chore: http/2 benchmark

[mkaufmaner]

Start the server:

PORT=3000 node --experimental-wasm-simd ./benchmarks/server-http2.js

Run the benchmark:

PORT=3000 ERROR_THRESHOLD=100 CONNECTIONS=10 PARALLEL=10 node --experimental-wasm-simd ./benchmarks/benchmark-http2.js 

Note: May have to run benchmark-http2.js multiple times, and in quick succession, in order to get a result.

Result:

│ Tests             │ Samples │          Result │ Tolerance │ Difference with slowest │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ http2 - request   │       1 │  449.75 req/sec │  ± 0.00 % │                       - │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - fetch    │       1 │  509.34 req/sec │  ± 0.00 % │               + 13.25 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - request  │       1 │ 1086.15 req/sec │  ± 0.00 % │              + 141.50 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - stream   │       1 │ 1168.15 req/sec │  ± 0.00 % │              + 159.73 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - pipeline │       1 │ 1200.61 req/sec │  ± 0.00 % │              + 166.95 % │

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.10 🎉

Comparison is base (e1d9717) 93.19% compared to head (7cd4203) 93.29%.

❗ Current head 7cd4203 differs from pull request most recent head 2e49522. Consider uploading reports for the commit 2e49522 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##            http2      #35      +/-   ##
==========================================
+ Coverage   93.19%   93.29%   +0.10%     
==========================================
  Files          41       41              
  Lines        3144     3147       +3     
==========================================
+ Hits         2930     2936       +6     
+ Misses        214      211       -3     

Impacted Files	Coverage Δ
lib/client.js	95.23% <100.00%> (+0.34%)	⬆️
lib/core/connect.js	80.00% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[metcoder95]

The results indeed looks promising, though if the streams are better managed, the things can go even better as we'll be reusing them to the same origin.
Let me fix some stuff to make it work with dispatch and focus on the performance improvements later on. Great work btw 🚀

[mkaufmaner]

The results indeed looks promising, though if the streams are better managed, the things can go even better as we'll be reusing them to the same origin.
Let me fix some stuff to make it work with dispatch and focus on the performance improvements later on. Great work btw 🚀

@metcoder95 Thank you! Looking forward to seeing the main PR in a mergable state. We 100% need more tests too especially around the closing of connections and sessions for the server and the client.

[metcoder95]

Hey @mkaufmaner, can you rebase with the branch and enable the benchmarks?
Feel free to open a new PR if the conflicts are complex to resolve 🙂

[mkaufmaner]

@metcoder95

Hey @mkaufmaner, can you rebase with the branch and enable the benchmarks? Feel free to open a new PR if the conflicts are complex to resolve 🙂

Merged down from your http2 branch, but I am not sure what you mean by "enable the benchmarks". Could you elaborate?

Secondly, I updated the benchmarks with allowH2: true, enabled the dispatch benchmark, and reran the http2 benchmark. I am still encountering the same issue before where sometimes the benchmark just doesn't print anything so you have to run it in succession to get a result.

Nonetheless,

│ Tests             │ Samples │          Result │ Tolerance │ Difference with slowest │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ http2 - request   │       1 │  459.30 req/sec │  ± 0.00 % │                       - │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - fetch    │       1 │  518.96 req/sec │  ± 0.00 % │               + 12.99 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - request  │       1 │  951.29 req/sec │  ± 0.00 % │              + 107.12 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - pipeline │       1 │ 1116.78 req/sec │  ± 0.00 % │              + 143.15 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - stream   │       1 │ 1169.42 req/sec │  ± 0.00 % │              + 154.61 % │
|───────────────────|─────────|─────────────────|───────────|─────────────────────────|
│ undici - dispatch │       1 │ 1230.43 req/sec │  ± 0.00 % │              + 167.89 % │

[metcoder95]

Could you elaborate?

Sure, I was meaning those that were commented out

Secondly, I updated the benchmarks with allowH2: true, enabled the dispatch benchmark, and reran the http2 benchmark. I am still encountering the same issue before where sometimes the benchmark just doesn't print anything so you have to run it in succession to get a result.

Those are good numbers, thanks! 🙇
I'm not sure what the error might be, I'm guessing that is due to how the request is done, as they differ from the way undici usually works, although not really important if we can get the numbers

Would you like to go further and do the cleanup to merge?

[mkaufmaner]

@metcoder95

Could you elaborate?

Sure, I was meaning those that were commented out

Secondly, I updated the benchmarks with allowH2: true, enabled the dispatch benchmark, and reran the http2 benchmark. I am still encountering the same issue before where sometimes the benchmark just doesn't print anything so you have to run it in succession to get a result.

Those are good numbers, thanks! 🙇 I'm not sure what the error might be, I'm guessing that is due to how the request is done, as they differ from the way undici usually works, although not really important if we can get the numbers

Would you like to go further and do the cleanup to merge?

Cleaned up the leftover comments so this should be ready to go.

[metcoder95]

Linter seems failing :(

[mkaufmaner]

Linter seems failing :(

Fixed the lint errors.