On Fri, Jul 31, 2015 at 12:31 PM, De Schepper, Koen (Koen) <[email protected]> wrote: > Surprisingly high for a an experiment of all combinations from 0 to 10 flows > of each (=120 combinations of up to max 20 flows together). Indeed using > birthday paradox, collision counting we calculated that on average each > experiment (the 120 combinations) has 6.5 collisions. The one in the > presentation had (by accident) the least (5), but others had more than 10. > > We did also experiments with an extra 20MBps UDP flow, and in the first run, > one of the collisions was in combination 1Cubic-1Dctcp. So with only 3 > flows, the DCTCP and UDP flow got in the same queue, and DCTCP completely > starved… For large amounts, statistics apply; for small amounts, Murphy ;-).
Was this the linux version or the ns2 version? The ns2 version had no direct support for accessing the packet headers and thus you had to configure the flows carefully and simulate that. Secondly in a more real world, additional entropy in the form of vastly different destaddrs would be around. If this was the linux version... interesting! I had complained about the quality of the hash before, notably on the lack of entropy in the proto field, and the xor in ipv6 (enough so that an entire new hashing architecture has been arriving in newer versions of linux with support for a much broader range of hash targets, and support for mpls, and macaddrs and so on) However, it is a shame to be unlucky, no matter the algorithm. 8 way set associativity is working pretty well elsewhere. > > Koen. > > > > > > From: aqm [mailto:[email protected]] On Behalf Of Jonathan Morton > Sent: dinsdag 28 juli 2015 11:45 > To: Dave Taht > Cc: Scheffenegger, Richard; De Schepper, Koen (Koen); [email protected] > Subject: Re: [aqm] Minutes of the AQM WG session > > > >> I don't know what a "surprisingly" high number of collisions is... > > Surprising to people unfamiliar with the birthday paradox, perhaps. The set > associative hash produces results that are much more intuitively reasonable. > > - Jonathan Morton -- Dave Täht worldwide bufferbloat report: http://www.dslreports.com/speedtest/results/bufferbloat And: What will it take to vastly improve wifi for everyone? https://plus.google.com/u/0/explore/makewififast _______________________________________________ aqm mailing list [email protected] https://www.ietf.org/mailman/listinfo/aqm
