Re: [LARTC] HTB for multiple users, good or bad approach?

Linux Advanced Routing and Traffic Control

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

 



On Friday 01 August 2003 19:28, Andreas Klauer wrote:
> Am Friday 01 August 2003 17:39 schrieb Stef Coene:
> > > The script I've come up with so far can be found there:
> > > http://www.metamorpher.de/files/ipshape.sh
>
> I updated the script.
>
> > Some remarks.  You define a burst on your 1:1 class.  You
> > don't need this.
>
> I bet there are lots of thinks which I don't need.
> I'm not familiar with any of these parameters, mainly I'm just
> more or less playing around with it. I removed this.
>
> > And make sure that the sum of the rate of
> > child classes = rate of parent class.
>
> I tried that. I noticed that I was wrong at one spot, but
> it should fit now, unless there are some rounding errors
> (which, hopefully, do not have such a bad impact on things).
>
> > Also, limit all bandwidth too less then your link
> > bandwidth so YOU are the botleneck
>
> How much less would you recommend?
> I subbed 5% of link bandwith for now.
5% is fine.

> > Using different prio's for you htb classes can be tricky.
> > Remove the prio parameters untill you understand what the
> > different prio's does.
>
> Yes, now that I read the FAQ on your page, I guess my
> prio values were all to high. They're all between 0-7 now,
> I hope thats better. If I understand the prio right, I need
> them, because I want the interactive connections to take
> as much bandwidth as they can get, even if the users are
> quite active themselves too... I also raised their rates.
> I reckon, though, that these settings still require lots
> of tweaking...
The problem with low prio (and so higher priority classes) happens when you 
overlimit the class (sending more data then the rate of the class).  If you 
overlimit it, the latency can be very high.

> > For the quantum problem, see the faq pages on www.docum.org
>
> I read your FAQ before, and I'm sure I still don't understand
> it fully. I use some (mostly randomly chosen) quantum values
> now, just to make these messages vanish... hopefully they make
> any sense...
You need to take r2q so for the lowest rate, rate / r2q > 1500 and for the 
highest rate, rate / r2q < 60000.  This is only for leaf classes (class with 
no child class).  If you don't find a r2q that can do this, you have to 
choose the best r2q and overrule the quantum for the classes where the rules 
are not satisfied
You can ignore the quantum errors, but it's better to make sure they you don't 
get them.

Stef

-- 

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



[Index of Archives]     [LARTC Home Page]     [Netfilter]     [Netfilter Development]     [Network Development]     [Bugtraq]     [GCC Help]     [Yosemite News]     [Linux Kernel]     [Fedora Users]
  Powered by Linux