How to Diagnose Nanosecond Network Latencies in Rich End-Host Stacks

Explore a conference talk from NSDI '22 that delves into diagnosing nanosecond network latencies in complex end-host stacks. Learn about a novel latency diagnosis tool with full-stack coverage and low overhead, capable of identifying sources of latency in end hosts with nanosecond precision. Discover how the presented methodology reconciles CPU and NIC hardware profiling traces across multiple time domains to reconstruct network-message lifetimes. Gain insights into unexpected latency sources in both kernel and user-space stacks, and see a practical demonstration of the tool's capabilities in systematically identifying and removing performance overheads in memcached, resulting in a significant reduction of 99.9th percentile latency.