Re: [LARTC] HTB and theory

[Stef Coene <stef.coene@docum.org>]

On Monday 09 December 2002 23:18, Abraham van der Merwe wrote:
> Hi Stef!
>
> > > I'll let HTB to automatically compute the values for 'burst' and
> > > 'cburst'. The problem is elsewhere. What are the correct values for
> > > 'rate' and 'ceil' of 11:2 class in this case? In fact, total value of
> > > 'ceil's from all sub-classes exceeds 5000Kbit, so there are moments
> > > when the bandwidth that comes from my ISP is bigger than guaranteed
> > > bandwidth. Is there some kind a theory that says how to establish the
> > > values of 'rate's and 'ceil's from the parent and its sub-classes?
> >
> > There are some rules : ceil of child <= ceil of parent, sum (child rates)
> > <= rate of parent ....  You don't have to follow this rules, but the
> > final shaping result can be strange.
> > See the faq page on www.docum.org.
>
> What would be nice is if you could document what happens when>
>      a) child ceil > parent ceil
Child ceil is used as maximum.

>      b) child rate > parent ceil
It depends on the child ceil.  Parent ceil is never used to limit the 
bandwidth.

>      c) child ceil > parent rate
>      d) child rate > parent rate
It's more complicated then this.  Other classes are also important.

> How does this influence the shaping results and the shaping of other
> classes? I've actually asked a similar question before: If you have
>
>            1:1
>            / \
>          /     \
>        1:2     1:3
>        /|\       \------
>      /	|  \       \     \
>    1:4 1:5 1:6     1:7   1:8
>
> if 1:7 and 1:8 respects the rate/ceil of 1:3 and 1:3 respects the rate/ceil
> of 1:1, but 1:4 does not respect the rate/ceil of 1:2 and 1:2 respects the
> rate/ceil of 1:1, does the shaping of 1:3 work as normal and only those
> classes under 1:2 are adversely affected or not? from your previous answer,
> it seems that the classes under 1:3 would not be affected, but I would love
> to know why. what happens with 1:4 requests some bandwidth which 1:2 does
> not have...
It's easier if you add some numbers of ceil/rate to understand the question.

I'm going to create an extra page on www.docum.org with your questions and my 
answer to explain how rate and ceil of classes and parent classes are used.  
Ok?

Stef

-- 

stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.oftc.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/