IDR Working Group C. Lin Internet-Draft New H3C Technologies Updates: 5004 (if approved) 6 July 2026 Intended status: Standards Track Expires: 7 January 2027 BGP Enhancements for ECMP EBGP Scenarios draft-lin-idr-bgp-ecmp-ebgp-enhancements-00 Abstract This document proposes extensions to BGP to apply the RFC 5004 route persistence algorithm across parallel EBGP sessions and to suppress unnecessary advertisements between EBGP peers in the same AS. This document updates RFC 5004. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on 7 January 2027. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Lin Expires 7 January 2027 [Page 1] Internet-Draft BGP ECMP EBGP Enhancements July 2026 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3 3. Problem Statement . . . . . . . . . . . . . . . . . . . . . . 4 3.1. Best Path Persistence Across Parallel EBGP Sessions . . . 4 3.2. Redundant Advertisement to Same-AS EBGP Peers . . . . . . 4 4. Proposed Solutions . . . . . . . . . . . . . . . . . . . . . 4 4.1. Route Selection Extension for Parallel EBGP Sessions . . 4 4.2. Suppression of Redundant Advertisement to Same-AS EBGP Peers . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5. Protocol Changes . . . . . . . . . . . . . . . . . . . . . . 5 5.1. BGP Path Selection Modification . . . . . . . . . . . . . 5 5.2. Advertisement Filtering Rule . . . . . . . . . . . . . . 5 6. Implementation Considerations . . . . . . . . . . . . . . . . 5 7. Security Considerations . . . . . . . . . . . . . . . . . . . 6 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 6 9. Normative References . . . . . . . . . . . . . . . . . . . . 6 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 6 1. Introduction In Spine-Leaf topologies of data center or AI network, devices frequently establish multiple parallel EBGP sessions to build Equal- Cost Multi-Path (ECMP) for redundancy and increased capacity, as shown in Figure 1. In Figure 1, Spine belongs to AS 100; Leaf1 through Leaf8 belong to AS 1 through 8 respectively. Spine has multiple parallel links with Leaf1 through Leaf8, and establishes multiple parallel EBGP sessions with each of Leaf1 through Leaf8. External path (a) is sent to Leaf1 through Leaf8 respectively. Lin Expires 7 January 2027 [Page 2] Internet-Draft BGP ECMP EBGP Enhancements July 2026 +---------------------------------------+ |AS 100 +-------+ | | +---------| |-----------+ | | | +-------| |---------+ | | | | | +-----| Spine |-------+ | | | | | | | +---| |-----+ | | | | | | | | | +-------+ | | | | | | | | | | | | | | | +------|-|-|-|-----------------|-|-|-|--+ | | | | | | | | +------|-|-|-|--+ +------|-|-|-|--+ |AS 1 +-------+ | |AS 8 +-------+ | | | Leaf1 | | ... | | Leaf8 | | | +-------+ | | | +-------+ | +-------------\-+ | +-----/---------+ \ | / \ | / (a) Figure 1: Spine-Leaf Network with Multiple Parallel EBGP Sessions [RFC5004] defines a mechanism to retain the current best path when a route is re-advertised by another EBGP peer without changes, thus improving the overall network stability. However, this mechanism does not apply across parallel sessions, which can lead to frequent best path oscillations in data center or AI network environments. Additionally, in such multiple-session configurations, a prefix received from one EBGP peer may be re-advertised to another EBGP peer in the same AS. This is unnecessary and increases network load. This document specifies an extension to [RFC5004] to apply route persistence algorithm cross parallel EBGP sessions, and a new advertisement rule to prevent sending a prefix to an EBGP peer if that prefix was originally received from another EBGP peer in the same AS. 2. Terminology 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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. Lin Expires 7 January 2027 [Page 3] Internet-Draft BGP ECMP EBGP Enhancements July 2026 3. Problem Statement 3.1. Best Path Persistence Across Parallel EBGP Sessions In a Spine-Leaf network using parallel EBGP sessions, Spine devices receive the same prefix from multiple Leaf peers through different BGP sessions. Since the path persistence algorithm defined in [RFC5004] does not apply to scenarios with identical BGP identifiers (i.e., parallel sessions), the best path will be reselected based on the comparison of peer addresses when the same route is received from another parallel peer. This can lead to unnecessary best path churn and may cause persistent routing oscillations. 