QoS Policy placement on WAN interfaces?

slandreth · ‎11-15-2007

This question is in regards to the proper placement of a QoS policy for WAN interfaces.

I am running Frame Relay between our CE-PE connections where we interface with the provider for MPLS services. In doing this, we end up with a subinterface, and I'm curious about the application of the QoS policies... I've heard two schools of thought on this; one that feels you should always configure a map-class under the PVC and apply the QoS policy there, and the other that feels you can just apply the QoS policy under the physical interface. Since we only run a single PVC, we apply the QoS policy under the physical interface... I haven't seen any issues so far, and results are identical when doing side-by-side comparisons.

Creating a 'map-class', ensuring the values are set properly, plus enabling traffic-shaping ends up being a lot of work for no perceived value.

All of the devices are at least 2811 series, ranging up to 7206's at higher-speed sites, and the results have been the same across all of the platforms.

(Tests were done using various router models, interface types, and an Agilent N2X Traffic Generator.)

Thanks!

owillins · ‎11-21-2007

Traffic shaping and QOS applies to both permanent virtual circuits (PVCs) and SVCs .defining separate VCs for different types of traffic and specifying queueing and an outbound traffic rate for each VC, you can provide guaranteed bandwidth for each type of traffic. By specifying different traffic rates for different VCs over the same line, you can perform virtual time-division multiplexing

jarredtaylor · ‎11-21-2007

Assuming your CE-PE connections just use frame-relay for the encapsulation (i.e. there are no frame switches and the routers connect back to back), then applying the policy on the physical interface is appropriate. I'm guessing that during your map-class tests you configured CIR=MINCIR=Port speed and applied the same output service policy.

slandreth · ‎11-22-2007

That's correct, just for encapsulation. For the map-class tests, I used fully configured traffic-shaping to make it as realistic as possible (same for ATM when tested.) Essentially, we actually had better test results under the physical interface.

Thanks for the feedback!