The Power of Profiling
summary
CPU profiling (works with perf) reveals heavy kernel time caused by block_in_place on every task, which shuffles tasks across threads and spends cycles parking and unparking them. Refactoring to keep blocking work off async futures and treating futures as async-only cuts average write latency by 30%, shown on a columnar database ingesting observability data. Use CPU profiler to find syscalls-heavy hotspots and prefer dedicated blocking threads or true async alternatives. CPU profiling (perf compatible)
AI-written summary · about
related talks
05m“Using async-profiler to Find Tail Latency Issues”
55mThe performance Addict's Toolbox
01h 06mUtilize your CPU power
27mImproving Rust Performance Through Profiling and Benchmarking
01h 03m...Scaling Visualization in concurrent C++ programs
59mParallel Programming in Rust: Techniques for Blazing Speed
01h 09mComputer Architecture, C++, and High Performance
22mUsing Tokio for CPU-Bound Tasks (Works Really Well)