I'm working on a QoS configuration for an MPLS backbone, and have a question on the use of WRED within a policy map. Reading the ciscopress End to End QoS Network design book, it implies that a queuing policy on an MPLS egress link can use WRED to drop traffic based on DSCP preference.

But, I thought the whole point of MPLS QoS was to use the EXP bits in the MPLS header, so the SP QoS policy will have no more than 7 classes, as this is the limit imposed by the EXP bits.

I set up a test in GNS3 to prove/disprove my thoughts using 7200 ipservices code, where I marked telnet traffic with DSCP EF, on a router connected to the CE.

Inbound to the CE I configured a policy matching EF, copying the EF DSCP to EXP 5, MPLS short pipe mode. On the MPLS Egress interface I configured a policy with 2 classes, one matching EXP 5 and the other DSCP EF, the idea being to see if the egress policy can actually look into the IP header past the MPLS header.

I got matches against EXP5, which I expected, no matches against DSCP EF, I wasnt sure about this result. I suppose I need to check if the original DSCP marking is unchanged.

So, this test indicates that if I'm using a 7200 I cannot use WRED on an MPLS encapsulated link, is this the same for other hardware platforms? The eventual solution will be 7600 based, very different to the 7200, I'm just trying to get an understanding of how the MPLS and QoS policies interact and am seeking guidence from anyone who has used this for real.