[{"author": "Jonathan Lennox", "text": "

<pedantry>If it's September it's OK DT

", "time": "2024-03-18T05:39:11Z"}, {"author": "Randell Jesup", "text": "

@martin seeman - We have working WebTransport in Firefox (https://wpt.fyi/results/webtransport?label=experimental&label=master&aligned). Is there any help you need to get Firefox added?

", "time": "2024-03-18T05:39:41Z"}, {"author": "Jonathan Lennox", "text": "

<pedantry>If it's September it's PDT</pedantry>

", "time": "2024-03-18T05:40:06Z"}, {"author": "Mike English", "text": "

Definitely need longer names

", "time": "2024-03-18T05:44:54Z"}, {"author": "David Schinazi", "text": "

W3C is clearly beating us in the name length department

", "time": "2024-03-18T05:45:18Z"}, {"author": "Mike English", "text": "

Which is funny, really, because they have a 3 in their name to get down to 3 characters.

", "time": "2024-03-18T05:45:57Z"}, {"author": "Martin Thomson", "text": "

@Randell Jesup we probably need to deliver a docker image with a Firefox build in it

", "time": "2024-03-18T05:46:11Z"}, {"author": "Martin Thomson", "text": "

I like the idea that we use a short label for the protocol, like \"wt\"

", "time": "2024-03-18T05:49:25Z"}, {"author": "Martin Thomson", "text": "

pronounced \"wet\"

", "time": "2024-03-18T05:49:29Z"}, {"author": "Martin Thomson", "text": "

or \"wat\"

", "time": "2024-03-18T05:50:11Z"}, {"author": "David Schinazi", "text": "

wut_transport

", "time": "2024-03-18T05:50:26Z"}, {"author": "Martin Thomson", "text": "

wot_stream

", "time": "2024-03-18T05:50:37Z"}, {"author": "Martin Thomson", "text": "

wit_session_close

", "time": "2024-03-18T05:50:51Z"}, {"author": "Martin Thomson", "text": "

wat_session_drain

", "time": "2024-03-18T05:50:59Z"}, {"author": "Victor Vasiliev", "text": "

I prefer \"wit\"

", "time": "2024-03-18T05:51:02Z"}, {"author": "Martin Thomson", "text": "

wet_data_blocked

", "time": "2024-03-18T05:51:25Z"}, {"author": "Jonathan Lennox", "text": "

Woot

", "time": "2024-03-18T05:51:47Z"}, {"author": "Victor Vasiliev", "text": "

I'm not sure zero implementations is \"good shape\", unless the Mozilla one is already done

", "time": "2024-03-18T05:51:52Z"}, {"author": "Marten Seemann", "text": "

@Randell Jesup: Here's the Chrome image: https://github.com/quic-interop/chrome-webtransport-interop-runner. The Firefox image can most likely reuse the JavaScript (maybe we should move it to a separate repo?), but will need to have a Firefox-specific Selenium setup

", "time": "2024-03-18T05:52:33Z"}, {"author": "Martin Thomson", "text": "

@Victor Vasiliev good shape might be overstating things ahead of implementation work, but we have plans for wth2

", "time": "2024-03-18T05:53:55Z"}, {"author": "Marten Seemann", "text": "

@Victor I implemented the client's check of the server SETTINGS in webtransport-go

", "time": "2024-03-18T05:54:02Z"}, {"author": "Jonathan Lennox", "text": "

I briefly parsed \"Web Transport overview\" as parallel to \"Web Transport over H3\" and wondered what protocol \"view\" was.

", "time": "2024-03-18T05:54:03Z"}, {"author": "Mike English", "text": "

@Jonathan same. I was about to start writing a \"View\" transport draft

", "time": "2024-03-18T05:54:58Z"}, {"author": "Mike English", "text": "

Looking for co-authors

", "time": "2024-03-18T05:55:15Z"}, {"author": "Randell Jesup", "text": "

@marten seemann Thanks; I'll take a look

", "time": "2024-03-18T06:00:00Z"}, {"author": "Martin Thomson", "text": "

The idea that Marten has here is pretty interesting, but I don't know if webtransport can take a dependency on it. It's non-trivial to work through all the implications in QUIC.

", "time": "2024-03-18T06:03:41Z"}, {"author": "Martin Thomson", "text": "

And we don't use the control/session stream much in wth3, so HOLB is probably a non-issue.

", "time": "2024-03-18T06:04:06Z"}, {"author": "Martin Thomson", "text": "

@Victor Vasiliev this is what Marten is proposing: a generic mechanism in QUIC that would allow WT to do good things with this signaling.

