Clos or fat tree network that is reachable southwards of it, including all specific prefixes. This means that a packet received from a node at the same or lower level whose destination is covered by one of those specific prefixes will be routed directly towards the node advertising that prefix, rather than sending the packet to a node at a higher level. A node's South Node TIEs, consisting of all node's adjacencies and South Prefix TIEs limited to those related to default IP prefix and disaggregated prefixes, are flooded southbound in order to inform nodes one level down of connectivity of the higher level as well as reachability to the rest of the fabric. In order to allow an E-W disconnected node in a given level to receive the South TIEs of other nodes at its level, every South Node TIE is "reflected" northbound to the level from which it was received. It should be noted that East-West links are included in South TIE flooding (except at the ToF level); those TIEs need to be flooded to satisfy the algorithms described in . In that way, nodes at same level can learn about each other without using a lower level except in case of leaf level. The precise, normative flooding scopes are given in . Those rules also govern what SHOULD be included in TIDEs on the adjacency. Again, East-West flooding scopes are identical to southern flooding scopes, except in case of ToF East-West links (rings), which are basically performing northbound flooding. South Node TIE "south reflection" enables support of positive disaggregation on failures, as described in , and flooding reduction, as described in . Normative Flooding Scopes
Type / Direction South North East-West
South Node TIE flood if the level of the originator is equal to this node flood if the level of the originator is higher than this node flood only if this node is not ToF
non-Node South TIE flood self-originated only flood only if the neighbor is the originator of TIE flood only if it is self-originated and this node is not ToF
all North TIEs never flood flood always flood only if this node is ToF
TIDE include at least all non-self-originated North TIE headers and self-originated South TIE headers and South Node TIEs of nodes at same level include at least all South Node TIEs and all South TIEs originated by a peer and all North TIEs if this node is ToF, then include all North TIEs; otherwise, only include self-originated TIEs
TIRE as Request request all North TIEs and all peer's self-originated TIEs and all South Node TIEs request all South TIEs if this node is ToF, then apply north scope rules; otherwise, apply south scope rules
TIRE as Ack Ack all received TIEs Ack all received TIEs Ack all received TIEs
If the TIDE includes additional TIE headers beside the ones specified, the receiving neighbor must apply the corresponding filter to the received TIDE strictly and MUST NOT request the extra TIE headers that were not allowed by the flooding scope rules in its direction. To illustrate these rules, consider using the topology in , with the optional link between spine 111 and spine 112, and the associated TIEs given in . The flooding from particular nodes of the TIEs is given in . Flooding Some TIEs from Example Topology
Local Node Neighbor Node TIEs Flooded from Local to Neighbor Node
Leaf111 Spine 112 Leaf111 North TIEs, Spine 111 South Node TIE
Leaf111 Spine 111 Leaf111 North TIEs, Spine 112 South Node TIE
... ... ...
Spine 111 Leaf111 Spine 111 South TIEs
Spine 111 Leaf112 Spine 111 South TIEs
Spine 111 Spine 112 Spine 111 South TIEs
Spine 111 ToF 21 Spine 111 North TIEs, Leaf111 North TIEs, Leaf112 North TIEs, ToF 22 South Node TIE
Spine 111 ToF 22 Spine 111 North TIEs, Leaf111 North TIEs, Leaf112 North TIEs, ToF 21 South Node TIE
... ... ...
ToF 21 Spine 111 ToF 21 South TIEs
ToF 21 Spine 112 ToF 21 South TIEs
ToF 21 Spine 121 ToF 21 South TIEs
ToF 21 Spine 122 ToF 21 South TIEs
... ... ...
RAIN: RIFT Adjacency Inrush Notification The optional RIFT Adjacency Inrush Notification (RAIN) mechanism helps to prevent adjacencies from being overwhelmed by flooding on restart or bring-up with many southbound neighbors. In its LIEs, a node MAY set the corresponding you_are_sending_too_quickly flag to indicate to the neighbor that it SHOULD flood Node TIEs with normal speed and significantly slow down the flooding of any other TIEs. The flag SHOULD be set only in the southbound direction. The receiving node SHOULD accommodate the request to lessen the flooding load on the affected node if it is south of the sender and should ignore the indication if it is north of the sender. The distribution of Node TIEs at normal speed, even at high load, guarantees correct behavior of algorithms like disaggregation or default route origination. Furthermore though, the use of this bit presents an inherent trade-off between processing load and convergence speed since significantly slowing down flooding of northbound prefixes from neighbors for an extended time will lead to traffic losses.
Initial and Periodic Database Synchronization The initial exchange of RIFT includes periodic TIDE exchanges that contain descriptions of the LSDB and TIREs, which perform the function of requesting unknown TIEs as well as confirming the reception of flooded TIEs. The content of TIDEs and TIREs is governed by .
Purging and Rollovers When a node exits the network, if "unpurged", residual stale TIEs may exist in the network until their lifetimes expire (which in case of RIFT is by default a rather long period to prevent ongoing reorigination of TIEs in very large topologies). RIFT does not have a "purging mechanism" based on sending specialized "purge" packets. In other routing protocols, such a mechanism has proven to be complex and fragile based on many years of experience. RIFT simply issues a new, i.e., higher sequence number, empty version of the TIE with a short lifetime given by the purge_lifetime constant and relies on each node to age out and delete each TIE copy independently. Abundant amounts of memory are available today, even on low-end platforms, and hence, keeping those relatively short-lived extra copies for a while is acceptable. The information will age out and, in the meantime, all computations will deliver correct results if a node leaves the network due to the new information distributed by its adjacent nodes breaking bidirectional connectivity checks in different computations. Once a RIFT node issues a TIE with an ID, it SHOULD preserve the ID as long as feasible (also when the protocol restarts), even if the TIE looses all content. The re-advertisement of an empty TIE fulfills the purpose of purging any information advertised in previous versions. The originator is free to not reoriginate the corresponding empty TIE again or originate an empty TIE with a relatively short lifetime to prevent a large number of long-lived empty stubs polluting the network. Each node MUST time out and clean up the corresponding empty TIEs independently. Upon restart, a node MUST be prepared to receive TIEs with its own System ID and supersede them with equivalent, newly generated, empty TIEs with a higher sequence number. As above, the lifetime can be relatively short since it only needs to exceed the necessary propagation and processing delay by all the nodes that are within the TIE's flooding scope. TIE sequence numbers are rolled over using the method described in . The first sequence number of any spontaneously originated TIE (i.e., not originated to override a detected older copy in the network) MUST be a reasonably unpredictable random number (for example, ) in the interval [0, 230-1], which will prevent otherwise identical TIE headers to remain "stuck" in the network with content different from the TIE originated after reboot. In typical link-state protocols, this is delegated to a 16-bit checksum on packet content. RIFT avoids this design due to the CPU burden presented by computation of such checksums and additional complications tied to the fact that the checksum must be "patched" into the packet after the generation of the content, which is a difficult proposition in binary, hand-crafted formats already and highly incompatible with model-based, serialized formats. The sequence number space is hence consciously chosen to be 64-bits wide to make the occurrence of a TIE with the same sequence number but different content as much or even more unlikely than the checksum method. To emulate the "checksum behavior", an implementation could choose to compute a 64-bit checksum or hash function over the TIE content and use that as part of the first sequence number after reboot.
Southbound Default Route Origination Under certain conditions, nodes issue a default route in their South Prefix TIEs with costs as computed in . A node X that
  1. is not overloaded and
  2. has southbound or East-West adjacencies
SHOULD originate such a default route in its South Prefix TIE if and only if
  1. all other nodes at X's level are overloaded or
  2. all other nodes at X's level have NO northbound adjacencies, or
  3. X has computed reachability to a default route during N-SPF.
The term "all other nodes at X's level" obviously describes just the nodes at the same level in the PoD with a viable lower level (otherwise, the South Node TIEs cannot be reflected; the nodes in PoD 1 and PoD 2 are "invisible" to each other). A node originating a southbound default route SHOULD install a default discard route if it did not compute a default route during N-SPF. This basically means that the top of the fabric will drop traffic for unreachable addresses.
Northbound TIE Flooding Reduction RIFT chooses only a subset of northbound nodes to propagate flooding and, with that, both balances it (to prevent "hot" flooding links) across the fabric as well as reduces its volume. The solution is based on several principles:
  1. a node MUST flood self-originated North TIEs to all the reachable nodes at the level above, which is called the node's "parents";
  2. it is typically not necessary that all parents reflood the North TIEs to achieve a complete flooding of all the reachable nodes two levels above, which we call the node's "grandparents";
  3. to control the volume of its flooding two hops north and yet keep it robust enough, it is advantageous for a node to select a subset of its parents as "Flood Repeaters" (FRs), which when combined, deliver two or more copies of its flooding to all of its parents, i.e., the originating node's grandparents;
  4. nodes at the same level do not have to agree on a specific algorithm to select the FRs, but overall load balancing should be achieved so that different nodes at the same level should tend to select different parents as FRs (consideration of possible strategies in an unrelated but similar field can be found in );
  5. there are usually many solutions to the problem of finding a set of FRs for a given node; the problem of finding the minimal set is (similar to) an NP-Complete problem, and a globally optimal set may not be the minimal one if load balancing with other nodes is an important consideration;
  6. it is expected that sets of equivalent nodes at a level L will often exist, defined as having a common set of parents at L+1. Applying this observation at both L and L+1, an algorithm may attempt to split the larger problem in a sum of smaller, separate problems; and
  7. it is expected that there will be a broken link between a parent and a grandparent from time to time, and in that case, the parent is probably a poor FR due to its lower reliability. An algorithm may attempt to eliminate parents with broken northbound adjacencies first in order to reduce the number of FRs. Albeit it could be argued that relying on higher fanout FRs will slow flooding due to higher replication, load reliability of FR's links is likely a more pressing concern.
In a fully connected Clos network, this means that a node selects one arbitrary parent as the FR and then a second one for redundancy. The computation can be relatively simple and completely distributed without any need for synchronization among nodes. In a "PoD" structure, where the level L+2 is partitioned into silos of equivalent grandparents that are only reachable from respective parents, this means treating each silo as a fully connected Clos network and solving the problem within the silo. In terms of signaling, a node has enough information to select its set of FRs; this information is derived from the node's parents' South Node TIEs, which indicate the parent's reachable northbound adjacencies to its own parents (the node's grandparents). A node may send a LIE to a northbound neighbor with the optional boolean field you_are_flood_repeater set to false to indicate that the northbound neighbor is not a flood repeater for the node that sent the LIE. In that case, the northbound neighbor SHOULD NOT reflood northbound TIEs received from the node that sent the LIE. If you_are_flood_repeater is absent or you_are_flood_repeater is set to true, then the northbound neighbor is a flood repeater for the node that sent the LIE and MUST reflood northbound TIEs received from that node. The element you_are_flood_repeater MUST be ignored if received from a northbound adjacency. This specification provides a simple default algorithm that SHOULD be implemented and used by default on every RIFT node.
  • let |NA(Node) be the set of northbound adjacencies of node Node and CN(Node) be the cardinality of |NA(Node);
  • let |SA(Node) be the set of southbound adjacencies of node Node and CS(Node) be the cardinality of |SA(Node);
  • let |P(Node) be the set of node Node's parents;
  • let |G(Node) be the set of node Node's grandparents. Observe that |G(Node) = |P(|P(Node));
  • let N be the child node at level L computing a set of FRs;
  • let P be a node at level L+1 and a parent node of N, i.e., bidirectionally reachable over adjacency ADJ(N, P);
  • let G be a grandparent node of N, reachable transitively via a parent P over adjacencies ADJ(N, P) and ADJ(P, G). Observe that N does not have enough information to check bidirectional reachability of ADJ(P, G);
  • let R be a redundancy constant integer; a value of 2 or higher for R is RECOMMENDED;
  • let S be a similarity constant integer; a value in range 0 .. 2 for S is RECOMMENDED, and the value of 1 SHOULD be used. Two cardinalities are considered as equivalent if their absolute difference is less than or equal to S, i.e., |a-b|<=S;
  • let RND be a 64-bit random number (for example, as described in ) generated by the system once on startup.
The algorithm consists of the following steps:
  1. Derive a 64-bit number by XORing N's System ID with RND.
  2. Derive a 16-bit pseudo-random unsigned integer PR(N) from the resulting 64-bit number by splitting it into 16-bit-long words W1, W2, W3, W4 (where W1 are the least significant 16 bits of the 64-bit number, and W4 are the most significant 16 bits) and then XORing the circularly shifted resulting words together:
    1. (W1<<1) xor (W2<<2) xor (W3<<3) xor (W4<<4); where << is the circular shift operator.
  3. Sort the parents by decreasing number of northbound adjacencies (using decreasing System ID of the parent as a tie-breaker): sort |P(N) by decreasing CN(P), for all P in |P(N), as the ordered array |A(N)
  4. Partition |A(N) in subarrays |A_k(N) of parents with equivalent cardinality of northbound adjacencies (in other words, with equivalent number of grandparents they can reach):
    1. set k=0; // k is the ID of the subarray
    2. set i=0;
    3. while i < CN(N) do
      1. set j=i;
      2. while i < CN(N) and CN(|A(N)[j]) - CN(|A(N)[i]) <= S:
        1. place |A(N)[i] in |A_k(N) // abstract action, maybe noop
        2. set i=i+1;
      3. /* At this point, j is the index in |A(N) of the first member of |A_k(N) and (i-j) is C_k(N) defined as the cardinality of |A_k(N). */ set k=k+1;
    /* At this point, k is the total number of subarrays, initialized for the shuffling operation below. */
  5. Shuffle each subarrays |A_k(N) of cardinality C_k(N) within |A(N) individually using the Durstenfeld variation of the Fisher-Yates algorithm that depends on N's System ID:
    1. while k > 0 do
      1. for i from C_k(N)-1 to 1 decrementing by 1 do
        1. set j to PR(N) modulo i;
        2. exchange |A_k[j] and |A_k[i];
      2. set k=k-1;
  6. For each grandparent G, initialize a counter c(G) with the number of its southbound adjacencies to elected flood repeaters (which is initially zero):
    1. for each G in |G(N), set c(G) = 0;
  7. Finally, only keep FRs as parents that are needed to maintain the number of adjacencies between the FRs and any grandparent G equal or above the redundancy constant R:
    1. for each P in reshuffled |A(N):
      1. if there exists an adjacency ADJ(P, G) in |NA(P) such that c(G) < R, then
        1. place P in FR set;
        2. for all adjacencies ADJ(P, G') in |NA(P) increment c(G')
  8. If any c(G) is still < R, it was not possible to elect a set of FRs that covers all grandparents with redundancy R.
Additional rules for flooding reduction:
  1. The algorithm MUST be re-evaluated by a node on every change of local adjacencies or reception of a parent South TIE with changed adjacencies. A node MAY apply a hysteresis to prevent an excessive amount of computation during periods of network instability just like in the case of reachability computation.
  2. Upon a change of the flood repeater set, a node SHOULD send out LIEs that grant flood repeater status to newly promoted nodes before it sends LIEs that revoke the status to the nodes that have been newly demoted. This is done to prevent transient behavior where the full coverage of grandparents is not guaranteed. Such a condition is sometimes unavoidable in case of lost LIEs, but it will correct itself at possible transient reduction in flooding propagation speeds. The election can use the LIE FSM FloodLeadersChanged event to notify LIE FSMs of the necessity to update the sent LIEs.
  3. A node MUST always flood its self-originated TIEs to all its neighbors.
  4. A node receiving a TIE originated by a node for which it is not a flood repeater SHOULD NOT reflood such TIEs to its neighbors, except for the rules described in .
  5. The indication of flood reduction capability MUST be carried in the Node TIEs in the flood_reduction element and MAY be used to optimize the algorithm to account for nodes that will flood regardless.
  6. A node generates TIDEs as usual, but when receiving TIREs or TIDEs resulting in requests for a TIE of which the newest received copy came on an adjacency where the node was not a flood repeater, it SHOULD ignore such requests on only the first request. Normally, the nodes that received the TIEs as flooding repeaters should satisfy the requesting node and, with that, no further TIREs for such TIEs will be generated. Otherwise, the next set of TIDEs and TIREs MUST lead to flooding independent of the flood repeater status. This solves a very difficult "incast" problem on nodes restarting with a very wide fanout, especially northbound. To retrieve the full database, they often end up processing many inrushing copies, whereas this approach load balances the incoming database between adjacent nodes and flood repeaters and should guarantee that two copies are sent by different nodes to ensure against any losses.
Special Considerations First, due to the distributed, asynchronous nature of ZTP, it can create temporary convergence anomalies where nodes at higher levels of the fabric temporarily become lower than where they ultimately belong. Since flooding can begin before ZTP is "finished" and in fact must do so given there is no global termination criteria for the unsynchronized ZTP algorithm, information may temporarily end up in wrong layers. A special clause when changing level takes care of that. More difficult is a condition where a node (e.g., a leaf) floods a TIE north towards its grandparent, then its parent reboots, partitioning the grandparent from the leaf directly, and then the leaf itself reboots. That can leave the grandparent holding the "primary copy" of the leaf's TIE. Normally, this condition is resolved easily by the leaf reoriginating its TIE with a higher sequence number than it notices in the northbound TIEs; here however, when the parent comes back, it won't be able to obtain the leaf's North TIE from the grandparent easily, and with that, the leaf may not issue the TIE with a higher sequence number that can reach the grandparent for a long time. Flooding procedures are extended to deal with the problem by the means of special clauses that override the database of a lower level with headers of newer TIEs received in TIDEs coming from the north. Those headers are then propagated southbound towards the leaf to cause it to originate a higher sequence number of the TIE, effectively refreshing it all the way up to ToF.
