BIER

Summary

The BIER Working Group session at IETF 125 focused on the status of active drafts, recent multi-vendor interoperability testing results, and new proposals for monitoring, AI-driven use cases, and data plane optimizations. Key highlights included a report from EANTC testing, a new draft on IPFIX for BIER, a discussion on Large Language Model (LLM) synchronization requirements, and updates to the BIER-FRR framework.

Key Discussion Points

Working Group Status and Administrivia

Presenters: Sandy Zhang, Tony Przygienda
Draft Status:
- Two documents are in the RFC Editor queue.
- draft-ietf-bier-ping is in the IESG; Gunter Van de Velde (Routing AD) noted that all DISCUSSions are resolved. However, a "probe" to the WG is needed to determine if the draft should be limited to the MPLS data plane or cover all data planes. Tony Przygienda agreed to review the thread.
- draft-ietf-bier-ospfv3-extensions is currently in Last Call in the LSR working group.
LDP Signaling: Hooman Bidgoli inquired about the status of LDP/MLDP signaling over BIER drafts. Sandy Zhang noted that these drafts need to be refreshed by the authors before moving forward.
Charter Compliance: Gunter Van de Velde reminded the group that the IESG is strictly enforcing charter compliance. Any new or progressing documents must clearly align with the current BIER charter.

Interoperability Testing (EANTC 2026)

Presentation: 1-EANTC-2026
Presenter: Hooman Bidgoli
Discussion:
- Testing showed successful interop between Huawei, Juniper, and Nokia for BIER forwarding and NG-MVPN (I-PMSI).
- A technical hurdle was identified regarding the "BIER Next Protocol" field for VC labels. Juniper used Protocol 1 (downstream/global), while Huawei/Nokia used Protocol 2 (upstream/context).
- Hooman reported that Nokia implemented a "knob" to allow Protocol 1 to enable interop but suggested the WG should simplify this field to avoid future "chaos."
- Tony Przygienda and Jeffrey Zhang cautioned against unification, noting that the distinction between global and context-specific lookups provides significant scalability benefits for certain implementations.
- The group reached a consensus to clarify these behaviors in a forthcoming update/draft to ensure future implementations are aligned.

Export of BIER Information in IPFIX

Presentation: Export of BIER Information in IPFIX
Presenter: Sandy Zhang
Discussion:
- This draft proposes 11 new IPFIX Information Elements (IEs) to monitor BIER flows (BIFT ID, BitString, BFER ID, etc.).
- Tony Przygienda raised a concern regarding how monitoring stations correlate BIFT IDs and labels to sub-domains and sets without extra context. He suggested adding a "Considerations" section to the draft explaining how to reconstruct the BIER context for collected data.

Multicast for Large Language Model (LLM) Synchronization

Presentation: Multicast Use Cases for Large Language Model Synchronization
Presenter: Sandy Zhang
Discussion:
- The use case involves replicating models (70GB to 1TB+) across GPU clouds. BIER is identified as a potential solution due to its stateless nature.
- Hooman Bidgoli suggested exploring optimizations to remove IP/UDP headers and carry AI payloads directly over BIER.
- Tony Przygienda noted that BIER can already run directly over Ethernet, but the WG needs specific requirements from AI/HPC experts to determine if non-standard header optimizations are truly beneficial.

Scalable Data Plane Architecture for BIER

Presentation: 4-Scalable Data Plane Architecture for BIER
Presenter: Zhigiang Li
Discussion:
- The proposal shifts from a "bit-centric" to an "interface-centric" forwarding model using Replication Memory Tables (RMT) and Interface Bitmask Tables (IBMT).
- Jeffrey Zhang observed that the performance gain depends on whether the implementation is bottlenecked by the number of bits or the number of neighbors.
- Tony Przygienda noted the proposal is an implementation optimization (non-normative) and requested the authors include a section on ECMP equivalence and an analysis of ASIC update frequency during control plane changes.

BIER-FRR Update

Presentation: 5-BIER-FRR update (related to draft-ietf-bier-frr)
Presenter: Tony Przygienda
Discussion:
- The presenter focused on simplifying the framework for Fast Reroute (FRR).
- A key point was that BIER provides "node protection" more effectively than stateful multicast (PIM).
- Tony proposed removing the "tunnel mode" from the draft as it is technically redundant to LFA-based approaches.
- The discussion touched on the "partitioned Q space" problem. Tony suggested using "protected adjacencies" to handle rerouting without requiring massive expansion of the BIFT.

Decisions and Action Items

draft-ietf-bier-ping: Chairs/Authors to address the AD's question regarding data plane scope (MPLS vs. general).
LDP/MLDP Signaling: Authors (including Hooman Bidgoli) to refresh expired drafts.
draft-ietf-bier-frr: Tony Przygienda and co-authors to simplify the draft by removing "tunnel mode" and resolving pending reviews from the IESG.
IPFIX Draft: Sandy Zhang to add a section on metadata correlation considerations.
Scalable Data Plane: Zhigiang Li to add sections on ECMP and ASIC update metrics.

Next Steps

Continue refining use cases for AI/LLM synchronization to identify potential technical gaps in the BIER architecture.
Prepare for the next meeting in Vienna with updated drafts and refined interop documentation.

Automatic IETF Minutes