MPLS QoS

bretjaquish · ‎05-15-2009

Hey all,

I am at a new employer that uses MPLS as the WAN transport technology from AT&T. I am wondering... How do you prevent oversubscribing branch offices with MPLS?

For example, the HQ office has a OC-3 and the branch offices have varied connectivity (T1s, T3s, 10Mb, etc.). I am wondering, because we are thinking about explanding our current QoS policy.

vmiller · ‎05-15-2009

First thing is to get your providers policy for handling marked traffic. What ever you do will need to match up to what they do in the MPLS network.

Second Get your hands on The latest Enterprise QoS Solutions Reference Guide. Its a superb starting point.

If you want to prevent oversubscribing, where there is a bandwidth mismatch (i.e. sub rate ethernet) this should work:

shape average xxxxxxx xxxx= purchased b/w

bretjaquish · ‎05-15-2009

I have the information from the provider for the 5 queues that we are paying for.

I pulled down the SRND already for QoS.

How can I apply the shaping command to multiple sites when they are unequal bandwidth?

We have two routers with OC-3 connections into the MPLS "cloud". The branch offices with varying speeds have connections to the "cloud".

It seems apparent to me that the main site would have no problem oversubscribing the remote sites.

For example, the SMS server pushes a service pack update out to a remote facility and locks the link up solid (soaking up as much bandwidth as possible).

Is there a way to limit/restrict traffic to different sites over the cloud?

Ideally, I would like to say:

20% for voice

30% for apps

40% for best effort

10% scavenger

The problem is that if I apply that to the OC-3....

20% of an OC-3 is more than a whole T1?

I need to limit/restrict based on the policy (20% for this or whatever) based on the bandwidth of the remote facility (T1, T3, etc.). Is that possible?

Giuseppe Larosa · ‎05-15-2009

Hello Bret,

there are different QoS models that can be used.

How to condition traffic from HQ to single site could be even handled with hierarchical QoS.

The real problem is when multiple sites try to access to the same site.

Or also inbound to HQ.

Other models suitable for multiple sites exist like the Hose model or point-to-cloud model

see

http://www.cisco.com/en/US/products/ps6607/products_white_paper09186a00800a4455.shtml#wp26751

so an ICR and an ECR ingress and egress Committed Rate can be defined.

in some way the QoS can be implemented by cooperating you and the MPLS provider.

for hierarchical Qos policies see:

there is support for two levels policers and also support for three levels policers has been introduced.

http://www.cisco.com/en/US/docs/ios/qos/configuration/guide/three_level_policer_ps6441_TSD_Products_Configuration_Guide_Chapter.html#wp1037207

Hope to help

Giuseppe

Joseph W. Doherty · ‎05-15-2009

"How do you prevent oversubscribing branch offices with MPLS? "

With MPLS, often you don't (unless you're logically running a hub and spoke topology, rather than any-to-any). Instead, you utilize QoS provided by the MPLS vendor to control MPLS egress congestion. This is in addition to you you might utilize QoS for ingress to MPLS.

[from your 2nd post]

"I need to limit/restrict based on the policy (20% for this or whatever) based on the bandwidth of the remote facility (T1, T3, etc.). Is that possible?"

Yes it is, but it only practical if one (or very few) sites that would send traffic to the site in question.

For instance, you can often (easily) shape traffic on an OC-3 to match the far side's bandwidth. But if you have two OC-3, you would need to divide up the far side bandwidth between them (or force all traffic to that site to prefer one OC-3 vs. the other [if the path is available]). Problem with dividing up bandwidth, routers are "unaware" of bandwidth not being used by other routers sending to the site.

Again, you would utilize MPLS QoS. For something like a SMS push, you would mark the traffic so the MPLS provider will place the traffic into a egress class with a low bandwidth guarantee. This allows the SMS traffic to use all available egress bandwidth, but keeps it from (too) adversely competing for bandwidth against more important traffic.