Re: [LARTC] Prio queue with many bands

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

On Friday 06 June 2003 22:01, Michael Ulitskiy wrote:
> > > > On Friday 06 June 2003 18:30, Michael Ulitskiy wrote:
> > > > > Hi,
> > > > >
> > > > > I'm wondering if there's any simple classfull queue like prio that
> > > > > supports at least 256 bands? I don't care about priomap or some
> > > > > additional functionality, I just want to divide traffic in 256
> > > > > bands using hashing filters using as least resources as possible. I
> > > > > understand I can do that with htb for example, but I guess htb is
> > > > > much more resourse-consuming.
> > > >
> > > > Why do you think so?  I think you will surprised how many active
> > > > classes an old 486 can handle.  It also depends on the maximum
> > > > bandwidth the box has to shape.
> > >
> > > I think so because unlike simple prio queue htb has to maintain some
> > > timers and counters and buffers. I'm not saying that htb is heavy, but
> > > I'm quite sure that prio is lighter. I have to shape /19 at the moment
> > > and possibly more in the future. In this case I want to save as much
> > > resourses as possible and attach htb classes to the simple queue bands.
> >
> > /19 means 2^32 hosts = 8192 hosts.  I think a P2 can handle this load
> > with no problems.  But it depends on the bandwidth it has to shape and
> > the complexity of the filters.  I can do the test if you are interested.
>
> Sure, I'm interested. As well as I'm interested to see your configuration
> to do so. I'm still experimenting. Thanks Stef.
I have 1 server (AMD, 300Mhz) and 1 client (P, 120Mhz).  The shaping is done 
on the client.  The server starts the listen daemon from inetd.
Starting 30 sessions gives a load of 25 on the client.  The load is caused by 
the ttcp prog I use.  So I need a seperate shaper for some testing.  On the 
other hand, even with a load of 25, the sessions were shaped like it should 
be.  So first I have to find some hardware before I can test it with more 
sessions.

> The problem is that the machine this all intended for is not a pure
> router/shaper, but a server with current traffic about 10-15M. Its usage
> getting out of control and needs to be shaped. At this point I have no idea
> how much resoures the traffic shaping would consume and whether I have to
> introduce intermediate shaper device or I can configure qos on the same
> machine.
Only some testing can give you an idea about how scalable it will be.

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/