Markdown Version | Transcript | Session Recording | Session Materials
Session Date/Time: 16 Mar 2026 06:00
CATS
Summary
The CATS (Computing-Aware Traffic Steering) working group met at IETF 125 to discuss the status of its core documents and several new proposals. The CATS Framework has passed Working Group Last Call and is moving toward the IESG Telechat. The CATS Metrics document is nearing maturity, with discussions focusing on IANA registries and metric collection roles. The group also explored OAM, Security, and YANG models, as well as the applicability of existing protocols (BGP, PCEP) and transport technologies (OTN) to the CATS architecture. Emerging topics related to AI workloads and semantic networking were introduced for further consideration.
Key Discussion Points
1. Working Group Document Status
-
- Presenter: Cheng Li
- Status: Passed WGLC. Comments from IESG/Directorate reviews (TSVART, GENART, OPS, Security) have been addressed in versions 19–22.
- Discussion: Adrian Farrel noted the document is scheduled for the IESG Telechat on April 2nd.
-
2.2. CATS Metrics (draft-ietf-cats-metric-definition)
- Presenter: Kehan Yao
- Updates: The document is now on the Standards Track to allow for IANA metric registry entries. It follows templates from RFC 8911/8912.
- Discussion: The authors asked if the Service Metric Agent (CSMA) should also collect communication metrics. No immediate consensus was reached; the chairs encouraged discussion on the mailing list. The draft is considered nearly mature enough for WGLC.
2. Emerging CATS Components (Upcoming Drafts)
-
- Presenter: Fu Junsong
- Technical Details: Defines a four-layer OAM model. Introduces Service Instance OAM (SI-OAM) and Service OAM (SVC-OAM).
- Discussion: Huaimo Chen suggested aligning "layer" terminology with the "levels" used in the metrics draft. Adrian Farrel emphasized that the work should focus on CATS-specific system monitoring rather than underlying network "plumbing" OAM.
-
3.2. CATS Security Considerations
- Presenter: Jinyu Shi
- Technical Details: Outlines risks for CSMA, CNMA, C-TC, and Forwarders (e.g., man-in-the-middle, DoS, fake metric injection).
- Discussion: Jim Guichard (AD) advised the authors to cross-reference and align strictly with the security section in the CATS Framework draft. Peng Liu suggested prioritizing the listed risks.
-
- Presenter: Fu Junsong
- Technical Details: Augments the IETF routing YANG model (RFC 8349). Defines interfaces for CATS-SBI and CSMA-API.
- Discussion: The authors requested WG adoption. The chairs will develop a plan for subsequent adoptions once the current core drafts advance.
-
3.4. Computing metrics as a service (CMAS)
- Presenter: Ina
- Technical Details: Proposes "Globally Available Slots" (GAS) as a lightweight service-level abstraction to hide hardware complexity and preserve privacy.
3. Protocol and Technology Applicability
-
4.1. Protocols Applicability for CATS
- Presenter: Huijuan Yao
- Technical Details: Analysis of BGP-LS, PCEP, BGP FlowSpec, and SRv6 extensions for CATS.
- Discussion: Jeff Tantsura noted that the rate of change (churn) in computing metrics might be too fast for traditional BGP-LS and suggested adding performance considerations to the draft.
-
4.2. BGP Edge Metadata Path Applicability
- Presenter: Linda Dunbar
- Technical Details: Discusses the non-transitive BGP path attribute for metadata distribution within limited domains.
- Discussion: Identified gaps for CATS include the lack of service instance IDs and TTL/lifetime attributes in the current IDR specification.
-
5.1. Framework and Applicability of CATS in Optical Transport Networks (OTN)
- Presenter: Zhen Han
- Discussion: Adrian Farrel clarified that this is the "applicability of OTN to CATS" (using OTN as a CATS underlay). Coordination with CCAMP/TEAS is recommended.
4. Advanced Topics and AI Workloads
-
- Presenter: Zhen Han
- Technical Details: Focuses on joint orchestration of compute and optical resources for AI training.
- Discussion: Adrian Farrel pointed to "Layer 1 VPN" RFCs as relevant historical context for discovering edge points and providing optical trails.
-
- Presenter: Hanlin Wang
- Technical Details: Intelligence Delivery Network (IDN) for deep learning model deployment and Open Decentralized Scalable Inference (ODSI) using blockchain-inspired coordination.
-
6.2. In-Network Scheduling & Semantic Contract for AI Workloads (Intellinode)
- Presenter: [Speaker 4]
- Technical Details: Proposes a "semantic contract" between applications and the network to allow priority scheduling for specific AI traffic types (e.g., gradients vs. KV cache).
- Discussion: Kyrillidi Kompella and the chairs noted this may currently be outside the WG charter but is useful for architectural context.
Decisions and Action Items
- Decision: The CATS Framework is ready for IESG review.
- Action Item (Authors of draft-ietf-cats-metric-definition): Post a query to the mailing list regarding the role of CSMA in communication metric collection.
- Action Item (Chairs): Develop a roadmap for the adoption of OAM, Security, and YANG models following the advancement of current milestones.
- Action Item (Authors of Protocol Applicability): Update the draft to include analysis of state churn and the rate of change for control plane protocols (BGP-LS).
Next Steps
- Advancement of draft-ietf-cats-metric-definition toward Working Group Last Call.
- Detailed review of Security and OAM drafts against the finalized Framework.
- Ongoing discussion on the mailing list regarding AI-related extensions and protocol scaling.
Related Documents
draft-fu-cats-oam-framework-00, draft-ietf-cats-metric-definition, draft-yl-cats-data-model-05-00, draft-zhangb-cats-cmas-02-00