search

pion / sctp

A Go implementation of SCTP
https://pion.ly/
MIT License
223 stars 79 forks source link

Add simple PMTU calculation with heartbeats #12

Open kc5nra opened 6 years ago

kc5nra commented 6 years ago

Calculate the maximum MTU size that can travel the entire path machine1<->machine2.

The easiest way is probably to craft heartbeat messages starting at the maximum value until they are successfully returned. Potentially listen to ICMP PacketTooBig

Sean-Der commented 5 years ago

@enobufs is this still an issue/something that needs to be done?

If there is anything I can do to help please tell me :) but you know way more about SCTP then me, so defering to the expert

enobufs commented 5 years ago

SCTP RFC actually mentions RFC4821 which recommends using the heartbeat. It is an interesting approach. The current implementation does not fully support heartbeat. In WebRTC context, we have heartbeat at ICE layer, and I am not sure how critical it is not having heartbeat at SCTP layer..

I think we need to investigate how other WebRTC libraries deal with path MTU discovery.

nils-ohlmeier commented 3 months ago

I think we need to investigate how other WebRTC libraries deal with path MTU discovery.

AFAIK libwebrtc actually has the PMTU hard coded to a safe low value (I would have to lookup the exact value).

And yes the two options are either on the SCTP or the ICE level.

The ICE PMTU would have the advantage that the same information could get passed to the RTP stack as well. But the problem is that for ICE there was only a draft for figuring out the PMTU, never an RFC. And AFAIK there is no other implementation which supports the ICE way. The SCTP approach is way more interoperable.