The meeting focused on finalizing problem statements for orchestration, the common API framework, and request routing, while evaluating the use of the ALTO protocol for network information sharing and defining an inventory model for compute resources.
Â
Action Items
- Alan: Clean up and post a new version of the Request Routing problem statement
- Chris: Review and comment on the Orchestration and API Framework documents
- All: Finalize all problem statements before the upcoming Rennes face-to-face meeting
Antitrust Statement and Group Discussion
- Orchestration & API Framework: Alan detailed the need for a neutral approach supporting both containers and virtual machines to ensure portability across different Host Service Providers (HSPs). He emphasized a common framework for identity and access, service discovery, and compute inventory to treat HSPs as cloud regions.
-
- Chris noted that using ALTO provides a mature vehicle that avoids specifying many details, though some industry players may be "allergic" to it
- Chris and Alan discussed the security risk of network map disclosure; they concluded that while tenants can potentially scrape network info, the risk is inherent to any routing solution
- The team discussed handling non-publicly routable subnets (e.g., CGNAT), concluding that the client, cache, and request router must all reside behind the same horizon for this to be feasible
- Inventory Model: Alan proposed a model to track static attributes (hardware types, provisioned capacity) versus dynamic availability. Chris cautioned that "static" data can become dynamic at scale (e.g., hardware failure) and suggested the model should focus on what is "on the menu" rather than real-time availability
- Logistics: Alan noted a shift from alternating bi-weekly topics to a flexible, week-to-week agenda to better accommodate participant availability
Â
Takeaway / Understanding
The group is leaning toward using ALTO for request routing due to its maturity and implementation availability, despite potential privacy concerns regarding network disclosure. They have established a clear distinction between inventory (static capabilities) and reservation/availability (dynamic state) for compute orchestration.
Â
Next Meeting: Â Â Thursday, September 17th at 9am PDT.