Hacker News new | ask | show | jobs
by danielvaughn 1075 days ago
At first glance, 200ms INP is a pretty high latency for a "good" rating. As a comparison, I believe 200ms is an average https roundtrip. I'd expect most interactions to be much lower than that.
2 comments

I guess it depends on how your interactions are implemented. If it’s an SPA then 200ms is absurdly slow. But if it’s a more traditional form submit or something then it would take a lot longer for your next set of pixels to comes through.
Based on the fact that they're hooking into events like onclick, I'd say that they are not looking at traditional form submits, because then the metric would just effectively be First Contentful Paint or something. My interpretation is that they are indeed looking at first paint after an event handler has been fired on the same page.
This is correct (source: working in web perf for 5 years).

INP is the time between you click/press a key/etc and the moment the next paint happens. It’s only measured for on-page interactions, not for navigations.

It’s basically like http://danluu.com/input-lag/ but as a web metric.

Thanks for confirming, yeah that makes sense. Side note, that input-lag thing is a very cool resource.
That was my first impression too but then I thought about what it’s actually measuring: page responsiveness, not animation jank.

I’m not going to expect a 16ms response or anything for every animation but much slower & you see jank.

For page interactivity though? 0.2s is pretty damn fast. Human response time is 0.15-0.25s

So it’s pretty reasonable

.2s is slow for a redraw. Just because it might take you 200ms to click after something happens doesn't mean you can't see/notice when things take that long.
I’m not saying it’s fast. I’m saying that based on the goal of what it is measuring (user input responsiveness) it’s fast enough. For the purposes of the metric anyhow.

Plus, speaking from far too long of a career dealing with user testing, respond too fast and users thing you didn’t actually do anything.

So you’re kind of boned either way. This is just measuring programmatic delay.