Network queue policies

General information

Network queue policies are applied to network ports, uplink ports, or daughter cards.

Network queue policies determine:

• the default burst allocations for queues based on the queue’s forwarding class

• the CIR, PIR, FIR, and burst size parameters for the queue

Note: You cannot use the same network queue policy on devices at different releases.

For network egress, a network burst policy is associated with the network port buffer pool. For network ingress, the network burst policy is associated the network ingress buffer pool of the daughter card.

Network queue policies support multipoint queues and a variable number of forwarding classes. When you want to deploy a network queue policy to devices of different types, you may need to create an instance of the policy for each NE type. Policies that are configured for multipoint queues and a variable number of forwarding classes are deployed only to devices that support the functionality. Modification of an existing policy results in automatic deployment to participating NEs, so must be done with consideration of the policy features and devices involved.

CIR, PIR, and FIR

CIR defines the rate that packets are considered in-profile by the system. In-profile packets are preferentially queued by the system at egress and at subsequent next-hop devices on which the packet can traverse. To be properly handled as in- or out-of-profile throughout the network, the packets must be marked accordingly for profiling at each hop.

PIR specifies the peak information rate for an ingress or egress daughter card queue to transmit packets through the switch fabric or out an egress interface. On ingress, the PIR defines the maximum rate at which the queue can transmit packets through the switch fabric. On egress, the PIR defines the maximum rate at which the queue can transmit packets out an egress interface. The specified PIR value does not guarantee that the queue can transmit at the intended rate. The actual rate sustained by the queue can be limited by oversubscription factors or the available bandwidth.

FIR configuration allows the scheduling priority to be modified independently from the marking/drop precedence of the packets being scheduled from the queue. Specifically, a higher queue scheduling priority can be assigned without modifying the queue’s CIR.

Burst size

The burst size for a queue determines whether the queue has exhausted its reserved buffer pool space while queuing packets. When the queue has exceeded the amount of buffer pool space considered in reserve for the queue, it must contend with other queues for the available shared buffer space in the buffer pool. Access to this shared pool space is controlled by the RED slope.

Two RED slopes are maintained in each buffer pool. A high-priority slope is used by in-profile packets. A low-priority slope is used by out-of-profile packets. All nc forwarding class packets are considered in-profile. Assured packets are handled by their in-profile and out-of-profile markings. All be packets are considered out-of-profile.

Premium queues should be configured such that the committed burst size is sufficient to prevent shared buffering of packets. This is handled by the CIR scheduling of premium queues and the overall small amount of traffic on the class. Premium queues in a properly designed system drain before all others, limiting their buffer utilization. The RED slopes detect congestion conditions and work to discard packets and slow down random TCP session flows through the queue.

The resultant committed burst size can be larger than the maximum burst size. This results in a portion of the committed burst size for the queue unused and should be avoided.