On Fri, Feb 24, 2006 at 06:19:53PM -0300, Andreas wrote: > >This was fairly obvious looking at your tc statistics output, where it > >lists both 1:1 and 1:2 as roots with no parent. There can only be one > >valid root class. > > Why? I need two virtual circuits. I don't want the 90mbit class > interfere with the 200kbit class: no lending, no borrowing.
I think there can be more than just one root class - the question is just wether it makes sense or not. I prefer using one root class - after all, you only got one interface, and you have to make sure that you do not exceed the total interface capacity. Therefore, the root class is the interface limiter. You can add isolated circuits to that root class easily; as long as all child classes of the root class have the same rate and ceil, no lending or borrowing between them will be done, simply because it is not necessary. This way you get your desired features plus an overview on how much rate the physical interface actually has to offer - from my point of view, that's a win-win situation. > It actually works if I use a *leaf* class as the target of the filter > (see my subsequent email). But this contradicts the documentation, which > even mentions one could gain speed by adding further filters to other > classes besides a root one. I never got filters to work that do not point to leaf classes. Wether it is possible at all or not, I do not know. Maybe it was planned and turned out to be too complicated - maybe it is implemented but not working due to some undiscovered bug. I'm too lazy to look at the code right now. I usually end up using iptables for classification; I find it to be far more userfriendly than the tc filters, and you can group filters any way you want. Regards Andreas Klauer _______________________________________________ LARTC mailing list LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
