CATS

Summary

The Computing-Aware Traffic Steering (CATS) Working Group met at IETF 126 in Vienna. The session was chaired by Peng Liu and Adrian Farrel. The chairs provided status updates on active working group documents, highlighting that the CATS framework and use cases are currently in the RFC Editor Queue (RFC-Ed Queue), draft-ietf-cats-metric-definition is undergoing early review, and draft-ietf-cats-oam-fw has been newly adopted as a working group document. AD Jim Guichard recently approved new milestones, including those for a data model document, protocol applicability, and OAM.

The meeting featured progress updates on both active working group drafts, followed by presentations of individual submissions covering protocol applicability, data models, BGP-LS metadata distribution, service-specific metrics, service platform definitions, midhaul and optical transport network (OTN) applicability, and operational metric semantics. A key recurring theme was the necessity of defining protocol requirements within the CATS working group before bringing specific protocol extensions to downstream working groups (such as IDR, PCE, and LSR).

Key Discussion Points

1. Working Group Status & Admin

Presenters: Peng Liu and Adrian Farrel
Slides: 1-Chairs Slides
Discussion:
- Peng Liu presented the agenda and current draft statuses.
- Adrian Farrel initiated a poll regarding travel plans for IETF 127 to assist in session planning.

Session Poll:

Question: Do you already know your travel plans for IETF 127? Will you travel — yes: 3, no: 9, no_opinion: 6 (total: 39)

2. CATS Metrics Definition

Presenter: Kehan Wang
Slides: 2.1-cats-metrics-definition
Draft: draft-ietf-cats-metric-definition
Discussion:
- Kehan Wang summarized updates since IETF 125 (revisions -07 to -10), acknowledging feedback from Adrian Farrel (chair review), Jack Line (SEC early review), and early reviews from the OPS area.
- Key updates include clarifying the distinct behaviors of aggregation functions (which combine homogeneous or heterogeneous inputs—defined as spatial, temporal, and cross-category aggregation) and normalization functions (which adjust a single input into a unitless, bounded score, such as via sigmoid or min-max scaling).
- Level 1 metrics were renamed to "metrics combined in categories" to clarify that they may retain physical units or be unitless depending on the aggregation function used.
- Security considerations were revised to address five key areas: integrity, authenticity, controllability, freshness, and confidentiality.
- Q&A: Adrian Farrel asked if ideas from other metric-related drafts on the agenda should be merged or kept separate. Kehan Wang responded that ideas from those drafts would serve as valuable inputs for the operational and implementation considerations section.

3. CATS OAM Framework

Presenter: Juan Deng
Slides: 2.2-draft-ietf-cats-oam-fw
Draft: draft-ietf-cats-oam-fw
Discussion:
- Juan Deng presented updates to the OAM framework (revisions up to -01), which included removing "SI-OAM" and "SVC-OAM" terminologies, adding push/pull reporting modes, clarifying problem statements/gaps, refining the 4-layer OAM model (Service OAM, Instance OAM, Path OAM, Link OAM), and detailing security and deployment considerations.
- Q&A:
  - Luis M. Contreras asked whether it would be simpler to run an OAM probe in the virtualization environment collocated with the service instance rather than imposing requirements directly on the service instance itself. Juan Deng agreed that OAM should monitor metrics reports rather than looking inside the service instance itself.
  - Sheng Li commented that the draft currently lacks sufficient CATS-specific content and should incorporate more dedicated CATS text in future updates. Juan Deng agreed to clarify these aspects.
  - Boris Khasanov asked (via microphone and chat) why STAMP was not considered as an OAM tool alongside TWAMP or BFD. Juan Deng replied that existing measurement tools could be extended to support computing metrics and service instances.
  - Chat Discussion: Carolina Caeiro queried the definition of "Multi-Domain Fault Isolation" and whether TRACE mode is strictly intra-domain. Quan Xiong clarified that this refers to fault localization across AS domains or areas within an administrative domain. Adrian Farrel pointed out that sharing fault details across separate AS domains is atypical and suggested focusing on failure location between CATS-forwarders and service instances without emphasizing administrative boundaries. Carolina Caeiro noted that inter-domain operations might be outside the current charter scope.

