Hi, I've finally found some time to try to understand Dan Siemon's results on
http://www.coverfire.com/archives/2011/02/21/network-latency-experiments/ That's some excellent work, Dan, a lot of thanks for it. If I read it right, the BDP is 5 packets (4Mbit/s times 15ms), and SFB is being run with a per-flow queue length of 20 packets (the default); hence, it should be able to swallow the bursts necessary in order to fill the link. The chaotic variations in latency, as well as the high packet loss, indicate that either there's something wrong with the implementation, or that somebody somewhere is sending massive bursts of packets that sfb is discarding, as it is designed to do. It would be good to have a graph of packet loss as a function of time, in order to check if SFB is converging too slowly in this case. -- Juliusz _______________________________________________ Bloat mailing list [email protected] https://lists.bufferbloat.net/listinfo/bloat
