Updated benchmarks

[mcollina]

I was checking how better fetch() improved in Node v20.6.0, and I got this nice set of results.

Here are the results for Node v18 too, just in case you think they might be valuable for comparisons.

(1 connection)

│ Tests               │ Samples │        Result │ Tolerance │ Difference with slowest │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ http - no keepalive │      10 │  5.28 req/sec │  ± 2.68 % │                       - │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ http - keepalive    │      10 │  7.14 req/sec │  ± 2.76 % │               + 35.31 % │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ undici - fetch      │      25 │ 35.47 req/sec │  ± 2.69 % │              + 572.37 % │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ undici - pipeline   │      20 │ 49.14 req/sec │  ± 2.96 % │              + 831.45 % │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ undici - dispatch   │      15 │ 60.01 req/sec │  ± 2.91 % │             + 1037.61 % │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ undici - stream     │      40 │ 61.86 req/sec │  ± 2.93 % │             + 1072.67 % │
|─────────────────────|─────────|───────────────|───────────|─────────────────────────|
│ undici - request    │      25 │ 63.41 req/sec │  ± 2.93 % │             + 1101.90 % │

(50 connections)

│ Tests               │ Samples │           Result │ Tolerance │ Difference with slowest │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ undici - fetch      │      25 │  1805.62 req/sec │  ± 2.81 % │                       - │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ http - no keepalive │      15 │  2631.14 req/sec │  ± 2.85 % │               + 45.72 % │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ http - keepalive    │      30 │  4725.24 req/sec │  ± 2.95 % │              + 161.70 % │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ undici - pipeline   │      45 │  5573.65 req/sec │  ± 2.95 % │              + 208.68 % │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ undici - request    │      65 │  8122.34 req/sec │  ± 2.87 % │              + 349.84 % │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ undici - stream     │      70 │  9426.43 req/sec │  ± 2.92 % │              + 422.06 % │
|─────────────────────|─────────|──────────────────|───────────|─────────────────────────|
│ undici - dispatch   │      80 │ 10238.30 req/sec │  ± 2.90 % │              + 467.03 % │

[anonrig]

The ordering of the benchmark results are changing. Should we open a new pull request to persist the ordering?

[KhafraDev]

I don't think it matters necessarily. I am interested in what made fetch faster though

[metcoder95]

I'd love to see h2 once merged 😅

[jimmywarting]

I am interested in what made fetch faster though

dito

[rluvaton]

Can you please rerun the benchmark again with 20.8.0 release?

I did some improvement in web-streams performance (hope it will help here as it improved pipe to and async iterator in webstream and as far as I know, it's only been used in requests with body and non of the benchmark check that)