benchmarks.md

Benchmarks {#benchmarks}

I use google benchmark to benchmark the performance of the library. Just copy and change the compare.py script and gbench directory to compare the benchmark results.

http server: xsl::asio vs standalone asio

Coroutine HTTP server vs a hand-tuned asio reference, same load generator: parity at 1 connection (4 µs p50), ~3x throughput at 16 connections (SO_REUSEPORT multi-poller).

spsc