Priority Queueing

Some types of critical real-time applications that absolutely cannot wait for low priority traffic work well with PQ. However, there is an obvious problem with this strategy as well. If the volume of traffic in the higher priority queues is greater than the link capacity, then no traffic from the lower priority queues will be forwarded. PQ starves low-priority applications in these cases.

So a pure PQ implementation requires that you have an extremely good understanding of your traffic patterns. The high-priority traffic must represent a small fraction of the total, with the lowest priorities having the largest net volume. Further, you must have enough link capacity that the PQ algorithm is only used during peak bursts. If there is routine link congestion, PQ will give extremely poor overall performance.

However, Cisco has recently implemented a new hybrid queue type, called Low Latency Queueing (LLQ), which you can use with CBWFQ to give the best features of PQ while avoiding the queue starvation problem. The idea is simply to use CBWFQ for all of the traffic except for a small number sub-queues that are reserved strictly for relatively low-volume real-time applications. The router services the real-time queues using a strict priority scheme, and the others using CBWFQ. So, if there is a packet in one of the real-time queues, the router will transmit it before looking in one of the other queues. However, when there is nothing in the priority queues, the router will use normal CBWFQ for everything else.

LLQ also includes the stipulation that, if the volume of high priority traffic exceeds a specified rate, the router will stop giving it absolute priority. The guarantees that LLQ will never starve the low priority queues.

This model is best suited to applications like voice or video in which the real-time data comes in a fairly continuous, but low-bandwidth, stream of small packets, as opposed to more bursty applications such as file transfers.