I am planning a setup with thousands of classes in a HTB qdisc, say from 1:1000 to 1:2000, each with a very small rate and a big ceil, for fair sharing of a 45mbit link. I suspect some problems could be lurking in there. Anyone having good/bad experience with such number of classes? Simon _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Fri, Aug 27, 2004 at 10:46:59AM +0200, Simon Lodal wrote:> > I am planning a setup with thousands of classes in a HTB qdisc, say from > 1:1000 to 1:2000, each with a very small rate and a big ceil, for fair > sharing of a 45mbit link.Consider using HFSC. HTB is not the best solution for such a number of classes with small rate. Users will not be able to get the whole ceil even if there will be avaliable bandwidth. Secondly make sure your''re using kernel with double linked list patch applied to qdisc api. -- Tomasz Paszkowski
On 27 August 2004 pm 15:46, Simon Lodal wrote:> I am planning a setup with thousands of classes in a HTB qdisc, say from > 1:1000 to 1:2000, each with a very small rate and a big ceil, for fair > sharing of a 45mbit link. > I suspect some problems could be lurking in there. > Anyone having good/bad experience with such number of classes? > SimonI tried with 1:9999 as parents and child classes from 1:1000 until .. 1:9900 increments by 10, ... phuff .. nothin to worry about although checking it one by one will make ur head dizzy .. :)) - Rio.Martin - Powered by HTB since early 2003. _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
HFSC seems interesting, but does it really work well? I am not afraid of new stuff, but bleeding edge is probably too risky. That HTB problem, I guess you mean it is possible to have available bandwidth, but when some of it has been distributed between users, all users'' ceils go below their quantums? But that implies that users have low ceils? Wouldn''t it be solved by setting all users'' ceil to the full link bandwidth? The doubly linked list patch you mention, I believe it should be in vanilla kernel since 2.4.20-22 or something? I use 2.4.27 here. If not, do you have a link for it? Simon Tomasz Paszkowski skrev:> On Fri, Aug 27, 2004 at 10:46:59AM +0200, Simon Lodal wrote: > >>I am planning a setup with thousands of classes in a HTB qdisc, say from >>1:1000 to 1:2000, each with a very small rate and a big ceil, for fair >>sharing of a 45mbit link. > > > Consider using HFSC. HTB is not the best solution for such a number of classes > with small rate. Users will not be able to get the whole ceil even if > there will be avaliable bandwidth. > > Secondly make sure your''re using kernel with double linked list patch applied > to qdisc api. >_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Fri, Aug 27, 2004 at 12:22:13PM +0200, Simon Lodal wrote:> > HFSC seems interesting, but does it really work well? I am not afraid of > new stuff, but bleeding edge is probably too risky.I''am running HFSC for network with 4,5k users, handling about 100Mb/s internet traffic with 25kpacket/s without any problems.> > That HTB problem, I guess you mean it is possible to have available > bandwidth, but when some of it has been distributed between users, all > users'' ceils go below their quantums?no> > But that implies that users have low ceils? Wouldn''t it be solved by > setting all users'' ceil to the full link bandwidth?ceil is set to the full link bandwidth. I dosen''t help.> > The doubly linked list patch you mention, I believe it should be in > vanilla kernel since 2.4.20-22 or something? I use 2.4.27 here. If not, > do you have a link for it?http://oss.sgi.com/projects/netdev/archive/2004-08/msg00348.html Consider also using rbtree HFSC patches. -- Tomasz Paszkowski
[back from reading up on HFSC] I am much tempted to go with HFSC. But I am also much tempted to stay with HTB only because HFSC is so complex, and randomly documented. Understanding the shortcomings of algorithms is key to using them. What a shame, it seems like there is some real smartness in there. Would you like to share your HFSC configuration? (less than the 27k lines you refer to in other places will also do :) Thanks for the pointers, I will look into them. Simon Tomasz Paszkowski skrev:> On Fri, Aug 27, 2004 at 12:22:13PM +0200, Simon Lodal wrote: > >>HFSC seems interesting, but does it really work well? I am not afraid of >>new stuff, but bleeding edge is probably too risky. > > > I''am running HFSC for network with 4,5k users, handling about 100Mb/s internet > traffic with 25kpacket/s without any problems. > > >>That HTB problem, I guess you mean it is possible to have available >>bandwidth, but when some of it has been distributed between users, all >>users'' ceils go below their quantums? > > > no > >>But that implies that users have low ceils? Wouldn''t it be solved by >>setting all users'' ceil to the full link bandwidth? > > > ceil is set to the full link bandwidth. I dosen''t help. > > >>The doubly linked list patch you mention, I believe it should be in >>vanilla kernel since 2.4.20-22 or something? I use 2.4.27 here. If not, >>do you have a link for it? > > > > http://oss.sgi.com/projects/netdev/archive/2004-08/msg00348.html > > Consider also using rbtree HFSC patches. >_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/