On Wednesday 21 May 2003 04:53, Martin A. Brown wrote:
> : > : When configuring HTB classes, is it true that the rate of a parent
> : > : MUST be exactly equal to the sum of rates of it's children?
> : >
> : > No. This is not true. This is merely a recommendation for how to use
> : > HTB. When the parent class ceil matches the total of all of the ceil
> : > parameters of the children classes, there is no way for bandwidth
> : > usage to exceed the parent ceil.
> :
> : It's not the "total of all of the ceil", it's the maximum of all the
> : ceil parameters of the children classes.
>
> Hmmmm.....I think I see my error. But I think we have an challenge in
> accurately describing it. So, to be unbelievable verbose:
>
> If the parent class "ceil" parameter matches (or is greater than) the
> sum of the "ceil" parameter for all of the leaf children (terminal
> HTB children), then there is no way for this parent's "ceil" to be
> exceeded.
>
> This definition is halfway between the "maximum of all the ceil
> parameters of the children classes" and the "total of all of the ceil",
> both of which are inaccurate in different ways. Do you agree, Stef?
>
> ( Martin wonders if Stef enjoys splitting hairs. )
What about ...
"The ceil and the rate of a class are respected, but the rate for each class
is always satisfied."
I'm doing some tests to prove this. But I have some strange results. I think
it's a quantum problem. The restult is an error of 5-10% if I create 2
classes with the same rate and quantum. If a class sends more data then its
rate, you use quantum. And that's not accurate.
> : And you can exceed the parent ceil. See me other post on this subject.
> : It can because the rate of a class is the minimum the class always gets.
> : Even if it exceeds the parent ceil.
>
> Agreed.
>
> : > Miernik--I would recommend this type of approach. In your example,
> : > though, you are using a rate which is below Devik's recommend minimum
> : > rate for reliable operation of HTB [2]. Perhaps you could create 25
> : > classes, and put two IPs in each class. That gives you a 6kbit rate
> : > for each child class.
> :
> : I think Devik means that this is a minimu rate for active classes.
> : And he states that there are only some of the classes active. So the
> : real rate will always > 6kbit. As long as quantum is big enough, I
> : don't think there will be a problem.
>
> Ah-ha! This makes sense. Thanks for the clarification.
But to prevent problems, he should add quantum 1500 if he creates the classes
to prevent sharing problems (this is only needed if quantum < 1500 if you jtb
calculates the quantum).
Stef
--
stef.coene@xxxxxxxxx
"Using Linux as bandwidth manager"
http://www.docum.org/
#lartc @ irc.oftc.net