Reachability Computation A node has three possible sources of relevant information for reachability computation. A node knows the full topology south of it from the received North Node TIEs or alternately north of it from the South Node TIEs. A node has the set of prefixes with their associated distances and bandwidths from corresponding prefix TIEs. To compute prefix reachability, a node conceptually runs a northbound and a southbound SPF. Here, N-SPF and S-SPF notation denotes the direction in which the computation front is progressing. Since neither computation can "loop", it is possible to compute non-equal costs or even k-shortest paths and "saturate" the fabric to the extent desired. This specification however uses simple, familiar SPF algorithms and concepts as examples due to their prevalence in today's routing. For reachability computation purposes, RIFT considers all parallel links between two nodes to be of the same cost advertised in the cost element of NodeNeighborsTIEElement. In case the neighbor has multiple parallel links at different costs, the largest distance (highest numerical value) MUST be advertised. Given the range of Thrift encodings, infinite_distance is defined as the largest non-negative MetricType. Any link with a metric larger than that (i.e., the negative MetricType) MUST be ignored in computations. Any link with the metric set to invalid_distance MUST also be ignored in computation. In case of a negatively distributed prefix, the metric attribute MUST be set to infinite_distance by the originator, and it MUST be ignored by all nodes during computation, except for the purpose of determining transitive propagation and building the corresponding routing table. A prefix can carry the directly_attached attribute to indicate that the prefix is directly attached, i.e., should be routed to even if the node is in overload. In case of a negatively distributed prefix, this attribute MUST NOT be included by the originator, and it MUST be ignored by all nodes during SPF computation. If a prefix is locally originated, the attribute from_link can indicate the interface to which the address belongs to. In case of a negatively distributed prefix, this attribute MUST NOT be included by the originator, and it MUST be ignored by all nodes during computation. A prefix can also carry the loopback attribute to indicate the said property. Prefixes are carried in different types of TIEs indicating their type. For the same prefix being included in different TIE types, tie-breaking is performed according to . If the same prefix is included multiple times in multiple TIEs of the same type originating at the same node, the resulting behavior is unspecified.
Northbound Reachability SPF N-SPF MUST use exclusively northbound and East-West adjacencies in the computing node's North Node TIEs (since if the node is a leaf, it may not have generated a South Node TIE) when starting SPF. Observe that N-SPF is really just a one-hop variety since South Node TIEs are not reflooded southbound beyond a single level (or East-West), and with that, the computation cannot progress beyond adjacent nodes. Once progressing, the computation uses the next higher level's South Node TIEs to find corresponding adjacencies to verify backlink connectivity. Two unidirectional links MUST be associated to confirm bidirectional connectivity, a process often known as "backlink check". As part of the check, both Node TIEs MUST contain the correct System IDs and expected levels. The default route found when crossing an E-W link SHOULD be used if and only if
  1. the node itself does not have any northbound adjacencies and
  2. the adjacent node has one or more northbound adjacencies.
This rule forms a "one-hop default route split-horizon" and prevents looping over default routes while allowing for "one-hop protection" of nodes that lost all northbound adjacencies, except at the ToF where the links are used exclusively to flood topology information in multi-plane designs. Other south prefixes found when crossing E-W links MAY be used if and only if
  1. no north neighbors are advertising the same or a supersuming non-default prefix and
  2. the node does not originate a non-default supersuming prefix itself.
That is, the E-W link can be used as a gateway of last resort for a specific prefix only. Using south prefixes across an E-W link can be beneficial, e.g., on automatic disaggregation in pathological fabric partitioning scenarios. A detailed example can be found in .
Southbound Reachability SPF S-SPF MUST use the southbound adjacencies in the South Node TIEs exclusively, i.e., progresses towards nodes at lower levels. Observe that E-W adjacencies are NEVER used in this computation. This enforces the requirement that a packet traversing in a southbound direction must never change its direction. S-SPF MUST use northbound adjacencies in North Node TIEs to verify backlink connectivity by checking for the presence of the link beside the correct System ID and level.
East-West Forwarding Within a Non-ToF Level Using south prefixes over horizontal links MAY occur if the N-SPF includes East-West adjacencies in computation. It can protect against pathological fabric partitioning cases that leave only paths to destinations that would necessitate multiple changes of forwarding direction between north and south.
East-West Links Within a ToF Level E-W ToF links behave in terms of flooding scopes defined in like northbound links and MUST be used exclusively for control plane information flooding. Even though a ToF node could be tempted to use those links during southbound SPF and carry traffic over them, this MUST NOT be attempted since it may, in anycast cases, lead to routing loops. An implementation MAY try to resolve the looping problem by following on the ring strictly tie-broken shortest-paths only, but the details are outside this specification. And even then, the problem of proper capacity provisioning of such links when they become traffic-bearing in case of failures is vexing, and when used for forwarding purposes, they defeat statistical non-blocking guarantees that Clos is providing normally.
Automatic Disaggregation on Link & Node Failures
Positive, Non-Transitive Disaggregation Under normal circumstances, a node's South TIEs contain just the adjacencies and a default route. However, if a node detects that its default IP prefix covers one or more prefixes that are reachable through it but not through one or more other nodes at the same level, then it MUST explicitly advertise those prefixes in a South TIE. Otherwise, some percentage of the northbound traffic for those prefixes would be sent to nodes without corresponding reachability, causing it to be dropped. Even when traffic is not being dropped, the resulting forwarding could "backhaul" packets through the higher-level spines, clearly an undesirable condition affecting the blocking probabilities of the fabric. This specification refers to the process of advertising additional prefixes southbound as "positive disaggregation". Such disaggregation is non-transitive, i.e., its effects are always constrained to a single level of the fabric. Naturally, multiple node or link failures can lead to several independent instances of positive disaggregation necessary to prevent looping or bow-tying the fabric. A node determines the set of prefixes needing disaggregation using the following steps:
  1. A DAG computation in the southern direction is performed first. The North TIEs are used to find all of the prefixes it can reach and the set of next hops in the lower level for each of them. Such a computation can be easily performed on a fat tree by setting all link costs in the southern direction to 1 and all northern directions to infinity. The set of those prefixes is referred to as |R; for each prefix r in |R, its set of next hops is referred to as |H(r).
  2. The node uses reflected South TIEs to find all nodes at the same level in the same PoD and the set of southbound adjacencies for each. The set of nodes at the same level is termed |N, and for each node, n, in |N, its set of southbound adjacencies is defined to be |A(n).
  3. For a given r, if the intersection of |H(r) and |A(n), for any n, is empty, then that prefix r must be explicitly advertised by the node in a South TIE.
  4. An identical set of disaggregated prefixes is flooded on each of the node's southbound adjacencies. In accordance with the normal flooding rules for a South TIE, a node at the lower level that receives this South TIE SHOULD NOT propagate it southbound or reflect the disaggregated prefixes back over its adjacencies to nodes at the level from which it was received.
To summarize the above in simplest terms: If a node detects that its default route encompasses prefixes for which one of the other nodes in its level has no possible next hops in the level below, it has to disaggregate it to prevent traffic loss or suboptimal routing through such nodes. Hence, a node X needs to determine if it can reach a different set of south neighbors than other nodes at the same level, which are connected to it via at least one common south neighbor. If it can, then prefix disaggregation may be required. If it can't, then no prefix disaggregation is needed. An example of disaggregation is provided in . Finally, a possible algorithm is described here:
  1. Create partial_neighbors = (empty), a set of neighbors with partial connectivity to the node X's level from X's perspective. Each entry in the set is a south neighbor of X and a list of nodes of X.level that can't reach that neighbor.
  2. A node X determines its set of southbound neighbors X.south_neighbors.
  3. For each South TIE originated from a node Y that X has, which is at X.level, if Y.south_neighbors is not the same as X.south_neighbors but the nodes share at least one southern neighbor, for each neighbor N in X.south_neighbors but not in Y.south_neighbors, add (N, (Y)) to partial_neighbors if N isn't there or add Y to the list for N.
  4. If partial_neighbors is empty, then node X does not disaggregate any prefixes. If node X is advertising disaggregated prefixes in its South TIE, X SHOULD remove them and re-advertise its South TIEs.
A node X computes reachability to all nodes below it based upon the received North TIEs first. This results in a set of routes, each categorized by (prefix, path_distance, next-hop set). Alternately, for clarity in the following procedure, these can be organized by a next-hop set as ((next-hops), {(prefix, path_distance)}). If partial_neighbors isn't empty, then the procedure in describes how to identify prefixes to disaggregate.
Computation of Disaggregated Prefixes
Each disaggregated prefix is sent with the corresponding path_distance. This allows a node to send the same South TIE to each south neighbor. The south neighbor that is connected to that prefix will thus have a shorter path. Finally, to summarize the less obvious points partially omitted in the algorithms to keep them more tractable:
  1. All neighbor relationships MUST perform backlink checks.
  2. The overload flag as introduced in and carried in the overload schema element has to be respected during the computation. Nodes advertising themselves as overloaded MUST NOT be transited in reachability computation but MUST be used as terminal nodes with prefixes they advertise being reachable.
  3. All the lower-level nodes are flooded the same disaggregated prefixes since RIFT does not build a South TIE per node, which would complicate things unnecessarily. The lower-level node that can compute a southbound route to the prefix will prefer it to the disaggregated route anyway based on route preference rules.
  4. Positively disaggregated prefixes do not have to propagate to lower levels. With that, the disturbance in terms of new flooding is contained to a single level experiencing failures.
  5. Disaggregated South Prefix TIEs are not "reflected" by the lower level. Nodes within the same level do not need to be aware of which node computed the need for disaggregation.
  6. The fabric is still supporting maximum load balancing properties while not trying to send traffic northbound unless necessary.
In case positive disaggregation is triggered and due to the very stable but unsynchronized nature of the algorithm, the nodes may issue the necessary disaggregated prefixes at different points in time. For a short time, this can lead to an "incast" behavior where the first advertising router based on the nature of the longest prefix match will attract all the traffic. Different implementation strategies can be used to lessen that effect, but those are outside the scope of this specification. It is worth observing that, in a single-plane ToF, this disaggregation prevents traffic loss up to (K_LEAF * P) link failures in terms of or, in other terms, it takes at minimum that many link failures to partition the ToF into multiple planes.
Negative, Transitive Disaggregation for Fallen Leaves As explained in , failures in multi-plane ToF or more than (K_LEAF * P) links failing in single-plane design can generate fallen leaves. Such scenario cannot be addressed by positive disaggregation only and needs a further mechanism.
Cabling of Multiple ToF Planes Returning in this section to designs with multiple planes as shown originally in , highlights how the ToF is cabled in case of two planes by the means of dual-rings to distribute all the North TIEs within both planes.
Topologically Connected Planes
already describes how failures in multi-plane fabrics can lead to traffic loss that normal positive disaggregation cannot fix. The mechanism of negative, transitive disaggregation incorporated in RIFT provides the corresponding solution, and the next section explains the involved mechanisms in more detail.
Transitive Advertisement of Negative Disaggregates A ToF node discovering that it cannot reach a fallen leaf SHOULD disaggregate all the prefixes of that leaf. For that purpose, it uses negative South Prefix TIEs that are, as usual, flooded southwards with the scope defined in . Transitively, a node explicitly loses connectivity to a prefix when none of its children advertises it and when the prefix is negatively disaggregated by all of its parents. When that happens, the node originates the negative prefix further down south. Since the mechanism applies recursively south, the negative prefix may propagate transitively all the way down to the leaf. This is necessary since leaves connected to multiple planes by means of disjointed paths may have to choose the correct plane at the very bottom of the fabric to make sure that they don't send traffic towards another leaf using a plane where it is "fallen", which would make traffic loss unavoidable. When connectivity is restored, a node that disaggregated a prefix withdraws the negative disaggregation by the usual mechanism of re-advertising TIEs omitting the negative prefix.
Computation of Negative Disaggregates Negative prefixes can in fact be advertised due to two different triggers. This will be described consecutively. The first origination reason is a computation that uses all the North Node TIEs to build the set of all reachable nodes by reachability computation over the complete graph, including horizontal ToF links. The computation uses the node itself as the root. This is compared with the result of the normal southbound SPF as described in . The differences are the fallen leaves and all their attached prefixes are advertised as negative prefixes southbound if the node does not consider the prefix to be reachable within the southbound SPF. The second origination reason hinges on the understanding of how the negative prefixes are used within the computation as described in . When attaching the negative prefixes at a certain point in time, the negative prefix may find itself with all the viable nodes from the shorter match next hop being pruned. In other words, all its northbound neighbors provided a negative prefix advertisement. This is the trigger to advertise this negative prefix transitively south and is normally caused by the node being in a plane where the prefix belongs to a fabric leaf that has "fallen" in this plane. Obviously, when one of the northbound switches withdraws its negative advertisement, the node has to withdraw its transitively provided negative prefix as well.
