Load Balancing and Visualization for Parallel Processors

随着并行计算系统可用性的增加，与构建并行算法、将它们映射到并行体系结构以及确定算法和系统性能相关的问题变得越来越重要。不良映射的结果是低效的计算和系统性能，以及从给定的并行处理器中可能获得的加速的总体降低。所提议的研究旨在检查以确保有效利用可用资源的方式将任务分配给处理器的方法（即，由于处理器间通信导致的高处理器利用率和最小延迟）。提出了许多任务分配（负载均衡）算法。一种是基于利用新的并行模拟退火算法，而第三种是前两种方法的混合。将这些算法相互比较，并与文献中提出的其他算法进行比较是本研究的关键部分。

With the increasing availability of parallel computing systems, the problems associated with constructing parallel algorithms, mapping them onto parallel architectures, and determining algorithm and system performance have grown in importance. The results of poor mappings are inefficient computation and system performance, and an overall reduction in the speedups potentially obtainable from a given parallel processor. The proposed research is aimed at examining approaches to allocating tasks to processors in a manner that ensures effective use of the available resources (i.e., high processor utilization with minimal delays due to interprocessor communications). A number of task allocation (load balancing) algorithms are proposed. One is based on utilizing a new parallel simulated annealing algorithm, while a third is a hybrid of the former two approaches. Contrasting these algorithms with each other, and with other algorithms proposed in the literature is a key part of the proposed research.