bench: add websockets #3203

tsctx · 2024-05-05T10:36:15Z

Part of #3201

codecov-commenter · 2024-05-05T10:41:21Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 94.17%. Comparing base (5d54543) to head (53de211).
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #3203   +/-   ##
=======================================
  Coverage   94.17%   94.17%           
=======================================
  Files          90       90           
  Lines       24432    24432           
=======================================
  Hits        23009    23009           
  Misses       1423     1423

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

lpinca · 2024-05-05T11:27:45Z

To have reliable benchmarks you should use two processes, one for the server and one for the client and ensure that the server is faster than the client, otherwise you might end up benchmarking the server instead of the client. See the discussion in nodejs/node#50586.

KhafraDev

Going to block for the previous comments

benchmarks/websocket/websocket-echo.mjs

benchmarks/_util/websocket-simple-server.js

tsctx · 2024-10-01T12:00:19Z

> $ node ./benchmarks/websocket-benchmark.mjs
(node:9028) [UNDICI-WSS] Warning: WebSocketStream is experimental! Expect it to change at any time.
(Use `node --trace-warnings ...` to show where the warning was created)
undici [binary]: transferred 102.46MiB/s
undici [string]: transferred 99.43MiB/s
undici - stream [binary]: transferred 95.38MiB/s
undici - stream [string]: transferred 86.72MiB/s
ws [binary]: transferred 100.69MiB/s
ws [string]: transferred 95.50MiB/s

lpinca · 2024-10-01T16:07:02Z

@tsctx I'm a bit skeptical about the results

// server.js
const uws = require('uWebSockets.js');
const app = uws.App();

app.ws('/*', {
  compression: uws.DISABLED,
  maxPayloadLength: 512 * 1024 * 1024,
  maxBackpressure: 128 * 1024,
  message: (ws, message, isBinary) => {
    ws.send(message, isBinary);
  }
});

app.listen(8080, (listenSocket) => {
  if (listenSocket) {
    console.log('Server listening to port 8080');
  }
});

// ws-client.js
'use strict';

const { WebSocket } = require('ws');

const messages = +process.argv[2];
const payloadLength = +process.argv[3];
const data = Buffer.alloc(payloadLength, '_');

const ws = new WebSocket('ws://127.0.0.1:8080');

ws.binaryType = 'arraybuffer';

ws.on('open', function () {
  console.time(`${messages} messages of ${payloadLength} bytes`);
  ws.send(data);
});

let count = 0;

ws.on('message', function () {
  if (++count === messages) {
    console.timeEnd(`${messages} messages of ${payloadLength} bytes`);
    ws.close();
  } else {
    ws.send(data);
  }
});

// undici-client.js
'use strict';

const messages = +process.argv[2];
const payloadLength = +process.argv[3];
const data = Buffer.alloc(payloadLength, '_');

const ws = new WebSocket('ws://127.0.0.1:8080');

ws.binaryType = 'arraybuffer';

ws.addEventListener('open', function () {
  console.time(`${messages} messages of ${payloadLength} bytes`);
  ws.send(data);
});

let count = 0;

ws.addEventListener('message', function () {
  if (++count === messages) {
    console.timeEnd(`${messages} messages of ${payloadLength} bytes`);
    ws.close();
  } else {
    ws.send(data);
  }
});

$ node ws-client.js 100000 125
100000 messages of 125 bytes: 8.972s
$ node undici-client.js 100000 125
100000 messages of 125 bytes: 9.446s

$ node ws-client.js 100000 1024
100000 messages of 1024 bytes: 9.474s
$ node undici-client.js 100000 1024
100000 messages of 1024 bytes: 10.430s

$ node ws-client.js 100000 262144
100000 messages of 262144 bytes: 1:33.381 (m:ss.mmm)
$ node undici-client.js 100000 262144
100000 messages of 262144 bytes: 3:06.370 (m:ss.mmm)

This is without binary addons.

KhafraDev · 2024-10-01T16:22:53Z

Same here, but I'm happily surprised undici is that close in both benchmarks (other than very large messages).

Uzlopak · 2024-10-01T17:16:40Z

actually, masking seem to be the performance bottleneck, which I could improve based on this benchmarks...