Attaching Prefixes After an SPF is run, it is necessary to attach the resulting reachability information in the form of prefixes. For S-SPF, prefixes from a North TIE are attached to the originating node with that node's next-hop set and a distance equal to the prefix's cost plus the node's minimized path distance. The RIFT route database, a set of (prefix, prefix-type, attributes, path_distance, next-hop set), accumulates these results. N-SPF prefixes from each South TIE need to also be added to the RIFT route database. The N-SPF is really just a stub so the computing node simply needs to determine, for each prefix in a South TIE that originated from adjacent node, what next hops to use to reach that node. Since there may be parallel links, the next hops to use can be a set; the presence of the computing node in the associated South Node TIE is sufficient to verify that at least one link has bidirectional connectivity. The set of minimum cost next hops from the computing node X to the originating adjacent node is determined. Each prefix has its cost adjusted before being added into the RIFT route database. The cost of the prefix is set to the cost received plus the cost of the minimum distance next hop to that neighbor while considering its attributes such as mobility per . Then each prefix can be added into the RIFT route database with the next-hop set; ties are broken based upon type first and then distance and further on PrefixAttributes. Only the best combination is used for forwarding. RIFT route preferences are normalized by the enum RouteType in the Thrift model given in . An example implementation for node X follows:
Adding Routes from South TIE Positive and Negative Prefixes X.level next_hop_set = set of minimum cost links to the South TIE.originator next_hop_cost = minimum cost link to South TIE.originator end if for each prefix P in the South TIE P.cost = P.cost + next_hop_cost if P not in route_database: add (P, P.cost, P.type, P.attributes, next_hop_set) to route_database end if if (P in route_database): if route_database[P].cost > P.cost or route_database[P].type > P.type: update route_database[P] with (P, P.type, P.cost, P.attributes, next_hop_set) else if route_database[P].cost == P.cost and route_database[P].type == P.type: update route_database[P] with (P, P.type, P.cost, P.attributes, merge(next_hop_set, route_database[P].next_hop_set)) else // Not preferred route so ignore end if end if end for end for]]>
After the positive prefixes are attached and tie-broken, negative prefixes are attached and used in case of northbound computation, ideally from the shortest length to the longest. The next-hop adjacencies for a negative prefix are inherited from the longest positive prefix that aggregates it; subsequently, adjacencies to nodes that advertised negative disaggregation for this prefix are removed. The rule of inheritance MUST be maintained when the next-hop list for a prefix is modified, as the modification may affect the entries for matching negative prefixes of immediate longer prefix length. For instance, if a next hop is added, then by inheritance, it must be added to all the negative routes of immediate longer prefixes length unless it is pruned due to a negative advertisement for the same next hop. Similarly, if a next hop is deleted for a given prefix, then it is deleted for all the immediately aggregated negative routes. This will recurse in the case of nested negative prefix aggregations. The rule of inheritance MUST also be maintained when a new prefix of intermediate length is inserted or when the immediately aggregating prefix is deleted from the routing table, making an even shorter aggregating prefix the one from which the negative routes now inherit their adjacencies. As the aggregating prefix changes, all the negative routes MUST be recomputed, and then again, the process may recurse in case of nested negative prefix aggregations. Although these operations can be computationally expensive, the overall load on devices in the network is low because these computations are not run very often, as positive route advertisements are always preferred over negative ones. This prevents recursion in most cases because positive reachability information never inherits next hops. To make the negative disaggregation less abstract and provide an example ToP node, T1 with 4 ToF parents S1..S4 as represented in are considered further:
A ToP Node with 4 Parents +----+ +----+ +----+ +----+ N | S1 | | S2 | | S3 | | S4 | ^ +----+ +----+ +----+ +----+ W< + >E | | | | v |+--------+ | | S ||+-----------------+ | |||+----------------+---------+ |||| +----+ | T1 | +----+
If all ToF nodes can reach all the prefixes in the network, with RIFT, they will normally advertise a default route south. An abstract Routing Information Base (RIB), more commonly known as a routing table, stores all types of maintained routes, including the negative ones and "tie-breaks" for the best one, whereas an abstract forwarding table (FIB) retains only the ultimately computed "positive" routing instructions. In T1, those tables would look as illustrated in :
Abstract RIB +---------+ | Default | +---------+ | | +--------+ +---> | via S1 | | +--------+ | | +--------+ +---> | via S2 | | +--------+ | | +--------+ +---> | via S3 | | +--------+ | | +--------+ +---> | via S4 | +--------+
In case T1 receives a negative advertisement for prefix 2001:db8::/32 from S1, a negative route is stored in the RIB (indicated by a "~" sign), while the more specific routes to the complementing ToF nodes are installed in FIB. RIB and FIB in T1 now look as illustrated in Figures and , respectively:
Abstract RIB After Negative 2001:db8::/32 from S1 +---------+ +-----------------+ | Default | <-------------- | ~2001:db8::/32 | +---------+ +-----------------+ | | | +--------+ | +--------+ +---> | via S1 | +---> | via S1 | | +--------+ +--------+ | | +--------+ +---> | via S2 | | +--------+ | | +--------+ +---> | via S3 | | +--------+ | | +--------+ +---> | via S4 | +--------+
The negative 2001:db8::/32 prefix entry inherits from ::/0, so the positive, more specific routes are the complements to S1 in the set of next hops for the default route. That entry is composed of S2, S3, and S4, or in other words, it uses all entries in the default route with a "hole punched" for S1 into them. These are the next hops that are still available to reach 2001:db8::/32 now that S1 advertised that it will not forward 2001:db8::/32 anymore. Ultimately, those resulting next hops are installed in FIB for the more specific route to 2001:db8::/32 as illustrated below:
Abstract FIB After Negative 2001:db8::/32 from S1 +---------+ +---------------+ | Default | | 2001:db8::/32 | +---------+ +---------------+ | | | +--------+ | +---> | via S1 | | | +--------+ | | | | +--------+ | +--------+ +---> | via S2 | +---> | via S2 | | +--------+ | +--------+ | | | +--------+ | +--------+ +---> | via S3 | +---> | via S3 | | +--------+ | +--------+ | | | +--------+ | +--------+ +---> | via S4 | +---> | via S4 | +--------+ +--------+
To illustrate matters further, consider T1 receiving a negative advertisement for prefix 2001:db8:1::/48 from S2, which is stored in RIB again. After the update, the RIB in T1 is illustrated in :
Abstract RIB After Negative 2001:db8:1::/48 from S2 +---------+ +----------------+ +------------------+ | Default | <----- | ~2001:db8::/32 | <------ | ~2001:db8:1::/48 | +---------+ +----------------+ +------------------+ | | | | +--------+ | +--------+ | +---> | via S1 | +---> | via S1 | | | +--------+ +--------+ | | | | +--------+ | +--------+ +---> | via S2 | +---> | via S2 | | +--------+ +--------+ | | +--------+ +---> | via S3 | | +--------+ | | +--------+ +---> | via S4 | +--------+
Negative 2001:db8:1::/48 inherits from 2001:db8::/32 now, so the positive, more specific routes are the complements to S2 in the set of next hops for 2001:db8::/32, which are S3 and S4, or in other words, all entries of the parent with the negative holes "punched in" again. After the update, the FIB in T1 shows as illustrated in :
Abstract FIB After Negative 2001:db8:1::/48 from S2 +---------+ +---------------+ +-----------------+ | Default | | 2001:db8::/32 | | 2001:db8:1::/48 | +---------+ +---------------+ +-----------------+ | | | | +--------+ | | +---> | via S1 | | | | +--------+ | | | | | | +--------+ | +--------+ | +---> | via S2 | +---> | via S2 | | | +--------+ | +--------+ | | | | | +--------+ | +--------+ | +--------+ +---> | via S3 | +---> | via S3 | +---> | via S3 | | +--------+ | +--------+ | +--------+ | | | | +--------+ | +--------+ | +--------+ +---> | via S4 | +---> | via S4 | +---> | via S4 | +--------+ +--------+ +--------+
Further, assume that S3 stops advertising its service as a default gateway. The entry is removed from RIB as usual. In order to update the FIB, it is necessary to eliminate the FIB entry for the default route, as well as all the FIB entries that were created for negative routes pointing to the RIB entry being removed (::/0). This is done recursively for 2001:db8::/32 and then for 2001:db8:1::/48. The related FIB entries via S3 are removed as illustrated in .
Abstract FIB After Loss of S3 +---------+ +---------------+ +-----------------+ | Default | | 2001:db8::/32 | | 2001:db8:1::/48 | +---------+ +---------------+ +-----------------+ | | | | +--------+ | | +---> | via S1 | | | | +--------+ | | | | | | +--------+ | +--------+ | +---> | via S2 | +---> | via S2 | | | +--------+ | +--------+ | | | | | | | | | | | | | | | | | +--------+ | +--------+ | +--------+ +---> | via S4 | +---> | via S4 | +---> | via S4 | +--------+ +--------+ +--------+
Say that at that time, S4 would also disaggregate prefix 2001:db8:1::/48. This would mean that the FIB entry for 2001:db8:1::/48 becomes a discard route, and that would be the signal for T1 to disaggregate prefix 2001:db8:1::/48 negatively in a transitive fashion with its own children. Finally, the case occurs where S3 becomes available again as a default gateway, and a negative advertisement is received from S4 about prefix 2001:db8:2::/48 as opposed to 2001:db8:1::/48. Again, a negative route is stored in the RIB, and the more specific route to the complementing ToF nodes is installed in FIB. Since 2001:db8:2::/48 inherits from 2001:db8::/32, the positive FIB routes are chosen by removing S4 from S2, S3, S4. The abstract FIB in T1 now shows as illustrated in :
Abstract FIB After Negative 2001:db8:2::/48 from S4 +-----------------+ | 2001:db8:2::/48 | +-----------------+ | +---------+ +---------------+ +-----------------+ | Default | | 2001:db8::/32 | | 2001:db8:1::/48 | +---------+ +---------------+ +-----------------+ | | | | | +--------+ | | | +--------+ +---> | via S1 | | | +---> | via S2 | | +--------+ | | | +--------+ | | | | | +--------+ | +--------+ | | +--------+ +---> | via S2 | +---> | via S2 | | +---> | via S3 | | +--------+ | +--------+ | +--------+ | | | | +--------+ | +--------+ | +--------+ +---> | via S3 | +---> | via S3 | +---> | via S3 | | +--------+ | +--------+ | +--------+ | | | | +--------+ | +--------+ | +--------+ +---> | via S4 | +---> | via S4 | +---> | via S4 | +--------+ +--------+ +--------+
Optional Zero Touch Provisioning (RIFT ZTP) Each RIFT node can operate in Zero Touch Provisioning (ZTP) mode, i.e., it has no RIFT-specific configuration (unless it is a ToF or it is explicitly configured to operate in the overall topology as a leaf and/or support L2L procedures), and it will fully automatically derive necessary RIFT parameters itself after being attached to the topology. Manually configured nodes and nodes operating using RIFT ZTP can be mixed freely and will form a valid topology if achievable. The derivation of the level of each node happens based on offers received from its neighbors, whereas each node (with the possible exception of nodes configured as leaves) tries to attach at the highest possible point in the fabric. This guarantees that even if the diffusion front of offers reaches a node from "below" faster than from "above", it will greedily abandon an already negotiated level derived from nodes topologically below it and properly peer with nodes above. The fabric is very consciously numbered from the top down to allow for PoDs of different heights and to minimize the number of configurations necessary, in this case, just a TOP_OF_FABRIC flag on every node at the top of the fabric. This section describes the necessary concepts and procedures of the RIFT ZTP operation.
Terminology The interdependencies between the different flags and the configured level can be somewhat vexing at first, and it may take multiple reads of the glossary to comprehend them.
Automatic Level Derivation:
Procedures that allow nodes without a level configured to derive it automatically. Only applied if CONFIGURED_LEVEL is undefined.
UNDEFINED_LEVEL:
A "null" value that indicates that the level has not been determined and has not been configured. Schemas normally indicate that by a missing optional value without an available defined default.
LEAF_ONLY:
An optional configuration flag that can be configured on a node to make sure it never leaves the "bottom of the hierarchy". The TOP_OF_FABRIC flag and CONFIGURED_LEVEL cannot be defined at the same time as this flag. It implies a CONFIGURED_LEVEL value of leaf_level. It is indicated in the leaf_only schema element.
TOP_OF_FABRIC:
A configuration flag that MUST be provided on all ToF nodes. LEAF_FLAG and CONFIGURED_LEVEL cannot be defined at the same time as this flag. It implies a CONFIGURED_LEVEL value. In fact, it is basically a shortcut for configuring the same level at all ToF nodes, which is unavoidable since an initial "seed" is needed for other ZTP nodes to derive their level in the topology. The flag plays an important role in fabrics with multiple planes to enable successful negative disaggregation. It is carried in the top_of_fabric schema element. A standards-conforming RIFT implementation implies a CONFIGURED_LEVEL value of top_of_fabric_level in case of TOP_OF_FABRIC. This value is kept reasonably low to allow for fast ZTP reconvergence on failures.
CONFIGURED_LEVEL:
A level value provided manually. When this is defined (i.e., it is not an UNDEFINED_LEVEL), the node is not participating in ZTP in the sense of deriving its own level based on other nodes' information. The TOP_OF_FABRIC flag is ignored when this value is defined. LEAF_ONLY can be set only if this value is undefined or set to leaf_level.
DERIVED_LEVEL:
Level value computed via automatic level derivation when CONFIGURED_LEVEL is equal to UNDEFINED_LEVEL.
LEAF_2_LEAF:
An optional flag that can be configured on a node to make sure it supports procedures defined in . It is a capability that implies LEAF_ONLY and the corresponding restrictions. The TOP_OF_FABRIC flag is ignored when set at the same time as this flag. It is carried in the leaf_only_and_leaf_2_leaf_procedures schema flag.
LEVEL_VALUE:
With ZTP, the original definition of "level" in is both extended and relaxed. First, the level is defined now as LEVEL_VALUE and is the first defined value of CONFIGURED_LEVEL followed by DERIVED_LEVEL. Second, it is possible for nodes to be more than one level apart to form adjacencies if any of the nodes is at least LEAF_ONLY.
Valid Offered Level (VOL):
A neighbor's level received in a valid LIE (i.e., passing all checks for adjacency formation while disregarding all clauses involving level values) persisting for the duration of the holdtime interval on the LIE. Observe that offers from nodes offering the level value of leaf_level do not constitute VOLs (since no valid DERIVED_LEVEL can be obtained from those and consequently the not_a_ztp_offer flag MUST be ignored). Offers from LIEs with not_a_ztp_offer being true are not VOLs either. If a node maintains parallel adjacencies to the neighbor, VOL on each adjacency is considered as equivalent, i.e., the newest VOL from any such adjacency updates the VOL received from the same node.
Highest Available Level (HAL):
Highest-defined level value received from all VOLs received.
Highest Available Level Systems (HALS):
Set of nodes offering HAL VOLs.
Highest Adjacency ThreeWay (HAT):
Highest neighbor level of all the formed ThreeWay adjacencies for the node.
Automatic System ID Selection RIFT nodes require a 64-bit System ID that SHOULD be derived as EUI-64 MAC Address Block Large (MA-L) according to . The organizationally governed portion of this ID (24 bits) can be used to generate multiple IDs if required to indicate more than one RIFT instance. As matter of operational concern, the router MUST ensure that such identifier is not changing very frequently (or at least not without sending all its TIEs with fairly short lifetimes, i.e., purging them) since the network may otherwise be left with large amounts of stale TIEs in other nodes (though this is not necessarily a serious problem if the procedures described in are implemented).
Generic Fabric Example ZTP forces considerations of an incorrectly or unusually cabled fabric and how such a topology can be forced into a "lattice" structure that a fabric represents (with further restrictions). A necessary and sufficient physical cabling is shown in . The assumption here is that all nodes are in the same PoD.
Generic ZTP Cabling Considerations +---+ | A | s = TOP_OF_FABRIC | s | L = LEAF_ONLY ++-++ L2L = LEAF_2_LEAF | | +--+ +--+ | | +--++ ++--+ | E | | F | | +-+ | +-----------+ ++--+ | ++-++ | | | | | | | +-------+ | | | | | | | | | +----+ | | | | | | | ++-++ ++-++ | | I +-----+ J | | | | | +-+ | ++-++ +--++ | | | | | | | +---------+ | +------+ | | | | | | +-----------------+ | | | | | | | ++-++ ++-++ | | X +-----+ Y +-+ |L2L| | L | +---+ +---+
First, RIFT must anchor the "top" of the cabling and that's what the TOP_OF_FABRIC flag at node A is for. Then, things look smooth until the protocol has to decide whether node Y is at the same level as I, J (and as consequence, X is south of it), or X. This is unresolvable here until we "nail down the bottom" of the topology. To achieve that, the protocol chooses to use the leaf flags in X and Y in this example. In the case where Y does not have a leaf flag, it will try to elect the highest level offered and end up being in same level as I and J.
Level Determination Procedure A node starting up with UNDEFINED_VALUE (i.e., without a CONFIGURED_LEVEL or any leaf or TOP_OF_FABRIC flag) MUST follow these additional procedures:
  1. It advertises its LEVEL_VALUE on all LIEs (observe that this can be UNDEFINED_LEVEL, which in terms of the schema, is simply an omitted optional value).
  2. It computes HAL as the numerically highest available level in all VOLs.
  3. Then, it chooses MAX(HAL-1,0) as its DERIVED_LEVEL. The node then starts to advertise this derived level.
  4. A node that lost all adjacencies with the HAL value MUST holddown computation of the new DERIVED_LEVEL for at least one second unless it has no VOLs from southbound adjacencies. After the holddown timer expired, it MUST discard all received offers, recompute DERIVED_LEVEL, and announce it to all neighbors.
  5. A node MUST reset any adjacency that has changed the level it is offering and is in ThreeWay state.
  6. A node that changed its defined level value MUST re-advertise its own TIEs (since the new PacketHeader will contain a different level than before). The sequence number of each TIE MUST be increased.
  7. After a level has been derived, the node MUST set the not_a_ztp_offer on LIEs towards all systems offering a VOL for HAL.
  8. A node that changed its level SHOULD flush TIEs of all other nodes from its LSDB; otherwise, stale information may persist on "direction reversal", i.e., nodes that seemed south are now north or east-west. This will not prevent the correct operation of the protocol but could be slightly confusing operationally.
A node starting with LEVEL_VALUE being 0 (i.e., it assumes a leaf function by being configured with the appropriate flags or has a CONFIGURED_LEVEL of 0) MUST follow this additional procedure:
  1. It computes HAT per the procedures above but does not use it to compute DERIVED_LEVEL. HAT is used to limit adjacency formation per .
It MAY also follow this modified procedure:
  1. It may pick a different strategy to choose VOL, e.g., use the VOL value with highest number of VOLs. Such strategies are only possible since the node always remains "at the bottom of the fabric", while another layer could "invert" the fabric by picking its preferred VOL in a different fashion rather than always trying to achieve the highest viable level.
RIFT ZTP FSM This section specifies the precise, normative ZTP FSM and can be omitted unless the reader is pursuing an implementation of the protocol. For additional clarity, a graphical representation of the ZTP FSM is depicted in . It may also be helpful to refer to the normative schema in . The initial state is ComputeBestOffer.
RIFT ZTP FSM | ChangeLocalHierarchyIndications | | ChangeLocalConfiguredLevel | | BetterHAL | UpdatingClients | ComputationDone | HoldingDown | LostHAL | Enter | ChangeLocalHierarchyIndications | | ChangeLocalConfiguredLevel | | BetterHAL | | LostHAL | Enter | ChangeLocalHierarchyIndications | | ChangeLocalConfiguredLevel | | HoldDownExpired | | BetterHAL | | LostHAL | | ComputationDone | | NeighborOffer | | NeighborOffer | | NeighborOffer | | BetterHAT | | BetterHAT | | BetterHAT | | LostHAT | | LostHAT | | LostHAT | | ShortTic | | ShortTic | | ShortTic | ComputeBestOffer Enter | v +------------------+ | ComputeBestOffer | | |<----+ | | | ChangeLocalHierarchyIndications | | | ChangeLocalConfiguredLevel | | | NeighborOffer | | | BetterHAL | | | BetterHAT | | | LostHAT | | | ShortTic | |-----+ | | | |<--------------------- | |---------------------> (UpdatingClients) | | ComputationDone +------------------+ ^ | | | LostHAL | V (HoldingDown) (ComputeBestOffer) | ^ | | ChangeLocalHierarchyIndications | | ChangeLocalConfiguredLevel | | HoldDownExpired V | +------------------+ | HoldingDown | | |<----+ | | | NeighborOffer | | | BetterHAL | | | BetterHAT | | | LostHAL | | | LostHat | | | ComputationDone | | | ShortTic | |-----+ +------------------+ ^ | (UpdatingClients) (ComputeBestOffer) | ^ | | ChangeLocalHierarchyIndications | | ChangeLocalConfiguredLevel | | BetterHAL | | BetterHAT | | LostHAT V | +------------------+ | UpdatingClients | | |<----+ | | | | | | NeighborOffer | | | ShortTic | |-----+ +------------------+ | | LostHAL V (HoldingDown)
The following terms are used for well-known procedures:
  • PUSH Event: queues an event to be executed by the FSM upon exit of this action
  • COMPARE_OFFERS: checks whether, based on current offers and held last results, the events BetterHAL/LostHAL/BetterHAT/LostHAT are necessary and returns them
  • UPDATE_OFFER: store current offer with adjacency holdtime as lifetime and COMPARE_OFFERS, then PUSH corresponding events
  • LEVEL_COMPUTE: compute best offered or configured level and HAL/HAT, if anything changed, PUSH ComputationDone
  • REMOVE_OFFER: remove the corresponding offer and COMPARE_OFFERS, PUSH corresponding events
  • PURGE_OFFERS: REMOVE_OFFER for all held offers, COMPARE OFFERS, PUSH corresponding events
  • PROCESS_OFFER:
    1. if no level is offered, then REMOVE_OFFER
    2. else
      1. if offered level > leaf, then UPDATE_OFFER
      2. else REMOVE_OFFER
States:
  • ComputeBestOffer: Processes received offers to derive ZTP variables.
