Hi Tony, I agree that "minimal flooding time " is a good choice, which may differ from the traditional cc in layer 4, which is difficult to get the completion time for each flow. I am still a little confused about " fast brand X needs to not overrun slow brand Y while performing well with fast brand Z". Is this talking about interoperation between different company device, which may use different type of cc mechanism?
Best Regards Xuesong > -----Original Message----- > From: Lsr [mailto:[email protected]] On Behalf Of [email protected] > Sent: Thursday, May 7, 2020 12:02 AM > To: Gengxuesong (Geng Xuesong) <[email protected]> > Cc: [email protected]; Christian Hopps <[email protected]> > Subject: Re: [Lsr] Congestion (flow) control thoughts. > > > Hi Xuesong, > > > <Xuesong> I think there is no need to distinguish the concept of flow > > control > and congestion control, considering that the core idea is the same: monitor > the > sending rate to match the capability of the bottleneck, no matter there are > competitors or not. And the control loop is necessary in both case. > > > This matches my perspective. > > > > <Xuesong> Thank you for explaining about the bottleneck of flooding and > different cc solutions that are under discussion. It is helpful and I will > read the > drafts. > > There is still one more question left in the previous email: " What is the > criteria of comparing different solutions?". I think this is crucial for > further > discussion and comparative tests. I notice that in Bruno's data, some > parameters are mentioned, such as " Duration"," LSP/second"," avg inter-LSP > delay", "retransmission time". I'm wondering whether it is able to choose a > better solution through these parameters. If not, what should be added or > considered. (Some other issues are also considered when choosing cc > mechanism in layer 4, such as: fairness among flows, co-existence with other > cc mechanism .. ) > > > Sorry for missing the question. The criteria are, I think, quite clear: > minimize > overall flooding time. As Bruno’s figures show, that does NOT happen simply > by transmitting at the maximum rate. That causes overruns, resulting in > retransmissions and that ends up being quite slow (tho we can work on that > too). The question then is: what control mechanisms do we put in place to > ensure minimal flooding time? This needs to interoperate across the full > spectrum of implementations: fast brand X needs to not overrun slow brand Y > while performing well with fast brand Z. > > The floor is very much open. > > Tony > > _______________________________________________ > Lsr mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/lsr _______________________________________________ Lsr mailing list [email protected] https://www.ietf.org/mailman/listinfo/lsr
