The Future of AI Networking Starts Here
summary
Demonstrates a full‑stack debugging workflow that links on-host traces with switch sFlow and hardware counters to pinpoint interconnect congestion, identifies specific bottlenecks (packet drops, PCIe stalls, NIC firmware stalls), and shows how targeted fixes—firmware patches, queue tuning, and flow steering—shift the bottleneck off the network. Covers tools used (telemetry collectors, programmable NIC tracing, RDMA counters), concrete failure modes observed in multi‑rack AI clusters, and which fixes measurably reduced tail latency.
AI-written summary · about
related talks
19mAI/ML Networking Challenges: The Fast and the Finicky!
25mResilience for AI Workloads at Scale: The Fast and the Finicky!
20mTurbocharging AI/ML workloads: Revving Up Speed and Resilience
30mThe Gashlycrumb Tinies of AI Networking You Must Know (or Languish!)
51mAdvanced approaches to traffic inspection & network diagnosis w/
07mWelcome Remarks
15mHostping: Diagnosing Intra-host Network Bottlenecks in RDMA Servers
08mKeynote: The Challenge and Opportunity to Build a Transparent Network