From: Dennis Hartmann (sanester8@xxxxxxxxx)
Date: Wed Jul 24 2002 - 17:38:57 GMT-3
Answers inline with Questions.
> 1)If a Hub with a large CIR or AR (1544) and has a
> Multipoint circuit
> and defined several subinterfaces on networks with
> the spokes which
> have smaller CIR or AR (64), should each
> subinterface have traffic
> shaping applied in such a way that the hubs
> bandwidth is divided up,
> and spokes are not overwhelmed, or should each
> subinterface have
> traffic shaping applied in such a way that they
> could utilize the whole
> 1544 during times where the other PVC's have no
> traffic?
It depends. Do you have T-1 access at each of the
remote sites? May want to leave it alone and burst a
bit. If you're sending voice over these circuits, you
DEFINETELY want to traffic shape to CIR. It's a
double-edged sword. You want to make use of your
bursting capabilities, but you don't want to have
excessive egress queueing at the hub site because
you're overpowering all of these circuits with low
CIRs.
> 2) Is the basis for the Tc calculated by the router,
> by the configured
> CIR/Bc ?
>From CCO:
Committed Rate Measurement Interval (Tc)
The time interval during which the user can send only
Bc-committed amount of data and Be excess amount of
data. In general, the duration of Tc is proportional
to the "burstiness" of the traffic. Tc is computed
(from the subscription parameters of CIR and Bc) as Tc
=Bc/CIR. Tc is not a periodic time interval. Instead,
it is used only to measure incoming data, during which
it acts like a sliding window. Incoming data triggers
the Tc interval, which continues until it completes
its commuted duration. See also Committed Information
Rate (CIR) and committed Burst Size (Bc).
> 3) Can Be only be invoked if the CIR is less than
> the AR, and is it
> measured in what the burst would be {ie.(32000)}
> divided by the CIR/Tc
> (ie 8). Even though we can only burst in the first
> Tc and would only
> effectivly ever use a fraction of the 32000?
No. Be is only used if the average rate is excceded
and the Bc is exceeded. To use the Be, you need a
dual-leaky bucket algorithm.
> I know these have been discussed before on this
> list, and are written
> about in all the books mentioned on this list and
> are adressed on CCO
> but I have seen a couple of variations between
> explanations.
>
> Anthony Pace
>
>
> On Tue, 23 Jul 2002 11:40:51 -0700 (PDT), "Dennis
> Hartmann"
> <sanester8@yahoo.com> said:
> > Be VERY wary of the QoS books. A lot of them
> are
> > very bad. Cisco Press has a good one, but I
> forget
> > the name of it. I teach the DQoS and iQoS
> classes. I
> > would be happy to answer any questions.
> >
> > Sincerely,
> >
> > Dennis Hartmann
> > Instructor-Consultant
> > dh8@pobox.com
> >
> > --- David Wolsefer <DWolsefer@totality.com> wrote:
> > > I have found that the previous networkers
> > > presentations on QoS are good
> > > sources along with the QoS sections of IOS
> > > documentation. There are also
> > > several QoS books out there, but I haven't had a
> > > chance to read them yet.
> > >
> > > Regards,
> > >
> > > David Wolsefer, CCIE #5858
> > >
> > > -----Original Message-----
> > > From: Jaspreet Bhatia
> > > To: ccielab@groupstudy.com
> > > Sent: 7/22/2002 8:34 PM
> > > Subject: QOS resources
> > >
> > > Folks,
> > > Can anyone suggest some good
> > > resources for brushing
> > > up
> > > on QOS. Thanks
> > >
> > > Jaspreet
> > >
> >
>
This archive was generated by hypermail 2.1.4 : Sat Sep 07 2002 - 19:36:42 GMT-3