Skip to content

Add a headless 16x run of every framework with 25 samples, Chrome pinned to one CPU - #127

Closed
NullVoxPopuli-ai-agent wants to merge 2 commits into
NullVoxPopuli:mainfrom
NullVoxPopuli-ai-agent:nvp/results-headless-16x-25
Closed

NullVoxPopuli-ai-agent wants to merge 2 commits into
NullVoxPopuli:mainfrom
NullVoxPopuli-ai-agent:nvp/results-headless-16x-25

Conversation

@NullVoxPopuli-ai-agent

Copy link
Copy Markdown
Contributor

Adds results/public/results/9.json: every framework, headless, 16x CPU throttle, 25 samples, with Chrome pinned to one CPU.

pnpm bench --framework=all --bench=all --headless --cpu-throttle=16 --count=25 --timeout=300000 --yes

The run

  • Every Chrome process (browser, renderer, GPU, workers) ran on CPU 4 for every bench. A google-chrome wrapper on PATH ran taskset -c 4. The runner and the builds ran on CPUs 0-3 and 5-7.
  • All 11 frameworks finished. Each bench has 25 samples, and every sample reached :done.
  • The run took 3.6 hours. The 1-minute load average stayed at 3.6 or below on 8 CPUs.
  • TMPDIR was a folder in /dev/shm. Other jobs filled the root disk, and puppeteer's --disable-dev-shm-usage puts Chrome's shared memory in TMPDIR.

Runner fix

page.goto and page.waitForNetworkIdle used puppeteer's 30 s default instead of --timeout. The solid-1 fan-out "single burst" bench blocks the main thread for about 42 s in one task. Chrome cannot report the last network response during that task, so the run stopped with Timed out after waiting 30000ms. Both waits now use --timeout. This changes when the runner gives up, not what it measures.

The result file records the sha of that commit.

Why --timeout=300000

React "Incrementing Render Effect" and the solid-1 fan-out benches take 25 to 51 s per sample at this setting. One react sample went past the 60 s default.

🤖 Generated with Claude Code

NullVoxPopuli and others added 2 commits September 29, 2026 23:51
goto and waitForNetworkIdle used puppeteer's 30 s default. At
--cpu-throttle=16 with Chrome on one CPU, the solid-1 fan-out
"single burst" bench blocks the main thread for about 42 s in one
task. Chrome cannot report the last network response during that
task, so waitForNetworkIdle stopped the whole run with
"Timed out after waiting 30000ms" on sample 23 of 25.

Both waits now use --timeout. This changes when the runner gives
up, not what it measures.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Result file 9.json. The run used one invocation:

  pnpm bench --framework=all --bench=all --headless \
    --cpu-throttle=16 --count=25 --timeout=300000 --yes

CPU pinning: every Chrome process (browser, renderer, GPU, workers)
ran on CPU 4 for every bench, through a google-chrome wrapper on
PATH that runs `taskset -c 4`. The runner and the builds ran on
CPUs 0-3 and 5-7, so they did not compete with Chrome.

Other notes:
- TMPDIR was a folder in /dev/shm, because other jobs filled the
  root disk and Chrome keeps its shared memory in TMPDIR.
- The 1-minute load average stayed at 3.6 or below on 8 CPUs.
- The run took 3.6 hours. React "Incrementing Render Effect" and
  the solid-1 fan-out benches take 25 to 51 s per sample, so the
  run needs --timeout above the 60 s default.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@vercel

vercel Bot commented Sep 30, 2026

Copy link
Copy Markdown

@NullVoxPopuli is attempting to deploy a commit to the NullVoxPopuli's projects Team on Vercel.

A member of the Team first needs to authorize it.

@vercel

vercel Bot commented Sep 30, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
rere-benchmark Ready Ready Preview Sep 30, 2026 1:45pm UTC

This branch was successfully deployed

1 active deployment
Preview — 3e984eb1 Deployed Sep 30, 2026 by vercel[bot]
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.

2 participants