Application Level Modelling of Parallel Machines

这项研究是关于并行机器的应用级性能建模，该并行机器由大量以规则结构连接的处理单元(PE)组成。PE还可以通过单独的互连网络访问一些共享资源。假设PE网络的不相交部分可以同时用于求解不同的子问题。该模型考虑了这类机器的几个重要方面，包括对PE子结构的争用、对共享资源(如全局内存)的争用、由子问题之间的数据依赖引起的阻塞、处理单个子问题的进程之间的同步延迟以及与优化设计相关的一些问题。这项研究的主要目标是开发实用技术来设计和调试即将成熟的大型并行机器。

This research is concerned with application level performance modeling of parallel machines consisting of a large number of processing elements (PE's) connected in some regular structure. The PE's may also have access to some shared resources via a separate interconnection network. It is assumed that disjoint portions of the PE network can be used simultaneously for solving different subproblems. The modeling considers several important aspects of such machines including contention for PE substructures, contention for shared resources such as global memory, blocking due to data dependencies between subproblems, synchronization delays between processes working on a single subproblem, and some issues relating to the optimal design. The main goal of the research is to develop practical techniques for designing and tuning large parallel machines that are coming of age.