CISE Research Resources: A Low Cost Parallel and Distributed Systems Laboratory

吴杰佛罗里达大西洋大学cise研究资源：低成本并行和分布式系统实验室提议的项目包括Beowulf集群上的三个并行应用程序。贝奥武夫集群的主要优势是其卓越的性价比：提议的集群将获得3-5千兆次浮点运算范围内的性能，成本不到同等功率超级计算机的十分之一。低成本的并行和分布式系统实验室由一个16节点的Beowulf集群组成。每个节点都类似于一台没有显示器或键盘的现成PC。节点通过100兆快速以太网线路联网。每个节点将同时运行Linux和Windows/NT。提出了三种并行应用：(1)生态建模的并行系统，其重点是最小化并行生态模型的模拟时间。一个核心组件将把NOAA的NNT-SMS直线并行建模包移植到Beowulf体系结构上。(2)Beowulf集群上的Java运行时框架，用于并行执行多线程进程。针对大规模多线程计算的高效调度问题，研究了一种新的基于彩票的任务窃取算法。(3)提高可编程机床精度的优化配置选择。本文将研究遗传算法的解决方案，以提高可编程机器的精度。该项目的成功将证明Beowulf集群作为超级计算机的一种经济有效的替代方案的有效性。

EIA-0130806 Jie WuFlorida Atlantic UniversityCISE Research Resources: A Low Cost Parallel and Distributed Systems LaboratoryThe proposed project includes three parallel applications on a Beowulf cluster. The chief advantage of a Beowulf cluster is its superb price/performance ratio: the proposed cluster will obtain performance in the 3-5 Gigaflop range, for less than one tenth the cost of a comparably powered supercomputer. The low cost parallel and distributed systems laboratory consists of a 16-node Beowulf cluster. Each node is similar to an off-the-shelf PC without a monitor or keyboard. The nodes are networked by 100 Mbit Fast Ethernet lines. Each node will run both Linux and Windows/NT. Three parallel applications are proposed: (1) A parallel system for ecological modeling, with its focus on minimizing the simulation time of parallelized ecological models. A central component will be porting NOAA's NNT-SMS rectilinear parallel modeling package to the Beowulf architecture, (2) A Java runtime framework on Beowulf clusters for parallel execution of multithreaded processes. A new lottery-based job stealing algorithm will be studied for efficient scheduling of large-scale multithreaded computation. (3) Optimal configuration selection for accuracy enhancement of programmable machines. A genetic algorithm solution will be studied to enhance the accuracy of programmable machines. The success of this project will demonstrate the usefulness of Beowulf clusters as a cost-effective alternative to the supercomputer.