Uzlopak · 2024-10-01T17:24:23Z

diff --git a/benchmarks/websocket/generate-mask.mjs b/benchmarks/websocket/generate-mask.mjs
index 032f05d8..c74cab08 100644
--- a/benchmarks/websocket/generate-mask.mjs
+++ b/benchmarks/websocket/generate-mask.mjs
@@ -1,20 +1,8 @@
-import { randomFillSync, randomBytes } from 'node:crypto'
+import { randomBytes } from 'node:crypto'
 import { bench, group, run } from 'mitata'
+import { generateMask } from "../../lib/web/websocket/frame.js"
 
-const BUFFER_SIZE = 16384
-
-const buf = Buffer.allocUnsafe(BUFFER_SIZE)
-let bufIdx = BUFFER_SIZE
-
-function generateMask () {
-  if (bufIdx === BUFFER_SIZE) {
-    bufIdx = 0
-    randomFillSync(buf, 0, BUFFER_SIZE)
-  }
-  return [buf[bufIdx++], buf[bufIdx++], buf[bufIdx++], buf[bufIdx++]]
-}
-
-group('generate', () => {
+group(function ()  {
   bench('generateMask', () => generateMask())
   bench('crypto.randomBytes(4)', () => randomBytes(4))
 })
diff --git a/lib/web/websocket/frame.js b/lib/web/websocket/frame.js
index e773b33e..c0b5d779 100644
--- a/lib/web/websocket/frame.js
+++ b/lib/web/websocket/frame.js
@@ -4,6 +4,8 @@ const { maxUnsigned16Bit, opcodes } = require('./constants')
 
 const BUFFER_SIZE = 8 * 1024
 
+const FIN = /** @type {const} */ (0x80)
+
 /** @type {import('crypto')} */
 let crypto
 let buffer = null
@@ -59,10 +61,7 @@ class WebsocketFrameSend {
 
     const buffer = Buffer.allocUnsafe(bodyLength + offset)
 
-    // Clear first 2 bytes, everything else is overwritten
-    buffer[0] = buffer[1] = 0
-    buffer[0] |= 0x80 // FIN
-    buffer[0] = (buffer[0] & 0xF0) + opcode // opcode
+    buffer[0] = FIN + opcode
 
     /*! ws. MIT License. Einar Otto Stangvik <einaros@gmail.com> */
     buffer[offset - 4] = maskKey[0]
@@ -70,21 +69,38 @@ class WebsocketFrameSend {
     buffer[offset - 2] = maskKey[2]
     buffer[offset - 1] = maskKey[3]
 
-    buffer[1] = payloadLength
+    buffer[1] = payloadLength | 0x80
 
-    if (payloadLength === 126) {
-      buffer.writeUInt16BE(bodyLength, 2)
-    } else if (payloadLength === 127) {
-      // Clear extended payload length
-      buffer[2] = buffer[3] = 0
-      buffer.writeUIntBE(bodyLength, 4, 6)
+    if (payloadLength > 125) {
+      if (payloadLength === 126) {
+        buffer.writeUInt16BE(bodyLength, 2)
+      } else if (payloadLength === 127) {
+        // Clear extended payload length
+        buffer[2] = buffer[3] = 0
+        buffer.writeUIntBE(bodyLength, 4, 6)
+      }
     }
 
-    buffer[1] |= 0x80 // MASK
+    const rest = bodyLength & 3
+    const p4 = bodyLength - rest
 
+    let i = 0
     // mask body
-    for (let i = 0; i < bodyLength; ++i) {
-      buffer[offset + i] = frameData[i] ^ maskKey[i & 3]
+    while (i < p4) {
+      buffer[offset + i] = frameData[i++] ^ maskKey[0]
+      buffer[offset + i] = frameData[i++] ^ maskKey[1]
+      buffer[offset + i] = frameData[i++] ^ maskKey[2]
+      buffer[offset + i] = frameData[i++] ^ maskKey[3]
+      i += 4
+    }
+
+    switch (rest) {
+      case 3:
+        buffer[offset + i + 2] = frameData[i + 2] ^ maskKey[2]
+      case 2:
+        buffer[offset + i + 1] = frameData[i + 1] ^ maskKey[1]
+      case 1:
+        buffer[offset + i] = frameData[i] ^ maskKey[0]
     }
 
     return buffer
@@ -134,5 +150,6 @@ class WebsocketFrameSend {
 }
 
 module.exports = {
+  generateMask,
   WebsocketFrameSend
 }

tsctx · 2025-06-14T11:50:01Z

@lpinca
I finally understand why the results were different. It turns out my script was benchmarking the latest undici websocket, whereas yours was benchmarking the global websocket.

