Session Date/Time: 10 Nov 2023 08:30

cats

Summary

The cats (Compute-Aware Traffic Steering) working group met at IETF 118 to discuss use cases, requirements, framework architectures, and gap analysis related to compute-aware traffic steering. Presentations were given on an AI for Media use case from the BBC, exposing compute metrics, updates to the use case and requirements draft, an aggregation scheme for service metrics, and hybrid routing architectures. The working group discussed consolidating drafts and focusing discussion on the mailing list.

Key Discussion Points

AI for Media Use Case (BBC):
- Personalized media delivery requires efficient compute resource utilization.
- Migration of services between core, edge, and user devices based on compute availability and cost.
- Need for standard metrics for compute resource availability, utilization, and network performance.
- Discussion on whether metric sharing would be in-band or out-of-band.
Exposure of Compute Metrics:
- Presented the idea of exposing communication and compute information for infrastructure as a service deployment and selection.
- Identified two service lifecycle stages: service deployment and service selection.
- Questions raised about the viability of exposing compute metrics, gaps in the service lifecycle, defining a common metric set, and where this work should be done.
- Discussion on a unified computing metric model for deployment and traffic steering.
Use Cases and Requirements Draft Updates:
- Introduction rewritten, terminology aligned with architecture document.
- Service instance selection requirements updated with timeout limitation, availability determination, and contention resolution.
- Requirements added for metric collection.
- Discussion on flexible metric definition and interoperation with nodes that do not support specific metrics.
- All instance affinity terms were unified.
- Computing aware AI loss model use case added with three scenarios: cloud edge co-inference, cloud-edge-device co-inference, and edge-only inference.
Aggregation Scheme for Service Metrics:
- Focused on use cases and problem statements, including a computing-aware AR/VR scenario.
- Identified potential problems related to diverse and dynamic computing-related metrics.
- Proposed an aggregation scheme to reduce the amount of data required and the frequency of updates.
Hybrid Routing Architecture:
- Compared centralized and distributed routing models.
- Proposed a hybrid architecture with cloud management platform, network controller, and both distributed and centralized routing decisions.
- Recommendation to utilize an easier and less costly approach to compute information awareness.

Decisions and Action Items

ACTION: BBC (Rajeev Ramani and Dan King) to share information about the AI for Media use case on the mailing list.
ACTION: Authors of different drafts with architecture proposals, requirements, and gap analysis to work together to consolidate the work into a single or fewer drafts. Chairs offered assistance in facilitating these discussions.
ACTION: Review the recent copy of the LDBC CAPT framework draft and relate service instance work to the Service Contact Instant concept.
ACTION: Update gap analysis to avoid introducing new requirements and align with the use cases and requirements draft.

Next Steps

Focus discussion on the mailing list to facilitate collaboration and consolidate drafts.
Merge drafted if possible.
Address outstanding comments in the use cases and requirements draft, focusing on the correspondence between use cases and requirements, metric definition, and security considerations.
Evaluate the two liaison communications from ITUT.
Chairs will generate responses as necessary.