摘要
本文提出了一种新颖的并行程序配置优化算法。这种算法利用黑板系统将配置优化问题分解组织为不同层次的知识领域,并利用A*算法对决策树进行搜索。研究了并行程序的任务调度、存储服务器的数据分配、自适应分片、协同I/O和数据筛选五个知识领域。
A novel optimal configuration algorithm of combined parallel computing and parallel storage is proposed. The algorithm utilizes blackboard system to solve the running time optimization problem of parallel program. The problem is divided into different knowledge areas and organized them as tree. Task scheduling, data distribution of storage server, adaptive disk striping,collective I/O and data sieving are considered. The experimental results show that the processor distribution and task scheduling using CPR has much advantage over static task scheduling. The advantage of dynamic I/O configuration is not very great over static I/O configuration because the I/O access time is only a small percentage of total application running time.
出处
《计算机应用与软件》
CSCD
北大核心
2005年第11期92-95,共4页
Computer Applications and Software