3.2. Redundant Advertisement to Same-AS EBGP Peers When a BGP speaker receives a prefix from an EBGP peer and then re- advertises it to another EBGP peer in the same AS, it creates unnecessary advertisement and overhead due to the AS_PATH loop detection. In multiple EBGP sessions topologies, this can occur frequently and SHOULD be avoided. 4. Proposed Solutions 4.1. Route Selection Extension for Parallel EBGP Sessions To address unnecessary best path churn, the BGP speaker SHOULD maintain best path persistence state for each prefix across all parallel EBGP sessions to the same peer AS. When a new UPDATE for the same prefix is received from any parallel EBGP peer, the speaker SHOULD compare it with the current best path using the extended algorithm: * If the new route remains equivalent to the current best path before BGP identifier comparison, and both routes are received from peers that are identified as belonging to the same set of parallel EBGP sessions, the current best path SHOULD be retained, regardless of which parallel EBGP peer advertised it. * This persistence state SHOULD be reset only if the route selection attributes change or the route is withdrawn. This mechanism requires the BGP speaker to be able to identify parallel EBGP sessions. Such identification MAY be based on explicit configuration or MAY be derived automatically by detecting that multiple EBGP peers share the same BGP Identifier. Lin Expires 7 January 2027 [Page 4] Internet-Draft BGP ECMP EBGP Enhancements July 2026 4.2. Suppression of Redundant Advertisement to Same-AS EBGP Peers Before advertising a prefix to an EBGP peer, the BGP speaker SHOULD check the AS_PATH of the prefix as stored in its Adj-RIB-In. If the first AS in the AS_PATH matches the peer's AS (indicating the prefix was originally received from that AS via another EBGP session), the speaker SHOULD NOT advertise the prefix to that peer. This check prevents unnecessary propagation of the prefix back into the same AS via a different EBGP session. 5. Protocol Changes 5.1. BGP Path Selection Modification Base on the best path persistence algorithm defined in [RFC5004], a new step is also added before BGP identifier comparison (Section 9.1.2.2 of [RFC4271]) of the best path selection process: "If the speaker supports the parallel-session persistence extension defined in this document, and if a new route is received from a peer that is identified as part of a set of parallel EBGP sessions, and this new route is determined to be equivalent to the current best path prior to the comparison of BGP Identifiers, and the current best path was also received from a peer within the same set of parallel sessions, then the speaker MUST retain the current best path and SHALL NOT select a new best path." 5.2. Advertisement Filtering Rule During the process of route advertisement, prior to transmitting an UPDATE message to a peer, a BGP speaker SHOULD apply the following rule: When advertising a prefix to an EBGP peer, if the first AS in the prefix's AS_PATH equals the peer's own AS number, the BGP speaker SHOULD NOT advertise that prefix to the peer. 6. Implementation Considerations Implementations SHOULD provide a configuration option to enable/ disable: * Route Selection Extension for Parallel EBGP Sessions. * Same-AS EBGP advertisement suppression. Lin Expires 7 January 2027 [Page 5] Internet-Draft BGP ECMP EBGP Enhancements July 2026 Defaults MAY be vendor-specific but SHOULD favor enabling these features in network with multiple Parallel EBGP sessions. 7. Security Considerations These extensions defined in this document affect route stability and advertisement scope but do not introduce new security threats beyond those inherent in BGP. Operators SHOULD ensure that parallel session detection is correctly configured to avoid incorrect route persistence. 8. IANA Considerations This document has no IANA requirements. 9. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC4271] Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A Border Gateway Protocol 4 (BGP-4)", RFC 4271, DOI 10.17487/RFC4271, January 2006, . [RFC5004] Chen, E. and S. Sangli, "Avoid BGP Best Path Transitions from One External to Another", RFC 5004, DOI 10.17487/RFC5004, September 2007, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . Author's Address Changwang Lin New H3C Technologies China Email: linchangwang.04414@h3c.com Lin Expires 7 January 2027 [Page 6]