> $ node ws-client.js 100000 262144
6324 req/10s
6684 req/10s
6525 req/10s
5947 req/10s
6365 req/10s
6328 req/10s
6487 req/10s
6177 req/10s
6482 req/10s
5969 req/10s
6128 req/10s
6677 req/10s
6531 req/10s
6798 req/10s
6332 req/10s
100000 messages of 262144 bytes: 2:36.519 (m:ss.mmm)
> $ node undici-client.js 100000 262144
5849 req/10s
6128 req/10s
6476 req/10s
6570 req/10s
6649 req/10s
6599 req/10s
5920 req/10s
5633 req/10s
5625 req/10s
6377 req/10s
6377 req/10s
6413 req/10s
6403 req/10s
6429 req/10s
6391 req/10s
100000 messages of 262144 bytes: 2:39.189 (m:ss.mmm)

Script

// ws-client.js
'use strict'

const { WebSocket } = require('ws')

const messages = +process.argv[2]
const payloadLength = +process.argv[3]
const data = Buffer.alloc(payloadLength, '_')

const ws = new WebSocket('ws://127.0.0.1:8080')

ws.binaryType = 'arraybuffer'

let n = 0
let timer

ws.on('open', function () {
  console.time(`${messages} messages of ${payloadLength} bytes`)
  timer = setInterval(() => {
    console.log(`${count - n} req/10s`)
    n = count
  }, 10000)
  ws.send(data)
})

let count = 0

ws.on('message', function () {
  if (++count === messages) {
    console.timeEnd(`${messages} messages of ${payloadLength} bytes`)
    ws.close()
    clearInterval(timer)
  } else {
    ws.send(data)
  }
})

// undici-client.js
'use strict'

const messages = +process.argv[2]
const payloadLength = +process.argv[3]
const data = Buffer.alloc(payloadLength, '_')

const { WebSocket } = require('./index')

const ws = new WebSocket('ws://127.0.0.1:8080')

ws.binaryType = 'arraybuffer'

let n = 0
let timer

ws.addEventListener('open', function () {
  console.time(`${messages} messages of ${payloadLength} bytes`)
  timer = setInterval(() => {
    console.log(`${count - n} req/10s`)
    n = count
  }, 10000)
  ws.send(data)
})

let count = 0

ws.addEventListener('message', function () {
  if (++count === messages) {
    console.timeEnd(`${messages} messages of ${payloadLength} bytes`)
    ws.close()
    clearInterval(timer)
  } else {
    ws.send(data)
  }
})

// server.js
const uws = require('uWebSockets.js')
const app = uws.App()

app.ws('/*', {
  compression: uws.DISABLED,
  maxPayloadLength: 512 * 1024 * 1024,
  maxBackpressure: 128 * 1024,
  message: (ws, message, isBinary) => {
    ws.send(message, isBinary)
  }
})

app.listen(8080, (listenSocket) => {
  if (listenSocket) {
    console.log('Server listening to port 8080')
  }
})

lpinca · 2025-06-14T12:56:10Z

FWIW, results are similar to ws now and I see no difference between global WebSocket and installed undici WebSocket. Aren't them the same thing?

tsctx · 2025-06-14T13:06:00Z

FWIW, results are similar to ws now and I see no difference between global WebSocket and installed undici WebSocket. Aren't them the same thing?

When we ran the benchmark, the global version of undici was older, so its performance didn't reflect the latest results. Now, we've updated to the most recent version, and its performance is comparable.

Ethan-Arrowood · 2025-06-14T14:07:11Z

benchmarks/_util/index.js

+ * @param {number} num
+ * @returns {string}
+ */
+function formatBytes (num) {


love this implementation 🌟

Ethan-Arrowood · 2025-06-14T14:09:48Z

benchmarks/_util/runner.js

+      let resolve = (value) => {}
+      let reject = (reason) => {}
+      const promise = new Promise(
+        (_resolve, _reject) => { resolve = _resolve; reject = _reject }
+      )


Suggested change

let resolve = (value) => {}

let reject = (reason) => {}

const promise = new Promise(

(_resolve, _reject) => { resolve = _resolve; reject = _reject }

)

const { promise, resolve, reject } = Promise.withResolvers();

http://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Promise/withResolvers

I guess this might only be valid Node v22+

We still support v20, so let's keep it that way.

mcollina

lgtm

I'm glad these are almost equal.

