HTTP3 Explained(http3-explained.haxx.se)
145 points byweinzierl8 hours ago |9 comments
akdor11546 hours ago
Damn it's nice reading a simple static site like this. Links open instantly to the next fully laid out page of content. If only the rest of the web could be like this..
fny1 hour ago
Gitbook is not a simple static site generator.

There are a also ton of outbound requests for JS on first load.

[0]: view-source:https://http3-explained.haxx.se/

Nifty392950 minutes ago
+1000

I need fancy javascript crap like I need a hole in my head.

INTPenis6 hours ago
Agreed but where is the actual git repo? I see a text saying this "contents get updated automatically on every commit to this git repository" but where is "this git repository"?

I can't find a link to the source anywhere.

rchard2scout6 hours ago
The introduction has a "help out" section which links to the github repo: https://github.com/bagder/http3-explained
dobladov6 hours ago
Cthulhu_6 hours ago
After a quick google: https://github.com/bagder/http3-explained

(using a search engine is faster than asking for a link on HN)

Zambyte5 hours ago
I found it on HN faster than I could have with a search engine because they asked :)
madeofpalk4 hours ago
Worth nothing, that react application (using React Server Components?)! If you have javascript enabled, it renders as a single page app, fetching each additional page via an API. If you disable JS, it renders it all on the server.
mb21002 hours ago
yes, that's why performance metric and on low-powered phones is so terrible. Look at that: https://pagespeed.web.dev/analysis/https-http3-explained-hax...
flykespice1 hour ago
That is a striking difference between mobile and desktop, why is that? (Also that is a very interesting site)
mb21001 hour ago
That's because on mobile, PageSpeed (which is a hosted version of the Ligthhouse dev tools you also have in Chrome) simulates a low-end Android device on a slow 3G network, which is what a lot of website visitors actually use (as opposed to the web developer using the newest iPhone on great WiFi).

That's why content-driven websites should not be an SPA, and why I built https://mastrojs.github.io

cwillu1 hour ago
Ugh, that explains why it hangs for a quarter second any time I scroll with the mousewheel.
tomalbrc3 hours ago
Wow almost as good as handwritten HTML!
Razengan1 hour ago
But how will the author know the last 500 websites you visited and where your eyes are looking right now and what you ate last Tuesday? They should put some AnAlYtIcS in.
fkyoureadthedoc1 hour ago
Damn it's nice to log onto Hacker News and see yet another top comment on an interesting article be bike shedding about webshit. And also wrong because if you crack open your react dev tools and have a peak inside the 2MB of javascript you'll see that this site is still everything you despise.
sedatk5 hours ago
The document is now five years old and full of statements like “we’ll see that in the upcoming years”. I think it’s due for an update.
lsaferite47 minutes ago
I was personally bugged by it claiming that QUIC wasn't an acronym.
esnard2 hours ago
Link for anyone willing to contribute: https://github.com/bagder/http3-explained

Looks unmaintained, though.

code_martial6 hours ago
Here’s a conceptual background about how and why HTTP/3 came to be (recollected from memory):

HTTP/1.0 was built primarily as a textual request-response protocol over the very suitable TCP protocol which guaranteed reliable byte stream semantics. The usual pattern was to use a TCP connection to exchange a request and response pair.

As websites grew more complex, a web page was no longer just one document but a collection of resources stitched together into a main document. Many of these resources came from the same source, so HTTP/1.1 came along with one main optimisation — the ability to reuse a connection for multiple resources using Keep Alive semantics.

This was important because TCP connections and TLS (nee SSL) took many round-trips to get established and transmitting at optimal speed. Latency is one thing that cannot be optimised by adding more hardware because it’s a function of physical distance and network topology.

HTTP/2 came along as a way to improve performance for dynamic applications that were relying more and more on continuous bi-directional data exchange and not just one-and-done resource downloads. Two of its biggest advancements were faster (fewer round-trips) TLS negotiation and the concept of multiple streams over the same TCP connection.

HTTP/2 fixed pretty much everything that could be fixed with HTTP performance and semantics for contemporary connected applications but it was still a protocol that worked over TCP. TCP is really good when you have a generally stable physical network (think wired connections) but it performs really badly with frequent interruptions (think Wi-Fi with handoffs and mobile networks).

Besides the issues with connection reestablishment, there was also the challenge of “head of the line blocking” — since TCP has no awareness of multiplexed HTTP/2 streams, it blocks everything if a packet is dropped, instead of blocking only the stream to which the packet belonged. This renders HTTP/2 multiplexing a lot less effective.

In parallel with HTTP/2, work was also being done to optimise the network connection experience for devices on mobile and wireless networks. The outcome was QUIC — another L4 protocol over UDP (which itself is barebones enough to be nicknamed “the null protocol”). Unlike TCP, UDP just tosses data packets between endpoints without much consideration of their fate or the connection state.

QUIC’s main innovation is to integrate encryption into the transport layer and elevate connection semantics to the application space, and allow for the connection state to live at the endpoints rather than in the transport components. This allows retaining context as devices migrate between access points and cellular towers.

