RFC-139: Enabling HTTP/3 on GOV.UK

[Nooshu]

✨ Rendered ✨

Deadline: 2022-07-14

[karlbaker02]

It's a very interesting read, very informative about the history of HTTP and the benefits that HTTP/3 brings - thanks for helping me to learn something new! I can see that HTTP/3 is definitely of value to us in making GOV.UK as responsive and as accessible to as broad a range of users as possible.

Out of curiosity and in no way to diminish the benefits described as we need to cater to all users, do we have any idea of the percentage of users coming to GOV.UK which are likely to be experiencing the slow connection speeds which HTTP/3 will specifically help out with?

On browsers, I noticed that there isn't any mention of Internet Explorer (though there is of Edge). I may be behind the times here, but have we stopped officially supporting Internet Explorer? Related to this, and to the previous point around the users that are experiencing the slow connection speeds, do we have a rough idea of the percentages using HTTP/3 enabled browsers?

[Nooshu]

Out of curiosity and in no way to diminish the benefits described as we need to cater to all users, do we have any idea of the percentage of users coming to GOV.UK which are likely to be experiencing the slow connection speeds which HTTP/3 will specifically help out with?

This is a really interesting question. The only way I can see us being able to measure this is via the Fastly CDN logs since this is network level information. They provide a set of metrics we can use to measure this. Even then I doubt we could pinpoint an actual number of users. But we would have an idea in terms of percentage across all connections. e.g. 1% of connections experience some form of packet loss / connection instability.

On browsers, I noticed that there isn't any mention of Internet Explorer (though there is of Edge). I may be behind the times here, but have we stopped officially supporting Internet Explorer? Related to this, and to the previous point around the users that are experiencing the slow connection speeds, do we have a rough idea of the percentages using HTTP/3 enabled browsers?

We still support IE11, but it is no longer updated by Microsoft. Being such an old browser IE11 only supports HTTP/2 on Windows 10+ OS's. It won't be effected by a transition to HTTP/3 since it won't understand the protocol.

As for the percentage of users who could currently use HTTP/3, we are looking at 55% examining the March 2021 stats from GA.

[ChrisBAshton]

That was an interesting read - HTTP/3 sounds great, and the direction we should be heading in! 👏

I can't see the proposal being at all contentious, but I do have a few suggestions to polish up the RFC:

[Nooshu]

Added updates to the RFC as Fastly has now made HTTP/3 + QUIC available to all customers, it will just require a CNAME change since we aren't using dedicated Fastly IP's.

Also proposing we enable it after RFC-147 has been added to allow us to collect RUM data for HTTP/2 & HTTP/3 for comparison.

Fastly documentation about the CNAME change can be found here

[Nooshu]

It's now also worth noting that global HTTP/3 availability in browsers has now jumped to 73% according to CaniUse.com.

[injms]

It's now also worth noting that global HTTP/3 availability in browsers has now jumped to 73% according to CaniUse.com.

You can also plug the list of browsers tracked by analytics into CanIUse - from the last 30 days HTTP/3 availability in browsers is just under 65%.

[richardTowers]

We've started talking to Fastly about this. It turns out that we are using dedicated IPs after all (apologies for spreading misinformation about that). That means their TLS team need to make a change ahead of time, and then we can choose to enable this on a per-service level by calling h3.alt_scv() in our VCL as per Fastly's documentation

[Nooshu]

We've started talking to Fastly about this. It turns out that we are using dedicated IPs after all (apologies for spreading misinformation about that). That means their TLS team need to make a change ahead of time, and then we can choose to enable this on a per-service level by calling h3.alt_scv() in our VCL as per Fastly's documentation

This is excellent news thanks @richardTowers! Once ready, I'm happy to follow your timelines and direction on how to safely deploy it to production 😍

[injms]

I'm approving this as the deadline for discussion has past, any questions and comments have been resolved - and also because this looks to be a solid improvement.

Thanks for your hard work and investigation on this @Nooshu!