On Wed, 28 Mar 2001, Christian Worm Mortensen wrote: > Hi, > > > > Yes, but in the howto also a qdisc named WRR is mentioned. > > > > What HOWTO? > > This thread is on two mailing lists. One of the lists is for the howto > located on http://ds9a.nl/2.4Routing/. This howto mentions the WRR qdisc > (see http://wipl-wrr.dkik.dk/wrr/) which is a qdisc that is _not_ > included in the standard kernel and has nothing to do with CBQ. > Ah, this is something Bert Hubert was trying to put together a few months back. It seems to be going well (although it didnt seem like he changed the ingress qdisc stuff like i suggested ;->) > > WRR works well when you apriori know the packet/cell sizes (eg in ATM). > > If you cant do this, then WRR is unfair once you start having a lot of > > flows going or you mistweak your weights etc. DRR fixes this. > > Hmm... Maybe you talk about how WRR/DRR is implemented in CBQ? > A pure WRR scheduler works perfect no matter what size the packets have. > If, of course, the scheduler takes packet sizes into account. > What exactly is the problem with a WRR scheduler? There are other minor details (hence my suggestion to read the paper because i cant remember details), but fixing the deficit such that you take into consideration 'byte credit' a queue has when you preempt it makes a WRR implementation closer to DRR. If you are already doing this just read the paper and fill up the other minor details. You might also wanna hook up with Martin aka 'devik' -- if i am not mistaken he also wrote a DRR scheduler > > Another thing: Unless you have a need to give special traffic very low delay I don't see any reason why you would want to use CBQ instead of pure WRR? I.e.: If you use CBQ with all prio parameters set to the same, why not use pure WRR instead? > > > Why dont you read the classical paper at: > > http://www.acm.org/sigcomm/sigcomm95/papers/shreedhar.html > > What I really need is a paper describing CBQ in Linux - the original > article desribing CBQ is very generel. And when I experimented with CBQ > the last time I did not see the behaviour I would exepect from the > article assuming that the generel scehudler was a WRR scheduler. The original CBQ implementation is the classical WRR; linux is DRR. So the behavior might not be exactly the same; if the results make intuitional sense then that should suffice. You might have been a victim of bugs or mistuning of parameters. Note there have been attempts to document Linux CBQ; search the mailing list. cheers, jamal
