Live Q+A 1
summary
Panel argues that traditional CDNs remain but AI shifts work to compute‑heavy edge nodes with many GPUs and less emphasis on SSD/NVMe caching; operators deploy GPU virtualization, overlay networks and caps for tenancy, and lean on RDMA/offload techniques reused from HPC. Key metrics: target 40 ms average round trip to edge, sometimes needing ~15 Mbps at ~30 ms per user. Power, orchestration and mismatched collectives vs fabric remain the practical bottlenecks.
AI-written summary · about
chapters (13)
- 0:00AI and immersive network evolution
- 2:11CDN versus edge computing for AI
- 4:19HPC and AI cross-layer optimization
- 6:44Backbone requirements for AI workloads
- 9:34High-level vs low-level ML APIs mismatch
- 12:42Security challenges at the edge
- 15:51AI inference scalability challenges
- 18:42Power constraints in AI clusters
- 21:08AI workload variety on backbone
- 23:18HPC and data center networking convergence
- 26:08Mobile edge vs AI edge computing
- 28:38One-word wishlist for network challenges
- 31:06Closing remarks
related talks
13mLive Q&A Session #2 – Shashi Gandham
07mWelcome Remarks
15mLive Q&A with Speakers – Joseph Provine
23mLive Q&A with Speakers – James Zeng
31mLive Panel: Scale-Out Networking for AI Supercomputers
18mArchitecting Multi-tenant Data-center Networks for GPU Customers
27mLIVE Q&A #4
17mMeta’s Network Journey to Enable AI