Hi! > So I played a bit with the DOWNLINK value and found out that down to and > including 555 no packets are dropped even if a download fully saturates the > link while 554 already drastically caps the download to ~65 kbytes/s, and > there are packets being dropped: It seems that your setup suffers from timer inaccuracy. This topic has been dealt with some times on this list. Read (for example) this thread: http://www.mail-archive.com/lartc@xxxxxxxxxxxxxxx/msg02795.html In short: it might be that the timer of your qdisc runs at to low a resolution to measure exactly how much packets to let pass since the last time it could send. Therefore it seems that the bandwith cap of the qdisc does not correlate linear with the given kbps value but more or less like a staircase (ie. a dramatic drop on a certain kbps value - like you described it at 554kbps) > Lower values than 554 drop even more packets, just as I'd expect; there > seems to be no other threshold as hard as from 554 to 555. The stairs should get smaller with the kbps value specified but should remain stairs and not become a linear slope - assuming I've guessed the source of your problem correctly. The solution (as mentioned in the thread above) is to either change the HZ value to for example 1024 (you don't have to do this by manually patching your kernel source because your gentoo kernel has a patch applied which lets you specify the HZ value from kernel config - I think its somewhere in "Processor type and features") or to change PSCHED_CLOCK_SOURCE to PSCHED_CPU in your /usr/src/linux/include/net/pkt_sched.h (which I personally had some problems with in the past - can't remember exactly though). I hope this helps - let us know... Andreas _______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
