Each step was A/B tested against the commit just before it (requests/sec at 16 clients, median of 3–5 alternating reps). Every step had its own baseline, so the percentages don't multiply into the totals exactly. A dash means no notable change on that route.
| Change | Room | Messages | Sidebar | Search | Post |
|---|---|---|---|---|---|
No Rack::Deflater, Redis compress: false, no Rack::ETag, 4 workers (measured together) |
+56% | +52% | +19% | +31% | +19% |
| Puma → Falcon | +1% | +3% | +10% | +12% | +1% |
Static-file checks skipped + tags once per process + Current.account once (together) |
+10% | — | +14% | +14% | — |
| Per-process fragment copies + cache versions without a DB checkout (together) | +13% | +20% | +6% | +8% | — |
| Boost forms as strings | — | +2% | — | — | +8% |
Sec-Fetch-Site instead of CSRF tokens |
+14% | +16% | — | +9% | — |
| Keep each page's gzip by digest | +7% | +20% | +7% | — | — |
Read cache (PRAGMA data_version) |
— | +10% | +16% | +14% | — |
| Keep the finished sidebar | +227% | ||||
| Keep finished room / messages / search | +352% | +129% | +254% | ||
| Cookies only when they change | +17–23% on every read route | ||||
| Push job query fixes | +1% | ||||
Restore Rack::Deflater (needed for parity) |
−1–3% on every route |
From Falcon + config fixes to the final optimized app: room +742%, messages +416%, sidebar +491%, search +565%, post +12%. Almost all of the read gain comes from keeping finished pages until the next write, and the CSRF change is what made that possible.