Evaluation Techniques for Hypercube and MIN-Based Architectures

Hypercube和基于多级互连网络（multi - stage interconnection network， MIN）的架构是近年来备受关注的两类极具发展前景的并行计算机。本研究的目的是开发用于预测这两类多处理器的性能、可靠性和与性能相关的可靠性行为的分析评估技术。性能模型基于三层方法——网络层、任务（作业）层和系统层。网络级分析给出了平均通信延迟，可用于在任务级查找任务完成时间。使用作业完成时间，可以从适当的队列模型计算吞吐量和响应时间等系统级参数。可靠性分析将考虑计算元素和通信网络的退化，以寻找基于任务的可靠性或可用性。可靠性模型旨在为任务的执行找到合适的子多维数据集。该分析涵盖了唯一路径和多路径min。通过将适当的性能度量与多处理器的结构状态相关联，开发了与性能相关的可靠性模型。本项目将提供一套完整的工具，用于分析这两类多处理器，这两类多处理器对不同的应用程序有很大的希望。

Feng Hypercube and multistage interconnection network (MIN)-based architectures are two promising classes of parallel computers that have received considerable attention in recent years. The objective of this research is to develop analytical evaluation techniques for predicting performance, dependability, and performance-related dependability behavior of these two classes of multiprocessors. The performance model is based on a three-level approach - network level, task (job) level, and system level. The network level analysis gives the average communication delay which can be used in finding job completion time at the task level. Using the job completion time, system level parameters such as throughput and response time could be computed from an appropriate queing model. Dependability analysis would consider the degradation of the computing elements and communication network for finding task-based reliability or availability. The dependability model is aimed at finding a proper subcube for the execution of a task. The analysis covers both unique-path and multi-path MINs. Performance-related dependability models are being developed by associating suitable performance measures with the structure states of a multiprocessor. This project will make available a complete set of tools for analyzing these two classes of multiprocessors that have a great promise for different applications.