  • HoldingDown: Holding down while receiving updates.
  • UpdatingClients: Updates other FSMs on the same node with computation results.
Events:
  • ChangeLocalHierarchyIndications: Node locally configured with new leaf flags.
  • ChangeLocalConfiguredLevel: Node locally configured with a defined level.
  • NeighborOffer: A new neighbor offer with optional level and neighbor state.
  • BetterHAL: Better HAL computed internally.
  • BetterHAT: Better HAT computed internally.
  • LostHAL: Lost last HAL in computation.
  • LostHAT: Lost HAT in computation.
  • ComputationDone: Computation performed.
  • HoldDownExpired: Holddown timer expired.
  • ShortTic: One-second timer tick. This event is provided to the FSM once a second by an implementation-specific mechanism that is outside the scope of this specification. This event is quietly ignored if the relevant transition does not exist.
Actions:
  • on ChangeLocalConfiguredLevel in HoldingDown finishes in ComputeBestOffer: store configured level
  • on BetterHAT in HoldingDown finishes in HoldingDown: no action
  • on ShortTic in HoldingDown finishes in HoldingDown: remove expired offers, and if holddown timer expired, PUSH_EVENT HoldDownExpired
  • on NeighborOffer in HoldingDown finishes in HoldingDown: PROCESS_OFFER
  • on ComputationDone in HoldingDown finishes in HoldingDown: no action
  • on BetterHAL in HoldingDown finishes in HoldingDown: no action
  • on LostHAT in HoldingDown finishes in HoldingDown: no action
  • on LostHAL in HoldingDown finishes in HoldingDown: no action
  • on HoldDownExpired in HoldingDown finishes in ComputeBestOffer: PURGE_OFFERS
  • on ChangeLocalHierarchyIndications in HoldingDown finishes in ComputeBestOffer: store leaf flags
  • on LostHAT in ComputeBestOffer finishes in ComputeBestOffer: LEVEL_COMPUTE
  • on NeighborOffer in ComputeBestOffer finishes in ComputeBestOffer: PROCESS_OFFER
  • on BetterHAT in ComputeBestOffer finishes in ComputeBestOffer: LEVEL_COMPUTE
  • on ChangeLocalHierarchyIndications in ComputeBestOffer finishes in ComputeBestOffer: store leaf flags and LEVEL_COMPUTE
  • on LostHAL in ComputeBestOffer finishes in HoldingDown: if any southbound adjacencies present, then update holddown timer to normal duration, else fire holddown timer immediately
  • on ShortTic in ComputeBestOffer finishes in ComputeBestOffer: remove expired offers
  • on ComputationDone in ComputeBestOffer finishes in UpdatingClients: no action
  • on ChangeLocalConfiguredLevel in ComputeBestOffer finishes in ComputeBestOffer: store configured level and LEVEL_COMPUTE
  • on BetterHAL in ComputeBestOffer finishes in ComputeBestOffer: LEVEL_COMPUTE
  • on ShortTic in UpdatingClients finishes in UpdatingClients: remove expired offers
  • on LostHAL in UpdatingClients finishes in HoldingDown: if any southbound adjacencies are present, then update holddown timer to normal duration, else fire holddown timer immediately
  • on BetterHAT in UpdatingClients finishes in ComputeBestOffer: no action
  • on BetterHAL in UpdatingClients finishes in ComputeBestOffer: no action
  • on ChangeLocalConfiguredLevel in UpdatingClients finishes in ComputeBestOffer: store configured level
  • on ChangeLocalHierarchyIndications in UpdatingClients finishes in ComputeBestOffer: store leaf flags
  • on NeighborOffer in UpdatingClients finishes in UpdatingClients: PROCESS_OFFER
  • on LostHAT in UpdatingClients finishes in ComputeBestOffer: no action
  • on Entry into ComputeBestOffer: LEVEL_COMPUTE
  • on Entry into UpdatingClients: update all LIE FSMs with computation results
Resulting Topologies The procedures defined in will lead to the RIFT topology and levels depicted in .
Generic ZTP Topology Autoconfigured +---+ | A | | 24| ++-++ | | +--+ +--+ | | +--++ ++--+ | E | | F | | 23+-+ | 23+-----------+ ++--+ | ++-++ | | | | | | | +-------+ | | | | | | | | | +----+ | | | | | | | ++-++ ++-++ | | I +-----+ J | | | 22| | 22| | ++--+ +--++ | | | | +---------+ | | | | | ++-++ +---+ | | X | | Y +-+ | 0 | | 0 | +---+ +---+
In the case where the LEAF_ONLY restriction on Y is removed, the outcome would be very different however and result in . This basically demonstrates that autoconfiguration makes miscabling detection hard and, with that, can lead to undesirable effects in cases where leaves are not "nailed" by the appropriately configured flags and arbitrarily cabled.
Generic ZTP Topology Autoconfigured +---+ | A | | 24| ++-++ | | +--+ +--+ | | +--++ ++--+ | E | | F | | 23+-+ | 23+-------+ ++--+ | ++-++ | | | | | | | +-------+ | | | | | | | | | +----+ | | | | | | | ++-++ ++-++ +-+-+ | I +-----+ J +-----+ Y | | 22| | 22| | 22| ++-++ +--++ ++-++ | | | | | | +-----------------+ | | | | +---------+ | | | | | ++-++ | | X +--------+ | 0 | +---+
Further Mechanisms
Route Preferences Since RIFT distinguishes between different route types, such as external routes from other protocols, and additionally advertises special types of routes on disaggregation, the protocol MUST tie-break internally different types on a clear preference scale to prevent traffic loss or loops. The preferences are given in the schema type RouteType. contains the route type as derived from the TIE type carrying it. Entries are sorted from the most preferred route type to the least preferred route type. TIEs and Contained Route Types
TIE Type Resulting Route Type
None Discard
Local Interface LocalPrefix
S-PGP South PGP
N-PGP North PGP
North Prefix NorthPrefix
North External Prefix NorthExternalPrefix
South Prefix and South Positive Disaggregation SouthPrefix
South External Prefix and South Positive External Disaggregation SouthExternalPrefix
South Negative Prefix NegativeSouthPrefix
Overload Bit The overload attribute is specified in the packet encoding schema in the overload flag. The overload flag MUST be respected by all necessary SPF computations. A node with the overload flag set SHOULD advertise all locally hosted prefixes, both northbound and southbound; all other southbound prefixes SHOULD NOT be advertised. Leaf nodes SHOULD set the overload attribute on all originated Node TIEs. If spine nodes were to forward traffic not intended for the local node, the leaf node would not be able to prevent routing/forwarding loops as it does not have the necessary topology information to do so.
Optimized Route Computation on Leaves Leaf nodes only have visibility to directly connected nodes and therefore are not required to run "full" SPF computations. Instead, prefixes from neighboring nodes can be gathered to run a "partial" SPF computation in order to build the routing table. Leaf nodes SHOULD only hold their own N-TIEs and, in cases of L2L implementations, the N-TIEs of their East-West neighbors. Leaf nodes MUST hold all S-TIEs from their neighbors. Normally, a full network graph is created based on local N-TIEs and remote S-TIEs that it receives from neighbors, at which time, necessary SPF computations are performed. Instead, leaf nodes can simply compute the minimum cost and next-hop set of each leaf neighbor by examining its local adjacencies. Associated N-TIEs are used to determine bidirectionality and derive the next-hop set. The cost is then derived from the minimum cost of the local adjacency to the neighbor and the prefix cost. Leaf nodes would then attach necessary prefixes as described in .
Mobility The RIFT control plane MUST maintain the real time status of every prefix, to which port it is attached, and to which leaf node that port belongs. This is still true in cases of IP mobility where the point of attachment may change several times a second. There are two classic approaches to explicitly maintain this information, "timestamp" and "sequence counter", which are defined as follows:
timestamp:
With this method, the infrastructure SHOULD record the precise time at which the movement is observed. One key advantage of this technique is that it has no dependency on the mobile device. One drawback is that the infrastructure MUST be precisely synchronized in order to be able to compare timestamps as the points of attachment change. This could be accomplished by utilizing the Precision Time Protocol (PTP) (IEEE Std. 1588 or 802.1AS ), which is designed for bridged LANs. Both the precision of the synchronization protocol and the resolution of the timestamp must beat the shortest possible roaming time on the fabric. Another drawback is that the presence of a mobile device may only be observed asynchronously, such as when it starts using an IP protocol like ARP , IPv6 Neighbor Discovery , IPv6 Stateless Address Configuration , DHCP , or DHCPv6 .
sequence counter:
With this method, a mobile device notifies its point of attachment on arrival with a sequence counter that is incremented upon each movement. On the positive side, this method does not have a dependency on a precise sense of time, since the sequence of movements is kept in order by the mobile device. The disadvantage of this approach is the need for support for protocols that may be used by the mobile device to register its presence to the leaf node with the capability to provide a sequence counter. Well-known issues with sequence counters, such as wrapping and comparison rules, MUST be addressed properly. Sequence numbers MUST be compared by a single homogenous source to make operation feasible. Sequence number comparison from multiple heterogeneous sources would be extremely difficult to implement.
RIFT supports a hybrid approach by using an optional 'PrefixSequenceType' attribute (which is also called a monotonic_clock in the schema) that consists of a timestamp and optional sequence number field. In case of a negatively distributed prefix, this attribute MUST NOT be included by the originator and it MUST be ignored by all nodes during computation. When this attribute is present (observe that per data schema, the attribute itself is optional, but in case it is included, the "timestamp" field is required):
  • The leaf node MAY advertise a timestamp of the latest sighting of a prefix, e.g., by snooping IP protocols or the node using the time at which it advertised the prefix. RIFT transports the timestamp within the desired North Prefix TIEs as the timestamp.
  • RIFT MAY interoperate with "Registration Extensions for 6LoWPAN Neighbor Discovery" , which provides a method for registering a prefix with a sequence number called a Transaction ID (TID). In such cases, RIFT SHOULD transport the derived TID without modification.
  • RIFT also defines an abstract negative clock (ASNC) (also called an "undefined" clock). The ASNC MUST be considered older than any other defined clock. By default, when a node receives a North Prefix TIE that does not contain a 'PrefixSequenceType' attribute, it MUST interpret the absence as the ASNC.
  • Any prefix present on the fabric in multiple nodes that have the same clock is considered as anycast.
  • The RIFT specification assumes that all nodes are being synchronized within at least 200 milliseconds or less. This is achievable through the use of NTP . An implementation MAY provide a way to reconfigure a domain to a different value and provides a variable called MAXIMUM_CLOCK_DELTA for this purpose.
Clock Comparison All monotonic clock values MUST be compared to each other using the following rules:
  1. The ASNC is older than any other value except ASNC and
  2. Clocks with timestamps differing by more than MAXIMUM_CLOCK_DELTA are comparable by using the timestamps only and
  3. Clocks with timestamps differing by less than MAXIMUM_CLOCK_DELTA are comparable by using their TIDs only, and
  4. An undefined TID is always older than any other TID, and
  5. TIDs are compared using rules of .
Interaction Between Timestamps and Sequence Counters For attachment changes that occur less frequently (e.g., once per second), the timestamp that the RIFT infrastructure captures should be enough to determine the most current discovery. If the point of attachment changes faster than the maximum drift of the timestamping mechanism (i.e., MAXIMUM_CLOCK_DELTA), then a sequence number SHOULD be used to enable necessary precision to determine currency. The sequence counter in is encoded as one octet and wraps around using the arithmetic defined in . Within the resolution of MAXIMUM_CLOCK_DELTA, sequence counter values captured during 2 sequential iterations of the same timestamp SHOULD be comparable. This means that with default values, a node may move up to 127 times in a 200-millisecond period and the clocks will remain comparable. This allows the RIFT infrastructure to explicitly assert the most up-to-date advertisement.
Anycast vs. Unicast A unicast prefix can be attached to one leaf at most, whereas an anycast prefix may be reachable via more than one leaf. If a monotonic clock attribute is provided on the prefix, then the prefix with the newest clock value is strictly preferred. An anycast prefix does not carry a clock, or all clock attributes MUST be the same under the rules of . In mobility events, it is important that the leaf is reflooding as quickly as possible to communicate the absence of the prefix that moved. Without support for , movements on the fabric within intervals smaller than 100 msec will be interpreted as anycast.
Overlays and Signaling RIFT is agnostic to any overlay technologies and their associated control and transports that run on top of it (e.g., Virtual eXtensible Local Area Network (VXLAN)). It is expected that leaf nodes and possibly ToF nodes can perform necessary data plane encapsulation. In the context of mobility, overlays provide another possible solution to avoid injecting mobile prefixes into the fabric as well as improving scalability of the deployment. It makes sense to consider overlays for mobility solutions in IP fabrics. As an example, a mobility protocol such as the Locator/ID Separation Protocol (LISP) may inform the ingress leaf of the location of the egress leaf in real time. Another possibility is to consider that mobility is an underlay service and support it in RIFT to an extent. The load on the fabric increases with the amount of mobility since a move forces flooding and computation on all nodes in the scope of the move so tunneling from the leaf to the ToF may be desired to speed up convergence times.
Key/Value (KV) Store
Southbound RIFT supports the southbound distribution of key-value pairs that can be used to distribute information to facilitate higher levels of functionality (e.g., distribution of configuration information). KV South TIEs may arrive from multiple nodes and therefore MUST execute the following tie-breaking rules for each key:
  1. Only KV TIEs received from nodes to which a bidirectional adjacency exists MUST be considered.
  2. For each valid KV South TIEs that contains the same key, the value within the South TIE with the highest level will be preferred. If the levels are identical, the highest originating System ID will be preferred. In the case of overlapping keys in the winning South TIE, the behavior is undefined.
Consider that if a node goes down, nodes south of it will lose associated adjacencies, causing them to disregard corresponding KVs. New KV South TIEs are advertised to prevent stale information being used by nodes that are further south. KV advertisements southbound are not a result of independent computation by every node over the same set of South TIEs but a diffused computation.
Northbound Certain use cases necessitate distribution of essential KV information that is generated by the leaves in the northbound direction. Such information is flooded in KV North TIEs. Since the originator of the KV North TIEs is preserved during flooding, the corresponding mechanism will define, if necessary, tie-breaking rules depending on the semantics of the information. Only KV TIEs from nodes that are reachable via multi-plane reachability computation mentioned in SHOULD be considered.
Interactions with BFD RIFT MAY incorporate Bidirectional Forwarding Detection (BFD) to react quickly to link failures. In such case, the following procedures are introduced:
  1. After RIFT ThreeWay hello adjacency convergence, a BFD session MAY be formed automatically between the RIFT endpoints without further configuration using the exchanged discriminators that are equal to the local_id in the LIEPacket. The capability of the remote side to support BFD is carried in the LIEs in LinkCapabilities.
  2. In case an established BFD session goes Down after it was Up, RIFT adjacency SHOULD be re-initialized and subsequently started from Init after it receives a consecutive BFD Up.
  3. In case of parallel links between nodes, each link MAY run its own independent BFD session or they MAY share a session. The specific manner in which this is implemented is outside the scope of this document.
  4. If link identifiers or BFD capabilities change, both the LIE and any BFD sessions SHOULD be brought down and back up again. In case only the advertised capabilities change, the node MAY choose to persist the BFD session.
  5. Multiple RIFT instances MAY choose to share a single BFD session; in such cases, the behavior for which discriminators are used is undefined. However, RIFT MAY advertise the same link ID for the same interface in multiple instances to "share" discriminators.
  6. The BFD TTL follows .
Fabric Bandwidth Balancing A well understood problem in fabrics is that, in case of link failures, it would be ideal to rebalance how much traffic is sent to switches in the next level based on the available ingress and egress bandwidth. RIFT supports a light-weight mechanism that can deal with the problem based on the fact that RIFT is loop-free.
Northbound Direction Every RIFT node SHOULD compute the amount of northbound bandwidth available through neighbors at a higher level and modify the distance received on the default route from these neighbors. The bandwidth is advertised in the NodeNeighborsTIEElement element, which represents the sum of the bandwidths of all the parallel links to a neighbor. Default routes with differing distances SHOULD be used to support weighted ECMP forwarding. Such a distance is called Bandwidth Adjusted Distance (BAD). This is best illustrated by a simple example.
