Cisco QoS algorithm - classification process

Question

Question about the classification process of the Cisco QoS.

I understand the concept, but I have some misunderstanding regarding the Cisco QoS algorithm.

Am I true with the following reasoning:

The interface A is configured with a service policy SP. SP is a policy map, containing 4 class map. Each class map match a different dscp value.

So, when a packet arrive in the interface A, switch retrieve the policy map associated. Then, the switch, check if packet match with each class map configured, successively, until a match occur or not (like ACL) When it match, the packet is considered as classified, and, the policy defined (policing, shaping, etc) in the class map section is applied.

Is it true ?

Could we say that the process classification is performed from the order of the class map appear in the policy map ?

Or, for each packet arriving in an interface, regardless the interface, do Cisco try to classify them independently from the policy map configured, based on all class map available in the switch ? and then, when the classification is done, the switch take the policy map associated, and know already which class map is relevant ?

Answer 1

The interface A is configured with a service policy SP. SP is a policy map, containing 4 class map.

So, when a packet arrive in the interface A, switch retrieve the policy map associated.

When it match, the packet is considered as classified, and, the policy defined (policing, shaping, etc) in the class map section is applied.

Is it true ?

Partially. Service Policies can be applied ingress or egress. You are describing the behavior of an ingress service policy; however, ingress service policies do not support shaping.

Shaping is an egress-only feature.

Could we say that the process classification is performed from the order of the class map appear in the policy map ?

Yes

Or, for each packet arriving in an interface, regardless the interface, do Cisco try to classify them independently from the policy map configured, based on all class map available in the switch ?

No, the only class-maps used are those configured in the ingress / egress policy-map.

Answer 2

Classification and marking is separate from performing a policy action based on the marking.

IPv4 packet headers have the DSCP (formerly ToS) field, and IPv6 packet headers have an equivalent Traffic Class field. The packets can be marked anywhere along the path, including at the source host. You mark the packets at the network trust point so that you can treat the packets according to your service policies within the trusted area of your network. In general, you want to classify and mark packets as close to the source as possible For packets originating on your own network, either on the source host, or on the access switch to which the source host is connected, and for packets entering your network, at the point where the packets enter your network.

The service policies are what carries out your treatment (shaping, policing, queuing, etc.) of the marked packets. You apply service policies to congestion points in your network, or at the edge of your network where packets exit your network.

A typical scenario is that you classify and mark packets on the access switch, then create service policies to shape and queue the traffic on a router interface where the traffic leaves your network.