[ ] @nicorusti Send email to Eliot, Harald and ask if external reference to SCMP is enough (within the context of link down messages)
Feedback details
Harald Alvestrand harald@alvestrand.no
Once sent, a packet’s path cannot be changed; path changes are handled by routers discovering that the path no longer exists and (I assume, this is likely documented in -controlplane) sending an alert back “don’t use this path”.
Form my lunch with Harald at IETF120:
Path revocation in case of link down should be clarified, the mechanism to signal endpoints how to react quickly to failures should be described better. Right now the draft only hints to this. This topic is very relevant also for webRTC. I mentioned SCMP Interface Down messages: This has been a topic of discussion within the SCION community for a long time. Currently, in the deployment routers send a non-authenticated SCMP message, while it should be. There are some proposals to leverage the PKI to authenticate this. Harald proposed to say out loud that we don’t have support for authentication in current deployment but that this should be authenticated, showing this is a known problem. We should avoid saying in the draft that a feature exists but then it is not specified. Ok to say some problems still need to be solved but not yet a concern in current deployment. (he shared story about irc and ICMP destination unreachable)
Path revocation in case of link failures is not clear. We only mention something in controlplane-02 section 2.4:
We should make this clearer in -dataplane too.
TODO:
Feedback details
Relates to https://github.com/scionassociation/scion-research_I-D/issues/25