Markdown Version | Session Recording

Session Date/Time: 03 Nov 2025 14:30

CATS Session Minutes - IETF 118 Montreal

Summary

The CATS (Computing-Aware Traffic Steering) working group met to discuss the status of its core documents and several individual drafts exploring solutions and applicability. The use case and framework documents are progressing towards Working Group Last Call (WGLC), with the framework document targeted for IESG publication as informational. The metrics document is proposing a shift to Standard Track and introducing key concepts like normalization and aggregation functions. Several individual drafts presented early solution ideas, including YANG models, OAM frameworks, BGP extensions for metadata, ALTO for metrics, DNS-based discovery, and applicability to AI Agent Communication Networks (ACN) and 5G Edge. Chairs emphasized the working group's charter limitations regarding protocol-specific solutions and the need for coordination with other working groups where appropriate.

Key Discussion Points

Working Group Documents

• CATS Use Cases and Requirements (draft-ietf-cats-use-cases-ps-08)
  • Updates: Incorporate shepherd's review, add a new potential use case (ISAC) in Appendix B, include a new requirement for CATS extensibility in non-CATs environments, add discussion on relevance to other SDOs (ETSI, 3GPP), and update contributors.
  • Status: Currently undergoing Working Group Last Call (WGLC), closing on November 18th. An IPR poll is also underway, with responses still needed from three contributors.
• CATS Framework (draft-ietf-cats-framework)
  • Updates: Six revisions made to address comments on terminology, passive voice, metric distribution, example illustrations, and clarifying that the CATS environment is for a single network provider. Major issues from Adrian's review were addressed, including scalability considerations for per-service instance metrics (to be aggregated), clarifying a single CPS in hybrid mode, modifying "centralized control plan" to "centralized control element," emphasizing uniform normalization functions across vendors, and moving metric distribution examples to an appendix.
  • Status: Authors believe it is ready for WGLC.
• CATS Metrics (draft-ietf-cats-metrics)
  • Proposed Status Change: Document proposes changing its status from Informational to Standard Track. A poll indicated general support, with one participant wanting to review the latest version before agreeing.
  • New Concepts: Introduced a new Level One metric class: "service type." Formally defined "normalization" and "aggregation" functions to transform raw/derived metrics into "pseudo-metrics" for decision-making. Aggregation combines multiple metrics into one (e.g., mean, max, weighted average). Normalization converts metrics (with or without units) into unit-less scores (e.g., sigmoid, min-max scaling).
  • Clarification of Scores: Added notes for implementers (provide precise definitions of metrics) and operators (exercise caution when making decisions based on unclear/underspecified unit-less metrics in heterogeneous environments).
  • Discussion:
    • Standardization Scope: Clarified the intent to define Level 1 classes of metrics (compute, communication, service, composition) and potentially additional templates, but not standardize Level 0 metrics.
    • IPPM Alignment: A suggestion was made to follow RFC 8911 and add new metrics to the IANA performance registry, referencing existing IPPM metrics where applicable.
    • OAM Overheads: Concerns were raised regarding the potential overhead of CATS OAM and how to measure its cost and associated SLAs.
    • Non-CATS Environments: The document does not currently address integration with non-CATs environments, though the use-case document has a requirement for extensibility.

Individual Drafts (Solution Space)

