Tim,

Does the class default maps the remianing 75% of the bandwidth to it?

Actually i have an 4 MB MPLS circuit connecting to the fastethernet interface (ehthernet handoff) and that is why the bandwidth statement.

My topology looks something like this

Access---Core/distribution---WAN edge router

I am planning to mark all the voice traffic to DSCP ef and the rest to scavenger class at the core switch and forward it across.

Narayan

Other than a small amount of router traffic that is not in any class all traffic that does not fall in another class goes into the default class.

In your case it is confusing to say 75% but in a very broad sense it will. Since the ethernet port can run at 100m and you have reserved 1m (ie 25% of 4096k) the remaining 99m goes into class default which is 99%. So you reserved 1m which is 1% of the interface but 25% of the declared bandwidth. I tend to use explict reservations when I do priority queueing rather than percent to avoid someone changing my bandwidth to try to influence the routing and screwing up the QoS.

In your case as long your MPLS provider will honor you EF markings and you have at least 1m of guarenteed low latency in your contract it will work. If you were to send 10m of traffic for a sustained time the provider should always pass the 1m voice traffic and drop up to 7m of the other. If your MPLS provider limits you to 4m before it looks at the packet markings you could have issues.

I tend to like to control which traffic is dropped myself so I would shape the traffic to 4m before it left my router.

To limit the traffic to 4m you would do

policy-map shape01

class class-default

shape average 4000000

service policy MPLS-EDGE

You would apply the shape01 to the fast ethernet interface.

Even the above example is using the class-default which since there are no other class statements is 100%.

With the shaping in place your priority class still has 1m but the class-default within your MPLS-EDGE policy is now limited to 3m which is correctly corresponds to 75% if you want to look at the number that way.

Thanks Tim,

Mostly i think the service provider rate limits the bandwidth before looking at the packet markings.

I always run into issues even with 3-4 calls during peak utilization

So should i reconfigure the QoS to something like this

policy-map MPLS-EDGE

class MPLS-VOICE

priority percent 25

class-map match-any MPLS-VOICE

match ip dscp ef

match ip precedence 5

int fa 0/1

bandwidth 4096

shape average 4000000

service-policy output MPLS-EDGE

I think my QoS settings are anyway not getting implemeted properly as i always receive complaints during peak traffic.

I have 8 different sites connecting to the MPLS cloud each at a different rate. I think i should shape the traffic at each location to the bandwidth limited by the provider.

I can provide the network topolgy if you want.

Narayan

That is more or less correct. I don't think you can apply the shaping directly to the interface.

This is a document discussing how to apply QoS to a ethernet subinterface. It is done the same way on a physical interface.

http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a0080114326.shtml

Look at the example under the title

Applying a Hierarchical Policy

This is similar to your case.

Tim,

That was a typo.. i would shape the traffic via a policy & a Class.

Mostly when i read about shaping it mentioned as to be used for non real time traffic. so if i use priority as well as the shape command, would it affect my VoIP traffic if the interafce starts queuing the voice packets

Narayan