> On 17 Jun 2015, at 15:59, Agarwal, Anil <anil.agar...@viasat.com> wrote: > > Michael, > > Assuming fq-aqm performs aqm independently on each queue, the estimation of > the number of recently active flows must be done for each queue independently. > Not sure how fq would help. > Perhaps, the very fact that we are detecting packets as belonging to flows > and mapping them to queues can be used in such an algorithm. > Note that fq does not keep track of flows or their beginnings or ends. It > could.
it could have some notion of recently active flows... and that would of course make the implementation of aqm and fq less independent (who cares) > Non-responsive UDP flows and UDP-based TCP-like-transports (e.g., QUIC and > utp) would also need to be handled properly. > > Alternatively, the current packet drop rate of an adaptive AQM scheme itself > could be used to estimate the number of recently active flows. Just a thought. true, good idea... anyway this is getting off into research... but nice thoughts i think! cheers michael _______________________________________________ aqm mailing list aqm@ietf.org https://www.ietf.org/mailman/listinfo/aqm
