Performance Analysis of the Bidirectional Ring-Based Multiprocessor

In this paper, we propose and evaluate a bidirectional slotted ring architecture for the distributed shared memory multiprocessor model. We provide a performance analysis of the proposed architecture by deriving an analytical model and studying the effects of varying system and workload parameters. The performance of the bidirectional ring architecture is compared against the unidirectional slotted ring model. The effects of workload parameters such as miss rate, communication locality, and write probability, and system parameters such as relative processor speed and message length on miss latency, processor utilization, and effective computation power are analyzed in detail. The analytical model was evaluated using workload parameters extracted from four parallel application traces. For a 32 node configuration, with the same link width, the bidirectional ring is shown to yield 15% better performance on the average compared to the unidirectional ring