quicwg/ack-frequency

Response to Long Idle periods

gloinul opened this issue · 5 comments

It is important to receive timely feedback after long idle periods, e.g. update stale RTT measurements. When no acknowledgement has been sent in over one smoothed round trip time, receivers are encouraged to send an acknowledgement soon after receiving an ack-eliciting packet.

Why is the formulation here "soon after receiving an ack-eliciting packet"? Would it not be better to clarify that this is a SHOULD Immediately ACK if no ACK has been sent for more than S_RTT? This would ensure a more rapid recovery from any temporary failures of the path to deliver any packets.

mirjak commented

This text was added to address issue #209 and added by PR #217.

As you can see in the PR, we purposefully didn't say immediate because it doesn't have to be immediate. However, we also thought that it might be sufficient to be vague here.

As part of this discussion are there reasons for not sending an ACK as soon as possible on the first ack-eliciting packets after a period of inactivity?

Just trying to understand if we create more uncertainty by not being prescriptive in some of these cases?

mirjak commented

I don't think we discussed specific scenarios but we felt that this is a more general issue and not something that should be "solved" in this extension. That's why we thought normative language might be inappropriate here but we still wanted to note the issue.

I think the core of my concern here qa if this document actually results in a change or uncertainty about what the default behavior are in implementations. But if you don't think, I am fine with closing the issue.

mirjak commented

Having closed this issue because I agree about the uncertainty but that is a generally issue as I said and not specific to this draft. If needed I think it needs to be addresses in a more general way and is therefore out of scope for this draft