I want to know how accurate CBQ can split the bandwidth in 2 differnt classes. You can do it on 2 different ways: you provide no weight parameter and uses the rates or you can provide a weight parameter. You also have the possibility to use an extra qdisc connected to a bounded class.
- no weight parameter: traffic division = proportional to rates of classes (See tests with rate) but not so accurate as with the weight parameter
- weight parameter: traffic division = proportional to weights of classes (See tests with weight parameter) but only accurate when the proportion of the rates you provide is smaller than the proportion of the weights.
If you want to split the traffic in different classes so the classes can borrow traffic from each other, create a bounded class with the total bandwidth, attach a qdisc to it and attach the classes who may share the bandwidth to this qdisc.
The qdisc attached to the bounded class will prevent that the other classes can get more bandwith then they may get.
I made this overview from all the splitting-tests. For more explanation, visit the two splitting pages and read the conclusions.
- Rates + no weight
- qdisc: OK (Splitting proportional to rates)
- NO qdisc
- Filter : 10 - 10:2 - 10:10: Each class gets 50%
- Filter : 10 - 10:10: Full bandwidth (no QOS)
- Rates + weight
- qdisc: OK (Splitting proportional to weights)
- NO qdisc
- Filter : 10 - 10:2 - 10:10: Each class gets 50%
- Filter : 10 - 10:10: OK (Splitting proportional to rates)