Providing performance guarantees for buffered crossbar switches without speedup

Abstract

Buffered crossbar switches are special crossbar switches with each crosspoint equipped with a small exclusive buffer. The crosspoint buffers decouple input ports and output ports, and simplify switch scheduling. In this paper, we propose a scheduling algorithm called Fair and Localized Asynchronous Packet Scheduling (FLAPS) for buffered crossbar switches, to provide tight performance guarantees. FLAPS needs no speedup for the crossbar and handles variable length packets without segmentation and reassembly (SAR). With FLAPS, each input port and output port independently make scheduling decisions and rely on only local queue statuses. We theoretically show that a crosspoint buffer size of 4L is sufficient for FLAPS to avoid buffer overflow, where L is the maximum packet length. In addition, we prove that FLAPS achieves strong stability, and provides bounded delay guarantees. Finally, we present simulation data to verify the analytical results. © Institute for Computer Science, Social-Informatics and Telecommunications Engineering 2009.