Balancing Bandwidth All Leaf to Spine Links 10 Mbit/s All Leaf to Spine Links 100 Mbit/s 100 x 100 100 Mbit/s | x | | +-+---+-+ +-+---+-+ | | | | |Spin111| |Spin112| +-+---+++ ++----+++ |x || || || || |+---------------+ || || +---------------+| || || || || || || || || || -----All Links 10 Mbit/s----- || || || || || || || || || +------------+| || || || |+------------+ || || |x || || || +-+---+++ +--++-+++ | | | | |Leaf111| |Leaf112| +-------+ +-------+
depicts an example topology where links between leaf and spine nodes are 10 Mbit/s and links from spine nodes northbound are 100 Mbit/s. It includes parallel link failure between Leaf 111 and Spine 111, and as a result, Leaf 111 wants to forward more traffic towards Spine 112. Additionally, it includes an uplink failure on Spine 111. The local modification of the received default route distance from the upper level is achieved by running a relatively simple algorithm where the bandwidth is weighted exponentially, while the distance on the default route represents a multiplier for the bandwidth weight for easy operational adjustments. On a node, L, use Node TIEs to compute 3 values from each non-overloaded northbound neighbor, N:
  1. L_N_u: sum of the bandwidth available from L to N (to account for parallel links)
  2. N_u: sum of the uplink bandwidth available on N
  3. T_N_u: L_N_u * OVERSUBSCRIPTION_CONSTANT + N_u
For all T_N_u, determine the corresponding M_N_u as log_2(next_power_2(T_N_u)) and determine MAX_M_N_u as the maximum value of all such M_N_u values. For each advertised default route from a node N, modify the advertised distance D to BAD = D * (1 + MAX_M_N_u - M_N_u) and use BAD instead of distance D to balance the weight of the default forwarding towards N. For the example above, a simple table of values will help in understanding the concept. The implicit assumption here is that all default route distances are advertised with D=1 and that OVERSUBSCRIPTION_CONSTANT=1. BAD Computation
Node N T_N_u M_N_u BAD
Leaf111 Spine 111 110 7 2
Leaf111 Spine 112 220 8 1
Leaf112 Spine 111 120 7 2
Leaf112 Spine 112 220 8 1
If a calculation produces a result exceeding the range of the type, e.g., bandwidth, the result is set to the highest possible value for that type. BAD SHOULD only be computed for default routes. A node MAY compute and use BAD for any disaggregated prefixes or other RIFT routes. A node MAY use a different algorithm to weight northbound traffic based on the bandwidth. If a different algorithm is used, its successful behavior MUST NOT depend on uniformity of the algorithm or synchronization of BAD computations across the fabric. For example, it is conceivable that leaves could use real time link loads gathered by analytics to change the amount of traffic assigned to each default route next hop. A change in available bandwidth will only affect, at most, two levels down in the fabric, i.e., the blast radius of bandwidth adjustments is constrained no matter the fabric's height.
Southbound Direction Due to its loop-free nature, during South SPF, a node MAY account for the maximum available bandwidth on nodes in lower levels and modify the amount of traffic offered to the next level's southbound nodes. It is worth considering that such computations may be more effective if they are standardized, but they do not have to be. As long as a packet continues to flow southbound, it will take some viable, loop-free path to reach its destination.
Label Binding In its LIEs, a node MAY advertise a locally significant, downstream-assigned, interface-specific label. One use of such a label is a hop-by-hop encapsulation allowing forwarding planes to be easily distinguished among multiple RIFT instances.
L2L Procedures RIFT implementations SHOULD support special East-West adjacencies between leaf nodes. Leaf nodes supporting these procedures MUST:
  1. advertise the LEAF_2_LEAF flag in its node capabilities,
  2. set the overload flag on all leaf's Node TIEs,
  3. flood only a node's own North and South TIEs over E-W leaf adjacencies,
  4. always use E-W leaf adjacency in all SPF computations,
  5. install a discard route for any advertised aggregate routes in a leaf's TIE, and
  6. never form southbound adjacencies.
This will allow the E-W leaf nodes to exchange traffic strictly for the prefixes advertised in each other's North Prefix TIEs since the southbound computation will find the reverse direction in the other node's TIE and install its north prefixes.
Address Family and Multi-Topology Considerations Multi-Topology (MT) and Multi-Instance (MI) concepts are used today in link-state routing protocols to support several domains on the same physical topology. RIFT supports this capability by carrying transport ports in the LIE protocol exchanges. Multiplexing of LIEs can be achieved by either choosing varying multicast addresses or ports on the same address. BFD interactions in are implementation-dependent when multiple RIFT instances run on the same link.
One-Hop Healing of Levels with East-West Links Based on the rules defined in Sections and and given the presence of E-W links, RIFT can provide a one-hop protection for nodes that have lost all their northbound links. This can also be applied to multi-plane designs where complex link set failures occur at the ToF when links are exclusively used for flooding topology information. outlines this behavior.
Security
Security Model An inherent property of any security and ZTP architecture is the resulting trade-off in regard to integrity verification of the information distributed through the fabric vs. provisioning and autoconfiguration requirements. At a minimum, the security of an established adjacency should be ensured. The stricter the security model, the more provisioning must take over the role of ZTP. RIFT supports the following security models to allow for flexible control by the operator:
  • The most security-conscious operators may choose to have control over which ports interconnect between a given pair of nodes, such a model is called the "Port-Association Model" (PAM). This is achievable by configuring each pair of directly connected ports with a designated shared key or public/private key pair.
  • In physically secure data center locations, operators may choose to control connectivity between entire nodes, called here the "Node-Association Model" (NAM). A benefit of this model is that it allows for simplified port sparing.
  • In the most relaxed environments, an operator may only choose to control which nodes join a particular fabric. This is denoted as the "Fabric-Association Model" (FAM). This is achievable by using a single shared secret across the entire fabric. Such flexibility makes sense when servers are considered as leaf devices, as those are replaced more often than network nodes. In addition, this model allows for simplified node sparing.
  • These models may be mixed throughout the fabric depending upon security requirements at various levels of the fabric and willingness to accept increased provisioning complexity.
In order to support the cases mentioned above, RIFT implementations supports, through operator control, mechanisms that allow for:
  • a specification of the appropriate level in the fabric,
  • discovery and reporting of missing connections, and
  • discovery and reporting of unexpected connections while preventing them from forming insecure adjacencies.
Operators may only choose to configure the level of each node but not explicitly configure which connections are allowed. In this case, RIFT will only allow adjacencies to establish between nodes that are in adjacent levels. Operators with the lowest security requirements may not use any configuration to specify which connections are allowed. Nodes in such fabrics could rely fully on ZTP and established adjacencies between nodes in adjacent levels. illustrates inherent trade-offs between the different security models. Some level of link quality verification may be required prior to an adjacency being used for forwarding. For example, an implementation may require that a BFD session comes up before advertising the adjacency. For the cases outlined above, RIFT has two approaches to enforce that a local port is connected to the correct port on the correct remote node. One approach is to piggyback on RIFT's authentication mechanism. Assuming the provisioning model (e.g., YANG) is flexible enough, operators can choose to provision a unique authentication key for the following conceptual models:
  • each pair of ports in "port-association model",
  • each pair of switches in "node-association model", or
  • the entire fabric in "fabric-association model".
The other approach is to rely on the System ID, port-id, and level fields in the LIE message to validate an adjacency against the expected cabling topology and optionally introduce some new rules in the FSM to allow the adjacency to come up if the expectations are met.
Security Model ^ /\ | /|\ / \ | | / \ | | / PAM \ | Increasing / \ Increasing Integrity +----------+ Flexibility & / NAM \ & Increasing +--------------+ Less Provisioning / FAM \ Configuration | / \ | | +--------------------+ \|/ | / Zero Configuration \ v +------------------------+
Security Mechanisms RIFT security goals are to ensure:
  • authentication,
  • message integrity,
  • the prevention of replay attacks,
  • low processing overhead, and
  • efficient messaging
unless no security is deployed by means of using 'undefined_securitykey_id' as key identifiers (key ID). Message confidentiality is a non-goal. The model in the previous section allows a range of security key types that are analogous to the various security association models. PAM and NAM allow security associations at the port or node level using symmetric or asymmetric keys that are preinstalled. FAM argues for security associations to be applied only at a group level or to be refined once the topology has been established. RIFT does not specify how security keys are installed or updated, though it does specify how the key can be used to achieve security goals. The protocol has provisions for "weak" nonces to prevent replay attacks and includes authentication mechanisms comparable to those described in and .
Security Envelope A serialized schema ProtocolPacket MUST be carried in a secure envelope as illustrated in . The ProtocolPacket MUST be serialized using the default Thrift's binary protocol. Any value in the packet following a security fingerprint MUST be used by a receiver only after the fingerprint generated based on an acceptable, advertised key ID has been validated against the data covered by the bare exceptions arising from operational exigencies. Based on local configuration, a node MAY allow for the envelope's integrity checks to be skipped and for the procedure specified in to be implemented. This means that for all packets, in case the node is configured to validate the outer fingerprint based on a key ID, an unexpected key ID or fingerprint not validating against the expected key ID will lead to packet rejection. Further, in case of reception of a TIE and the receiver being configured to validate the originator by checking the TIE Origin Security Envelope Header fingerprint against a key ID, an incorrect key ID or inner fingerprint not validating against the key ID will lead to the rejection of the packet. For reasons of clarity, it is important to observe that the specification uses the words "fingerprint" and "signature" interchangeably since the specific properties of the fingerprint part of the envelope depend on the algorithms used to insure the payload integrity. Moreover, any security chosen never implies encryption due to performance impact involved but only fingerprint or signature generation and validation. An implementation MUST implement at least both sending and receiving HMAC-SHA256 fingerprints as defined in to ensure interoperability but MAY use 'undefined_securitykey_id' by default.
Security Envelope 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 UDP Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Source Port | RIFT destination port | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | UDP Length | UDP Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Outer Security Envelope Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | RIFT MAGIC | Packet Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | RIFT Major | Outer Key ID | Fingerprint | | | Version | | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Security Fingerprint covers all following content ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Weak Nonce Local | Weak Nonce Remote | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Remaining TIE Lifetime (all 1s in case of LIE) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ TIE Origin Security Envelope Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | TIE Origin Key ID | Fingerprint | | | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Security Fingerprint covers all following content ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Serialized RIFT Model Object +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Serialized RIFT Model Object ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
RIFT MAGIC:
16 bits Constant value of 0xA1F7 that allows easy classification of RIFT packets independent of the UDP port used.
Packet Number:
16 bits An optional, per-adjacency, per-packet type number set using the sequence number arithmetic defined in . If the arithmetic in is not used, the node MUST set the value to undefined_packet_number. This number can be used to detect losses and misordering in flooding for either operational purposes or in implementation to adjust flooding behavior to current link or buffer quality. This number MUST NOT be used to discard or validate the correctness of packets. Packet numbers are incremented on each interface and within that for each type of packet independently. This allows parallelizing packet generation and processing for different types within an implementation, if so desired.
RIFT Major Version:
8 bits This value MUST be set to "protocol_major_version", which is defined in the schema and used to serialize the object contained. It allows checking whether protocol versions are compatible on both sides, i.e., which schema version is necessary to decode the serialized object. An implementation MUST drop packets with unexpected values and MAY report a problem. The specification of how an implementation may negotiate the schema's major version is outside the scope of this document.
Outer Key ID:
8 bits A simple, unstructured value acting as indirection into a structure holding an algorithm and any related secrets necessary to validate any provided outer security fingerprint or signature. The value undefined_securitykey_id means that no valid fingerprint was computed or is provided; otherwise, one of the algorithms in MUST be used to compute the fingerprint. This key ID scope is local to the nodes on both ends of the adjacency.
TIE Origin Key ID:
24 bits A simple, unstructured value acting as indirection into a structure holding an algorithm and any related secrets necessary to validate any provided inner security fingerprint or signature. The value undefined_securitykey_id means that no valid fingerprint was computed; otherwise, one of the algorithms in MUST be used to compute the fingerprint. This key ID scope is global to the RIFT instance since it may imply the originator of the TIE so the contained object does not have to be deserialized to obtain the originator.
Fingerprint Length:
8 bits Length in 32-bit multiples of the following fingerprint (not including lifetime or weak nonces). It allows the structure to be navigated when an unknown key type is present. To clarify, a common corner case when this value is set to 0 is when it signifies an empty (0 bytes long) security fingerprint.
Security Fingerprint:
32 bits * Fingerprint Length This is a signature that is computed over all data following after it. If the significant bits of the fingerprint are fewer than the 32-bit padded length, then the significant bits MUST be left aligned and the remaining bits on the right are padded with 0s. When using Public Key Infrastructure (PKI), the security fingerprint originating node uses its private key to create the signature. The original packet can then be verified, provided the public key is shared and current. Methodology to negotiate, distribute, or rollover keys is outside the scope of this document.
Remaining TIE Lifetime:
32 bits In case of anything but TIEs, this field MUST be set to all ones and the Origin Security Envelope Header MUST NOT be present in the packet. For TIEs, this field represents the remaining lifetime of the TIE and the Origin Security Envelope Header MUST be present in the packet.
Weak Nonce Local:
16 bits Local Weak Nonce of the adjacency, as advertised in LIEs.
Weak Nonce Remote:
16 bits Remote Weak Nonce of the adjacency, as received in LIEs.
TIE Origin Security Envelope Header:
It MUST be present if and only if the Remaining TIE Lifetime field is not all ones. It carries through the originator's key ID and corresponding fingerprint of the object to protect TIE from modification during flooding. This ensures origin validation and integrity (but does not provide validation of a chain of trust).
Observe that, due to the schema migration rules per , the contained model can always be decoded if the major version matches and the envelope integrity has been validated. Consequently, description of the TIE is available to flood it properly, including unknown TIE types.
Weak Nonces The protocol uses two 16-bit nonces to salt generated signatures. The term "nonce" is used a bit loosely since RIFT nonces are not being changed in every packet, which is common in cryptography. For efficiency purposes, they are changed at a high enough frequency to dwarf practical replay attack attempts. And hence, such nonces are called from this point on "weak" nonces. Any implementation using a different outer key ID from 'undefined_securitykey_id' MUST generate and wrap around local nonces properly and SHOULD do it even if not using any algorithm from . When a nonce increment leads to the undefined_nonce value, the value MUST be incremented again immediately. All implementations MUST reflect the neighbor's nonces. An implementation SHOULD increment a chosen nonce on every LIE FSM transition that ends up in a different state from the previous one and MUST increment its nonce at least every nonce_regeneration_interval if using any algorithm in (such considerations allow for efficient implementations without opening a significant security risk). When flooding TIEs, the implementation MUST use recent (i.e., within allowed difference) nonces reflected in the LIE exchange. The schema specifies in maximum_valid_nonce_delta the maximum allowable nonce value difference on a packet compared to reflected nonces in the LIEs. Any packet received with nonces deviating more than the allowed delta MUST be discarded without further computation of signatures to prevent computation load attacks. The delta is either a negative or positive difference that a mirrored nonce can deviate from the local value to be considered valid. If nonces are not changed on every packet, but at the maximum interval on both sides, this opens statistically a maximum_valid_nonce_delta/2 window for identical LIEs, TIE, and TI(x)E replays. The interval cannot be too small since LIE FSM may change states fairly quickly during ZTP without sending LIEs, and additionally, UDP can both loose as well as misorder packets. In cases where a secure implementation does not receive signatures or receives undefined nonces from a neighbor (indicating that it does not support or verify signatures), it is a matter of local policy as to how those packets are treated. A secure implementation MAY refuse forming an adjacency with an implementation that is not advertising signatures or valid nonces, or it MAY continue signing local packets while accepting a neighbor's packets without further security validation. As a necessary exception, an implementation MUST advertise the remote nonce value as undefined_nonce when the FSM is not in TwoWay or ThreeWay state and accept an undefined_nonce for its local nonce value on packets in any other state than ThreeWay. As an optional optimization, an implementation MAY send one LIE with a previously negotiated neighbor's nonce to try to speed up a neighbor's transition from ThreeWay to OneWay and MUST revert to sending undefined_nonce after that.
Lifetime Reflooding the same TIE version quickly with small variations in its lifetime may lead to an excessive number of security fingerprint computations. To avoid this, the application generating the fingerprints for flooded TIEs MAY round the value down to the next rounddown_lifetime_interval on the packet header to reuse previous computation results. TIEs flooded with such rounded lifetimes will only limit the amount of computations necessary during transitions that lead to advertisement of the same TIEs with the same information within a short period of time.
Security Association Changes No mechanism is specified to convert a security envelope for the same key ID from one algorithm to another once the envelope is operational. The recommended procedure to change to a new algorithm is to take the adjacency down, make the necessary changes to the secret and algorithm used by the according key ID, and bring the adjacency back up. Obviously, an implementation MAY choose to stop verifying the security envelope for the duration of the algorithm change to keep the adjacency up, but since this introduces a security vulnerability window, such rollover SHOULD NOT be recommended. Other approaches, such as accepting multiple algorithms for same key ID for a configured time window, are possible but in the realm of implementation choices rather than protocol specification.
