Re: [quicwg/base-drafts] Which DCID do Handshake retransmissions use? (#3348)

[Martin Thomson <[e-mail]>]

Artificial constraints on when the frame can be sent would be terrible to enforce and would introduce a performance cost for using NEW_CONNECTION_ID.  In a great many cases, the 0.5RTT from the server is idle, which makes it a great way to get things like NEW_CONNECTION_ID sent without taking capacity from real work.  Asking the server to defer sending would mean that NEW_CONNECTION_ID could compete with HTTP responses and the like.

I see no reason not to allow use of connection IDs when they are available.  They apply to the connection as a whole.  You will note that we allow changing them with the Initial/Retry mini-protocol in ways that don't correlate with other connection-level events, so this isn't any different in that regard.  Forcing the connection ID state to synchronize with other state changes seems more likely to complicate things than help.

The caution @kazuho mentions is always true.  Whenever an endpoint is forced to retire connection IDs, a short supply is always a risk to the connection.

-- 
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/3348#issuecomment-574965824