CISE Postdoctoral Research Associateship in Experimental Computer Science: Processor Allocation in Hierarchical Heterogeneous Networks of Workstations

9625875 Lilja，大卫·J·明尼苏达大学，双城：实验计算机科学博士后研究助理：工作站分层异质网络中的处理器分配工作站网络的广泛可用性和合理成本使其成为传统超级计算机系统的有吸引力的替代方案。然而，异质系统中并行应用在执行和通信时间上的低一致性会对可能的并行任务调度解决方案造成很大的限制。这项工作的主要目标是开发技术，以确定要分配给多编程工作站的异类网络中的每个应用的适当数量和类型的处理器，并使不同的硬件平台与特定应用任务的计算和通信要求相匹配。拟议的“匹配系数”为专用环境中的硬件和应用程序配对提供了性能指标，它将扩展为具有不同处理器类型和互连网络的多程序设计工作站网络的调度启发式方法。这项工作的高度实验性将极大地受益于与在明尼苏达大学工作的计算科学家和工程师的互动，以及该大学独特的硬件和软件资源的可用性。***

9625875 Lilja, David J. University of Minnesota, Twin Cities CISE Postdoctoral Research Associateships in Experimental Computer Science: Processor Allocation in Hierarchical Heterogeneous Networks of Workstations The wide availability and reasonable cost of networks of workstations makes them an attractive alternative to traditional supercomputer systems. However, the low degree of uniformity in the execution and communication times of parallel applications in heterogeneous systems can impose significant constraints on the possible parallel task scheduling solutions. The primary goals of this work are to develop techniques for determining the appropriate number and type of processors to be allocated to each application in a heterogeneous network of multiprogrammed workstations, and to match the diverse hardware platforms to the computational and communication requirements of the specific application tasks. The proposed "matching factor," which provides a performance indicator for pairing hardware and applications in a dedicated environment, will be extended with scheduling heuristics for a network of multiprogrammed workstations with heterogeneous processor types and interconnection networks. The highly experimental nature of this work will greatly benefit from the interaction with computational scientists and engineers working at the University of Minnesota, and from the availability of the University's unique hardware and software resources. ***