The Edge Working Group call focused on advancing request routing specifications for hosted edge caching environments, with primary emphasis on live event delivery. Key discussions centered around network data exposure requirements, DNS routing challenges at the edge, and the balance between tenant-owned versus hosting system provider (HSP)-owned request routing models. The group addressed critical technical issues including subnet-to-location mapping, capacity sharing, DNS granularity limitations, and the inadequacy of third-party geolocation databases for edge network routing. The working group operates on two parallel tracks—orchestration and request routing—with weekly alternating sessions to advance both specifications.
Action Items
- Follow up with JT on sending out the membership survey
- Find replacement co-chair for the Edge WG (Alan expressed unofficial candidacy)
- Alan to add commentary on DNS routing challenges and reference Brian’s request routing strategies document
- Continue work on request routing problem statement document
- Next week’s session: orchestration discussion (same time)
Antitrust Statement and Group Discussion
- Project Scope and Use Cases
- Primary focus on live event delivery to handle traffic surges
- Enabling content providers, CDNs (Netflix, YouTube) to run caches at network edge in hosted environments
- Initial focus on live events, but evaluating VOD and broader use cases for future expansion
- Solution designed to support multiple tenants running workloads close to end users
- Parallel Work Streams
- Two focus areas: orchestration and request routing
- Orchestration: addressing instantiation, planning, and deployment of cache instances across edge locations
- Considering both Kubernetes and virtual machine orchestration solutions as reference platforms
- Request routing: defining how clients are directed to appropriate edge locations
- Weekly alternating sessions between the two domains
- Network Data Exposure Requirements
- End user footprints (subnet-to-edge-location mappings) must be shared by network operators
- Class C subnet granularity typically sufficient and reasonable for IPv4
- Privacy concerns, particularly in Europe, around sharing network topology data
- Sharing aggregated subnet info with general edge site locations (not GPS coordinates) considered reasonable
- Capacity information sharing discussed—tenant-specific capacity allocations preferred over exposing overall network capacity
- Network operators may provide different capacity values to different tenants without exposing total capacity
- Data sharing typically governed by business agreements between ISPs and tenants
- DNS Routing Challenges
- DNS granularity may be insufficient for deeply distributed edge deployments (cache per BRAS/CMTS/PGW)
- Corporate DNS servers often don’t propagate EDNS Client Subnet (ECS) information
- Some networks use geographically distant DNS resolvers for load balancing, making resolver location unreliable for user location inference
- Network operators must indicate whether DNS resolver location can be used as indicator for end user location
- For public DNS servers: ECS information is critical for routing decisions
- For operator DNS servers: either location affinity between resolvers and users must exist, or ECS must be provided
- Without proper DNS information or ECS, effective routing becomes impossible
- Tenant DNS servers must be able to see client IP addresses (load balancers sometimes obfuscate incoming end user IPs)
- Third-Party Geolocation Database Limitations
- Third-party databases (e.g., MaxMind) discussed as insufficient for edge network routing
- These databases may be extremely precise for geographic location but don’t account for network topology
- Network distance and geographic distance often differ significantly at the edge
- Using such databases can result in harmful routing decisions for network operators
- Network operator information exposure is necessary for proper edge routing
- Infrastructure and Security Constraints
- Anycast routing works well at central/regional locations but becomes complex or impossible in edge networks
- Not all networks support Anycast, particularly in L2 segment environments with non-routable IP addresses
- CDNs typically support both Anycast and DNS routing options
- Anycast cannot be offered as mandatory solution for all edge deployment cases
- BGP security concerns raised, particularly for bring-your-own-IP scenarios
- BGP integration manageable between HSPs and network operators (vendor relationships) but complex for third parties
- Request Routing Models
- Two models under consideration: HSP-owned request routing and tenant-owned request routing
- HSP-owned: orchestration layer owns routing with tenant providing guidance
- Tenant-owned: content publisher or CDN tenant is responsible for request routing
- Integration with existing technologies like content steering being evaluated
- Content steering may need enhancement to support edge use cases
- Business and Relationship Considerations
- Working group aims to stay focused on technical specifications, not business structures
- In content delivery context, tenant (CDN or content provider) holds relationship with end users
- Structures may vary: CDN serves content publisher who serves end users
- Network operator and tenant may both have relationships with end users, but tenant owns content delivery relationship
- PII and customer data considerations raised but left to business agreements
Organizational Updates
- Yoav is leaving Qwilt but will continue as its advisor for a while and will remain involved in the working group, however, we are looking for a new co-chair to replace him
- Alan expressed interest in stepping in