Skip to content

README.md

Latest commit

 

History

History
92 lines (68 loc) · 3.5 KB

README.md

File metadata and controls

92 lines (68 loc) · 3.5 KB

Benchmarks

This folder contains a tiny, allocation-free hot-loop benchmark client:

  • It opens one or more TCP connections.
  • It repeatedly writes a pre-rendered HTTP/1.1 request string (minimal request by default).
  • It discards a fixed number of bytes per response (no HTTP parsing in the hot loop).

zhttp (in-process)

zig build bench -Doptimize=ReleaseFast -- --mode=zhttp --conns=1 --iters=100000 --warmup=10000

Convenience runner (environment-driven defaults):

CONNS=1 ITERS=100000 WARMUP=10000 zig run benchmark/run_zhttp.zig

zhttp (external)

Runs zhttp-bench-server and benchmarks it via the same client in --mode=external:

zig run benchmark/run_zhttp_external.zig
# or
zig build bench

run_zhttp_external writes benchmark/results/bench_latest.json and benchmark/results/bench_latest.md, and refreshes the README fetch section.

FaF (external)

FaF is benchmarked against the faf-example app (/plaintext on port 8080). zig run benchmark/run_faf.zig clones both faf-example and faf into .zig-cache/faf-example and .zig-cache/faf, pins FaF to the revision in Cargo.lock, and applies a tiny compatibility patch (for newer Rust nightly intrinsics) before building. It expects zhttp-bench to be available; set BENCH_BIN to a built bench binary path or build it first.

BENCH_BIN=./zig-out/bin/zhttp-bench zig run benchmark/run_faf.zig

Compare (external)

Runs both servers with the same benchmark client settings (defaults: FULL_REQUEST=1, CONNS=16, REUSE=1):

  • identical host/path/conns/iters/warmup/full_request/reuse
  • fixed response bytes discovered twice per target and pinned for the timed run
zig run benchmark/run_compare.zig
# or
zig build bench-compare

run_compare writes:

  • benchmark/results/latest.json
  • benchmark/results/latest.md

and refreshes README comparison/fetch sections.

To point at an already running server:

zig build bench -Doptimize=ReleaseFast -- --mode=external --host=127.0.0.1 --port=8080 --path=/plaintext

Notes:

  • --host is currently expected to be an IPv4 literal.
  • If --fixed-bytes is not provided, the benchmark auto-discovers Content-Length once (outside the hot loop).
  • run_zhttp_external.zig and run_faf.zig both validate response size stability by discovering fixed bytes twice before timing.
  • run_compare.zig now uses a dedicated FaF no-reuse example template so the no-reuse comparison matches zhttp's close-response byte count too.
  • Use --full-request to send Host: and Connection: headers (default is the minimal GET ...\r\n\r\n request).
  • --reuse=1 (default) reuses one keep-alive connection per worker; use --reuse=0 / --no-reuse to reconnect every request.
  • zhttp-bench-server uses cpu_count permanent workers in both reuse modes.
  • For zhttp server config, prefer .temp_worker_spawn = .polling for keep-alive-heavy (--reuse=1) traffic and .temp_worker_spawn = .signaled for reconnect-heavy (--reuse=0) traffic.
  • For the helper scripts, set FULL_REQUEST=1 to pass --full-request:
    • FULL_REQUEST=1 zig run benchmark/run_zhttp_external.zig
    • FULL_REQUEST=1 zig run benchmark/run_faf.zig
  • For helper scripts, set REUSE=0 to disable reuse:
    • REUSE=0 zig run benchmark/run_zhttp_external.zig
    • REUSE=0 zig run benchmark/run_faf.zig

perf helper

run_perf.zig replaces the old shell helper and records/report profiles by driving zhttp-bench:

MODE=zhttp CONNS=1 ITERS=100000 WARMUP=10000 zig run benchmark/run_perf.zig