Research On Cloud Task Scheduling Algorithms Based On Mapreduce

The cluster's performance and user experience (UX) are largely determined by Cloud Task Scheduling Algorithms, in which the selection of data locality plays an important role.Morever, Delay Scheduling is a principal solution to select local task. However, it still has the problem of idling nodes, on the other hand, some important points may be missed when multiple local tasks are executed simultaneously in the head-of-line job.In this paper, the process of selecting local task is analyzed comprhensively, and the probability of task localization and the free time of next local node are summarized. On this basis, the Task Selection Decision Function is proposed, whose strategy of selecting local task is improved so as to optimize the data locality. In addition, when the maximum latency of the node without local data was set, the nodes were assigned tasks if the the waiting time of the node exceeded the maximum latency,or it was not. It was a practicable way to solve the problem of the idle node in Delay Scheduling in some degree.Based on the experimental results, it is conclued that the improved algorithms could raise data locality effectively as compared to the existing Fair scheduling algorithms, and the response time of the job is reduced in some degree.