Session Date/Time: 28 Mar 2023 08:00

rift

Summary

This meeting covered updates on the RIFT base specification, key value registry, auto-VPN, flood reflection, and the proposed new charter. The base spec is nearing completion, and the key value registry is proving valuable for building new technologies. Discussions focused on implementation feedback, debugging, and the future direction of RIFT, including potential use cases in machine learning clusters.

Key Discussion Points

Base Specification Update: Alvaro is finishing the AD review of version 17. The draft includes clarifications on tie ordering algorithms, F/N statements, and IPv4 forwarding. Empty MTU mismatch detection is clarified, and broadcast support has been added with security considerations. Distance metric terminology has been tightened.
Applicability Draft: The applicability draft needs to cover deployment aspects like subnet checking and best practices (e.g., using /31 vs. /32). It should serve as a mild implementation and deployment guide.
Key Value Registry: The southbound key value store is being used extensively. Debugging is crucial due to potential tie-breaking issues. A well-known key type is used for debugging purposes.
Auto-VPN and Flood Reflection: Work is progressing, with Jordan driving the drafts and code. Star-works are advertised and showing no problems. The base spec stability is crucial for these drafts to roll out.
Implementation and Interoperability: There was discussion about encouraging implementation and interoperability testing. Hector is being looked at as a test platform for the base spec. It was stated, “Unless people go and implement and play with it…deployment or flexibility document makes sense, but just cooking one up because we cook one up. No. There be nothing in it really.”
New Charter: A new charter is proposed to include ongoing work items like key value registry, segment routing support, and using zero-touch provisioning for auto-VPN and flood reflection. Additional items include multi-support and policy-guided prefixes. Questions were raised about the scope of zero touch provisioning within the charter, and it was mentioned that work was previously presented in BESS-LSR, and should be coordinated with the chairs beforehand.
Machine Learning: RIFT is being considered as a potential protocol for massive deployments, such as those found in machine learning clusters, due to its properties.
Scope of RIFT: The intended routing protocol extension nature of RIFT was discussed, emphasizing that any new work should align with this core function.

Decisions and Action Items

Applicability Draft: Authors to incorporate deployment aspects and best practices.
Base Specification: A quick working group last call may be necessary after recent changes.
Key Value Registry: Consider reserving a key value for generalized prefixes(GP) support. Sandy to add reserved key for north bound to show the key value store works.
New Charter: Review and provide feedback on the proposed charter text.
Zero Touch Provisioning: Coordinate with the chairs to ensure there is agreement on the scope of this work.
Segment Routing: Ensure that any segment routing support extends routing protocol functionality.

Next Steps

Finalize the base specification.
Focus on implementing and testing auto-VPN and flood reflection, then deploy.
Further refine the applicability draft.
Finalize the new charter proposal based on feedback.