Skip to content

docs: update real-world findings — grpc-go, sql.DB benchmarks, abandoned CLs - #49

Open
gidotencate wants to merge 1 commit into
mainfrom
docs/real-world-findings-update
Open

docs: update real-world findings — grpc-go, sql.DB benchmarks, abandoned CLs#49
gidotencate wants to merge 1 commit into
mainfrom
docs/real-world-findings-update

Conversation

@gidotencate

Copy link
Copy Markdown
Owner

Summary

  • Add grpc-go clientStream to summary table (288B→256B sizeclass reduction, PR #9281 pending)
  • Add full grpc-go section with sizeclass context and PR status
  • Add database/sql.DB benchmark numbers: 7–25× improvement under 1–4 concurrent stressors (CL 767580 pending)
  • Mark net/http.Transport (CL 767600) as abandoned: end-to-end RoundTrip benchmarks showed ~28 µs network round-trip completely dominates mutex acquisition cost; no measurable production impact
  • Mark net/http.response (CL 767581) as abandoned: Go team policy (code clarity over byte savings); sizeclass savings were confirmed (232B→216B, different allocator classes) but synthetic benchmarks only showed 1–5%
  • Add "When findings are actionable" guidance: I/O-dominated paths (HTTP, DB wire protocol) dwarf cache-line benefits; findings matter most on in-memory connection pools, scheduler queues, and CPU-bound hot paths

Test plan

  • Verify all CL/PR links resolve correctly
  • Confirm benchmark numbers match what was posted on Gerrit CL 767580

…ndoned CLs

- Add grpc-go clientStream to summary table (288B→256B, PR #9281 pending)
- Add grpc-go section with full sizeclass context
- Add sql.DB benchmark numbers (7–25× under 1–4 concurrent stressors, CL 767580)
- Mark net/http.Transport (CL 767600) abandoned: end-to-end benchmarks showed
  ~28 µs round-trip dominates mutex cost; no measurable production impact
- Mark net/http.response (CL 767581) abandoned: Go team policy (code clarity >
  byte savings), confirmed sizeclass savings, but only 1–5% in synthetic benchmarks
- Add "When findings are actionable" note: I/O-dominated paths (HTTP, DB wire
  protocol) dwarf cache-line benefits; findings matter most on in-memory pools
  and CPU-bound hot paths
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