Add benchmark suite comparing sarin, wrk, and bombardier

benchmark/README.md

@@ -0,0 +1,42 @@
+# Benchmark
+
+Compares [sarin](https://github.com/aykhans/sarin), [wrk](https://github.com/wg/wrk), and [bombardier](https://github.com/codesenberg/bombardier) against a minimal C HTTP server using epoll.
+
+## Requirements
+
+- `sarin`, `wrk`, `bombardier` in PATH
+- `gcc`
+
+## Usage
+
+```bash
+./benchmark/run.sh
+```
+
+Configuration is at the top of `run.sh`:
+
+```bash
+DURATION="30s"
+CONNECTIONS=(50 100 200)
+ITERATIONS=3
+```
+
+## Structure
+
+```
+benchmark/
+  run.sh          - benchmark script
+  server/
+    server.c      - C epoll HTTP server (returns "ok")
+  results/        - output directory (auto-created)
+```
+
+## Output
+
+Each run produces per-tool files:
+
+- `*.out` - tool stdout (throughput, latency)
+- `*.time` - `/usr/bin/time -v` output (peak memory, CPU time)
+- `*_resources.csv` - sampled CPU/memory during run
+
+A summary table is printed at the end with requests/sec, total requests, elapsed time, and peak memory.