The team is discussing the use of the ALTO framework to enable tenants to perform request routing by mapping subscribers to cache locations, while addressing network operator concerns regarding security, network topology exposure, and the poor historical reputation of ALTO deployments.
Action Items
- Yoav: Merge the two request routing problem statement documents into one
- Alan: Start Foundation Document v1.1 capturing the hotel/network operator model discussion
- Chris and Glenn: Socialize the ALTO subscriber-to-cache mapping proposal internally with edge routing experts
- Alan: Add geo-based routing as a supported (but scoped) option in the problem statement
- Ensure problem statement comments are reviewed before the next meeting in Rennes
Antitrust Statement and Group Discussion
Document consolidation:
- Yoav proposed merging the two request routing documents into one (two chapters), keeping the speccing approach document separate as it remains a living document.
- Alan agreed on merging to reach three finalized problem statement documents and suggested starting a new Foundation Document v1.1 to capture the hotel/network operator discussion.
ALTO framework — security and exposure concerns:
- Chris raised that querying subscriber-to-cache mappings one-by-one effectively reconstructs the full network map, making ALTO’s selective exposure moot.
- Glenn confirmed operators’ two core concerns: (1) network topology as an attack vector, and (2) competitive intelligence (sizing/resource allocation visibility).
- Glenn added a third concern: operators need to preserve flexibility for dynamic routing and topology changes.
- Alan countered that ALTO here only exposes subscriber-to-cache mappings (not subscriber-to-subscriber), and that publicly known IP ranges limit the sensitivity of the disclosure.
ALTO deployment reputation:
- Glenn warned that the name “ALTO” alone triggers rejection from senior operator leadership, citing its historically failed RFC deployment.
- Chris agreed ALTO is not being used idiomatically and that the team inherits its negative reputation without necessarily needing all its complexity.
- Alan suggested branding it as an “SVTA Request Routing Service” referencing the RFC number to avoid triggering resistance.
DNS and IP exposure paths:
- Alan walked through three resolution methods (protocol/IP, DNS, ECS subnet); concluded that DNS does not solve the exposure problem because ~50% of US queries use public DNS with ECS /24 subnets, still revealing network structure.
- Mitigation proposed: selectively deny mapping access to untrusted tenants; trusted tenants get full mapping, others are routed by the operator on their behalf.
Geo-based alternative (Sanjay’s proposal):
- Sanjay asked whether tenants could use commercial geo-DBs (e.g., MaxMind) with only cache IP/location shared, eliminating the need for ALTO.
- Alan acknowledged this works at metro/core granularity but fails at the edge, where S1 and S2 may share the same geo qualifier and finer-grained IP mapping is required.
- Yoav challenged whether MaxMind is reliable enough for deep-edge deployments.
BGP vs. ALTO:
- Yoav noted that Netflix/Google demand BGP stream access for similar routing decisions today.
- Alan argued ALTO is superior to direct BGP exposure because it allows policy enforcement on permissible pairings, whereas raw BGP gives tenants uncontrolled network visibility.
Planning use case:
- Yoav reminded the group that mapping data is also needed for the planning phase — tenants deciding where to deploy workloads based on expected user volumes and capacity.
- Alan elaborated: tenants need to query “if I deploy at E1, how many users do I serve?” and check inventory (RAM, egress capacity) before committing.
Takeaway / Understanding
The core tension is unresolved: enabling tenant-driven request routing inherently requires exposing subscriber-to-cache mappings, and no technical mechanism (DNS filtering, selective ALTO queries, geo abstraction) fully eliminates that exposure at the edge — the real lever is trust-based access control per tenant. ALTO’s architecture fits ~80% of the problem, but its reputational baggage may require rebranding and scoping to avoid deployment rejection.
Next Meeting: Thursday, September 10th at 9am PDT.