The documentation (2.4routing) seems to imply that SFQ balances on a per connection basis. It would seem to be more fair to balance on a source ip basis as that would evenly divide bandwidth among users. Is this correct? If so, is there an alternative qdisc that would provide the above? Regards, David Koski david@KosmosIsland.com
Hi ! David Koski schrieb:> > The documentation (2.4routing) seems to imply that SFQ balances on a per > connection basis. It would seem to be more fair to balance on a source ip basisSFQ places packets in slots based on a hash function. Then each slot is dequeud in a round-robin fashion.> as that would evenly divide bandwidth among users. Is this correct? If so, is > there an alternative qdisc that would provide the above?I don''t think there is a qdiscs for this but SFQ should be easy to change. Changing the hash function to something like iph->saddr % SFQ_HASH_DIVISOR should be enough .. Bye, Patrick
On Wed, 10 Apr 2002, Patrick McHardy wrote:> Hi ! > > David Koski schrieb: > > > > The documentation (2.4routing) seems to imply that SFQ balances on a per > > connection basis. It would seem to be more fair to balance on a source ip basis > > SFQ places packets in slots based on a hash function. Then each slot is > dequeud in > a round-robin fashion. > > > as that would evenly divide bandwidth among users. Is this correct? If so, is > > there an alternative qdisc that would provide the above? > > I don''t think there is a qdiscs for this but SFQ should be easy to > change.The WRR (http://wipl-wrr.sourceforge.net/wrr.html) scheduler can do this job. Wojtek Sawasciuk