Skip to content

README, BUILDING, API reference + benchmark numbers - #12

Merged
AgentDudu merged 1 commit into
developfrom
feature/docs-and-benchmark
Oct 20, 2026
Merged

AgentDudu merged 1 commit into
developfrom
feature/docs-and-benchmark

Conversation

@AgentDudu

Copy link
Copy Markdown
Owner

What

  • README.md (new): what works, quick start (Linux + MinGW + MSVC), limits table (431/413/505 mapping verified against connection.cpp), layout, verification matrix.
  • docs/BUILDING.md (new): toolchains, presets, all four gates, Docker/Git-Bash/TSan-ASLR/clang-format gotchas recorded.
  • docs/API.md (new): header-exact reference per layer — parser usage loop, ownership rules, threading model (one connection ⇨ one reactor, never shared), error-handling conventions.
  • tools/bench-inside.sh: reproducible numbers (Release build, wrk + ab).

Numbers (Ubuntu 24.04 container, gcc 13 -O2, server shares host with clients)

Scenario 1 reactor 4 reactors
1 KiB keep-alive (wrk -t4 -c100) 9,770 req/s · p99 19.8 ms 47,275 req/s · p99 4.3 ms
64 KiB keep-alive 5,683 req/s · 356 MB/s 25,919 req/s · 1.59 GB/s
1 KiB fresh conns (ab -c50) 8,027 req/s 8,443 req/s

Honest gap

Windows numbers were not recorded — no wrk/ab on the Windows box; PLAN.md records the deviation from the original acceptance criterion ("both platforms") rather than faking it.

Benchmarks recorded on Linux (tools/bench-inside.sh): 1->4 reactors scale
9.8k -> 47.3k req/s keep-alive on 1 KiB; 64 KiB serves 1.59 GB/s; fresh
connections acceptor-bound at ~8.4k req/s. Windows numbers not recorded
(no wrk/ab on the Windows host) - deviation noted in PLAN.md.
@AgentDudu
AgentDudu merged commit 7dc4c44 into develop Oct 20, 2026
4 checks passed
@AgentDudu
AgentDudu deleted the feature/docs-and-benchmark branch October 20, 2026 18:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant