Data and thoughts after enabling HTTP/3 at @Wix for our user domains #webperf #http3 #quic
Positive impact from TTFB to later phases such as FCP and LCP
1. TTFB p75 for CDN hits is considerably faster (~2x), especially in countries with weak devices or connections
Positive impact from TTFB to later phases such as FCP and LCP
1. TTFB p75 for CDN hits is considerably faster (~2x), especially in countries with weak devices or connections
2. Browser Adoption - Chrome/Firefox currently support the latest h3, Edge/Opera support h3-29
60% - 70% of requests are using HTTP/3 in Chromium based browsers (Chrome, Edge, Opera), and 40% of Firefox requests
Only 2% of Safari (experimental, will it be the default in iOS 15?)
60% - 70% of requests are using HTTP/3 in Chromium based browsers (Chrome, Edge, Opera), and 40% of Firefox requests
Only 2% of Safari (experimental, will it be the default in iOS 15?)
3. Session breakdown per protocol and browser for first page views tells a bit of a different story.
Only ~15%-20% of HTML requests are using HTTP/3
This can probably be explained by the connection upgrade behavior of the alt-svc header - first time visitors initially use h2
Only ~15%-20% of HTML requests are using HTTP/3
This can probably be explained by the connection upgrade behavior of the alt-svc header - first time visitors initially use h2
4. Did you know..?
Time to first byte (TTFB) is the time between the navigationStart (start of the navigation) and responseStart, (when the first byte of response data is received) - developer.mozilla.org/en-US/docs/Web…
ttfb = time.responseStart - time.navigationStart
Time to first byte (TTFB) is the time between the navigationStart (start of the navigation) and responseStart, (when the first byte of response data is received) - developer.mozilla.org/en-US/docs/Web…
ttfb = time.responseStart - time.navigationStart
5. This also includes DNS time, and the time between navigationStart and domainLookupStart.
NavigationStart: when the prompt for unload terminates on the previous document in the same browsing context. If there is no previous document, this value will be the same as fetchStart
NavigationStart: when the prompt for unload terminates on the previous document in the same browsing context. If there is no previous document, this value will be the same as fetchStart
6. Redirects leading to your URL are also counted as part of the TTFB, but only if the redirects are from the same origin.
Surprisingly, these parts account for a non-trivial amount of time of TTFB at the higher percentiles, especially for mobile.
Surprisingly, these parts account for a non-trivial amount of time of TTFB at the higher percentiles, especially for mobile.
7. The use of the alt-svc header means that all h3 sessions have previously visited the domain (within the configured max-age).
This fact, combined with the definition of TTFB, makes for an unfair comparison between h2 and h3 - a repeat visitor is more likely to use DNS caching.
This fact, combined with the definition of TTFB, makes for an unfair comparison between h2 and h3 - a repeat visitor is more likely to use DNS caching.
8. A fairer comparison between h2 and h3 would be of the time between connectStart and responseStart: Connection + SSL handshake + request to first byte of response
Attached are the p75 and p90 gains for this metric, which are more inline with the expected gains considering RTT
Attached are the p75 and p90 gains for this metric, which are more inline with the expected gains considering RTT
• • •
Missing some Tweet in this thread? You can try to
force a refresh