4. CATS Protocols Applicability

Presenter: Linda Dunbar
Slides: 3.1-CATS Protocols Applicability
Discussion:
- Linda Dunbar presented an analysis of protocols for CATS (including BGP, IGP, BGP-LS, and PCE).
- The draft concludes that BGP is highly useful for policy-constrained distribution within limited domains but unsuitable for rapidly fluctuating L0 raw metrics. IGP flooding is also unsuitable for detailed L0 metrics, but BGP-LS remains valuable for centralized topology collection. PCE can be used for centralized path computation.
- Q&A:
  - Aihua Liu suggested adding an analysis of how multiple protocols (e.g., BGP and IGP) collaborate within a hybrid CATS system. Linda Dunbar welcomed the suggestion.
  - Sheng Li suggested consulting experts from BGP, IGP, and PCE working groups, expressing personal skepticism about the suitability of PCE.
  - Boris Khasanov suggested considering YANG Push/telemetry within the management plane analysis.
  - Jim Guichard (AD) cautioned the group against "jumping the gun." He emphasized that the draft should focus strictly on outlining protocol requirements based on the CATS framework, rather than making premature protocol selection assumptions or spawning a proliferation of solution drafts in other working groups (like IDR) before CATS metrics and requirements are stable.
  - Jeff Haas (speaking as IDR Co-Chair) supported Jim Guichard's remarks, advising CATS to first refine the behaviors and metrics it wants to expose before taking concrete proposals to IDR.

5. CATS Data Model

Presenter: Juan Deng
Slides: 3.2-draft-yl-cats-data-model-07
Discussion:
- Juan Deng presented the YANG data model for CATS, detailing the CATS Computing Information Base (CIB), the CATS Network Information Base (NIB), the southbound interface (SBI), and the CSMA API.
- Recent changes aligned the YANG tree with RFC 8340, changed the traffic classifier container to a list, and adjusted metric configurations and read-only statistics counters.
- Q&A:
  - Adrian Farrel asked if the data model was stable enough for a YANG Doctors review. Juan Deng confirmed it is ready.
  - Jeff Haas noted that while a YANG Doctors review will validate syntax and structure, the CATS working group must actively perform the semantic review of the data model.

6. Distribution of Service Metadata in BGP-LS

Presenter: Mo Young Han
Slides: 3.3-Distribution of Service Metadata in BGP-LS
Discussion:
- Mo Young Han presented a centralized metadata collection model using BGP-LS to carry service metadata northbound from Edge sites to a Centralized Path Selector (CPS), showcasing real-world trial deployment results in China.
- Q&A:
  - Susan Hares (IDR Co-Chair) requested that when proposals are brought to IDR, they should have explicit confirmation from the CATS WG that they are required for the CATS framework, ideally documented in an appendix.
  - Jeff Haas requested the authors and WG to analyze various proposals to see if common L1/L2 CATS metrics can be consolidated and exposed cleanly through BGP or BGP-LS, rather than introducing fragmented attributes.

7. Computing Service Metric Definition and Operation

Presenter: Lina Ji
Slides: 3.4-draft-zhangb-cats-service-metrics-op
Discussion:
- Lina Ji presented service-oriented metrics ("guess"—representing available concurrent capacity, and "computing time"—representing expected processing time) designed to bridge the gap between raw hardware metrics and the control plane.
- Q&A:
  - Kehan Wang suggested integrating these metrics directly into the main draft-ietf-cats-metric-definition document (under operational considerations) rather than maintaining a separate draft. Lina Ji argued that this work is service-specific and should remain independent.
  - Susan Hares asked if the "computing time" metric assumes network-wide synchronization. Lina Ji clarified that computing time is derived locally by the service instance and reported as an elapsed delta, meaning global timestamp synchronization is not required.

