[c-nsp] Improved queuing in 12.4(20)T?

[Rodney Dunn]

On Tue, Aug 26, 2008 at 01:32:23PM +0200, Oliver Boehmer (oboehmer) wrote:
> Per Carlson <mailto:perc69 at gmail.com> wrote on Tuesday, August 26, 2008
> 1:16 PM:
> 
> > Hi Oli.
> > 
> >> I haven't looked at HQF for a while, but I recall the H-QoS scenario
> >> you're using benefits especially from HQF as the parent shaper is
> >> aware of the LLQ within the child, but not entirely sure about this.
> >> It would explain the improved behaviour, though.
> > 
> > I have always had the impression that the parent shaper *is* aware of
> > the child policy. Otherwise the whole H-QoS scheme is rather useless,
> > at least with respect to LLQ/PQ.
> 
> Hmm, I think I would need to do some digging here, but I was thinking
> about a different kind of "awareness" here. Obviously the shaper in the
> parent is aware of a child, so when the shaper has to queue a packet
> (i.e. signals connections), it'll use the child policy (which, as you
> write, is the whole point of H-QoS).

The difference I suspect is how the time intervals are handled. I worked
on an issue once where we actually would slightly burst above the
configured shape rate and it got worse the smaller the Tc was.
With HQF the excess is handled different and resulted in more accurate
shaping rates. I forgot all the nitty gritty details of how we did it.

> 
> Maybe Rodney can comment more, but I would still assume HQF being the
> reason for the different behavior in your environment.

Yep. I agree with you.

Rodney

> 
> 	oli
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/