Information Elements Schema This section introduces the schema for information elements. The Interface Description Language (IDL) is Thrift. On schema changes that
  1. change field numbers or
  2. add new required fields or
  3. remove any fields or
  4. change lists into sets, unions into structures or
  5. change multiplicity of fields or
  6. changes type or name of any field or
  7. change data types of the type of any field or
  8. adds, changes or removes a default value of any existing field or
  9. removes or changes any defined constant or constant value or
  10. changes any enumeration type except extending `common.TIETypeType` (use of enumeration types is generally discouraged) or
  11. adds new TIE type to TIETypeType with flooding scope different from prefix TIE flooding scope
the major version of the schema MUST increase. All other changes MUST increase the minor version within the same major. Introducing an optional field does not cause a major version increase even if the fields inside the structure are optional with defaults. All signed integers, as forced by Thrift support, must be cast for internal purposes to equivalent unsigned values without discarding the signedness bit. An implementation SHOULD try to avoid using the signedness bit when generating values. The schema is normative.
Backwards-Compatible Extension of Schema The set of rules in guarantees that every decoder can process serialized content generated by a higher minor version of the schema, and with that, the protocol can progress without a 'flag-day'. Contrary to that, content serialized using a major version X is not expected to be decodable by any implementation using a decoder for a model with a major version lower than X. Schema negotiation and translation within RIFT is outside the scope of this document. Additionally, based on the propagated minor version in encoded content and added optional node capabilities, new TIE types or even de facto mandatory fields can be introduced without progressing the major version, albeit only nodes supporting such new extensions would decode them. Given the model is encoded at the source and never re-encoded, flooding through nodes not understanding any new extensions will preserve the corresponding fields. However, it is important to understand that a higher minor version of a schema does not guarantee that capabilities introduced in lower minors of the same major are supported. The node_capabilities field is used to indicate which capabilities are supported. Specifically, the schema SHOULD add elements to the NodeCapabilities field's future capabilities to indicate whether it will support interpretation of schema extensions on the same major revision if they are present. Such fields MUST be optional and have an implicit or explicit false default value. If a future capability changes route selection or generates conditions that cause packet loss if some nodes are not supporting it, then a major version increment will be unavoidable. NodeCapabilities shown in LIE MUST match the capabilities shown in the Node TIEs; otherwise, the behavior is unspecified. A node detecting the mismatch SHOULD generate a notification. Alternately or additionally, new optional fields can be introduced into, e.g., NodeTIEElement, if a special field is chosen to indicate via its presence that an optional feature is enabled (since capability to support a feature does not necessarily mean that the feature is actually configured and operational). To support new TIE types without increasing the major version enumeration, TIEElement can be extended with new optional elements for new 'common.TIETypeType' values as long the scope of the new TIE matches the prefix TIE scope. In case it is necessary to understand whether all nodes can parse the new TIE type, a node capability MUST be added in NodeCapabilities to prevent a non-homogenous network.
common.thrift This schema references , , and . empty_set_of_nodeids = {} /** default lifetime of TIE is one week */ const LifeTimeInSecType default_lifetime = 604800 /** default lifetime when TIEs are purged is 5 minutes */ const LifeTimeInSecType purge_lifetime = 300 /** optional round down interval when * TIEs are sent with security signatures * to prevent excessive computation. */ const LifeTimeInSecType rounddown_lifetime_interval = 60 /** any 'TieHeader' that has a smaller lifetime difference than this constant is equal (if other fields equal). */ const LifeTimeInSecType lifetime_diff2ignore = 400 /** default UDP port to run LIEs on */ const UDPPortType default_lie_udp_port = 914 /** default UDP port to receive TIEs on, which can be peer specific */ const UDPPortType default_tie_udp_flood_port = 915 /** default MTU link size to use */ const MTUSizeType default_mtu_size = 1400 /** default link being BFD capable */ const bool bfd_default = true /** type used to target nodes with key value */ typedef i64 KeyValueTargetType /** default target for key value are all nodes. */ const KeyValueTargetType keyvaluetarget_default = 0 /** value for _all leaves_ addressing. Represented by all bits set. */ const KeyValueTargetType keyvaluetarget_all_south_leaves = -1 /** undefined nonce, equivalent to missing nonce */ const NonceType undefined_nonce = 0; /** outer security key ID, MUST be interpreted as in implementation as unsigned */ typedef i8 OuterSecurityKeyID /** security key ID, MUST be interpreted as in implementation as unsigned */ typedef i32 TIESecurityKeyID /** undefined key */ const TIESecurityKeyID undefined_securitykey_id = 0; /** Maximum delta (negative or positive) that a mirrored nonce can deviate from local value to be considered valid. */ const i16 maximum_valid_nonce_delta = 5; const TimeIntervalInSecType nonce_regeneration_interval = 300; /** Direction of TIEs. */ enum TieDirectionType { Illegal = 0, South = 1, North = 2, DirectionMaxValue = 3, } /** Address family type. */ enum AddressFamilyType { Illegal = 0, AddressFamilyMinValue = 1, IPv4 = 2, IPv6 = 3, AddressFamilyMaxValue = 4, } /** IPv4 prefix type. */ struct IPv4PrefixType { 1: required IPv4Address address; 2: required PrefixLenType prefixlen; } /** IPv6 prefix type. */ struct IPv6PrefixType { 1: required IPv6Address address; 2: required PrefixLenType prefixlen; } /** IP address type. */ union IPAddressType { /** Content is IPv4 */ 1: optional IPv4Address ipv4address; /** Content is IPv6 */ 2: optional IPv6Address ipv6address; } /** Prefix advertisement. @note: For interface addresses, the protocol can propagate the address part beyond the subnet mask and on reachability computation that has to be normalized. The non-significant bits can be used for operational purposes. */ union IPPrefixType { 1: optional IPv4PrefixType ipv4prefix; 2: optional IPv6PrefixType ipv6prefix; } /** Sequence of a prefix in case of move. */ struct PrefixSequenceType { 1: required IEEE802_1ASTimeStampType timestamp; /** Transaction ID set by the client in, e.g., 6LoWPAN. */ 2: optional PrefixTransactionIDType transactionid; } /** Type of TIE. */ enum TIETypeType { Illegal = 0, TIETypeMinValue = 1, /** first legal value */ NodeTIEType = 2, PrefixTIEType = 3, PositiveDisaggregationPrefixTIEType = 4, NegativeDisaggregationPrefixTIEType = 5, PGPrefixTIEType = 6, KeyValueTIEType = 7, ExternalPrefixTIEType = 8, PositiveExternalDisaggregationPrefixTIEType = 9, TIETypeMaxValue = 10, } /** RIFT route types. @note: The only purpose of those values is to introduce an ordering, whereas an implementation can internally choose any other values as long the ordering is preserved. */ enum RouteType { Illegal = 0, RouteTypeMinValue = 1, /** First legal value. */ /** Discard routes are most preferred */ Discard = 2, /** Local prefixes are directly attached prefixes on the * system, such as interface routes. */ LocalPrefix = 3, /** Advertised in S-TIEs */ SouthPGPPrefix = 4, /** Advertised in N-TIEs */ NorthPGPPrefix = 5, /** Advertised in N-TIEs */ NorthPrefix = 6, /** Externally imported north */ NorthExternalPrefix = 7, /** Advertised in S-TIEs, either normal prefix or positive disaggregation */ SouthPrefix = 8, /** Externally imported south */ SouthExternalPrefix = 9, /** Negative, transitive prefixes are least preferred */ NegativeSouthPrefix = 10, RouteTypeMaxValue = 11, } enum KVTypes { Experimental = 1, WellKnown = 2, OUI = 3, }]]>
encoding.thrift address_families; } /** Unique ID of a TIE. */ struct TIEID { /** direction of TIE */ 1: required common.TieDirectionType direction; /** indicates originator of the TIE */ 2: required common.SystemIDType originator; /** type of the tie */ 3: required common.TIETypeType tietype; /** number of the tie */ 4: required common.TIENrType tie_nr; } /** Header of a TIE. */ struct TIEHeader { /** ID of the tie. */ 2: required TIEID tieid; /** Sequence number of the tie. */ 3: required common.SeqNrType seq_nr; /** Absolute timestamp when the TIE was generated. */ 10: optional common.IEEE802_1ASTimeStampType origination_time; /** Original lifetime when the TIE was generated. */ 12: optional common.LifeTimeInSecType origination_lifetime; } /** Header of a TIE as described in TIRE/TIDE. */ struct TIEHeaderWithLifeTime { 1: required TIEHeader header; /** Remaining lifetime. */ 2: required common.LifeTimeInSecType remaining_lifetime; } /** TIDE with *sorted* TIE headers. */ struct TIDEPacket { /** First TIE header in the TIDE packet. */ 1: required TIEID start_range; /** Last TIE header in the TIDE packet. */ 2: required TIEID end_range; /** _Sorted_ list of headers. */ 3: required list headers; } /** TIRE packet */ struct TIREPacket { 1: required set headers; } /** neighbor of a node */ struct NodeNeighborsTIEElement { /** level of neighbor */ 1: required common.LevelType level; /** Cost to neighbor. Ignore anything equal/larger than 'infinite_distance' or equal 'invalid_distance' */ 3: optional common.MetricType cost = common.default_distance; /** can carry description of multiple parallel links in a TIE */ 4: optional set link_ids; /** total bandwidth to neighbor as sum of all parallel links */ 5: optional common.BandwidthInMegaBitsType bandwidth = common.default_bandwidth; } /** Indication flags of the node. */ struct NodeFlags { /** Indicates that node is in overload, do not transit traffic through it. */ 1: optional bool overload = common.overload_default; } /** Description of a node. */ struct NodeTIEElement { /** Level of the node. */ 1: required common.LevelType level; /** Node's neighbors. Multiple node TIEs can carry disjoint sets of neighbors. */ 2: required map neighbors; /** Capabilities of the node. */ 3: required NodeCapabilities capabilities; /** Flags of the node. */ 4: optional NodeFlags flags; /** Optional node name for easier operations. */ 5: optional string name; /** PoD to which the node belongs. */ 6: optional common.PodType pod; /** Optional startup time of the node */ 7: optional common.TimestampInSecsType startup_time; /** If any local links are miscabled, this indication is flooded. */ 10: optional set miscabled_links; /** ToFs in the same plane. Only carried by ToF. Multiple Node TIEs can carry disjoint sets of ToFs that MUST be joined to form a single set. */ 12: optional set same_plane_tofs; /** It provides the optional ID of the fabric configured */ 20: optional common.FabricIDType fabric_id = common.default_fabric_id; } /** Attributes of a prefix. */ struct PrefixAttributes { /** Distance of the prefix. */ 2: required common.MetricType metric = common.default_distance; /** Generic unordered set of route tags, can be redistributed to other protocols or used within the context of real time analytics. */ 3: optional set tags; /** Monotonic clock for mobile addresses. */ 4: optional common.PrefixSequenceType monotonic_clock; /** Indicates if the prefix is a node loopback. */ 6: optional bool loopback = false; /** Indicates that the prefix is directly attached. */ 7: optional bool directly_attached = true; /** Link to which the address belongs to. */ 10: optional common.LinkIDType from_link; /** Optional, per-prefix significant label. */ 12: optional common.LabelType label; } /** TIE carrying prefixes */ struct PrefixTIEElement { /** Prefixes with the associated attributes. */ 1: required map prefixes; } /** Defines the targeted nodes and the value carried. */ struct KeyValueTIEElementContent { 1: optional common.KeyValueTargetType targets = common.keyvaluetarget_default; 2: optional binary value; } /** Generic key value pairs. */ struct KeyValueTIEElement { 1: required map keyvalues; } /** Single element in a TIE. */ union TIEElement { /** Used in case of enum common.TIETypeType.NodeTIEType. */ 1: optional NodeTIEElement node; /** Used in case of enum common.TIETypeType.PrefixTIEType. */ 2: optional PrefixTIEElement prefixes; /** Positive prefixes (always southbound). */ 3: optional PrefixTIEElement positive_disaggregation_prefixes; /** Transitive, negative prefixes (always southbound) */ 5: optional PrefixTIEElement negative_disaggregation_prefixes; /** Externally reimported prefixes. */ 6: optional PrefixTIEElement external_prefixes; /** Positive external disaggregated prefixes (always southbound). */ 7: optional PrefixTIEElement positive_external_disaggregation_prefixes; /** Key-Value store elements. */ 9: optional KeyValueTIEElement keyvalues; } /** TIE packet */ struct TIEPacket { 1: required TIEHeader header; 2: required TIEElement element; } /** Content of a RIFT packet. */ union PacketContent { 1: optional LIEPacket lie; 2: optional TIDEPacket tide; 3: optional TIREPacket tire; 4: optional TIEPacket tie; } /** RIFT packet structure. */ struct ProtocolPacket { 1: required PacketHeader header; 2: required PacketContent content; }]]>
Further Details on Implementation
Considerations for Leaf-Only Implementation RIFT can and is intended to be stretched to the lowest level in the IP fabric to integrate ToRs or even servers. Since those entities would run as leaves only, it is worth it to observe that a leaf-only version is significantly simpler to implement and requires much less resources:
  1. Leaf nodes only need to maintain a multipath default route under normal circumstances. However, in cases of catastrophic partitioning, leaf nodes SHOULD be capable of accommodating all the leaf routes in their own PoD to prevent traffic loss.
  2. Leaf nodes only hold their own North TIEs and the South TIEs of level 1 nodes they are connected to.
  3. Leaf nodes do not have to support any type of disaggregation computation or propagation.
  4. Leaf nodes are not required to support the overload flag.
  5. Leaf nodes do not need to originate S-TIEs unless optional L2L features are desired.
Considerations for Spine Implementation Nodes that do not act as ToF are not required to discover fallen leaves by comparing reachable destinations with peers and therefore do not need to run the computation of disaggregated routes based on that discovery. On the other hand, non-ToF nodes need to respect disaggregated routes advertised from the north. In the case of negative disaggregation, spines nodes need to generate southbound disaggregated routes when all parents are lost for a fallen leaf.
Security Considerations
General One can consider attack vectors where a router may reboot many times while changing its System ID and pollute the network with many stale TIEs or TIEs that are sent with very long lifetimes and not cleaned up when the routes vanish. Those attack vectors are not unique to RIFT. Given large memory footprints available today, those attacks should be relatively benign. Otherwise, a node SHOULD implement a strategy of discarding contents of all TIEs that were not present in the SPF tree over a certain, configurable period of time. Since the protocol is self-stabilizing and will advertise the presence of such TIEs to its neighbors, they can be re-requested again if a computation finds that it has an adjacency formed towards the System ID of the discarded TIEs. The inner protection configured based on any of the mechanisms in guarantees the integrity of TIE content, and when combined with the outer part of the envelope, using any of the mechanisms in , guarantees protection against replay attacks as well. If only outer protection (i.e., an outer key ID different from 'undefined_securitykey_id') is applied to an adjacency by the means of any mechanism in , the integrity of the packet and replay protection is guaranteed only over the adjacency involved in any of the configured directions. Further considerations can be found in Sections and .
Time to Live and Hop Limit Values RIFT explicitly requires the use of a TTL/HL value of 1 or 255 when sending/receiving LIEs and TIEs so that implementors have a choice between the two. Using a TTL/HL value of 255 does come with security concerns, but those risks are addressed in . However, this approach may still have difficulties with some forwarding implementations (e.g., incorrectly processing TTL/HL, loops within the forwarding plane itself, etc.). It is for this reason that RIFT also allows implementations to use a TTL/HL of 1. Attacks that exploit this by spoofing it from several hops away are indeed possible but are exceptionally difficult to engineer. Replay attacks are another potential attack vector, but as described in the subsequent security sections, RIFT is well protected against such attacks if any of the mechanisms in are applied. Additionally, for link-local scoped multicast addresses used for LIE, the value of 1 presents a more consistent choice.
Malformed Packets The protocol protects packets extensively through optional signatures and nonces, so if the possibility of maliciously injected malformed or replayed packets exist in a deployment, algorithms in must be applied. Even with the security envelope, since RIFT relies on Thrift encoders and decoders generated automatically from IDL, it is conceivable that errors in such encoders/decoders could be discovered and lead to delivery of corrupted packets or reception of packets that cannot be decoded. Misformatted packets normally lead to the decoder returning an error condition to the caller, and with that, the packet is basically unparsable with no other choice but to discard it. Should the unlikely scenario occur of the decoder being forced to abort the protocol, this is neither better nor worse than today's behavior of other protocols.
RIFT ZTP presents many attack vectors in untrusted environments, starting with nodes that oscillate their level offers to the possibility of nodes offering a ThreeWay adjacency with the highest possible level value and a very long holdtime trying to put itself "on top of the lattice", thereby allowing it to gain access to the whole southbound topology. Session authentication mechanisms are necessary in environments where this is possible, and RIFT provides the security envelope to ensure this, if so desired, if any mechanism in is deployed.
Lifetime RIFT removes lifetime modification and replay attack vectors by protecting the lifetime behind a signature computed over it and additional nonce combination, which results in the inability of an attacker to artificially shorten the remaining_lifetime. This only applies if any mechanism in is used.
Packet Number A packet number is an optional defined value number that is carried in the security envelope without any fingerprint protection and is hence vulnerable to replay and modification attacks. Contrary to nonces, this number must change on every packet and would present a very high cryptographic load if signed. The attack vector packet number present is relatively benign. Changing the packet number by a man-in-the-middle attack will only affect operational validation tools and possibly some performance optimizations on flooding. It is expected that an implementation detecting too many "fake losses" or "misorderings" due to the attack on the packet number would simply suppress its further processing.