", "time": "2024-03-18T06:05:29Z"}, {"author": "Marten Seemann", "text": "

Here's the issue I was asked to create: https://github.com/ietf-wg-webtrans/draft-ietf-webtrans-http3/issues/152

", "time": "2024-03-18T06:05:56Z"}, {"author": "Victor Vasiliev", "text": "

@MT I meant that other capsule-using protocols like MASQUE might benefit from this

", "time": "2024-03-18T06:06:27Z"}, {"author": "Marten Seemann", "text": "

@Luke: This only applies to MAX_STREAMS, but stream flow control credit is granted within the RTT during which it is consumed.

", "time": "2024-03-18T06:06:48Z"}, {"author": "Victor Vasiliev", "text": "

I think the important thing here it remind implementations that they should prioritize control streams over WT data streams, otherwise it might lead to some really weird performance situations

", "time": "2024-03-18T06:07:28Z"}, {"author": "Luke Curley", "text": "

if a single lost packet causes flow control to block, your flow control windows are waaay too tight

", "time": "2024-03-18T06:07:49Z"}, {"author": "Luke Curley", "text": "

regardless of HoLB

", "time": "2024-03-18T06:08:13Z"}, {"author": "Martin Thomson", "text": "

@Luke Curley I can send redundant copies of those frames to safeguard against that

", "time": "2024-03-18T06:08:21Z"}, {"author": "Martin Thomson", "text": "

@Victor Vasiliev yeah, I get that, but it's a more complicated thing to use, so I don't want to jump on it aggressively

", "time": "2024-03-18T06:08:51Z"}, {"author": "Luke Curley", "text": "

I suppose you can send MAX_STREAM_DATA with each STREAM frame

", "time": "2024-03-18T06:09:39Z"}, {"author": "Luke Curley", "text": "

with QUIC

", "time": "2024-03-18T06:09:43Z"}, {"author": "Christian Huitema", "text": "

There are tons of way to implement looser flow control in QUIC and not be bothered...

", "time": "2024-03-18T06:09:53Z"}, {"author": "Luke Curley", "text": "

do implementations do this? or do they use the fact that MAX_STREAM_DATA was lost and retransmit?

", "time": "2024-03-18T06:10:06Z"}, {"author": "Mike Bishop", "text": "

I will note that if we didn't assume all HTTP Datagrams are datagrams being sent within the WT session, you could have control messages without HOLB sent using QUIC Datagrams.

", "time": "2024-03-18T06:10:41Z"}, {"author": "Martin Thomson", "text": "

Take it to QUIC, definitely.

", "time": "2024-03-18T06:10:42Z"}, {"author": "Marten Seemann", "text": "

quic-go sends 5 MAX_STREAM_DATA per RTT (for a high-throughput stream)

", "time": "2024-03-18T06:10:45Z"}, {"author": "Christian Huitema", "text": "

There is no rule that says that the flow control window should be just the BDP. Nothing prevents an implementation to set the limit to 4 times that...

", "time": "2024-03-18T06:11:30Z"}, {"author": "Martin Thomson", "text": "

:shrug: / \uff3c\uff08\u3007_\uff4f\uff09\uff0f

", "time": "2024-03-18T06:11:35Z"}, {"author": "Luke Curley", "text": "

tell the QUIC layer to optimistically retransmit the control stream xd

", "time": "2024-03-18T06:11:49Z"}, {"author": "Victor Vasiliev", "text": "

FEC for capsules?

", "time": "2024-03-18T06:12:07Z"}, {"author": "Marten Seemann", "text": "

@Christian: Sure, but now you're commiting 4x memory. You might now want to do that for high BDP connections

", "time": "2024-03-18T06:12:17Z"}, {"author": "Christian Huitema", "text": "

Or apply FEC to the control stream

", "time": "2024-03-18T06:12:18Z"}, {"author": "Luke Curley", "text": "

heh yeah FEC for the control stream, exactly

", "time": "2024-03-18T06:12:32Z"}, {"author": "Christian Huitema", "text": "

The control stream is not high bandwidth

", "time": "2024-03-18T06:12:48Z"}, {"author": "Mike English", "text": "

I'm not going to have coffee, David, but thank you for the time.

", "time": "2024-03-18T06:12:58Z"}, {"author": "Alex Chernyakhovsky", "text": "

IIRC there was an experimenet with QUIC doing FEC way back when by Google and I thought it turned out to not have enough benefits?

", "time": "2024-03-18T06:13:04Z"}, {"author": "Martin Thomson", "text": "

Yeah, send it once every RTT/n where n > 1

", "time": "2024-03-18T06:13:07Z"}, {"author": "Alan Frindell", "text": "

Maybe a virtual interop on the new flow control?

", "time": "2024-03-18T06:13:41Z"}, {"author": "Mike English", "text": "

:sleeping:

", "time": "2024-03-18T06:13:47Z"}, {"author": "Marten Seemann", "text": "

@Christian: fair enough. I could use a large flow control window for the control stream. But that also means I have to build a separate API for that...

", "time": "2024-03-18T06:14:07Z"}]