> They get a different speed for traffic staying on our network than for > traffic towards/from the internet, > so that's a master class and 2 child classes per customer per interface. Ok. > I made a test setup with cbq which worked, but wasn't too reliable I > measured a tolerance of about 30%. > I read that cbq is not maintained, htb is much more reliable, and I > believe I can do the same classful > stuff mentioned above with htb. Correct. > The box I will use to limit the traffic on has 3 ethernet connections > with customers and 1 uplink. > I read somewhere that only outgoing traffic can be limited. Only outgoing traffic can be classfully shaped or limited. > Is that correct or will limiting of incoming traffic work but isn't it > just as reliable? It's just not very flexible. > If I would filter outgoing traffic from the customer on the box, I would > have to do that on every > interface except for the one the customer is on. Therefore the client > will be able to sent out > more traffic than allowed, if it is spread over multiple outgoing > interfaces. What is your concern, inter-customer traffic ? Or even the Internet traffic can go thru more than one interface ? > Is there a solution to this? Yes, IMQ. You can do ingress shaping with it, or you can bundle output traffic from various interfaces and shape at a single point. > If I use ip aliasing (a la eth0:1), does that mean I would have to make > qdiscs/classes for eth0:1 > or will the traffic be covered by the qdisc/classes on eth0? No qdiscs for aliases, all traffic is covered at that the main interface. Rubens _______________________________________________ LARTC mailing list / LARTC@xxxxxxxxxxxxxxx http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
