Re: [quicwg/base-drafts] compensation of ack_delay is fragile against errors (#2060)

[Kazuho Oku <[e-mail]>]

> If there is a high rate of ACK loss or if the peer's alarms are persistently waking up late, then the SRTT and/or RTTVar will increase, thereby increasing these timers. If this happens very infrequently, then the EWMA will forget them relatively quickly. I think that seems appropriate?

I had not thought about that, and admittedly I am unfamiliar with loss recovery, however I wonder if that's appropriate.

IIUC, the reason we use X * SRTT where X is slightly greater than one is because we expect an ACK to arrive in SRTT with some jitter.

However, in case of ACK loss, the SRTT using the computation would become too large for the case where there is no loss, or too small for the case where there *is* a loss. For eaxmaple, in the case of having ACK loss at 33%, SRTT becomes 33% greater than the RTT. I am not sure if that is a desirable behavior.

If we can assume that ACK loss happens infrequently, IMO the correct thing to do would be to ignore the effect of ACK retransmissions.

![loss - 1](https://user-images.githubusercontent.com/41567/49265498-a5935200-f495-11e8-85e4-e2d394d9e1e1.jpg)

-- 
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/2060#issuecomment-443072299