Interesting, well, I guess I have my research for today!  Thanks again.

nb

On Wed, Jul 16, 2003 at 11:56:41AM +0159, Henning Brauer wrote:
> On Tue, Jul 15, 2003 at 06:55:58PM +0200, Daniel Hartmeier wrote:
> > On Tue, Jul 15, 2003 at 09:44:32AM -0500, Nicholas D . Buraglio wrote:
> > 
> > > Multiple groups with different base rates; say the level 1 group will allow you 
> > > a base rate of 128k and 5Mb of queuing.  The first 5Mb you download would be at 
> > > 3Mb, after that the speed would drop to 128k.  Here is the twist.  Let's say you 
> > > download 4Mb out of your queue and then you continue to use around 56k 
> > > constantly (obviously this would fluctuate).  The queue would then build back to 
> > > the 5Mb at the rate of the difference between your base rate (128k) and what you 
> > > are using (56k).  So in our example we would be charging our queue at 56k.
> > 
> > I don't think this is doable. AltQ has no concept of 'the first 5Mb you
> > download'. Firstly, it doesn't know about users, just source/destination
> > addresses (so that would make it 'the first 5Mb sent/received by a
> > host'). Next, it doesn't know about connections, either, so you'd be
> > down to 'the first 5Mb of the total traffic sent/received by the host,
> > since reboot'. And AltQ doesn't even keep counters like these, so it's
> > totally unexpressable :)
> > 
> > Currently, AltQ does not keep any per-host statistics/counters. All it
> > can read and update is per-queue values. And even if you'd create one
> > queue per host, it wouldn't care about the amount of past traffic, all
> > it cares about is rates.
> > 
> > If I'm missing something, please correct me.
> 
> no, you are entirely right.
> 
> However, what Nicholas describes is basically how hfsc works... it's 
> probably not fine-tuneable to exactly match his description, but 
> should be reasonable close.
> 
> -- 
> Henning Brauer, BS Web Services, http://bsws.de
> [EMAIL PROTECTED] - [EMAIL PROTECTED]
> Unix is very simple, but it takes a genius to understand the simplicity.
> (Dennis Ritchie)

Reply via email to