Re: [LARTC] equal bandwidth for all IPs

[Andy Furniss <andy.furniss@dsl.pipex.com>]

Peter Surda wrote:
> On Sun, 22 May 2005 18:51:36 +0100 Andy Furniss <andy.furniss@dsl.pipex.com>
> wrote:
> 
> 
>>You could aswell as wrr consider esfq - you can only roughly divide what
>>bandwidth you have, but the advantage over wrr is that you can choose a
>>queue length for your link speed.
> 
> ESFQ is good, but isn't a panacea. You still can use esfq as a queueing
> discipline for wrr's subclasses. In fact it's a perfect match, because you have
> both fair division among sources AND among the connections made from the same
> IP.

I agree wrr is better than esfq in many ways. I also haven't used it or 
looked at it for some time, so may be totally wrong about what you can 
and can't do with it :-)

> 
>>I don't think you can di it with wrr and if you are shaping download
> 
>>from the internet it could be better.
> You don't have to do this directly with WRR, but you can do it with HTB "above"
> WRR. As for download, you can use IMQ.

What I meant was I don't think you can choose one length for the whole 
wrr class - you can add queues to wrr like you can htb classes and you 
can limit those, but you have one for each user - so with many users and 
shaping from the wrong end of a slow link you end up with a total queue 
length that is too long to keep good control of the link.

The above may not be such an issue depending on the speed of the link 
and the number of users active at any one time.

Andy.
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc