Evolving Meta's Edge Architecture
summary
Meta evolves a globally distributed Edge into a decentralized multi‑tenant compute platform that places GPUs and NVMe closer to users, claiming global average user‑to‑edge latency under 40 ms and a fleet scale of 100,000 GPUs and over 7,000 clusters. The team builds layered scheduling (targeting + resource-aware subfiltering), a demand forecasting pipeline and an edge autoscaler, enforces ephemeral hosts and network isolation, and reports reduced backbone traffic while enabling low‑latency AI inference, gaming and metaverse rendering.
AI-written summary · about
chapters (6)
- 0:00Introduction to Meta's edge
- 2:46Edge architecture overview
- 6:39Evolving edge requirements
- 10:01Edge compute use cases
- 13:02Orchestration and security
- 18:10Future hardware and summary
related talks
18mThe AI-First Data Infrastructure
29mEnhancing AI with Edge Computing: A Leap Towards Affordable Performance
47mEnergy Proportional Edge Computing Infrastructure
30mDesigning a Hybrid AI Platform on Kubernetes
16mKeynote: First Generation of Meta’s Datacenter Accelerators
23mKeynote
01mEdge Computing
28mSmall Changes, Big Results