So HTTP/3? Well, one way to think about it is that it is HTTP/2 semantics over QUIC transport. So you get excellent latency characteristics over frequently interrupted networks and you get true stream multiplexing semantics because QUIC doesn’t try to enforce delivery order or any such thing.

Is HTTP/3 the default option going forward? Maybe not until we get the level of support that TCP enjoys at the hardware level. Currently, managing connection state in application software means that over controlled environments (like E-W communications within a data centre), HTTP/3 may not have as good a throughput as HTTP/2.

newpavlov4 hours ago
Thank you for a great overview! I wish HTTP3/QUIC was the "default option" and had much wider adoption.

Unfortunately, software implementations of QUIC suffer from dealing with UDP directly. Every UDP packet involves one syscall, which is relatively expensive in modern times. And accounting for MTU further makes the situation ~64 times worse.

In-kernel implementations and/or io-uring may improve this unfortunate situation, but today in practice it's hard to achieve the same throughput as with plain TCP. I also vaguely remember that QUIC makes load-balancing more challenging for ISPs, since they can not distinguish individual streams as with TCP.

Finally, QUIC arrived a bit too late and it gets blocked in some jurisdictions (e.g. Russia) and corporate environments similarly to ESNI.

lelanthran1 hour ago
> In-kernel implementations and/or io-uring may improve this unfortunate situation, but today in practice it's hard to achieve the same throughput as with plain TCP.

This would depend on how the server application is written, no? Using io-uring and similar should minimise context-switches from userspace to kernel space.

> I also vaguely remember that QUIC makes load-balancing more challenging for ISPs, since they can not distinguish individual streams as with TCP.

Not just for ISPs; IIRC (and I may be recalling incorrectly) reverse proxies can't currently distinguish either, so you can't easily put an application behind Nginx and use it as a load-balancer.

The server application itself has to be the proxy if you want to scale out. OTOH, if your proxy for UDP is able to inspect the packet and determine the corresponding instance to send a UDP packet too, it's going to be much fewer resources required on the reverse proxy/load balancer, as they don't have to maintain open connections at all.

It will also allow some things more easily; a machine that is getting overloaded can hand-off (in userspace) existing streams to a freshly created instance of the server on a different machine, because the "stream" is simply related UDP packets. TCP is much harder to hand-off, and even if you can, it requires either networking changes or kernel functions to hand-off.

kccqzy1 hour ago
Why would every UDP packet involve one syscall when you can use sendmmsg(2) instead of sendmsg(2)? And similarly recvmmsg(2) instead of recvmsg(2).

EDIT: I found https://news.ycombinator.com/item?id=45387462 which is a way better discussion than what I wrote.

NeutralForest5 hours ago
Thanks for taking the time to make this, that was helpful!
code_martial4 hours ago
Glad you found it helpful! Most of it is distilled from High Performance Browser Networking (https://hpbn.co/). It’s a very well organised, easy to follow book. Highly recommended!

Unfortunately, it’s not updated to include QUIC and HTTP/3 so I had to piece together the info from various sources.

kevg1234 hours ago
> As the packet loss rate increases, HTTP/2 performs less and less well. At 2% packet loss (which is a terrible network quality, mind you), tests have proven that HTTP/1 users are usually better off - because they typically have up to six TCP connections to distribute lost packets over. This means for every lost packet the other connections can still continue.

Why doesn't HTTP/2 use more than one socket?

thwarted3 hours ago
Because one thing it tries to optimize for is avoiding TLS session negotiation.
kevg1233 hours ago
Makes sense. One idea would be if the browser could detect packet loss (e.g. netstat -s and look for TCP retransmissions, and equivalent on other OSes) and open more sockets if there is.
gramakri25 hours ago
Where can I download the pdf? It seems the link points to itself
panki275 hours ago
It's hidden in the "Copy" drop down at the top right.

https://http3-explained.haxx.se/~gitbook/pdf?limit=100

sharts3 hours ago
Will there be HTTP/4 ?
derelicta4 hours ago
It's still crazy how quickly http3 got adopted by web actors. Can't wait til we do the same for IMAP and SMTP
immibis1 hour ago
Email is mostly dead - we use Gmail (or Microsoft 365) now. It is to email what Slack is to IRC. With only one or two vendors, the need for widely interoperable protocols is gone - they only need to interoperate between a few large service providers, and that can be done by private agreement.
lsaferite35 minutes ago
You realize those ESPs use and support the industry standard open protocols under the hood, right? Slack is 100% proprietary and does not use industry standard protocols for interchange or federation. These are not even remotely comparable. Slack would need to use industry standard and open protocols (i.e. XMPP) to allow federation with products like Teams and Discord for the situations to be comparable.
ahoka7 hours ago
Anyone else blocks UDP 80/443 due to privacy concerns?
detaro7 hours ago
What privacy concern do you have that does not apply to TCP 80/443?
ahoka7 hours ago
Tracking sessions across different physical connections has some non-trivial privacy implications:

https://http3-explained.haxx.se/en/quic/quic-connections#con...

