[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: TC egress traffic rates



> > http://www.davin.ottawa.on.ca/ols/img17.htm
> > should help
> 
> To continue the discussion for learning purposes, the next logical
> question would be, why does the bandwidth setting have to be the
> device's true physical BW (the slide doesn't give an explaination)?
> If for example, you wish to simulate a lower bandwidth pipe (say
> ISDN), then subdivide *that* using cbq classes, what would be the
> procedure to follow?

Hi,
it should be clear from my previous discussion with jamal. Because
we can't compute real EOS times of netif we need to know physical
bw of device to be able to estimate EOSes by integrating time.
( Yes Jamal, after all I agree with you see my next email ;-] )

You can simulate narrow pipe by inserting one child class 
under top level cbq class and give it "bounded" parameter.
Then connect other classes under it.
Probably you will see that "simulation" will not be too 
precise. CBQ does good enough job to divide trafic on physical
link but is not too precise in bounding.
Read Hui Zhang's papers about HFSC vs. CBQ for more details.

regards, devik