On implementation of scheduling algorithms in high speed input queuing cell switches

It is well known that the phenomenon of head-of-the-line (HOL) blocking limits the maximum achievable throughput of a pure input queuing cell switch. At each input, if instead of one queue for all the outputs, one queue for each output is used, and then 100% throughput can be achieved. This type of queuing, known as virtual output queuing (VOQ), removes the HOL blocking problem, but, in order to achieve 100% throughput, requires a scheduling algorithm to determine at each slot which input cells should be matched to their outputs. As the port speed is increased, the matching between inputs and outputs of the switch has to be found in a smaller time. Hence the speed of the scheduler has to be increased, which may not be possible owing to restrictions imposed by the hardware. We study a solution of this problem by the simple technique of skipping. In skipping, a new matching is found every k (> 1) slots instead of every slot. The effectiveness of skipping is shown using various traffic models.