NavinF6 hours ago
How do you imagine other protocols handle switching physical connections? With HTTP 1, you send your session ID as a cookie after wasting time creating a new TCP connection
ahoka6 hours ago
Yes, obviously, but we already know how that is used. This is a more complex protocol that might enable attack vectors that were not possible before and we do not think about when accessing websites:

But see the notes taken from the HTTP/3 RFC itself, written by the authors:

10.11. Privacy Considerations

   Several characteristics of HTTP/3 provide an observer an opportunity
   to correlate actions of a single client or server over time.  These
   include the value of settings, the timing of reactions to stimulus,
   and the handling of any features that are controlled by settings.

   As far as these create observable differences in behavior, they could
   be used as a basis for fingerprinting a specific client.

   HTTP/3's preference for using a single QUIC connection allows
   correlation of a user's activity on a site.  Reusing connections for
   different origins allows for correlation of activity across those
   origins.

   Several features of QUIC solicit immediate responses and can be used
   by an endpoint to measure latency to their peer; this might have
   privacy implications in certain scenarios.
MallocVoidstar7 hours ago
No.
frmdstryr3 hours ago
Yes, no performance difference either.
ckbkr107 hours ago
Sounds overly complicated, I doubt this will have a widespread adoption
gucci-on-fleek6 hours ago
About 30% percent of traffic to Cloudflare uses HTTP/3 [0], so it seems pretty popular already. For comparison, this is 3× as much traffic as HTTP/1.1.

[0]: https://radar.cloudflare.com/adoption-and-usage#http1x-vs-ht...

mgaunard6 hours ago
and then cloudflare converts that to http/2 or even 1.1 for the backend
vanviegen6 hours ago
So? Those protocols work fine within the reliable low latency network of a datacenter.
wongarsu5 hours ago
I'd even go as far as claiming that on reliable wired connections (like between cloudflare and your backend) HTTP/2 is superior to HTTP/3. Choosing HTTP/3 for that part of the journey would be a downgrade
klempner5 hours ago
At the very least, the benefits of QUIC are very very dubious for low RTT connections like inside a datacenter, especially when you're losing a bunch of hardware support and moving a fair bit of actual work to userspace where threads need to be scheduled etc. On the other hand Cloudflare to backend is not necessarily low RTT and likely has nonzero congestion.

With that said, I am 100% in agreement that the primary benefits of QUIC in most cases would be between client and CDN, whereas the costs are comparable at every hop.

hshdhdhehd4 hours ago
Is CF typically serving from the edge, or serving from the nearest to the server? I imagine it would be from the edge so that it can CDN what it can. So... most of the time it wont be a low latency connection from CF to backend. Unless your back end is globally distributed too.
immibis1 hour ago
Also, within a single server, you should not use HTTP between your frontend nginx and your application server - use FastCGI or SCGI instead, as they preserve metadata (like client IP) much better. You can also use them over the network within a datacenter, in theory.
ofrzeta6 hours ago
"As of September 2024, HTTP/3 is supported by more than 95% of major web browsers in use and 34% of the top 10 million websites."

https://en.wikipedia.org/wiki/HTTP/3

pimterry6 hours ago
Yes and, at the same time practical support within programming language standard libraries & common tooling lags way behind: https://httptoolkit.com/blog/http3-quic-open-source-support-...
taffer5 hours ago
You will get most of the benefits of HTTP 3 even if your app libraries run HTTP 1.1, as long as the app is behind a reverse proxy that speaks HTTP 3.
theandrewbailey4 hours ago
kunley2 hours ago
Yep, for example, Caddy (zero special configuration to enable HTTP 3)
karel-3d6 hours ago
A lot of servers still don't support that.

Go http webserver doesn't support http 3 without external libraries. Nginx doesn't support http 3. Apache doesn't support http 3. node.js doesn't support http 3. Kubernetes ingress doesn't support http 3.

should I go on?

edit: even curl itself - which created the original document linked above - has http 3 just in an experimental build.

samueloph13 minutes ago
> edit: even curl itself - which created the original document linked above - has http 3 just in an experimental build.

It's not experimental when built with ngtcp2, which is what you will get on distros like Debian 13-backports (plain Debian 13 uses OpenSSL-QUIC), Debian 14 and onward, Arch Linux and Gentoo.

Reference: https://curl.se/docs/http3.html

miyuru6 hours ago
>Nginx doesn't support http 3

nginx do support it.

https://nginx.org/en/docs/quic.html

karel-3d5 hours ago
And I see I was not that wrong; the module is still marked as "experimental" and not built by default.

https://nginx.org/en/docs/http/ngx_http_v3_module.html

karel-3d6 hours ago
ah okay i was wrong there, mea culpa
dotancohen6 hours ago
The guy's point still stands - lots of popular software do not yet support http3.
aleks_me24 hours ago
Well this statement have to be precised.

caddyserver v2 supports HTTP/3 and it's an webserver written in go https://caddyserver.com/features

FYI: There is also an rust webserver which supports HTTP/3. https://v2.ferronweb.org/

karel-3d4 hours ago
Go built-in webserver.
frmdstryr3 hours ago
Also apparently slower over fast connections https://arxiv.org/pdf/2310.09423