Sender-driven congestion control

Homa: The End of TCP for AI Clusters — John Ousterhout, Stanford

John Ousterhout

AI inference networking · GPU synchronization and tail latency · Incast