7600 MPLS QoS

juniper84 · ‎07-06-2008

I've been going through all the documentation regarding MPLS and configuring MPLS QoS on PFC's and I'm stumped on this question.

In the past MPLS networks I've used Pipe Mode with Explicit Null LSP to configure QoS within the MPLS network. The benefit of this for a carrier network was that it preserved the customer markings and allowed us to control the treatment of the traffic right up to (and including) the egress of the PE to the carrier owned CPE.

From reading the documentation on the 7600 I don't see anywhere the ability to use Pipe Mode with Explicit-Null LSP...I only see Uniform Mode and Short Pipe Mode. Right away Uniform Mode is out of the question, and Short Pipe mode is the best alternative, but it only allows you to control the treatment of the traffic until it reaches the final PE, at which point the traffic has no MPLS EXP bits left on it and only the original customer markings are left.

So my question is...am I just missing something regarding the 7600 and its ability to support Pipe mode with Explicit-Null? I'm asking this because I also noticed that 'set qos-groups' is not available to do on ingress MPLS-MPLS interfaces...

And if this method of MPLS QoS is not supported on the 7600, whats the next best thing?...Lastly, if Short Pipe Mode is the only alternative, then how can the SP still control treatment on the Egress of the final PE...as all MPLS EXP bits will be stripped during the final 'pop'.

Thanks in advance MPLS guru's!!!!

n.nandrekar · ‎07-06-2008

hi!

You are right! Pipe mode (w or w/o explicit null) is not supported on 7600 platform. So short-pipe is the closest thing that you have.

Also "qos-group" is supported only on SIP-600, and that too "Supported only for software-based EoMPLS on an input SPA switchport interface." So not of any help.

Regards,

Niranjan

chintan-shah · ‎07-24-2008

Hi,

I got to know that 7600 doesn't suppport WRED based on discard-class or EXP and only option to use DSCP/IP Precendace. But question is for MPLS VPN traffic on packet will have already one label even at PE to P and in that case how would it possible to do WRED based on DSCP/IP Precendance.

Regards,

Chintan.

prakashj · ‎07-31-2008

Hi Chintan,

Please see below

When precedence-based weighted random early detection (WRED) is configured on an output policy map and outgoing packets are MPLS packets, the router drops the MPLS packets based on the three EXP bits in the MPLS label, instead of using the three bits of the IP precedence field in the underlying IP packets.

When DSCP-based WRED is configured on an output policy map and outgoing packets are MPLS packets, the router drops the MPLS packets based on the three EXP bits in the MPLS label, instead of using the six bits of the DSCP field in the underlying IP packets. The router left shifts the three EXP bits and makes it six bits. For example, if the value of the EXP bits is 5 (binary 101), the router converts them to binary 101000 (makes it looks like six DSCP bits), and drops packets based on this value.

Regards,

Prakash

chintan-shah · ‎07-31-2008

Hi Prakash,

When you configure WRED based on IP precedenace or DSCP why should it look MPLS EXP bits otherwise what is use of WRED based on EXP bit ?

Regards,

Chintan.