Onboarding session and progress recap for the EDGE Working Group, focused on building a technical framework for hosting third-party cache workloads at distributed deep edge locations close to end users, including off-net options.
Action Items
- Circulate Rennes session slides to all participants (Alan)
- Arrange access to Edge Working Group Google Drive for new members (Florencia)
- Progress D2 spec (tenant-driven routing with HSP guidance interface) as priority deliverable (Working Group)
- Organize a dedicated BGP-at-the-edge feasibility session with BGP experts (Glenn to arrange)
- Publish updated Foundation Spec covering: network operator information-sharing confidentiality, tenant traffic attribution/incident response, and dynamic compute capacity handling
- Document Kubernetes orchestration architectural options to supplement the existing compute orchestration problem statement
Antitrust Statement and Group Discussion
Alan provided a full recap of the working group’s scope and progress:
- Three parties defined in the architecture: Tenant (cache workload owner), Hosting Service Provider (HSP) (compute infrastructure), and Network Operator (access network).
- Hosted Caches Foundation Spec published in April outlines motivation, architecture, and main workflows.
- Three problem statement areas identified: Request Routing, Service Architecture, and Compute Orchestration.
- Request Routing covers two models: HSP delegates routing on behalf of tenant (D1), or tenant does its own routing with HSP/network operator sharing guidance (D2); D2 work is already underway.
- Existing geo/BGP-based routing breaks down at sub-metro density (e.g., 10+ cache sites in one city), necessitating new integrations.
- Compute Orchestration will adopt Kubernetes first, with OpenStack/VMs also in scope; scaffolding (SSH keys, inventory, authentication) around cluster setup is part of the problem.
- Key design principle: multi-tenancy and elasticity — operators can host multiple tenants dynamically, scaling capacity during events rather than relying on static dedicated boxes.
- Differentiation from Open Caching: Edge WG hosts the tenant’s own workload/protocol stack directly; Open Caching redirects traffic to operator-managed caches.
Kathyayani asked whether compute orchestration scope extends to non-cache workloads (e.g., AI inference):
- Alan confirmed the framework is theoretically broader but practically scoped to media delivery and live streaming first.
Suresh raised questions about deep-edge ISP traffic routing using ASN/prefix tables and GPU/AI workloads at the edge:
- Alan clarified the distinction from Open Caching and confirmed GPU/AI inference is not in current scope; AI is characterized as a latency problem, not a bandwidth problem on broadband networks today.
Yasser asked about AI traffic bursts; Kathyayani and Suresh noted VLMs and video AI as emerging traffic sources to watch.
Takeaway / Understanding
The Edge Working Group has a clear near-term roadmap anchored in live streaming use cases, with three defined problem areas progressing at different rates. AI/GPU workloads and broader compute use cases are acknowledged as future possibilities but are explicitly deferred to keep scope tight and delivery timely.
Next Meeting: Thursday, October 29th at 9am PDT.