[cisco-voip] Looking for a sample 6513 IOS-based QoS config

[Andre Beck]

On Wed, Dec 06, 2006 at 09:12:55AM -0600, Jonathan Charles wrote:
> We are in the process of adding a new 6500 core and I am looking for a QoS

I wouldn't call that a core (phones are access), but anyway ;)

> config for VoIP... I will have some phones plugged into one blade, so I need
> the port configs as well... just looking for a sample, it is kind of
> difficult to pull out what I need from Cisco's web site...

The basic problem here is that switches are not really flexible in their
QoS designs. They can do exactly what the ASICs were designed for, and
that sometimes requires a lot of reading for the user to understand, too.

For the 3560/3750 there is a quite good description of how they work
in the manual, I assume that there will be even more and deeper for
the flagship. Don't expect to grok it in less than a day worth of
reading time, though.
 
> For instance, I want to trust the DSCP markings of the phone and I want a
> priority queue for VoIP bearer traffic (dscp ef)...

I'd recommend to have a look into the "auto qos voip cisco-phone" and
"auto qos voip trust" automation statements. They configure a rather
complicated global setup and apply everything needed to the interface
in the context of which they were called. The configuration that results
from these statements should be studied, understand and then changed
according to local needs. For instance, the "cisco-phone" variant on
a switched port will generate statements for CoS trust, not DSCP trust.
Further, it is a good question whether you need priority queuing at
all, at 100Mbps and especially 1Gbps, a shared or shaped queue with a
certain guaranteed bandwidth share will usually have low enough latency
and jitter impact for VoIP to work perfectly. On the 3560/3750 e.g. the
auto qos statements do not generate any priority queueing on the egress
ports, they rather create a shaped queue with 10% guarantee of the total
interface bandwidth. They use a priority queue in the ingress path though,
due to the switch design (potentially oversubscribed stack ring).

HTH,
Andre.
-- 
                  The _S_anta _C_laus _O_peration
  or "how to turn a complete illusion into a neverending money source"

-> Andre Beck    +++ ABP-RIPE +++    IBH Prof. Dr. Horn GmbH, Dresden <-