Virtual output queues

The switch uses a virtual output queue (VOQ) architecture where most packet buffering occurs on the ingress line module. Traffic destined for one port (unicast), uses different buffering and scheduling than traffic destined for multiple ports (flood). The relative priority and the amount of packet transmission for these two types of traffic differ.

For unicast traffic, each line module contains eight VOQs for every destination port in the chassis (one per local priority). The queue profile determines which VOQ is used to buffer a local priority. The schedule profile determines the order of VOQ servicing. Unicast packets wait in VOQs until the scheduler selects them to cross the fabric. Each destination port has a shallow egress transmit queue that buffers unicast packets.

Broadcast, multicast, and unknown-unicast packets, collectively called flooded traffic, use a separate path to the destination ports. Each line module contains eight VOQs per destination line module (including itself) to buffer traffic to be flooded (one VOQ per local priority). A copy of the packet to be flooded is buffered in VOQs for each destination line module. The queue profile determines which VOQ is used for each local priority.

Flooded traffic VOQs use a fixed strict schedule profile to determine the order of VOQ servicing. Flooded packets wait in VOQs until the scheduler selects them to cross the fabric. On the destination line module, they are replicated to one or more destination ports. Each destination port has a second shallow egress queue for replicated packets buffered for transmit.

A WFQ scheduler is used to select packets for transmission from the unicast and replicated traffic egress queues. When selecting packets between two non-empty queues, WFQ uses a fixed data weight of four for unicast traffic, and a weight of one for replicated traffic. As long as both queues are non-empty, replicated (flooded) packets comprise approximately 20% of the transmitted traffic, independent of the unicast scheduled percentages.