LLQ Priority command in multiple classes

[Alexmpj376]

Hi Guys,

I would like to know if I could use LLQ Priority command in several classes under the same Policy-Map and if yes, what is the benefit of it

[Peter Paluch]

Hi Alex,

Yes, it is possible to use the priority command in multiple classes of a single policy-map. As you know, the priority command has two primary aspects of operation: It creates a priority queue that is served whenever there is data in it, and also, it works as an implicit policer - when a congestion on an interface occurs, the priority class will be automatically policed down to the rate specified by the priority command. Note that if there is no congestion, the priority class can also consume more bandwidth and go above the rate specified in the priority command; that is why the policing related to a priority class is referred to as implicit.

If you have multiple classes of a single policy-map configured with a priority command, there will still be only one priority queue, meaning that all data from all priority classes will be put into a single priority queue indiscriminately. There will is no preference between them in the sense of scheduling. However, in times of congestion, each class will be independently policed to its own rate specified by its own priority command.

Admittedly, the use of multiple priority classes in a single policy-map is not a common occurrence, as by definition, all priority classes are equal in terms of scheduling which also means that you could have put them all into a single class from very beginning, and the only difference is the independent implicit policing of each priority class - but even that occurs only in times of congestion.

Perhaps Joseph Doherty who I am sure will join this thread will shed more light on this - Joe, thanks in advance ;)

Best regards,
Peter

[Umesh Shetty]

Hi Peter,

Thanks for all the information, I had a question. If I have two priority classes one for Voice set to 20 % and one for Video set for 20 % . Will the router treat both types of traffic into on single priority queue with a priority of 40% (A + B) ? and its not guaranteed that the 40% will have an equal share of voice and video traffic .

Regards

Umesh

[Peter Paluch]

Hi Umesh,

Hey, how are you? Haven't seen you here for ages :)

If I have two priority classes one for Voice set to 20 % and one for Video set for 20 % . Will the router treat both types of traffic into on single priority queue with a priority of 40% (A + B) ? and its not guaranteed that the 40% will have an equal share of voice and video traffic .

It is not that simple.

First of all, if the interface does not experience congestion with the current amount of traffic flowing out of it then the value of the priority command is not relevant - the implicit policer does not kick into action until the interface gets congested. The only effect the priority command has in this situation is that it causes the traffic to be put into a single priority queue which is dequeued until it is empty, and only then other non-priority queues are dequeued. So once again, without a congestion on an interface, the values of priority commands do not have an effect. The only effect is that the traffic in priority classes is dequeued first, and all other classes are served only when the single per-policy-map priority queue collecting traffic from all priority classes in that policy-map is empty.

When congestion occurs, the implicit policer starts policing each priority queue down to its predefined value. With enough traffic in each class, each one of them will be policed down to its configured value. That would mean that you would truly get only 20% of video and 20% of voice - yes, together, they constitute 40% of the interface capacity, but each class would be policed independently even though these both classes share the same priority queue.

The fact that you have a single priority queue for all priority classes of a single policy-map, and yet can do independent policing on those classes may be confusing at first, but this is exactly what happens.

So to answer your question: If enough traffic is available to congest the interface then your voice would be policed down to its independent 20% and video would be policed down to another independent 20%. When there is no congestion, individual classes including voice and video can go beyond their allocated 20%.

I am not certain if this truly answers your question so please feel welcome to ask further!

Best regards,
Peter

[Umesh Shetty]

Hi Peter,

Have been away for a while but and did miss the forum quite a lot, hope you are doing good :)

Your explanation answers my question , thanks again

Regards

Umesh

[Joseph W. Doherty]

Peter, I don't think I can add anything, beyond, perhaps, ASR 1Ks QoS policies, I believe, don't always behave like they do on ISRs.  For example, an ASR 1K might actually police a LLQ class when interface is not congested, but don't quote me on that.