Re: [quicwg/base-drafts] Invert connection ID (#493)

[MikkelFJ <[e-mail]>]

I must admit that I am not aware of all the details influencing how the connection id is being used, but I had this idea:

What if each endpoint is allowed to choose their own connection id and stick it? This would mean that packets send from client to server would use one ID, and packets sent from server to client would use another ID. This will impact middle boxes, but not necessarily in a bad way.

Although this is not the original intent for the connection ID such a design would make it possible to remove IP and UDP headers on custom networks, such as IoT mesh networks. Here the connection ID would be the routable address. That would not work without a notion of direction with the shared connection ID.

Wrt. rotating connection ID, it could be required that an endpoint does not start using a new ID before a frame has been ack'ed in which the transtion from old to new ID has been communicated - with special handling for initial setup where server echoes inital client ID in early packets. If this ACK is not used, an end-point may have to cache data until it can make an association from old to new data and this could be abused.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/pull/493#issuecomment-299479343