[c-nsp] MPLS-EXP Marking on 6k PE

[Phil Mayers]

On 12/07/2010 04:48 PM, cisconsp at SecureObscure.com wrote:

> 1)       Will a packet entering via a SVI in a VRF (MPLS L3 VPN) have its IP
> PREC bits automagically copied into the MPLS-EXP bits, or do I need to apply
> an ingress policy map on every PE-CE interface&  SVI translating
> IP-PREC/DSCP to MPLS-EXP?

Yes. Note that DFC/PFC-3B has a bug which means you must also enable 
"mls qos rewrite ip dscp" globally (this is the default). If you are 
using all -3C you might not need this, but may want it anyway.

> 2)       Do I need to configure "mls qos trust dscp" on the PE-CE ingress
> SVI in a VRF? Or just the switchport? Or Both?

You do need it. If you are doing vlan-based QoS (i.e. the port has "mls 
qos vlan-based") you need it on the SVI. If you are doing port-based, 
you need it on the switchport.

> 3)       Do I need to configure "mls qos trust something" on the PE-P
> interfaces so that MPLS-EXP is respected and not stripped off?

No. "trust EXP" is the default

> I don't currently use any L2 or L3 QOS on the 6500, and just want the IP
> PREC bits marked and passed upstream to the P routers for prioritization in
> the core.

Sure. Beware however - when you enable QoS without classification, you 
effectively reduce your buffer sizes, because the port buffers are now 
divided up. If you have any congestion, you will want to ensure the 
traffic is classified correctly and/or reconfigure the buffer sizes.