Re: [quicwg/base-drafts] Rework Key Update (#2237)

[Kazuho Oku <[e-mail]>]

kazuho commented on this pull request.



> @@ -5012,6 +5020,44 @@ Reason Phrase:
   This SHOULD be a UTF-8 encoded string {{!RFC3629}}.
 
 
+## KEYS_READY Frame {#frame-keys-ready}
+
+An endpoint sends a KEYS_READY frame (type=0x1e) to signal that it has installed
+keys for reading and writing packets.  Receipt of this frame in a packet
+indicates that all earlier keys can be safely discarded.

> I think that we just have to make some allowances in the congestion controller/loss recovery for that possibility. For packets that we abandon like this, we lose the positive signal (i.e., congestion window increase), but we don't necessarily create a negative signal, because abandonment means that you don't declare them lost.

I think you are correct in pointing out that there's no negative signal even though we are losing a positive signal. As implied in my previous comment, my preference goes to delaying dropping keys until all the transmissions for that epoch is complete because that allows us to decouple loss-recovery / congestion control and epochs (FWIW, we will not bee seeing ACKs for ServerHello and ClientFinished). But I would not argue strongly, because there are ways to compensate for the missing signal.

> I'll tweak the text about the dropping of keys requiring both sending and receiving of the frame.

:+1:

-- 
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/2237#discussion_r256235145