rfc-5004

Avoid BGP Best Path Transitions from One External to Another

E. Chen, S. Sangli
date2007-09 streamIETF areartg wgidr statusPROPOSED STANDARD pages6 canonicalhttps://www.rfc-editor.org/rfc/rfc5004 doi10.17487/RFC5004
In this document, we propose an extension to the BGP route selection rules that would avoid unnecessary best path transitions between external paths under certain conditions. The proposed extension would help the overall network stability, and more importantly, would eliminate certain BGP route oscillations in which more than one external path from one BGP speaker contributes to the churn. [STANDARDS-TRACK]

Extracted elements (11)

design-rationale §5

The algorithm is backward-compatible and can be deployed on a per-BGP-speaker basis without introducing any backward compatibility issues, making immediate incremental deployment possible.

bgp, routing

design-rationale §1

The BGP identifier of a route received from an external peer is effectively a random number from the local speaker's perspective, since identifiers are allocated by the remote AS. The last two tie-breaking steps of BGP route selection therefore produce arbitrary results for external-to-external comparisons, motivating the stability extension.

bgp, routing

design-rationale §3

The parallel-session exclusion is intentional: peering addresses for parallel sessions are typically allocated by one AS with route selection in mind, so applying the stability algorithm could disrupt existing routing setup and should not override that deliberate configuration.

bgp, routing

interoperability-note §5

The proposed extension introduces some non-determinism in BGP route selection compared to existing behavior; however, this non-determinism can be fully eliminated by using existing mechanisms such as LOCAL_PREF or MED if deterministic selection is required.

bgp, routing

interoperability-note §4

There are permanent BGP route oscillation scenarios described in RFC 3345 that this mechanism does not eliminate; the algorithm only addresses oscillations caused by multiple external paths from one BGP speaker contributing to churn.

bgp, routing

normative-requirement §3 SHOULD NOT

The algorithm SHOULD NOT be applied when both paths are from peers with an identical BGP identifier, i.e., when parallel BGP sessions exist between two BGP speakers, to preserve peering address allocation as a route selection tool.

bgp, routing

normative-requirement §3 SHOULD NOT

The best-path stability algorithm SHOULD NOT be applied when either path is from a BGP Confederation peer.

bgp, routing

normative-requirement §3 SHOULD

When the existing best path A and a new candidate best path B are both from external peers and neither is eliminated prior to the BGP identifier comparison step (Step f), the existing best path A SHOULD be kept as the best path to avoid unnecessary transitions.

bgp, routing

protocol-element §3

The algorithm targets the BGP route selection process defined in RFC 4271 Section 9.1.2.2, specifically the tie-breaking steps involving BGP identifier comparison (Step f) and peering address comparison, and proposes keeping the existing best external path when these steps would otherwise trigger a switch.

bgp, routing

protocol-element §4

The route oscillation example involves route reflectors (R1 with client R3, R2 with client R4) where interaction between route reflection (RFC 4456) and MED causes churning; applying the algorithm at R3 prevents switching from path (a) to (b) after R1 withdraws (c), stopping the oscillation.

bgp, routing

security-consideration §6

This extension does not introduce any security issues.

bgp, security