gorryfair
commented
6 months ago The answer to the first comment is that: yes this sender is constrained initially. We can add text to explain that clearly.
Second, Careful resume is unlikely to provide benefit if the CWND approaches the jump_cwnd, as you say. CR is more conservative when validating, and uses pacing (which can slow a sender). This means there is no benefit for CWND approaches jump_cwnd, I think the draft earlier notes not saving a CWND when (CWND/4)<IW, but we will for sure read the text and see if more can be said for this specific case.
joergdeutschmann-i7
Thanks for adding the helpful table and examples in appendix A. The table says the CWND is increased according to slow start rules in Reconnaissance phase, until the sender becomes CWND-limited and CR confirmed. Then example A.1 mentions the sender enters Unvalidated phase from Reconnaissance when it is CWND-limited with 29 packets in flight. I have two questions on that.
First, Is this because the sender also needs several RTTs to confirm the path (as said in the editor note in Sec 3.2)? 'Cuz if the sender has abundant data to send all the time, it becomes CWND-limited very quickly before CWND reaches 29?
Second, if the case in A.1 may happen, it is also possible that the CWND could be larger than the calculated jump_cwnd when entering Unvalidated phase (e.g., in this case, CWND = 29, while saved_cwnd = 4*IW = 40). In that case, careful resume will otherwise slow down the ramp up.