8. Public Service Platform for CATS

Presenter: Lina Ji
Slides: 3.5-draft-zhangb-cats-cmas
Discussion:
- Lina Ji introduced a catalog-style "Public Service Platform" designed to resolve service descriptions into CSIDs and reference contexts before traffic steering begins.
- Q&A: Yun Zi asked about the relationship of this platform to the Discovery of Agents and Workloads (DOM) BOF. Lina Ji clarified that DOM focuses on agent discovery, whereas this platform functions as a registry catalog for service entries matching the CATS framework.

9. Compute-Aware Traffic Steering for Midhaul Networks

Presenter: Luis M. Contreras
Slides: 3.6-Compute-Aware Traffic Steering for Midhaul Networks
Discussion:
- Luis M. Contreras explained how CATS can apply to ORAN midhaul networks to connect Distributed Units (DU) with Centralized Units (CU), focusing on the interface between the ORAN Service Management and Orchestrator (SMO) and CATS components.
- Q&A:
  - Carlos J. Bernardos pointed out a potential relationship with DMM work on mobile traffic steering. Luis M. Contreras agreed to review it.
  - Adrian Farrel asked if this represented a layered or stitched architecture. Luis M. Contreras described it as more akin to stitching, where the transport network dynamically provides the requested connectivity to entities governed by the SMO.
  - Juan Deng asked if there were gaps in the existing CATS data model for this use case. Luis M. Contreras expressed interest in collaborating to map potential gaps into the data model draft.

10. Computation-aware Traffic Steering in Optical Transport Networks (OTN)

Presenter: OTN Presenter
Slides: 3.7-Framework and Applicability of Computation-aware Traffic Steering in Optical Transport Networks
Discussion:
- The presenter proposed utilizing OTN and Fine-Grain OTN (FG-OTN) as a deterministic, hard-isolated underlay option for CATS, particularly to connect distributed accelerator/GPU clusters. The proposed workflow covers matrix distribution, joint selection, and deterministic container encapsulation.
- Next steps include aligning the analysis with official CATS use cases and performing a protocol gap analysis.

11. Operational Semantics for CATS Metric Consumption

Presenter: Longfei Xia
Slides: 3.8-draft-zhu-cats-metric-semantics
Discussion:
- Longfei Xia presented on the challenge of dealing with outdated metrics at the decision point. The draft proposes three operational semantics: freshness (temporal confidence), operational acceptability (what the consumer is allowed to do with the metric), and assurance exposure (operational visibility).
- Q&A: Kehan Wang suggested integrating the freshness concept into draft-ietf-cats-metric-definition under the operational considerations section. Adrian Farrel supported the concepts introduced in the draft.

Decisions and Action Items

YANG Doctor Review: The chairs agreed to request a YANG Doctors review for draft-yl-cats-data-model-07 to validate its structure and syntax.
Incorporate Operational Metrics/Semantics: Authors of draft-ietf-cats-metric-definition will work with the authors of draft-zhangb-cats-service-metrics-op and draft-zhu-cats-metric-semantics to fold relevant operational/freshness concepts into the main metrics draft prior to starting Working Group Last Call.
Protocol Requirements Priority: The working group reaffirmed its commitment to focus on defining and agreeing upon protocol requirements within the CATS WG before pursuing solution drafts in downstream protocol working groups (e.g., IDR, PCE, LSR).

Next Steps

Metrics Draft: Publish a new revision of draft-ietf-cats-metric-definition addressing the OPS directorate review comments and integrating the operational policies discussed, preparing the document for WG Last Call.
OAM Framework: Refine draft-ietf-cats-oam-fw to address comments regarding localized probing, STAMP integration, and the scope of multi-domain fault isolation.
IETF 127 Interim Planning: Due to expected low physical attendance by the chairs and participants at IETF 127, the chairs will solicit feedback on the mailing list regarding holding two virtual interim meetings (one before and one after IETF 127) in lieu of a physical session in San Francisco.

Automatic IETF Minutes