☂️ OYB Next Steps

❔ ❔ ❔ Curious what OYB is? ❔ ❔ ❔

Summary I did a quick repurposing of DZL to analyze where our time is going in aggregate. This is the top 30 average timings across ~400 runs today

timing-gather-loadblank                 1103.33
timing-gather-afterpass-accessibility    827.09
timing-gather-afterpass-optimizedimages  825.39
timing-gather-afterpass-fontsize         797.62
timing-init-config                       779.76
timing-gather-gettrace                   697.89
timing-audit-speed-index                 578.59
timing-gather-getbenchmarkindex          510.86
timing-audit-screenshot-thumbnails       500.83
timing-config-requireaudits              472.83
timing-gather-afterpass-starturl         248.76
timing-audit-unminified-javascript       243.60
timing-audit-metrics                     241.27
timing-gather-loadpage-offlinepass       234.72
timing-gather-loadpage-redirectpass      230.89
timing-config-requiregatherers           214.00
timing-gather-afterpass-scripts          170.01
timing-gather-afterpass-cssusage         165.33
timing-gather-afterpass-metarobots       146.49
timing-gather-afterpass-offline          122.49
timing-gather-afterpass-crawlablelinks   116.67
timing-audit-estimated-input-latency     116.23
timing-audit-first-contentful-paint      112.46
timing-audit-load-fast-enough-for-pwa    101.72
timing-audit-is-on-https                  90.27
timing-audit-unminified-css               86.65
timing-audit-offscreen-images             81.61

Looking closer at the top 10…

1. Load blank. #6369 helped, but it’s still a bigger impact to our runtime on average than any of our audits 😱, we might want to experiment with shorter blank times until we can revive data URIs (#4310)
2. axe. Not much we can do here immediately, I’ve already filed a couple issues on some of the worst offenders. Might indicate to us it’s worth investing a bit of resources to help improve their perf story?
3. OptimizedImages. We’ve already done the biggest thing here by moving it to Chrome side. Only obvious work I see here would reduce our functionality or move more into estimation/heuristics.
4. FontSize. @Hoten is already on the case!!
5. Init config. This is surprising. I’m on it!
6. Get trace. Already aware of it and on it #3596
7. Speed Index. @brendankenny was on it with wasm and I have some more ideas if that isn’t good enough.
8. Get Benchmark Index. We should move to running this only in #6162 settings.
9. Screenshot Thumbnails. Not a ton we can do here, maybe look into wasm or native browser resizing?
10. Require audits. Also a little surprising. I’m not sure there’s any obvious low hanging fruit, it’s probably just the evaluation/parsing of every script we have.

Summary of Near-term AIs:

• Shorter load blank times @patrickhulce (#6446)
• Reduce font size time @Hoten (#6391)
• Reduce time spent in init:config @patrickhulce (#6445)
• Further reduce trace size @paulirish ???
• Speed Index decoding moved to wasm @brendankenny
• Don’t run benchmark index if not automatic timing @patrickhulce (#6162, pending)

[paulirish]

I think you’re remembering what @patrickhulce wrote and actually landed 😃

Ha! Apparently so! 😃

aXe

I found one recent win that’s pretty easy: https://github.com/dequelabs/axe-core/pull/1172 (discussion has led to a diff solution, but also straightforward).

After this though is probably either table stuff or color contrast. Both rather spooky. 🎃

[patrickhulce]

Revisiting this ~20 months later, I think we’ve addressed the key AIs here already and much has changed since that analysis. If run time becomes an important issue again we can do another round of investigations and takeaways.