支持并行任务的多约束网格调度模型  

Multi-constrained grid scheduling module supporting parallel task

作　　者：臧雪柏[1] 张旭利[2] 吕巍[1] 唐觅[1] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学数学学院,长春130012

出　　处：《吉林大学学报（工学版）》2009年第S1期241-244,共4页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(60373097);吉林省科技发展计划项目(20030522-1)

摘　　要：针对资源负载瞬时过重情况下负载平衡和跨度的优化问题,提出了一种支持并行任务的多约束网格调度算法。考虑到网格环境资源的非专有性和性能差异较大的特点,提出了扩展分配概率的概念。算法将负载平衡分配概率与期望执行时间对任务分配造成的影响进行加权,依据扩展分配概率进行任务到节点的映射。在3个节点搭建的Globus网格环境中对算法进行了测试和评估。测试结果表明,多约束算法在节点初始负载不均时,算法能达到可接受的负载平衡,且跨度优于负载平衡算法,表明了算法的合理性。Aiming to the problems of load balance and makespan optimization in the case of resources load instant overweight,the thesis presents a multi-constrained grid scheduling algorithm supporting parallel task.Considering the resources characters of non-appropriation and larger performance difference in grid environment,defines extend distributing ratio.The algorithm mixes the load balance distributing ratio and infections resulted from ETC into a weighting expression,and gets the mapping between tasks and nodes based on the extend distributing ratio.The multi-constrained algorithm is tested and evaluated in a three-node Globus grid environment.The experiment data indicates that the multi-constrained algorithm is able to gets an acceptable load balance and a better makespan than load-balancing algorithm when the primary load of grid environment is lopsided.The multi-constrained algorithm is proved in reason.

关 键 词：计算机应用 网格计算 任务调度 跨度 负载平衡 

分 类 号：TP393.01[自动化与计算机技术—计算机应用技术]