遥感影像并行处理中基于优先级的任务分配策略  

Priority-based Task Allocation Strategy in Remote Sensing Image Parallel Processing

作　　者：付征叶[1] 凡高娟[2,3] 王俊岭[2,3] 

机构地区：[1]河南大学计算中心,河南开封475004 [2]河南大学计算机与信息工程学院,河南开封475004 [3]河南大学数据与知识工程研究所,河南开封475004

出　　处：《计算机工程》2014年第2期48-51,57,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(60973126;61272545);河南省科技攻关计划基金资助项目(122102210053);河南大学校内科研基金资助重点项目(2012ZRZD09)

摘　　要：对集群环境下大规模遥感影像并行计算中任务分配效率低、负载不均衡的问题进行分析讨论,在此基础上建立多机任务分配模型,提出一种基于计算节点优先级的任务分配算法。该算法综合考虑计算节点的负载和性能,在任务分配时实时地收集各个节点的信息,计算出各个计算节点的优先级,按照优先级的高低分配任务,保证在满足集群间负载均衡的前提下能合理地将任务分配到计算节点。实验结果表明,该算法能快速实时地进行任务分配,任务的分布更加合理和均匀,并且当任务个数增多时,算法的执行效率要比轮转调度算法高出约2倍。After a discussion of the low efficiency task allocation and load imbalance problem in parallel computing of remote sensing image, this paper gives the multi-task distribution model, and proposes a compute nodes＇ priority-based task distribution algorithm, which is in comprehensive consideration of load and performance of compute nodes. It collects real-time information on each node when the task is assigned. According to the formula to calculate the priority of each computing node and in accordance with the priority level of assign tasks, under the premise of load balancing, this algorithm can assign task reasonable to computer nodes. Evaluation result shows that this algorithm can process task allocation in reasonable time. The distribution of the tasks becomes more reasonable and uniform, and behaves better than round-robin scheduling algorithm by about 2 times when the number of tasks increases.

关 键 词：任务分配 集群 并行计算 负载均衡 任务调度模型 优先级 

分 类 号：TP391.12[自动化与计算机技术—计算机应用技术]