From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alexandru Matei Date: Mon, 23 Jan 2006 14:53:53 +0000 Subject: Re: [LARTC] multiple wrr as child of htb Message-Id: <43D4EE01.403@qb.ro> List-Id: References: <43D0B7B2.2050700@gmail.com> In-Reply-To: <43D0B7B2.2050700@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: lartc@vger.kernel.org Thanks all for your answer. I'll see how it will behave in practice and let you know. Thanks again, Alex Peter Surda wrote: > Alexandru Matei schrieb: > >> Ok, maybe I was not clear enough. > > > Yes you were :-) > >> What I'm interested in is the order of passing throught filters. WRR >> has an built-in filter. > > > Yes. But this isn't anything special per se, any classful qdisc > distributes packets somehow, you just usually "tune" it by using tc > filter. > >> HTB child classes are fed by matching certain ip addresses (in my >> case htb chlid1 is for x.x.x.0/29, htb child2 is for x.x.x.8/29 ) >> and so on. > > > Yes. > >> I want to attach wrr1 as child to htb child 1 so the traffic get >> even divided between clients x.x.x.0/29, wrr2 as child to htb child 2 >> and so on. > > > That isn't a problem (other than potential stability issues I > mentioned before). > >> What I don't know is: >> 1/ tc filter match for is evaluated first > > > Of course. > >> 2/ if wrr1 is going to classify only hosts seen in x.x.x.0/29 or all >> hosts x.x.x.0/24 > > > WRR classifies those packets that "flow" through it, like any other > qdisc. It doesn't care whether they are a /29 subnet or some arbitrary > IPs (it uses hashing in a way similar to conntrack for assigning > classes). > >> Regards, >> Alex > > > Yours sincerely, > Peter > _______________________________________________ LARTC mailing list LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc