IGP Flexible Algorithm with Link Loss

Abstract

This document proposes extensions to the IGP Flexible Algorithms framework defined in [RFC9350]. It introduces a mechanism to exclude links exceeding a specified packet loss rate threshold during path computation. The solution leverages existing link loss measurements advertised via IS-IS [RFC8570] and OSPF [RFC7471], and defines new constraints for Flex-Algorithm path calculation.¶

1. Introduction

Link packet loss rate (hereafter "link loss") refers to the percentage of data packets that are lost during transmission over a network. It is a critical metric for network performance evaluation. High loss rates directly impact service quality, congestion management, and operational efficiency. To maintain optimal forwarding paths, it is essential to avoid links with excessive packet loss during IGP path computation.¶

The IGP Flexible Algorithms enable IGPs to compute constraint-based paths [RFC9350]. Current path computation methods focus on determining the minimum cost of the path from the source to the destination. Flex-Algorithm already supports path computation based on IGP cost, minimum link delay, and traffic-engineering metrics. [I-D.ietf-lsr-flex-algo-bw-con] defines a family of generic metrics (e.g., bandwidth-based metric type) and bandwidth-related constraints to enable path computation based on bandwidth. However, current flexible algorithm definitions lack native support for path computation based on link loss, as the path cost should be defined as the maximum or minimum value among all links in the path.¶

To address this issue, two solutions are considered. First, new operators, such as a maximum value operator, can be defined to function as a step function. Specifically, when the link loss exceeds a threshold, the link cost is set to the maximum value. Second, new Flexible Algorithm Definition (FAD) constraints can be defined to exclude links that do not meet the link loss requirements during path calculation. The second method is specifically demonstrated in this document, and the general ideas are as follows:¶

• 1. The link loss is used as a link constraint for path computation. That is, links with a loss rate exceeding the specified value are excluded.¶

• 2. Metric-type remains unchanged: igp, te, and delay.¶

This document proposes the method to exclude links exceeding a specified packet loss rate by defining:¶

• a) A new Flexible Algorithm Definition (FAD) constraint to exclude links exceeding a configured maximum loss threshold (Section 2).¶

• b) Operational procedures for integrating loss constraints into Flex-Algorithm path computations (Section 3).¶

• c) Mechanisms to stabilize routing during loss metric fluctuations (Section 4).¶

The solution reuses existing link loss advertisements defined in [RFC8570] for IS-IS and [RFC7471] for OSPF, ensuring backward compatibility with deployed networks. The link packet loss rate can be measured using methods such as TWAMP [RFC5357] and STAMP [RFC8762]. However, these measurement techniques are beyond the scope of this document. It is important to ensure that link-loss measurements are consistent throughout the IGP routing domain.¶

2. Exclude Maximum Link Loss Sub-TLV

A new sub-TLV, the Exclude Maximum Link Loss Sub-TLV, is defined as part of the FAD TLV. To ensure loop-free forwarding, all routers participating in a Flex-Algorithm MUST agree on the FAD definition. Selected nodes within the IGP domain MUST advertise FADs by including them in their routing updates, as specified in Sections 5, 6, and 7 of [RFC9350].¶

The Exclude Maximum Link Loss Sub-TLV is introduced to define the maximum allowable link loss value. When this Sub-TLV is carried within the FAD TLV, all network links with packet loss rates exceeding the specified maximum value are excluded from the Flex-Algorithm path computation.¶

    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
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |      Type     |    Length     |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                Max Link Loss                  |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

      Type: 252(TBA by IANA)

      Length: 3 octets

      Max Link Loss:  24-bit unsigned integer representing the maximum
      allowable loss percentage. Encoded with a resolution of 0.000003%
      per unit, providing a maximum expressible value of 50.331642%
      (0xFFFFFF * 0.000003). Values exceeding this cap MUST be advertised
      as 0xFFFFFF.

    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
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |               Type            |            Length           |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                Max Link Loss                  |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

      Type: 252(TBA)

      Length: 3 octets

      Max Link Loss:  24-bit unsigned integer representing the maximum
      allowable loss percentage. Encoded with a resolution of 0.000003%
      per unit, providing a maximum expressible value of 50.331642%
      (0xFFFFFF * 0.000003). Values exceeding this cap MUST be advertised
      as 0xFFFFFF.

3. Calculation of Flexible Algorithm Paths

The following rule is added to the topology pruning rules in Section 13 of [RFC9350]:¶

• 1. Check if any exclude FAEML rule is part of the Flex-Algorithm definition. If such exclude rule exists and the link has link loss advertised, check if the link satisfies the FAEML rule. If not, the link MUST be pruned from the computation.¶

4. Operational Considerations

In certain scenarios, the link status may fluctuate between available and unavailable due to the link packet loss rate oscillating around the threshold value. Consequently, Flex-Algorithm computation may be triggered repeatedly. Several mechanisms are considered to address this issue:¶

• 1. Delayed collection: The IGP-advertised loss can be calculated over a sufficiently long interval, such as 10 minutes, to reduce the frequency of updates.¶

• 2. Averaging and normalization: The IGP-advertised loss should be derived from a form of averaging, such as an exponential weighted average, of the collected loss values. The advertised loss can be normalized to prevent the dissemination of non-significant changes in loss metrics.¶

• 3. Flapping suppression: If frequent changes in the IGP-advertised loss are detected, a timer can be implemented to delay the update process, thereby stabilizing the routing computations.¶

6. References