Worker Messages - Many Messages vs. Many Workers (v2)

Revision 2 of this benchmark created on

Description

In a research project, I noticed that many simultaneous Workers with small messages start to congest the main thread handling them all. So check the throughput of messages to a single Worker vs. a pool of Workers. More Workers appear to be slower for the same total number of messages.

Interestingly, when only using MessageChannels within a single thread, it is faster to spread the messages across multiple channels ...

Setup

let wrkr = 100;
let msgs = 100 * wrkr;

// create workers with inline script blobs
let script = "onmessage = ({data}) => postMessage({ id: data.id })";
let blob = URL.createObjectURL(new Blob([script]));
let workers = Array(wrkr).fill().map(() => new Worker(blob));

// pre-create promises with external resolvers
let resolver = Array(msgs).fill();
let promises = Array(msgs).fill().map((_, i) => new Promise(r => resolver[i] = r));

// register promise resolvers on workers
workers.forEach(p => p.onmessage = ({data}) => resolver[data.id]());

Teardown

workers.forEach(w => w.terminate());

Test runner

Ready to run.

Testing in
TestOps/sec
All messages to a single Worker
for (let i = 0; i < msgs; i++) {
	workers[0].postMessage({ id: i });
};
Promise.all(promises).then(() => deferred.resolve());
ready
Spread messages to some Workers
for (let i = 0; i < msgs; i++) {
	workers[i % (wrkr/10)].postMessage({ id: i });
};
Promise.all(promises).then(() => deferred.resolve());
ready
Spread messages to all Workers
for (let i = 0; i < msgs; i++) {
	workers[i % wrkr].postMessage({ id: i });
};
Promise.all(promises).then(() => deferred.resolve());
ready

Revisions

You can edit these tests or add more tests to this page by appending /edit to the URL.