vinext research · claim 3
From 32 to 128 in flight, vinext's rate moves −2.2% (446.9 → 436.9 rps). Next's rate moves −29.0% (375.3 → 266.6) and its p99 goes from 130.7 ms to 1,914.7 ms. Every run dropped zero requests.
Out of scope. Not issue #933. No 6 replicas, no nginx, no k6, no 1000 rps target. Servers ran one at a time.
vinext serves +19.1% more requests per second and the median is −19.0 ms. The p99 goes the other way (+26.2 ms), on a +22.8% larger body.
375.3 rps
446.9 rps
vinext is +63.9% on rate. The median is −84.7 ms and the p99 is −1,343.8 ms against Next — 3.35× tighter — on a +22.9% larger body.
266.6 rps
436.9 rps
| Framework | In flight | rps | p50 | p99 | ok | dropped | body | window |
|---|---|---|---|---|---|---|---|---|
| Next.js | 32 | 375.3 | 79.9 ms | 130.7 ms | 7,520 | 0 | 10,880 B | ≈20.0 s |
| vinext | 32 | 446.9 | 60.9 ms | 156.9 ms | 8,956 | 0 | 13,358 B | ≈20.0 s |
| Next.js | 128 | 266.6 | 347.8 ms | 1,914.7 ms | 5,364 | 0 | 10,880 B | ≈20.1 s |
| vinext | 128 | 436.9 | 263.1 ms | 570.9 ms | 8,778 | 0 | 13,371 B | ≈20.1 s |
Paste JSON printed by load.mjs, or a file with a runs array. Nothing leaves this browser.
The harness hits ten paths, round-robin, with Node fetch. The “10s abort” in the method line is the per-request timeout (10_000 ms in load.mjs), not the length of the run. Completed requests divided by rps lands on ≈20.0–20.1 s, which is that script's default duration.
//products/products/1/blog/blog/hello/dashboard/dashboard/analytics/docs/getting-started/settings/profile/aboutAverage body size stays put when concurrency changes (Next 10,880 B at both widths; vinext 13,358 B then 13,371 B), so the extra ~23% is payload, not queueing. The captured dashboard HTML snapshots differ by one byte — the extra weight is elsewhere in the mix. Bars use the worse of the pair as 100%. Lower is better for p50, p99, and body. Higher is better for rps.