Processor Allocation in Partitionable Parallel Architectures

可分区的并行架构允许同时 多个任务的执行，其中每个任务可以在 处理器数量。 这个模型引入了一个问题， 处理器分配给每个任务。 处理器分配 进程必须（1）确定所需的处理器数量， 处理每个任务;（2）在 系统中的处理器。 目标是最大限度地减少 完成所有的任务。 在目前大多数情况下， 分配给每个任务的处理器的数量是手动确定的， 导致处理器的利用不足。 这个项目 考虑设计有效的处理器分配算法， 一大类可划分的并行架构。 的 传统的调度问题，其中每个作业运行在一个 处理器，与处理器分配问题有很大不同 因此，他们的解决方案不能被期望充分执行， 处理器分配问题。 本项目的目标是 为了研究处理器分配问题，在不同的计算条件下， 和架构模型，并设计和评估算法， 处理器分配和处理器调度。 这些算法 测试，通过模拟和应用到真实的 问题 算法所带来的性能提升，以及 由操作系统引起的开销，通过 在不同的并行架构上进行实验。 的 这项研究的结果有望提供一个明确的步骤， 在理解资源分配和调度问题方面取得了进展 在可分区的架构中，从而导致更好的 利用并行架构。

A partitionable parallel architecture allows the simultaneous execution of a number of tasks, where each task can be executed on a number of processors. This model introduces the problem of how many processors to allocate to each task. The processor allocation process must (1) determine the number of processors required to process each task and (2) allocate and schedule these tasks on the processors in the system. The objective is to minimize the time to complete all the tasks. In most cases at the current time, the number of processors assigned to each task is determined manually thereby leading to under-utilization of the processors. This project considers the design of efficient processor allocation algorithms for a large class of partitionable parallel architectures. The conventional scheduling problems, where each job runs on one processor, differ significantly from the processor allocation problem and thus their solutions cannot be expected to perform adequately for this processor allocation problem. The objectives of this project are to study the processor allocation problem, under different computation and architecture models, and design and evaluate algorithms for processor assignment and processor scheduling. These algorithms are tested, through both a simulation and their application to real problems. The performance gains made by the algorithms, and the overheads incurred by the operating system, are measured through experiments performed on different parallel architectures. The results from this research are expected to provide a definite step forward in understanding resource allocation and scheduling problems in partitionable architectures, and consequently lead to a better utilization of parallel architectures.