I found out that running async-await can be much slower in some scenarios.
<html>
<script>
function makeAPromise() {
return Promise.resolve(Math.random());
}
function usingPromises() {
const before = window.performance.now();
return makeAPromise().then((num) => {
const after = window.performance.now();
console.log('Total (promises): ', after-before, 'ms');
return num;
})
}
async function usingAwait() {
const before = window.performance.now();
const num = await makeAPromise();
const after = window.performance.now();
console.log('Total (await): ', after-before, 'ms');
return num;
}
function runBoth() {
usingAwait();
usingPromises();
}
runBoth();
</script>
<button onclick="usingPromises()">usingPromises</button>
<button onclick="usingAwait()">usingAwait</button>
<button onclick="runBoth()">both</button>
</html>I found out that running async-await can be much slower in some scenarios.
<html>
<script>
function makeAPromise() {
return Promise.resolve(Math.random());
}
function usingPromises() {
const before = window.performance.now();
return makeAPromise().then((num) => {
const after = window.performance.now();
console.log('Total (promises): ', after-before, 'ms');
return num;
})
}
async function usingAwait() {
const before = window.performance.now();
const num = await makeAPromise();
const after = window.performance.now();
console.log('Total (await): ', after-before, 'ms');
return num;
}
function runBoth() {
usingAwait();
usingPromises();
}
runBoth();
</script>
<button onclick="usingPromises()">usingPromises</button>
<button onclick="usingAwait()">usingAwait</button>
<button onclick="runBoth()">both</button>
</html>IMO, the console.log in usingPromises should print similar results to the one in usingAwait.
But in reality, I get:
Total (promises): 0.25 ms
Total (await): 2.065 ms
Also, after the page load, if I click on 'usingPromises' or 'usingAwait' button I get similar results for each of them. (both are fast when running alone)
Total (promises): 0.060000000026775524 ms
Total (await): 0.08999999999650754 ms
But if I click on the 'both' button, the 'await' version is ~3-4 times slower than the promises version.
I have a real application running lots of promises / async-await function on initialisations, and I found out that replacing some of the async-await functions to their "equal" promises version can shave significant loading time (~200ms).
Can someone explain why is that? Isn't async-await also using the same job queue as promises (micro task)? Are there best practices to when should promises should be used instead of async-await?
- Running on chrome 62 on mac
Thanks
Share Improve this question asked Nov 11, 2017 at 10:47 avivravivr 2,5933 gold badges24 silver badges36 bronze badges 2-
1
Actually, running Chrome on Mac as well, async-await is much faster over here:
Total (await): 0.07500000000004547 msandTotal (promises): 0.75 ms. Could be a hardware related thing. Async-Await is using Promises internally. – NikxDa Commented Nov 11, 2017 at 10:54 -
1
NB: Calling
runBothgives twisted results, because the promise resolutions are sequenced in the event queue: so one gets to print withconsole.logbefore the other, and it is thatconsole.logthat brings additional delay to the second. It would already be an improvement if you would definerunBothasPromise.resolve().then(usingAwait).then(usingPromises). – trincot Commented Nov 11, 2017 at 11:05
1 Answer
Reset to default 6Your first result, when running with the button Both, is misleading. The promise resolutions are sequenced in the microtask event queue: so one gets to print with console.log before the other, but it is that console.log that brings additional delay to the second, because it happens between the creation of the second promise and the treatment of its resolution.
It would already be an improvement if you would define runBoth as:
Promise.resolve().then(usingAwait).then(usingPromises)
Now both promises will be created in microtasks, and the first one will be resolved and dealt with before the second promise is created. That will lead to a more fair parison where console.log is not measured in any of the timings.