WebTransport is almost here to allow UDP-like exchange in the browser

https://developer.mozilla.org/en-US/docs/Web/API/WebTransport_API

by BinaryIgor • 9 months ago

57 71 9 months ago

developer.mozilla.org

ktpsns• 8 months ago

I wonder why they did not extend the existing websocket API. Now if you want to be HTTP/1 or HTTP/2 backward compatible, you have to add another abstraction ontop of these two because WebTransports is HTTP/3 only.

silverwind• 8 months ago

Yeah, I don't see this or HTTP/3 gaining any widespread adoption because HTTP/3 needs obscure DNS records and messing with firewalls.

kaoD• 8 months ago

HTTP/3 does not need obscure DNS records (but it's greatly enhanced by them).

Messing with firewalls in what way?

baggy_trough• 8 months ago

All it really needs is a UDP port 443 hole punch. The DNS stuff is an optional optimization.

CharlesW• 8 months ago

I can't find 2025 stats, but HTTP/3 adoption was already widespread two years ago. https://blog.apnic.net/2023/09/25/why-http-3-is-eating-the-w...

chrismorgan• 8 months ago

WebSocket and WebTransport are pretty wildly incompatible in an API sense. One provides a single reliable bidirectional stream. The other provides arbitrarily many unreliable and reliable unidirectional and bidirectional streams, according to your own orchestration.

It’s like saying “I had a violin, why didn’t they make this new ‘orchestra’ thing behave the same way?”

If you’re willing to limit yourself to a single reliable bidirectional stream, abstracting over both WebTransport and WebSocket will be easy.

WebTransport, in its fullness, also pretty clearly depends on HTTP/3. Any implementation based on HTTP/1 or HTTP/2 (which is in progress) will lack unreliable delivery and independence of streams (due to using TCP), and possibly more features.

advisedwang• 8 months ago

There is a proposal to do this: https://datatracker.ietf.org/doc/html/draft-ietf-webtrans-ht...

KPGv22• 8 months ago

The websocket RFC is designed around TCP, and "extending" WS to work over UDP would essentially be a new protocol. Which is what this is.

You can think about this with HTTP/1 vs HTTP/3: HTTP/3 isn't an extension of HTTP/1. It's radically different.

mort9618• 8 months ago

I think WebTransport is cool.

What worries me is that some people involved in standardisation seem to be of the opinion that WebTransport supersedes WebSocket. WS has become my go-to transport when I just need to be able to reliably send messages back and forth, whether or not the web is involved at all, and I don't see WebTransport as a replacement for that use case. I hope WS sticks around forever.

Also, it's messed up that WT is only available in HTTPS. There are so many cool use cases for web technologies in local contexts where HTTPS is not a practical option, it's a shame most new technologies are arbitrarily banned from those use cases.

yladiz• 8 months ago

You should be able to use things like WebTransport locally, localhost is considered a secure context.

mort9618• 8 months ago

192.168.0.x is not though

super256• 8 months ago

I just read the MDN article of WS and you are right: WT is indeed treated as the successor of WS (in many use cases), which I did not expect at all: 'Additionally, the WebTransport API is expected to replace the WebSocket API for many applications. WebTransport is a versatile, low-level API that provides backpressure and many other features not supported by either WebSocket or WebSocketStream, such as unidirectional streams, out-of-order delivery, and unreliable data transmission via datagrams.'

https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_...

gr4vityWall• 8 months ago

Agreed. WebSocket works perfectly when someone wants a lightweight message-based protocol on top of TCP and doesn't want to implement it themselves, potentially in multiple languages. It being originally browser tech became a minor detail.

kixelated7• 8 months ago

I like to frame WebTransport as multiple WebSocket connections to the same host, but using a shared handshake.

It's common to multiplex a WebSocket connection but you don't need that with WebTransport, while also avoiding head-of-line-blocking.

But yeah I wish WebTransport had a better TCP fallback. I still use WebSocket for that.

mort9618• 8 months ago

The thing is, I very rarely (or arguably never) have a use case which requires that one client has multiple connections to the same server. The thing I want is almost always to have a bidirectional stream of messages where messages arrive in order. Essentially a simple message envelope protocol on top of TCP.

Aurornis2• 8 months ago

Designed a protocol to supersede an old one is a good design goal. It makes encompassing all of the previous functionality and use cases a design priority.

It doesn't imply that websockets would be deprecated by WebTransport. It's a goal to make WebTransport an easy choice.

Also, it's messed up that WT is only available in HTTPS. There are so many cool use cases for web technologies in local contexts where HTTPS is not a practical option

WebTransport is relatively heavy compared to websockets. Once you consider all of the layers, HTTP/3, possible HTTP/2 fallback, and transports, adding TLS is minimal overhead for code size.

For contexts where setting up certificates is not practical, I assume the workaround will be to generate a self-signed certificate and then tell the client to ignore certificate validation problems.

layla5alive• 8 months ago

There are material performance overheads to TLS

mort9618• 8 months ago

Performance isn't really my main concern; I have some projects with configuration web GUIs running on MCUs where it could matter (especially the code size to include mbedtls), but most of the stuff I care about is on Linux machines which are powerful enough that it doesn't matter.