mcollina · 2025-06-14T20:24:02Z

@KhafraDev ptal

metcoder95 approved these changes May 5, 2024

View reviewed changes

KhafraDev requested changes May 5, 2024

View reviewed changes

benchmarks/websocket/websocket-echo.mjs Outdated Show resolved Hide resolved

tsctx force-pushed the bench/add-websockets branch from 7af0bac to 7ef0e3a Compare May 5, 2024 21:13

tsctx marked this pull request as draft May 6, 2024 08:53

tsctx force-pushed the bench/add-websockets branch 2 times, most recently from f5204cf to 370f8cb Compare May 10, 2024 11:50

tsctx force-pushed the bench/add-websockets branch from 370f8cb to 943796a Compare June 20, 2024 23:21

github-advanced-security bot found potential problems Jun 20, 2024

View reviewed changes

benchmarks/_util/websocket-simple-server.js Fixed Show fixed Hide fixed

KhafraDev mentioned this pull request Aug 14, 2024

websocket: improve frame parsing #3447

Merged

tsctx force-pushed the bench/add-websockets branch 4 times, most recently from 2d4b1f8 to 6a6b336 Compare August 20, 2024 00:29

lpinca reviewed Aug 20, 2024

View reviewed changes

benchmarks/_util/websocket-simple-server.js Outdated Show resolved Hide resolved

tsctx force-pushed the bench/add-websockets branch 2 times, most recently from 841a95b to 1444870 Compare August 22, 2024 06:30

tsctx force-pushed the bench/add-websockets branch 4 times, most recently from e249b3b to 5e86207 Compare September 9, 2024 10:41

tsctx force-pushed the bench/add-websockets branch from 18e2482 to 6f1e7ef Compare September 18, 2024 11:34

tsctx force-pushed the bench/add-websockets branch from 6f1e7ef to 097cca7 Compare October 1, 2024 11:43

tsctx marked this pull request as ready for review October 1, 2024 11:50

tsctx force-pushed the bench/add-websockets branch from fdcd74a to d70f6c7 Compare October 1, 2024 11:52

tsctx force-pushed the bench/add-websockets branch from 3bb9ada to 09b62cf Compare June 14, 2025 11:49

tsctx force-pushed the bench/add-websockets branch 2 times, most recently from 478578c to 5c22800 Compare June 14, 2025 12:19

bench: add websockets

8e4991b

tsctx force-pushed the bench/add-websockets branch from 5c22800 to 8e4991b Compare June 14, 2025 12:25

Ethan-Arrowood approved these changes Jun 14, 2025

View reviewed changes

mcollina approved these changes Jun 14, 2025

View reviewed changes

KhafraDev approved these changes Jun 14, 2025

View reviewed changes

mcollina merged commit 9ff152f into nodejs:main Jun 15, 2025
28 of 32 checks passed

tsctx deleted the bench/add-websockets branch June 15, 2025 10:26

github-actions bot mentioned this pull request Jun 26, 2025

[Release] v7.11.0 #4300

Merged

renovate bot mentioned this pull request Jul 19, 2025

chore(deps): update all non-major dependencies rajput-hemant/qwik-template#78

Open

1 task

renovate bot mentioned this pull request Aug 5, 2025

Update All saboooor/saboooor#7

Open

1 task

Uh oh!

bench: add websockets #3203

bench: add websockets #3203

Uh oh!

Conversation

tsctx commented May 5, 2024

Uh oh!

codecov-commenter commented May 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

lpinca commented May 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KhafraDev left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tsctx commented Oct 1, 2024

Uh oh!

lpinca commented Oct 1, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KhafraDev commented Oct 1, 2024

Uh oh!

Uzlopak commented Oct 1, 2024

Uh oh!

Uzlopak commented Oct 1, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tsctx commented Jun 14, 2025

Uh oh!

lpinca commented Jun 14, 2025

Uh oh!

tsctx commented Jun 14, 2025

Uh oh!

Ethan-Arrowood Jun 14, 2025

Choose a reason for hiding this comment

Uh oh!

Ethan-Arrowood Jun 14, 2025

Choose a reason for hiding this comment

Uh oh!

Ethan-Arrowood Jun 14, 2025

Choose a reason for hiding this comment

Uh oh!

tsctx Jun 15, 2025

Choose a reason for hiding this comment

Uh oh!

mcollina left a comment

Choose a reason for hiding this comment

Uh oh!

mcollina commented Jun 14, 2025

Uh oh!

Uh oh!

Uh oh!

codecov-commenter commented May 5, 2024 •

edited

Loading

lpinca commented May 5, 2024 •

edited

Loading

lpinca commented Oct 1, 2024 •

edited

Loading

Uzlopak commented Oct 1, 2024 •

edited

Loading