Outer Fingerprint Attacks Even when a mechanism in is enabled to generate outer fingerprints, further attack considerations apply. A node can try to inject LIE packets observing a conversation on the wire by using the observed outer key ID, albeit it cannot generate valid signatures in case it changes the integrity of the message, so the only possible attack is DoS due to excessive LIE validation if any mechanism in is used. A node can try to replay previous LIEs with a changed state that it recorded, but the attack is hard to replicate since the nonce combination must match the ongoing exchange and is then limited to only a single flap since both nodes will advance their nonces in case the adjacency state changed. Even in the most unlikely case, the attack length is limited due to both sides periodically increasing their nonces. Generally, since weak nonces are not changed on every packet for performance reasons, a conceivable attack vector by a man in the middle is to flood a receiving node with the maximum bandwidth of recently observed packets, both LIEs as well as TIEs. In a scenario where such attacks are likely, maximum_valid_nonce_delta and nonce_regeneration_interval can be implemented as configurable and set to small values. This will likely present a significant computational load on large fabrics under normal operation.
TIE Origin Fingerprint DoS Attacks Even when a mechanism in is enabled to generate inner fingerprints or signatures, further attack considerations apply. In case the inner fingerprint could be generated by a compromised node in the network other than the originator based on shared secrets, the deployment must fall back on use of signatures that can be validated but not generated by any other node except the originator. A compromised node in the network can attempt to brute force "fake TIEs" using other nodes' TIE origin key ID without possessing the necessary secrets. Albeit the ultimate validation of the origin signature will fail in such scenarios and not progress further than immediately peering nodes, the resulting DoS attack seems unavoidable since the TIE origin key ID is only protected by the (here assumed to be compromised) node.
Host Implementations It can be reasonably expected that the proliferation of RotH servers, rather than dedicated networking devices, will represent a significant amount of RIFT devices. Given their normally far wider software envelope and access granted to them, such servers are also far more likely to be compromised and present an attack vector on the protocol. Hijacking of prefixes to attract traffic is a trust problem and cannot be easily addressed within the protocol if the trust model is breached, i.e., the server presents valid credentials to form an adjacency and issue TIEs. In an even more devious way, the servers can present DoS (or even DDoS) vectors from issuing too many LIE packets, flooding large amounts of North TIEs, and attempting similar resource overrun attacks. A prudent implementation forming adjacencies to leaves should implement threshold mechanisms and raise warnings when, e.g., a leaf is advertising an excess number of TIEs or prefixes. Additionally, such implementation could refuse any topology information except the node's own TIEs and authenticated, reflected South Node TIEs at their own level. To isolate possible attack vectors on the leaf to the largest possible extent, a dedicated leaf-only implementation could run without any configuration by:
  • hard-coding a well-known adjacency key (which can be always rolled over by means of, e.g., a well-known key-value distributed from top of the fabric),
  • hard-coding a leaf level value, and
  • always setting the overload flag.
IPv4 Broadcast and IPv6 All-Routers Multicast Implementations describes an optional implementation that supports LIE exchange over IPv4 broadcast addresses and/or the IPv6 all-routers multicast address. It is important to consider that if an implementation supports this, the attack surface widens as LIEs may be propagated to devices outside of the intended RIFT topology. This may leave RIFT nodes more susceptible to the various attack vectors already described in this section.
IANA Considerations As detailed below, multicast addresses and standard port numbers have been assigned. Additionally, registries for the schema have been created with initial values assigned.
Multicast and Port Numbers In the "IPv4 Multicast Address Space" registry, the value of 224.0.0.121 has been assigned for 'ALL_V4_RIFT_ROUTERS'. In the "IPv6 Multicast Address Space" registry, the value of ff02::a1f7 has been assigned for 'ALL_V6_RIFT_ROUTERS'. The following assignments have been made in the "Service Name and Transport Protocol Port Number Registry": RIFT LIE Port
Service Name:
rift-lies
Port Number:
914
Transport Protocol:
udp
Description:
Routing in Fat Trees Link Information Element
Assignee:
IESG (iesg@ietf.org)
Contact:
IETF Chair (chair@ietf.org)
Reference:
RFC 9692
RIFT TIE Port
Service Name:
rift-ties
Port Number:
915
Transport Protocol:
udp
Description:
Routing in Fat Trees Topology Information Element
Assignee:
IESG (iesg@ietf.org)
Contact:
IETF Chair (chair@ietf.org)
Reference:
RFC 9692
Registry for RIFT Security Algorithms A new registry has been created to hold the allowed RIFT security algorithms. No particular enumeration values are necessary since RIFT uses a key ID abstraction on packets without disclosing any information about the algorithm or secrets used and only carries the resulting fingerprint or signature protecting the integrity of the data. The registry applies the "Specification Required" policy per . The designated expert should ensure that the algorithms suggested represent the state of the art at a given point in time and avoid introducing algorithms that do not represent enhanced security properties or ensure such properties at a lower cost as compared to existing registry entries. RIFT Security Algorithms
Name Recommendation Reference
HMAC-SHA256 Simplest way to ensure integrity of transmissions across adjacencies when used as outer keys and integrity of TIEs when used as inner keys. Recommended for most interoperable security protection. and
HMAC-SHA512 Same as HMAC-SHA256 with stronger protection. and
SHA256-RSASSA-PKCS1-v1_5 Recommended for high security applications where private keys are protected by according nodes. Recommended as well in case not only integrity but origin validation is necessary for TIEs. Recommended when adjacencies must be protected without disclosing the secrets on both sides of the adjacency.
SHA512-RSASSA-PKCS1-v1_5 Same as SHA256-RSASSA-PKCS1-v1_5 with stronger protection.
Registries with Assigned Values for Schema Values This section requests registries that help govern the schema via the usual IANA registry procedures. The registry group "Routing in Fat Trees (RIFT)" holds the following registries. Registry values are stored with their minimum and maximum version in which they are available. All values not provided are to be considered "Unassigned". The range of every registry is a 16-bit integer. Allocation of new values is performed via "Expert Review" action only in the case of minor changes per the rules in . All other allocations are performed via "Specification Required". In some cases, the registries do not contain necessary information such as whether the fields are optional or required, what units are used, or what datatype is involved. This information is encoded in the normative schema itself by the means of IDL syntax or necessary type definitions and their names.
RIFTVersions Registry This registry stores all RIFT protocol schema major and minor versions, including the reference to the document introducing the version. This also means that, if multiple documents extend rift schema, they have to serialize using this registry to increase the minor or major versions sequentially.
Schema Version Reference
8.0 RFC 9692,
RIFTCommonAddressFamilyType Registry This registry has the following initial values. In addition to the columns shown below, the IANA registry also includes Comment and Reference columns. Address Family Type
Value Name Min. Schema Version Max. Schema Version
0 Illegal 8.0
1 AddressFamilyMinValue 8.0
2 IPv4 8.0
3 IPv6 8.0
4 AddressFamilyMaxValue 8.0
RIFTCommonHierarchyIndications Registry This registry has the following initial values. In addition to the columns below, the IANA registry also includes Comment and Reference columns. Flags Indicating Node Configuration in Case of ZTP
Value Name Min. Schema Version Max. Schema Version
0 leaf_only 8.0
1 leaf_only_and_leaf_2_leaf_procedures 8.0
2 top_of_fabric 8.0
RIFTCommonIEEE8021ASTimeStampType Registry This registry has the following initial values. In addition to the columns below, the IANA registry also includes Comment and Reference columns. The timestamp is per IEEE 802.1AS; all values MUST be interpreted in implementation as unsigned.
Value Name Min. Schema Version Max. Schema Version
0 Reserved 8.0 All Versions
1 AS_sec 8.0
2 AS_nsec 8.0
RIFTCommonIPAddressType Registry This registry has the following initial values. IP Address Type
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 ipv4address 8.0 Content is IPv4
2 ipv6address 8.0 Content is IPv6
RIFTCommonIPPrefixType Registry This registry has the following initial values. Prefix Advertisement
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 ipv4prefix 8.0
2 ipv6prefix 8.0
RIFTCommonIPv4PrefixType Registry This registry has the following initial values. IPv4 Prefix Type
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 address 8.0
2 prefixlen 8.0
RIFTCommonIPv6PrefixType Registry This registry has the following initial values. IPv6 Prefix Type
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 address 8.0
2 prefixlen 8.0
RIFTCommonKVTypes Registry This registry has the following initial values.
Value Name Min. Schema Version Max. Schema Version Comment
0 Unassigned
1 Experimental 8.0
2 WellKnown 8.0
3 OUI 8.0
RIFTCommonPrefixSequenceType Registry This registry has the following initial values. Sequence of a Prefix in Case of Move
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 timestamp 8.0
2 transactionid 8.0 Transaction ID set by client in, e.g., 6LoWPAN.
RIFTCommonRouteType Registry This registry has the following initial values. RIFT Route Types
Value Name Min. Schema Version Max. Schema Version Comment
0 Illegal 8.0
1 RouteTypeMinValue 8.0
2 Discard 8.0
3 LocalPrefix 8.0
4 SouthPGPPrefix 8.0
5 NorthPGPPrefix 8.0
6 NorthPrefix 8.0
7 NorthExternalPrefix 8.0
8 SouthPrefix 8.0
9 SouthExternalPrefix 8.0
10 NegativeSouthPrefix 8.0
11 RouteTypeMaxValue 8.0
RIFTCommonTIETypeType Registry This registry has the following initial values. In addition to the columns below, the IANA registry also includes Comment and Reference columns. Type of TIE
Value Name Min. Schema Version Max. Schema Version
0 Illegal 8.0
1 TIETypeMinValue 8.0
2 NodeTIEType 8.0
3 PrefixTIEType 8.0
4 PositiveDisaggregationPrefixTIEType 8.0
5 NegativeDisaggregationPrefixTIEType 8.0
6 PGPrefixTIEType 8.0
7 KeyValueTIEType 8.0
8 ExternalPrefixTIEType 8.0
9 PositiveExternalDisaggregationPrefixTIEType 8.0
10 TIETypeMaxValue 8.0
RIFTCommonTieDirectionType Registry This registry has the following initial values. Direction of TIEs
Value Name Min. Schema Version Max. Schema Version Comment
0 Illegal 8.0
1 South 8.0
2 North 8.0
3 DirectionMaxValue 8.0
RIFTEncodingCommunity Registry This registry has the following initial values. Prefix Community
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 top 8.0 Higher order bits
2 bottom 8.0 Lower order bits
RIFTEncodingKeyValueTIEElement Registry This registry has the following initial values. Generic Key Value Pairs
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 keyvalues 8.0
RIFTEncodingKeyValueTIEElementContent Registry This registry has the following initial values. It defines the targeted nodes and the value carried.
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 targets 8.0
2 value 8.0
RIFTEncodingLIEPacket Registry This registry has the following initial values. RIFT LIE Packet
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 name 8.0 Node or adjacency name.
2 local_id 8.0 Local link ID.
3 flood_port 8.0 UDP port to which we can receive flooded TIEs.
4 link_mtu_size 8.0 Layer 2 MTU, used to discover mismatch.
5 link_bandwidth 8.0 Local link bandwidth on the interface.
6 neighbor 8.0 Reflects the neighbor once received to provide 3-way connectivity.
7 pod 8.0 Node's PoD.
10 node_capabilities 8.0 Node capabilities supported.
11 link_capabilities 8.0 Capabilities of this link.
12 holdtime 8.0 Required holdtime of the adjacency, i.e., for how long a period adjacency should be kept up without valid LIE reception.
13 label 8.0 Optional, unsolicited, downstream assigned locally significant label value for the adjacency.
21 not_a_ztp_offer 8.0 Indicates that the level on the LIE must not be used to derive a ZTP level by the receiving node.
22 you_are_flood_repeater 8.0 Indicates to the northbound neighbor that it should be reflooding TIEs received from this node to achieve flood reduction and balancing for northbound flooding.
23 you_are_sending_too_quickly 8.0 Indicates to the neighbor to flood node TIEs only and slow down all other TIEs. Ignored when received from the southbound neighbor.
24 instance_name 8.0 Instance name in case multiple RIFT instances are running on the same interface.
35 fabric_id 8.0 It provides the optional ID of the fabric configured. This must match the information advertised on the node element.
RIFTEncodingLinkCapabilities Registry This registry has the following initial values. Link Capabilities
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 bfd 8.0 Indicates that the link is supporting BFD.
2 ipv4_forwarding_capable 8.0 Indicates whether the interface will support IPv4 forwarding.
RIFTEncodingLinkIDPair Registry The LinkID pair describes one of the parallel links between two nodes. This registry has the following initial values.
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 local_id 8.0 Node-wide unique value for the local link.
2 remote_id 8.0 Received the remote link ID for this link.
10 platform_interface_index 8.0 Describes the local interface index of the link.
11 platform_interface_name 8.0 Describes the local interface name.
12 trusted_outer_security_key 8.0 Indicates whether the link is secured, i.e., protected by outer key, absence of this element means no indication, undefined outer key means not secured.
13 bfd_up 8.0 Indicates whether the link is protected by an established BFD session.
14 address_families 8.0 Optional indication that address families are up on the interface.
RIFTEncodingNeighbor Registry This registry has the following initial values. Neighbor Structure
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 originator 8.0 System ID of the originator.
2 remote_id 8.0 ID of remote side of the link.
RIFTEncodingNodeCapabilities Registry This registry has the following initial values. Capabilities the Node Supports
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 protocol_minor_version 8.0 Must advertise supported minor version dialect that way.
2 flood_reduction 8.0 Indicates that node supports flood reduction.
3 hierarchy_indications 8.0 Indicates place in hierarchy, i.e., top of fabric or leaf only (in ZTP) or support for L2L procedures.
RIFTEncodingNodeFlags Registry This registry has the following initial values. Indication Flags of the Node
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 overload 8.0 Indicates that node is in overload; do not transit traffic through it.
RIFTEncodingNodeNeighborsTIEElement Registry This registry has the following initial values. Neighbor of a Node
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 level 8.0 Level of neighbor.
3 cost 8.0 Cost to neighbor. Ignore anything equal or larger than 'infinite_distance' and equal to 'invalid_distance'.
4 link_ids 8.0 Carries description of multiple parallel links in a TIE.
5 bandwidth 8.0 Total bandwidth to neighbor as sum of all parallel links.
RIFTEncodingNodeTIEElement Registry This registry has the following initial values. Description of a Node
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 level 8.0 Level of the node.
2 neighbors 8.0 Node's neighbors. Multiple node TIEs can carry disjoint sets of neighbors.
3 capabilities 8.0 Capabilities of the node.
4 flags 8.0 Flags of the node.
5 name 8.0 Optional node name for easier operations.
6 pod 8.0 Pod to which the node belongs.
7 startup_time 8.0 Optional startup time of the node.
10 miscabled_links 8.0 If any local links are miscabled, this indication is flooded.
12 same_plane_tofs 8.0 ToFs in the same plane. Only carried by ToF. Multiple node TIEs can carry disjoint sets of ToFs that must be joined to form a single set.
20 fabric_id 8.0 It provides the optional ID of the fabric configured.
RIFTEncodingPacketContent Registry This registry has the following initial values. Content of a RIFT Packet
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 lie 8.0
2 tide 8.0
3 tire 8.0
4 tie 8.0
RIFTEncodingPacketHeader Registry This registry has the following initial values. Common RIFT Packet Header
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 major_version 8.0 Major version of protocol.
2 minor_version 8.0 Minor version of protocol.
3 sender 8.0 Node sending the packet; in case of LIE/TIRE/TIDE also the originator of it.
4 level 8.0 Level of the node sending the packet, required on everything except LIEs. Lack of presence on LIEs indicates undefined_level and is used in ZTP procedures.
RIFTEncodingPrefixAttributes Registry This registry has the following initial values. Attributes of a Prefix
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
2 metric 8.0 Distance of the prefix.
3 tags 8.0 Generic unordered set of route tags, can be redistributed to other protocols or used within the context of real time analytics.
4 monotonic_clock 8.0 Monotonic clock for mobile addresses.
6 loopback 8.0 Indicates if the prefix is a node loopback.
7 directly_attached 8.0 Indicates that the prefix is directly attached.
10 from_link 8.0 Link to which the address belongs to.
12 label 8.0 Optional, per-prefix significant label.
RIFTEncodingPrefixTIEElement Registry This registry has the following initial values. TIE Carrying Prefixes
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 prefixes 8.0 Prefixes with the associated attributes.
RIFTEncodingProtocolPacket Registry This registry has the following initial values. RIFT Packet Structure
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 header 8.0
2 content 8.0
RIFTEncodingTIDEPacket Registry This registry has the following initial values. TIDE with Sorted TIE Headers
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 start_range 8.0 First TIE header in the TIDE packet.
2 end_range 8.0 Last TIE header in the TIDE packet.
3 headers 8.0 _sorted_ list of headers.
RIFTEncodingTIEElement Registry This registry has the following initial values. Single Element in a TIE
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 node 8.0 Used in case of
enum common.tietypetype.
nodetietype.
