benchmarks

Benchmarks

Reproducible performance harness for Velocix. All numbers in docs/PERFORMANCE.md are produced by these scripts.

Requirements

• granian (the intended server)
• ab (ApacheBench) for HTTP load
• fast-query-parsers, orjson (runtime deps, already in requirements.txt)
• py-spy + sudo for the live process split

In-process framework floor (no sockets)

python3 benchmarks/profile_inproc.py

Live over granian (4 workers, c=100, best of 3)

bash benchmarks/run_granian.sh

Comparing against other frameworks

benchmarks/bench_compare/ is a full Locust harness comparing Velocix against Starlette, FastAPI, Litestar, Falcon, BlackSheep, and Sanic — all serving byte-identical responses on 4 routes, under granian 4 workers. Two profiles: realistic (100 users, think time) and saturation (500 users, no think time).

bash benchmarks/bench_compare/run_all.sh   # whole battery, ~20 min
bash benchmarks/bench_compare/run_sat.sh velocix
bash benchmarks/bench_compare/run_real.sh velocix