Minimizing the communication overhead of iterative scheduling algorithms for input-queued switches

Abstract

Communication overhead should be minimized when designing iterative scheduling algorithms for input-queued packet switches. In general, the overall communication overhead is a function of the number of iterations required per time slot (M) and the data bits exchanged in an input-output pair per iteration (B). In this paper, we aim at maximizing switch throughput while minimizing communication overhead. We first propose a single-iteration scheduling algorithm called Highest Rank First (HRF). In HRF, the highest priority is given to the preferred input-output pair calculated in each local port at a RR (Round Robin) order. Only when the preferred VOQ(i,j) is empty, input i sends a request with a rank number r to each output. The request from a longer VOQ carries a smaller r. Higher scheduling priority is given to the request with a smaller r. To further cut down its communication overhead to 1 bit per request, we design HRF with Request Compression (HRF/RC). The basic idea is that we transmit a single bit code in request phase. Then r can be decoded at output ports from the current and historical codes received. The overall communication overhead for HRF/RC becomes 2 bits only, i.e. 1 bit in request phase and 1 bit in grant phase. We show that HRF/RC renders a much lower hardware cost than multi-iteration algorithms and a single-iteration algorithm π-RGA [11]. Compared with other iterative algorithms with the same communication overhead (i.e. SRR [10] and 1-iteration iSLIP [6]), simulation results show that HRF/RC always produces the best delay-throughput performance. © 2011 IEEE.