Re: [quicwg/base-drafts] Does it make sense to try 0-RTT after Retry? (#2842)

[Kazuho Oku <[e-mail]>]

@marten-seemann 
> Since we can assume that most (all?) server implementation will drop 0-RTT packets when sending a Retry, the only reasonable thing to do for a client is to treat all 0-RTT packets as lost and retransmit them. Considering that, it would make sense to require servers to drop all Retry packets from the first flight, as @kazuho suggests.

Thank you for suggesting a path forward.

I think my weak preference goes to something slightly different.

I think we are in agreement that most if not all the servers would not buffer the 0-RTT packets carrying the original DCIDs when they send Retrys. It makes sense to acknowledge the fact in the specification and suggest clients to retransmit 0-RTT data it has already sent.

What I am not sure is if there is a reason to forbid servers from buffering such 0-RTT packets.

>From client's point of view, sending 0-RTT data once per connection is an optimization. Retransmitting 0-RTT data when the server sends a Retry is further optimization. If we think that way, servers that buffer 0-RTT packets when sending a Retry has increased chance of utilizing 0-RTT data. Compared to that, there is no reason to mandate servers to drop 0-RTT packets.

To summarize, I think all we need to clarify is that servers responding with Retries are likely to discard 0-RTT packets that carry the original DCIDs and therefore that retransmitting 0-RTT data makes sense. But nothing more.

-- 
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/issues/2842#issuecomment-505390494