QoS on sub-interface, confusion!

aacole · ‎07-29-2010

I've been trying to understand how QoS applied to sub-interfaces works, and cannot find any clear explanation.

My understanding is that a router configured with a number of sub-interfaces, no QoS applied, then each can have up to the physical line speed as its based on the clock rate of the attached circuit. If more than 1 subinterface is transmiting at once then the line speed bandwidth is shared between those active subinterfaces.

I was asked What happens if QoS is applied to only 1 of those subinterfaces, a parent policer and priority queue in a child policy on 1 subinterface?

My initial response was that a QoS policy should be applied to all subinterfaces, that seemed the correct way to do this, but, after reading through documentation I'm not sure that QoS on every subinterface is required..

Would the priority queue applied to one subinterface give traffic in that queue precedence over all the rest of the subinterfaces?

The example is where all subinterfaces are data only, apart from 1 which has a mix of voice and data.

abaskara · ‎08-03-2010

Hi there,

The answer depends on the type of interface. Let's say you are talking about ATM sub-interfaces, most of our non-legacy linecards support per-vc queueing. So, qos takes effect (during congestion) on a per VC level, and traffic will be prioritised.

If the linecard in question does not support per VC/sub-interface queueing structure, then the preferred method would be to apply qos on all the sub-interfaces, to equally share the link.

Thanks,

Arun

aacole · ‎08-16-2010

Hi Arun,

The particular type of interface I'm working with is an Ethernet interface supporting sub-interfaces with vlan's, this is on a 7600 with ES+ cards.

The physical interface connects to a service provider that’s using 30-40 MPLS Pseudo Wire circuits to the remote sites, each PW has a policed bandwidth allocated on the SP interface.

Currently the service supports data only, and congestion is not a problem, with no QoS on the ES+. But, one remote site needs to support VoIP as well, so the question arose, should I configure ALL circuits with QoS or can I initially just enable priority queuing for VoIP on the particular sub-interface that is to support the voice?

As a best practice I would recommend applying queuing and shaping to each sub-interface, but as only one has to support VoIP I think this may be generating additional unnecessary configuration, but cannot decide after reading (several times!) through the documentation.

I'll take a look at the card capabilities on its data sheet, see if that provides an answer.