MPLS VPN QOS - Which tunneling mode to use?

asaykao73 · ‎07-19-2009

Hi...

What tunneling mode should I be using in my QoS implementation in a SP environment? We run only MPLS L3 VPNs.

I know the differences between the tunneling modes but can't figure out which one to use.

Ideally we want to preserve the customer's markings so I wasn't sure as the SP we look at implementing pipe mode (PE Edge (to CE) policies are based on

provider markings) or short pipe mode (PE Edge (to CE) policies are based on customer markings).

What do I need to look at to help me make my decision?

Thanks.

Andy

Rui Carlos Antunes · ‎07-20-2009

If you have detailed information on how the customer wants QoS to be implemented on the access links (links between PE and CE) you can use short pipe mode, which means basically applying on the PE the exactly same configuration that the customer has on the CE.

Otherwise, you will use pipe mode, and apply QoS policies similar to the ones applied on the rest of the network.

asaykao73 · ‎07-20-2009

Hi rantunes,

Thank you for your reply.

I've already defined a few QoS profiles which we will be offering customers but not sure about the tunneling mode to use.

Is there any reason to choose either pipe or short pipe because I assume once we implement one or the other, this will be the tunneling mode we offer to all future customers.

Cheers.

Andy

amit.bhagat · ‎07-22-2009

Hi Andy,

If you would like to preserve your customer's marking, that is a great reason to go for the UNIFORM Model where the customer's IP Precedence/DSCP bits are copied to MPLS EXP bits at the ingress. At the egress, you (as a SP) copy the EXP bits back to IP header's IP Precedence or DSCP bits.

HTH.

Rui Carlos Antunes · ‎08-20-2009

The difference between pipe and short-pipe is that, on the egress LSR (LER), the pipe model still uses the MPLS EXP value (of the MPLS label) on the service-policy (as match criteria) while the short-pipe uses the the IP Precedence / DSCP (on the IP packet) on the service-policy.

Short-pipe allows for more QoS granularity on the accesses facing the CPE's, if DSCP is used (6 bits of the DSCP against 3 bits of the MPLS EXP). Other than that, they are similar...

The Uniform model is not used often, because it implies possible remarcation of the ToS byte of the original IP packet header (thus, messing up the customer's own QoS policy). The Uniform model is, however, the only possible model on a non-MPLS backbone.

Hope this helped.