云计算环境下基于模糊聚类的并行调度策略研究  

Cloud Parallel Task Scheduling Algorithm Based on Fuzzy Clustering

作　　者：张千[1] 梁鸿[1] 郉永山 

机构地区：[1]中国石油大学计算机与通信工程学院,青岛266555

出　　处：《计算机科学》2014年第8期75-80,共6页Computer Science

基　　金：中国石油天然气集团公司石油科技中青年创新基金项目(07E1024);中央高校基本科研业务费专项资金(13CX02032A)资助

摘　　要：并行任务调度是分布式计算研究的核心问题之一,其结合大规模的石油地震勘探数据在处理过程中对高性能计算的需求,研究云计算环境下地震资料数据的并行调度问题。由于地震资料数据规模较大,因此通常将大作业进行分割,通过并行处理获得较高的处理效率。对任务进行并行处理的一个关键问题是如何将划分后的任务分配到合适的调度节点,最高效的情况是使云计算环境中的每一个资源节点都在进行计算,并且计算性能高的节点执行作业块大且复杂的任务,性能相对低的资源节点则运行对计算性能要求不高的任务或小任务,以达到整体上的负载平衡。因此基于模糊聚类思想,提出一种任务与资源混合聚类的调度优化策略,以作业与资源节点属性的匹配程度为基准,对并行作业进行聚类划分求解,在缩小任务调度规模的同时,为动态调度任务奠定基础。在划分完成后引入基于改进的贝叶斯分类调度算法,将资源节点依据其实时负载情况与队列中的作业进行快速的匹配。实验证实,此方案具有较高的执行效率。Parallel task scheduling is one of the key problems in the field of cloud computing research area,which mainly researches parallel scheduling problems in cloud computing environment by the reference to the high performance com- puting required by massive oil seismic exploration data processing. Because of the natural reparability of Seismic data,it can maximize the full use of computing resources to put the job file to the resource nodes,which can just meet the task computing requirements. This paper proposed scheduling optimization strategy of task and resource hybrid clustering based on fuzzy clustering. The strategy takes matching degree of task and resource nodes as reference and with the clus- tering partition solution of concurrent job, narrows task scheduling scale and at the same time, lays foundation for the dynamic scheduling of tasks. After the division is completed, improved Bayes classification algorithm is introduced to fast match tasks and computer according to real-time load and queue operations. In the end, the experiments verify that this scheme has higher efficiency.

关 键 词：云计算 并行调度 模糊聚类 任务资源混合聚类 贝叶斯分类算法 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]