[lxgr]

The speed problems interestingly seem to be caused by a bug in an implementation of SCTP used by most (all?) WebRTC browser implementations:

https://github.com/saljam/webwormhole/issues/55

The symptom seems to be that the SCTP data rate drops with increasing latency (which used to be a problem with very old TCP implementations too, but all modern ones handle high-latency networks much better).

[Sean-Der]

I think SCTP has everything we need, just more knobs need to be exposed [0]

I have gotten feedback about the performance about Pion's SCTP implementation as well. It is a hard problem. The amount of people who care about SCTP performance + WebRTC and are able to work on it is very small.

If anyone is interested in more about this [1] is a fun issue to read.

libwebrtc is also planning to stop using libwebrtc soon. That would mean all browsers (Chrome, Safari, FireFox) will be on something new. Could be good no idea yet. The ongoing work is here [2]

[0] https://github.com/w3c/webrtc-extensions/issues/71

[1] https://github.com/pion/sctp/issues/62

[2] https://webrtc.googlesource.com/src/+log/refs/heads/master/n...

[lxgr]

Thanks, this is great context! Especially the very different focus in these two projects explains a lot (lightweight control channel for A/V [1] vs. the entire reason for using WebRTC in all of these file transfer projects).

Curious to see how the new implementation will play out for the browsers!

[1] https://github.com/w3c/webrtc-extensions/issues/71#issuecomm...

[Sean-Der]

*libwebrtc is also planning to stop using usrsctp soon