The Impact of Architectural Features on the Performance of Parallel Algorithms

点击翻译按钮获取中文摘要

Ramachandran The main thrust of the research is to study the inter- relationship between parallel algorithms and architectures. The objective is to understand the impediments to the efficient implementation of parallel algorithms on realistic parallel architectures. This research involves studying the impact of architectural issues such as latency, network contention, granularity, synchronization, and communication on the performance of parallel algorithms. The approach to be taken is to identify frequently occurring parallel kernels in large-scale scientific/engineering applications; experiment with these kernels on target architectures of both SIMD and MIMD varieties; and augment the experimental work with simulation studies to extrapolate the results for future parallel architectures. The effects of input/output features on the performance of such parallel kernels are also being studied.