Performance of Parallel Systems

这项研究涉及并行系统的性能，特别是同构共享内存多处理器的性能。这些系统刚刚问世，有望在不久的将来提供高性价比的高性能计算，但也存在大量未解决的问题，其中许多问题也适用于其他系统架构。他们将研究三个具体领域。首先是软件结构对系统性能的影响；我们将探索并行系统中降级的来源以及这些降级来源之间的关系，为构建性能良好的并行软件提供指导。第二个是并行系统的软件控制策略；他们将开发和评估能够产生良好的程序加速比和良好的系统效率的技术。第三个是并行系统上的性能工具；他们将为并行程序构建性能调试工具，以及解决并行系统问题并利用并行性本身的性能建模工具。

This research concerns the performance of parallel systems, and in particular of homogeneous shared-memory multiprocessors. These systems are just now becoming available, show promise of proividing cost- effective high-performance computing for the near-term future, and present a wide range of unsolved problems, many of which apply to other system architectures as well. They will study three specific areas. The first is the influence of software structure on system performance; we will explore the sources of degradation in parallel systems and the relationships among these sources of degradation, providing guidelines for the construction of parallel software that performs well. The second is software control policies for parallel systems; they will develop and evaluate techniques that yield both good program speedup and good system efficiency. The third is performance tools on and for parallel systems; They will build performance debugging tools for parallel programs, and performance modelling tools that address the problems of parallel systems and that exploit parallelism themselves.