See my response to the certificates thing here: https://news.ycombinator.com/item?id=45820782#45956556

rustyconover• 8 months ago

Where are the server side implementations?

pphysch• 8 months ago

Like this?

Golang: https://github.com/quic-go/webtransport-go

KPGv22• 8 months ago

I would guess there aren't as many because, in order to implement this, your language must already have an HTTP/3 library. My language of choice doesn't even support QUIC yet (so I'm writing the library for it, then for HTTP/3). I wouldn't be surprised if other languages are similar.

As of one month ago, Java still didn't have HTTP/3 support. Though it's apparently coming in March (with JDK 26).

kixelated7• 8 months ago

I maintain https://github.com/kixelated/web-transport

But yeah the HTTP/3 integration definitely makes WebTransport harder to support. The QUIC connection needs to be shared between HTTP/3 and WebTransport.

mehagar• 8 months ago

We tested this against WebRTC data channels (which also uses UDP) and found that the congestion control algorithm used for WebTransport limits its use for videoconferencing. We'll probably look into it again if browsers start allowing this to be configured.

valorzard• 8 months ago

Did you try using the unreliable datagram extension?

Sean-Der2• 8 months ago

I haven't used WebTransport myself, how much can you control? I wonder how often libwebrtc does stuff that wouldn't be allowed in Javascript. Things like probing and overshooting the window?

Since libwebrtc is all in C++ and 'trusted' it can do stuff like faststart or have error correction/bandwidth estimation that is codec aware.

kixelated7• 8 months ago

There's no probing in any QUIC implementation but it's possible. There's a QUIC extension in the IETF similar to transport-wide-cc but it would still be up to the browser to use it for any upload CC.

kixelated7• 8 months ago

SCTP and by extension, WebRTC data channels, are supposed to use the same congestion control algorithms as TCP/QUIC. But I don't know which CC libsctp does these days.

WebTransport in Chrome currently uses CUBIC but the Google folks want to turn on BBR everywhere. It uses the same QUIC implementation as HTTP/3 so it's going to be more battle hardened.

greenavocado2• 8 months ago

SCTP: The FORWARD-TSN chunk was introduced to support selective unreliability: it allows the sender to tell the receiver that it will not retransmit some number of chunks, and requests that the receiver consider all these chunks as received.

ilaksh• 8 months ago

Okay so if sending raw binary is a normal use case, does that mean HTTP/3 is a misnomer? Or are they confusing me with the way they present Web transport ad being equivalent to HTTP/3 and WebTransport just sending binary data?

Also, what client libraries are there for Python, Rust etc. for WebTransport? Does it work with any HTTP/3/QUIC implementation?

kykat• 8 months ago

With all these new APIs, I worry what the web will look like in 10 years, will be have this huge API surface always will us? Will we start to see deprecations in the web api? I'm worried if the XSLT is setting a precedent on deprecating "complex" and "hard to maintain" APIs.

EDIT: Think also about webgl1/2, webgpu, webxr, websocket, webrtc, webauthn ...

user3939382• 8 months ago

10 years? We’re slowly re implementing yet another layer on the whole monster insane stack that already made no sense.

api• 8 months ago

Everything that is extensible eventually becomes an OS.

mattvr2• 8 months ago

While WebTransport is promising, it's limited to client-server communication unlike WebRTC.

WebRTC supports peer-to-peer UDP connections as well. Thus it's better for use cases like low-latency games, video calling, and secure direct communication between devices.

A better push might be to make WebRTC more simple and modern, but I'm not sure if any standards committees are working on this yet.

Sean-Der2• 8 months ago

What do you think a more simple/modern WebRTC looks like?

* RTP over QUIC is promising https://datatracker.ietf.org/doc/draft-ietf-avtcore-rtp-over...

Do you want a new PeerConnection API? Are you annoyed with the Offer/Answer model? Do you have beef with the protocols. The possibility are endless :)

mattvr2• 8 months ago

In a nutshell, ideally `peer.connect(remoteId)`. An API like peer-js/simple-peer. And symmetric negotiation would be great as well.

WebRTC should be the universal networking primitive for the next phase of the web, but the API exposes too many implementation details – its abstractions leak.

This plus the overall weight of integrations limit mass adoption by developers.

ranger_danger• 8 months ago

peer-to-peer UDP connections

Doesn't it require DTLS over UDP though?

kixelated7• 8 months ago

Yeah, technically it's SCTP over DTLS for data channels. Only the media layer gets to use raw UDP, limiting the scope.

r3tr0• 8 months ago

I have been counting down the days.

It will help a ton with the low latency streaming we need to do in our eBPF based Linux performance tool. (https://yeet.cx)

dabinat• 8 months ago

I’m completely unsurprised the holdout is Safari. I really wish Apple decoupled Safari updates from iOS updates. There are legitimate reasons why people might be stuck on an old OS (due to unsupported hardware) or simply not want to upgrade (e.g. because of the recent controversial UI changes). Tying updating your browser to updating your whole OS means web platform changes take much longer to become full baseline.

Plus Safari is often the last browser to get certain features anyway.

WebTransport is almost here to allow UDP-like exchange in the browser

Related Stories