• YANG Model for CATS (draft-ran-cats-yang-model)
  • Content: Revised to align with the framework document's functional components (CP, CSMA, CPS), added definitions for CATS Computing Information Base (CCIB) and CATS Network Information Base (CNIB), and updated terminology for interfaces (CASBI, CSMA API) for reporting.
  • Discussion: The working group is inclined to consider adoption. Concerns were raised about defining YANG models without a defined protocol for information distribution and the appropriateness of defining YANG for forwarding plane mechanisms, which CATS currently considers a local matter for vendors.
• OAM for CATS (draft-chen-cats-oam-framework)
  • Content: Outlined motivation for OAM in CATS (monitoring, fault detection, SLA verification, flow performance, troubleshooting). Proposed an OAM layering model (link, tunnel, service OAM) and functional aspects (service instance, classified packet, flow OAM). Described OAM requirements related to sub-second telemetry, resource visibility, billing, end-to-end quality, and fault localization.
  • Discussion: Chairs expressed appreciation for early consideration of OAM/manageability, suggesting it could influence YANG model development and referencing Ops Area work (5706bis). The document is currently a framework and does not propose protocol changes.
• 5G Edge Metadata in BGP (draft-li-idr-5g-edge-metadata-path-attribute)
  • Content: Defines an optional, non-transitive BGP path attribute to carry metrics and metadata for edge services. Emphasizes mechanisms to limit propagation to a "limited domain" for security (e.g., BGP OPEN capability negotiation, non-advertise communities, AS scope sub-TLVs, route filtering). Includes preliminary metrics like site preference and physical availability index.
  • Discussion: Chairs suggested exploring the applicability and non-applicability to CATS, particularly for multi-AS systems and dynamic metrics. A separate, short draft on "applicability of 5G edge metadata to CATS" was suggested.
• Service Metadata in BGP FlowSpec (draft-yao-bess-bgp-flowspec-service-metadata)
  • Content: Proposes using BGP Flow Specification for distributing service metadata in a centralized model. Aims to collect data via BGPLS, process it at a controller, and distribute aggregated metadata/path policies via BGPFS using TLV formats. Recommends aggregated values to reduce overhead.
  • Discussion: Chairs noted this work is more general than CATS and suggested pushing it in IDR or BESS, with a back-reference to CATS. The co-chair questioned alignment with the CATS metrics document and asked for clarification on whether "large scale" refers to single or multi-domain deployments, as CATS currently charters work for single domains.
• ALTO for CATS Metrics (draft-luis-alto-cats-metrics)
  • Content: Explores ALTO as an off-path solution to expose CATS metrics for service deployment and steering. Proposes ALTO servers collecting metrics and providing them to ALTO clients (e.g., service orchestrators or CATS path selectors). Requires ALTO extensions for CATS metric levels, COS models, and associating specific CATS metrics with endpoints.
  • Discussion: Confirmed the document is CATS-focused. Given the ALTO working group is closed, the challenge of finding a proper home for ALTO extensions was acknowledged, with the Ops Area Working Group being a potential venue.
• Reference Model for AI Agent Communication Networks (ACN) (draft-yu-cats-ai-agent-acn-ref-model)
  • Content: Proposes a reference model for AI Agent Communication Networks (ACN), highlighting their need for CATS optimization. Suggests extending CATS with a third metric category: "AI agent metrics" (e.g., functionality, model types, parameters of AI models), distinct from network and compute metrics, to cover application-level exchanges among AI agents.
  • Discussion: Concerns were raised that CATS is not currently chartered for AI agents and that AI agent metrics are application-level. It was suggested to keep CATS metrics separate and explore alignment between ACN and CATS frameworks in standalone documents or other working groups.
• Applicability of CATS Framework to Noteworthy Scenarios (draft-yu-cats-framework-applicability)
  • Content: Argues that the CATS framework's generic architecture is well-suited to demonstrate the applicability of CATS technologies. Illustrates this by mapping the framework to three scenarios: AI Agent Communication Networks (ACN), 5G Edge enhancement, and O-RAN middle-haul.
  • Discussion: This document served to illustrate potential future directions and relevance for the CATS framework beyond its initial scope.
• DNS-based Server Discovery for CATS (draft-liu-cats-dns-service-discovery)
  • Content: Proposes using DNS mechanisms (DNS-SD) for service registration and discovery within the CATS framework, mapping service identifiers to addresses. Defines a new "CATS Parameters" resource record type and outlines integration with a service registry and resolver within the CATS management plane.
  • Discussion: Chairs reiterated that CATS is not chartered for DNS work and this needs a proper home. Noted related work (AI agent discovery) presented in Dispatch, which is moving to a BOF. Suggested authors coordinate with Jim Mosley (Infoblox) to align approaches and structures for DNS-based discovery.

Decisions and Action Items

• Use Cases and Requirements Document:
  • Action: Authors to chase up outstanding IPR responses from contributors to complete the WGLC.
• Framework Document:
  • Action: Adrian (Chair) to review the latest revision this week to confirm that all review comments have been addressed.
  • Decision: Chairs plan to start the Working Group Last Call for the framework document almost immediately, aiming for completion within November.
  • Action: Adrian (Chair) to notify authors of missing IPR disclosures for follow-up.
• Metrics Document:
  • Action: Authors to facilitate further discussions on the mailing list to gather more reviews and feedback, particularly on the proposed shift to Standard Track and the scope of metric standardization.
  • Action: Seek confirmation from the WG on the proposed Standard Track status and the strategy to standardize Level 1 and Level 2 metrics only.
• YANG Model for CATS Document:
  • Action: Chairs to take Joel Halpern's concerns regarding protocol dependency and forwarding plane constraints into account and discuss internally for potential refinement before a formal adoption decision.
• OAM for CATS Document:
  • Action: Authors to send an email to the mailing list requesting more review and feedback.
• 5G Edge Metadata in BGP Document:
  • Action: Linda to consider writing a separate, short draft on the "applicability of 5G edge metadata to CATS," covering both applicable and non-applicable scenarios (e.g., multi-AS, dynamic metrics).
• Service Metadata in BGP FlowSpec Document:
  • Action: Authors to clarify the scope of "large scale" (single vs. multi-domain) and ensure alignment with the CATS metrics document. The primary push for this work should be in IDR or BESS.
• ALTO for CATS Metrics Document:
  • Action: Authors to collect feedback from CATS working group members.
  • Action: Chairs to investigate potential homes for ALTO extensions, possibly in the Ops Area Working Group.
• Reference Model for AI Agent Communication Networks (ACN) Document & Applicability of CATS Framework to Noteworthy Scenarios Document:
  • Action: Authors to continue working on these topics, engage with other AI Agent-focused groups, and explore potential future charter extensions for CATS or cross-WG coordination.
• DNS-based Server Discovery for CATS Document:
  • Action: Authors to connect with Jim Mosley (Infoblox) to discuss potential alignment of DNS-based discovery approaches and structures, especially given the upcoming BOF on AI agent discovery.
  • Action: Chairs to assist in finding a proper home for this DNS-related work, as CATS is not chartered for it.

Next Steps

• Working Group Last Calls: Complete the WGLC for the use cases document and initiate the WGLC for the framework document (within November).
• Metrics Document Progression: Intensify discussions and reviews for the metrics document on the mailing list, aiming for submission to the IESG for publication in the next three months.
• Individual Drafts: Authors of individual drafts are encouraged to continue their work, seek feedback on the mailing list, clarify scope, and explore appropriate working groups (e.g., IDR, BESS, Ops Area, Dispatch/BOFs) for protocol-specific or broader applicability work, especially where CATS charter limitations apply.
• Coordination: Chairs will facilitate coordination with other working groups for topics that overlap with CATS interests but fall outside its direct charter (e.g., DNS, ALTO, AI agents).