Session Date/Time: 22 Mar 2024 03:00

cats

Summary

The cats working group meeting focused on updates to existing drafts, particularly the use case and requirements document and the framework document. A significant portion of the meeting was dedicated to discussions on compute metrics and their distribution. Several new drafts were presented as flash talks, showcasing potential solutions for the CATS framework. The chair conducted polls to gauge attendee sentiment regarding flash presentations and the preferred method for progressing the metrics discussion.

Key Discussion Points

Use Case and Requirements Document:
- Updates included merging use cases and requirements from various contributors.
- Discussion on the definition of "service" in CATS versus SFC, clarifying the scope of the working group. CATS focuses on user traffic and client applications while SFC relates to operator-based services.
- Agreement on updating the computing service definition in the working group documents to clarify CATS' scope and relationship with SFC.
- Concerns raised about the collection of metrics from service instances and the need for aggregation.
Framework Document:
- Updates included merging content from a draft on awareness architecture.
- Clarifications on terms, definitions, and deployment models (distributed, centralized, hybrid).
- Discussion of where the service instance is positioned under the CATS architecture.
- Call for more reviewers and contributions to address pending issues.
Compute Modeling and Metrics:
- Emphasis on the need for simple and compatible computing metrics to facilitate joined routing decisions.
- Discussion of potential metrics (predicted computing delay, service capability, status indication) and optimization objectives (minimizing total delay, load balancing).
- Exploration of the service lifecycle and the need for understanding compute metrics for service deployment and selection.
- Concerns about potential for losing metrics relevant to overall compute/network interplay in efforts to simplify.
- Discussion on whether the metrics come from the edge into the network or from the network to the edge.
Metric Distribution:
- Analysis of metric distribution overhead, considering scope, frequency, and size.
- Discussion of optimization techniques such as notification domains, push/pull mechanisms, and caching.
- Formula presented for metric distribution overhead computation.
Solution Proposals (Flash Talks):
- Segment routing (SR) as a data plane option for CATS.
- Solutions for deploying multiple service contact instances behind a single CATS egress.
- Leveraging Application-aware Networking (APN) within the CATS framework.
- Utilizing IP Resource Saving (IPRS) encoding for CATS service mobility.

Decisions and Action Items

Action Item: Update the definition of "computing service" in working group documents. (Responsible: Authors of Use Case and Requirements Document, Framework Document. Due: ASAP).
Action Item: Authors of drafts on metric distribution and compute modeling to collaborate and revise WG documents (Responsible: Authors of these drafts. Due: Before IETF 120).
Action Item: The Chairs to determine how best to drive the discussion on Metrics given the preference indicated in the process poll.
Action Item: Jeff to provide pointers to related work happening elsewhere in IETF, either to the cats mailing list, or directly to the chairs. (Responsible: Jeff, Due: ASAP).

Next Steps

Authors to address pending issues and comments on the drafts.
Further discussion on the mailing list, particularly regarding the metric discussion.
Prepare new versions of the drafts incorporating feedback.
Chairs to decide on a strategy for progressing the metrics discussion (e.g., interim meeting, design team).