2 prefixes 8.0 Used in case of
enum common.tietypetype.
prefixtietype.
3 positive_disaggregation_
prefixes		 8.0 Positive prefixes
(always southbound).
5 negative_disaggregation_
prefixes		 8.0 Transitive, negative prefixes
(always southbound)
6 external_prefixes 8.0 Externally reimported
prefixes.
7 positive_external_
disaggregation_prefixes		 8.0 Positive external
disaggregated prefixes
(always southbound).
9 keyvalues 8.0 Key-value
store elements.
RIFTEncodingTIEHeader Registry This registry has the following initial values. Header of a TIE
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
2 tieid 8.0 ID of TIE.
3 seq_nr 8.0 Sequence number of TIE.
10 origination_time 8.0 Absolute timestamp when TIE was generated.
12 origination_lifetime 8.0 Original lifetime when TIE was generated.
RIFTEncodingTIEHeaderWithLifeTime Registry This registry has the following initial values. Header of a TIE as Described in TIRE/TIDE
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 header 8.0
2 remaining_lifetime 8.0 Remaining lifetime.
RIFTEncodingTIEID Registry This registry has the following initial values. Unique ID of a TIE
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 direction 8.0 Direction of TIE.
2 originator 8.0 Indicates originator of TIE.
3 tietype 8.0 Type of TIE.
4 tie_nr 8.0 Number of TIE.
RIFTEncodingTIEPacket Registry This registry has the following initial values. TIE Packet
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 header 8.0
2 element 8.0
RIFTEncodingTIREPacket Registry This registry has the following initial values. TIRE Packet
Value Name Min. Schema Version Max. Schema Version Comment
0 Reserved 8.0 All Versions
1 headers 8.0
References Normative References Apache Thrift Documentation Apache Software Foundation Secure Hash Standard (SHS) NIST Guidelines for Use of Extended Unique Identifier (EUI), Organizationally Unique Identifier (OUI), and Company ID (CID) IEEE Informative References Routing in Fat Trees (RIFT) Applicability and Operational Considerations IEEE Standard for a Precision Clock Synchronization Protocol for Networked Measurement and Control Systems IEEE IEEE Standard for Local and Metropolitan Area Networks - Timing and Synchronization for Time-Sensitive Applications in Bridged Local Area Networks IEEE On Nonblocking Folded-Clos Networks in Computer Communication Environments 2011 IEEE International Parallel & Distributed Processing Symposium A Note on Two Problems in Connexion with Graphs Numerische Mathematik, vol. 1, pp. 269-271 Dynamo: amazon's highly available key-value store ACM SIGOPS Operating Systems Review, vol. 41, no. 6, pp. 205-220 Finding the k Shortest Paths USC Fat-Trees: Universal Networks for Hardware-Efficient Supercomputing IEEE Transactions on Computers, vol. C-34, no. 10, pp. 892-901 A Scalable, Commodity Data Center Network Architecture USC USC USC ACM SIGCOMM Computer Communication Review, vol. 38, no. 4, pp. 63-74 Valley-free violation in Internet routing - Analysis based on BGP Community data University College (London, UK) University College (London, UK) 2012 IEEE International Conference on Communications (ICC) Day One: Routing in Fat Trees (RIFT) Juniper Networks Juniper Networks Juniper Networks Juniper Networks Juniper Networks Juniper Networks Juniper Networks Juniper Networks Juniper Network Books
Sequence Number Binary Arithmetic This section defines a variant of sequence number arithmetic related to explained over two complement arithmetic, which is easy to implement. Assuming straight two complement's subtractions on the bit width of the sequence numbers, the corresponding >: and =: relations are defined as:
  • U_1, U_2 are 12-bits aligned unsigned version number
  • D_f is ( U_1 - U_2 ) interpreted as two complement signed 12-bits
  • D_b is ( U_2 - U_1 ) interpreted as two complement signed 12-bits
  • U_1 >: U_2 IIF D_f > 0 and D_b < 0
  • U_1 =: U_2 IIF D_f = 0
The >: relationship is anti-symmetric but not transitive. Observe that this leaves >: of the numbers having maximum two complement distance, e.g., ( 0 and 0x800 ) undefined in the 12-bits case since D_f and D_b are both -0x7ff. A simple example of the relationship in case of 3-bit arithmetic follows as table indicating D_f/D_b values and then the relationship of U_1 to U_2:
U2 / U1 0 1 2 3 4 5 6 7
0 +/+ +/- +/- +/- -/- -/+ -/+ -/+
1 -/+ +/+ +/- +/- +/- -/- -/+ -/+
2 -/+ -/+ +/+ +/- +/- +/- -/- -/+
3 -/+ -/+ -/+ +/+ +/- +/- +/- -/-
4 -/- -/+ -/+ -/+ +/+ +/- +/- +/-
5 +/- -/- -/+ -/+ -/+ +/+ +/- +/-
6 +/- +/- -/- -/+ -/+ -/+ +/+ +/-
7 +/- +/- +/- -/- -/+ -/+ -/+ +/+
U2 / U1 0 1 2 3 4 5 6 7
0 = > > > ? < < <
1 < = > > > ? < <
2 < < = > > > ? <
3 < < < = > > > ?
4 ? < < < = > > >
5 > ? < < < = > >
6 > > ? < < < = >
7 > > > ? < < < =
Examples
Normal Operation
Normal Case Topology ^ N +--------+ +--------+ Level 2 | |ToF 21| |ToF 22| W <-*-> E ++-+--+-++ ++-+--+-++ | | | | | | | | | S v P111/2 P121/2 | | | | MHP MHP | | | | ^ ^ ^ ^ | | | | | | | | | | | | +--------------+ | +-----------+ | | | +---------------+ | | | | | | | | | +-----------------------------+ | | ^ | | | | | | | All 0/0 0/0 0/0 +-----------------------------+ TIEs v v v | | | | | | | +-+ +<-0/0----------+ | | | | | | | | | | +-+----++ Optional +-+----++ ++----+-+ ++-----++ Level 1 | | E/W Link | | | | | | |Spin111+----------+Spin112| |Spin121| |Spin122| +-+---+-+ ++----+-+ +-+---+-+ ++---+--+ | | | | | | | | | +---0/0--->-----+ 0/0 | +----------------+ | 0/0 | | | | | | | | +---<-0/0-----+ | v | +--------------+ | | v | | | | | | | +-+---+-+ +--+--+-+ +-+---+-+ +---+-+-+ Level 0 | | L2L | | | | | | |Leaf111+~~~~~~~~~~+Leaf112| |Leaf121| |Leaf122| +-+-----+ +-+---+-+ +--+--+-+ +-+-----+ + + \ / + + Prefix111 Prefix112 \ / Prefix121 Prefix122 Multi-homed Prefix +---------- PoD 1 ---------+ +---------- PoD 2 ---------+
This section describes RIFT deployment in the example topology given in without any node or link failures. The scenario disregards flooding reduction for simplicity's sake and compresses the node names in some cases to fit them into the picture better. First, the following bidirectional adjacencies will be established:
  1. ToF 21 (PoD 0) to Spine 111, Spine 112, Spine 121, and Spine 122
  2. ToF 22 (PoD 0) to Spine 111, Spine 112, Spine 121, and Spine 122
  3. Spine 111 to Leaf 111 and Leaf 112
  4. Spine 112 to Leaf 111 and Leaf 112
  5. Spine 121 to Leaf 121 and Leaf 122
  6. Spine 122 to Leaf 121 and Leaf 122
Leaf 111 and Leaf 112 originate N-TIEs for Prefix 111 and Prefix 112 (respectively) to both Spine 111 and Spine 112 (Leaf 112 also originates an N-TIE for the multihomed prefix). Spine 111 and Spine 112 will then originate their own N-TIEs, as well as flood the N-TIEs received from Leaf 111 and Leaf 112 to both ToF 21 and ToF 22. Similarly, Leaf 121 and Leaf 122 originate North TIEs for Prefix 121 and Prefix 122 (respectively) to Spine 121 and Spine 122 (Leaf 121 also originates a North TIE for the multihomed prefix). Spine 121 and Spine 122 will then originate their own North TIEs, as well as flood the North TIEs received from Leaf 121 and Leaf 122 to both ToF 21 and ToF 22. Spines hold only North TIEs of level 0 for their PoD, while leaves only hold their own North TIEs while, at this point, both ToF 21 and ToF 22 (as well as any northbound connected controllers) would have the complete network topology. ToF 21 and ToF 22 would then originate and flood South TIEs containing any established adjacencies and a default IP route to all spines. Spine 111, Spine 112, Spine 121, and Spine 122 will reflect all South Node TIEs received from ToF 21 to ToF 22 and all South Node TIEs from ToF 22 to ToF 21. South TIEs will not be re-propagated southbound. South TIEs containing a default IP route are then originated by both Spine 111 and Spine 112 towards Leaf 111 and Leaf 112. Similarly, South TIEs containing a default IP route are originated by Spine 121 and Spine 122 towards Leaf 121 and Leaf 122. At this point, IP connectivity across the maximum number of viable paths has been established for all leaves, with routing information constrained to only the minimum amount that allows for normal operation and redundancy.
Leaf Link Failure
Single Leaf Link Failure | | | | +-+---+-+ +-+---+-+ | | | | |Spin111| |Spin112| +-+---+-+ ++----+-+ | | | | | +---------------+ X | | | X | +-------------+ | X | | | | +-+---+-+ +--+--+-+ | | | | |Leaf111| |Leaf112| +-------+ +-------+ + + Prefix111 Prefix112
In the event of a link failure between Spine 112 and Leaf 112, both nodes will originate new Node TIEs that contain their connected adjacencies, except for the one that just failed. Leaf 112 will send a North Node TIE to Spine 111. Spine 112 will send a North Node TIE to ToF 21 and ToF 22 as well as a new South Node TIE to Leaf 111 that will be reflected to Spine 111. Necessary SPF recomputation will occur, resulting in Spine 112 no longer being in the forwarding path for Prefix 112. Spine 111 will also disaggregate Prefix 112 by sending new South Prefix TIE to Leaf 111 and Leaf 112. Though disaggregation is covered in more detail in the following section, it is worth mentioning in this example as it further illustrates RIFT's mechanism to mitigate traffic loss. Consider that Leaf 111 has yet to receive the more specific (disaggregated) route from Spine 111. In such a scenario, traffic from Leaf 111 towards Prefix 112 may still use Spine 112's default route, causing it to traverse ToF 21 and ToF 22 back down via Spine 111. While this behavior is suboptimal, it is transient in nature and preferred to dropping traffic.
Partitioned Fabric
Fabric Partition +--------+ +--------+ Level 2 |ToF 21| |ToF 22| ++-+--+-++ ++-+--+-++ | | | | | | | | | | | | | | | 0/0 | | | | | | | | | | | | | | | | +--------------+ | +--- XXXXXX + | | | +---------------+ | | | | | | | | | +-----------------------------+ | | | 0/0 | | | | | | | | 0/0 0/0 +- XXXXXXXXXXXXXXXXXXXXXXXXX -+ | | 1.1/16 | | | | | | | | +-+ +-0/0-----------+ | | | | | 1.1./16 | | | | +-+----++ +-+-----+ ++-----0/0 ++----0/0 Level 1 | | | | | 1.1/16 | 1.1/16 |Spin111| |Spin112| |Spin121| |Spin122| +-+---+-+ ++----+-+ +-+---+-+ ++---+--+ | | | | | | | | | +---------------+ | | +----------------+ | | | | | | | | | | +-------------+ | | | +--------------+ | | | | | | | | | | +-+---+-+ +--+--+-+ +-+---+-+ +---+-+-+ Level 3 | | | | | | | | |Leaf111| |Leaf112| |Leaf121| |Leaf122| +-+-----+ ++------+ +-----+-+ +-+-----+ + + + + Prefix111 Prefix112 Prefix121 Prefix122 1.1/16
shows more catastrophic scenario where ToF 21 is completely severed from access to Prefix 121 due to a double link failure. If only default routes existed, this would result in 50% of traffic from Leaf 111 and Leaf 112 towards Prefix 121 being dropped. The mechanism to resolve this scenario hinges on ToF 21's South TIEs being reflected from Spine 111 and Spine 112 to ToF 22. Once ToF 22 is informed that Prefix 121 cannot be reached from ToF 21, it will begin to disaggregate Prefix 121 by advertising a more specific route (1.1/16), along with the default IP prefix route to all spines (ToF 21 still only sends a default route). The result is Spine 111 and Spine 112 using the more specific route to Prefix 121 via ToF 22. All other prefixes continue to use the default IP prefix route towards both ToF 21 and ToF 22. The more specific route for Prefix 121 being advertised by ToF 22 does not need to be propagated further south to the leaves, as they do not benefit from this information. Spine 111 and Spine 112 are only required to reflect the new South Node TIEs received from ToF 22 to ToF 21. In short, only the relevant nodes received the relevant updates, thereby restricting the failure to only the partitioned level rather than burdening the whole fabric with the flooding and recomputation of the new topology information. To finish this example, the following list shows sets computed by ToF 22 using notation introduced in :
  • |R = Prefix 111, Prefix 112, Prefix 121, Prefix 122
  • |H (for r=Prefix 111) = Spine 111, Spine 112
  • |H (for r=Prefix 112) = Spine 111, Spine 112
  • |H (for r=Prefix 121) = Spine 121, Spine 122
  • |H (for r=Prefix 122) = Spine 121, Spine 122
  • |A (for ToF 21) = Spine 111, Spine 112
With that and |H (for r=Prefix 121) and |H (for r=Prefix 122) being disjoint from |A (for ToF 21), ToF 22 will originate a South TIE with Prefix 121 and Prefix 122, which will be flooded to all spines.
Northbound Partitioned Router and Optional East-West Links
North Partitioned Router + + + X N1 | N2 | N3 X | | +--+----+ +--+----+ +--+-----+ | |0/0> <0/0| |0/0> <0/0| | | A01 +----------+ A02 +----------+ A03 | Level 1 ++-+-+--+ ++--+--++ +---+-+-++ | | | | | | | | | | | +----------------------------------+ | | | | | | | | | | | | | +-------------+ | | | +--------------+ | | | | | | | | | | | +----------------+ | +-----------------+ | | | | | | | | | | | | +------------------------------------+ | | | | | | | | | | | ++-+-+--+ | +---+---+ | +-+---+-++ | | +-+ +-+ | | | L01 | | L02 | | L03 | Level 0 +-------+ +-------+ +--------+
shows a part of a fabric where level 1 is horizontally connected and A01 lost its only northbound adjacency. Based on N-SPF rules in , A01 will compute northbound reachability by using the link A01 to A02. However, A02 will not use this link during N-SPF. The result is A01 utilizing the horizontal link for default route advertisement and unidirectional routing. Furthermore, if A02 also loses its only northbound adjacency (N2), the situation evolves. A01 will no longer have northbound reachability while it receives A03's northbound adjacencies in South Node TIEs reflected by nodes south of it. As a result, A01 will no longer advertise its default route in accordance with .
Acknowledgments A new routing protocol in its complexity is not a product of a parent but of a village, as the author list already shows. However, many more people provided input and fine-combed the specification based on their experience in design, implementation, or application of protocols in IP fabrics. This section will make an inadequate attempt in recording their contribution. Many thanks to for some of the early discussions around the topic of using IGPs for routing in topologies related to Clos. is especially acknowledged for the key conversation on epistemology that tied the current asynchronous distributed systems theory results to a modern protocol design presented in this scope. , , , , , , , , , , and provided thoughtful comments that improved the readability of the document and found a good amount of corners where the light failed to shine. was first to mention single router, single arm default considerations. helped out with some initial thoughts on BFD interactions while corrected several misconceptions about BFD's finer points and helped to improve the security section around leaf considerations. pointed out many possible improvements and acted as a sounding board in regard to modern protocol implementation techniques RIFT is exploring. formalized the problem of partitioned spine and fallen leaves for the first time clearly on a (clean) napkin in Singapore that led to the very important part of the specification centered around multiple ToF planes and negative disaggregation. and others shared many thoughts on problems encountered in design and operation of large-scale data center fabrics. found a delicate error in the flooding procedures and a schema datatype size mismatch. Too many people to mention provided reviews from many directions in IETF, often pointing to critical defects, sometimes asking for things again that have been removed by one of the previous reviewers as objectionable or superfluous, and many times claiming the document being somewhere on the extremes between too crowded with the obvious and omitting introduction to cryptic concepts everywhere. The result is the best editors could do to find a balance of a document guiding the reader by into a specification tight enough to result in interoperable implementations while at the same time introducing enough operational context of IP routable fabrics to guarantee a concise, common language when facing unaccustomed concepts the protocol relies on. In the process, it was important to not end up carrying Aesop's donkey of course, so while the result may not be perceived as perfect by everyone, it should be practically speaking more than sufficient for everyone that ends up using it in the future. Last but not least, , , , and guided the undertaking as ADs by asking many necessary procedural and technical questions that did not only improve the content but also laid out the track towards publication. And is mentioned very last but not least for both his painstakingly detailed review and improvement of security aspects of the specification.
Contributors This work is a product of a list of individuals who are all to be considered major contributors, independent of the fact whether or not their name made it to the limited author list. Juniper Juniper Hudson River Trading Cisco Individual Individual LabN Individual Individual Nvidia