Research On Task Scheduling Algorithm Based On Task Backup For Cloud Computing

Task seheduling is the key technology of cloud computing, in order to reducing thetransmission overhead during task execution process under the limited networkbandwidth, it can place tasks on compute nodes that contain their input blocks. So theresearch of task scheduling algorithm for data locality become a hot topic in the cloudcomputing. Currently existing task scheduling algorithm which based on data localdrive mostly adopt delaying the scheduling of part of the tasks to make them waiting forthe right compute nodes to achieve higher data locality. In the case of large waitingoverhead, delay strategy will affect the job completion time, and can not guarantee thatthe system load balancing. In addition, the applications based on cloud computingplatform usually need to use a number of computing resources and storage resources tocompleting computing tasks, so the fault-tolerant capability of system has becomeincreasingly important. Most traditional fault-tolerant scheduling algorithm based onbackup by coping multiple backup tasks for each of the main tasks to guarantee systemfault tolerance, although it can ensure that the fault-tolerant capability of system, requirea lot of backup cost, and only considering how to schedule backup tasks when a singleprocessor failure at some point, do not consider how to schedule backup tasks whenmultiple processors to fail simultaneously.On the basis of the study of the characteristics of cloud computing, this paperpresent a based on data locality driven primary task scheduling algorithm--DLD(DataLocality Drive). The algorithm solves the problem that meeting customer servicesatisfaction and load balancing at the same time. In addition, consider single backuptask status, for the failure of more than one processor at the same time, present theminimum cost of backup scheduling algorithm, the algorithm to solve the problem thatrequire a lot of backup cost.In this paper, the work includes:①Introduce the emergence background of cloud computing, form and architecture,compare with parallel computing, grid computing and utility computing, summary thecharacteristics of cloud computing, and descible the existing cloud computing platform.②It present DLD algorithm under considering the data locally, network bandwidthand the cluster load. The task scheduling is divided into two stages. The first stage is"local" phase, placing all tasks on compute nodes which contain their input blocks. The second stage is the "reduce" phase, reduce the makespan iteratively.③On the basis of the study of backup overloading and synchronization dislocationtechniques, for the more than one processor failure at the same time case, use theconcept of boundary scheduling to presenting minimum cost of backup schedulingalgorithm, which is to back up the cost minimization main objective.④By simulation scheduler experimental data, analyzing the performance ofprimary task scheduling algorithms and backup task scheduling algorithm. The resultsshow that the proposed method has a good reference to task scheduling of cloudcomputing environment.