Custom Queueing

CQ is implemented as a round-robin queueing algorithm. The router takes a certain predetermined amount of data from each queue on each pass. You configure this as a number of bytes. This allows you to specify approximately how much of the bandwidth each queue will receive. For example, if you have four queues, all set to the same number of bytes per pass, you will expect to send roughly equal amounts of data for all of these applications. Since the queues are only used when the network link is congested, this means that each of the four applications will receive roughly one quarter of the available bandwidth.

However, it is important to remember that the router will always take data one packet at a time. So if, for example, you have a series of 1500 byte packets sitting in a particular queue, and you have configured the router to take 100 bytes from this queue on each pass, it will actually transmit one entire packet each time, and not one every 15 times. This is important because it can mean that your calculations of the relative amounts of bandwidth allocated to each queue might be different from what the router actually sends.

This difference tends to disappear as you increase the number of bytes taken each time the queues are serviced. But you don't want to let the number get too large, or you will cause unnecessary latency and jitter problems for your applications. For example, if the byte count for each of your four queues is 10,000 bytes, and all of the queues are full, the router will send 10,000 bytes from the first queue, then 10,000 bytes from the second queue, and so on. From the time it finished servicing the first queue until the time that it returns to service it again, it will have sent 30,000 bytes. It takes roughly 160 ms to send this much data through a T1 link. But the gap between the previous two packets in this queue was effectively zero. Variations in latency like this are called jitter, and they can cause serious problems for many real-time applications.