Using Split Queues to Improve the Performance of Parallel Switch

Xiaofeng Hu, Zhigang Sun, Xicheng Lu, Jinshu Su

Abstract: Parallel switch scales well with the growth of port density and line rate. PSIQC (Parallel Switch based on Input-Queued Crossbar) is a parallel switch that is scalable and simple to implement. But it needs large capacity high-speed memories to store cells, and the average cell latency is high under heavy load. This paper presents a revised version of PSIQC based on split queues that is initialed as SQ-PSIQC (Split Queued Parallel Switch based on Input-Queued Crossbar), and its scheduling algorithm SQ-RRDS (Split Queued Round Robin and Deterministic Sequence). SQ-PSIQC not only has all of the original characteristics, but also solves the two above-mentioned problems. In SQ-PSIQC the memory buffers are required to operate only at 1/m of the line rate, where m is the number of the middle switches. The simulation results show that SQ-PSIQC performs better than PSIQC in the average latency and throughput under any load, especially heavy load.