Research Of Multiple DAGs Scheduling On The Cloud

Cloud computing is the hot spot of commercial and scientific research field at home andabroad, it is a development of the grid computing, parallel computing and distributedcomputing and also a new type of commercial computing model. Using mature virtualizationtechnology, it can provide services according to user needs. Job scheduling and resourceallocation are two key techniques of cloud computing. Cloud computing pays great attentionto the Qos of users, its virtualization technology makes the resource allocation and jobscheduling different from parallel distributed computing. This dissertation introduces theconcept of cloud and the hot topic of the cloud research. Also summarizes the mechanism oftask scheduling in cloud environment systematically. To reduce the execution time of singletask, proposed the algorithm based on the failure rate of resources to select a part of theresource， Compare with the algorithm which does not consider the resource failure rate, thesimulation experimental results show that this algorithm can be better to reduce the executiontime. According to multiple DAGs workflow scheduling problem in heterogeneous distributedenvironments, proposed a scheduling algorithm based on minimize the data transportation andcompletion time, which can deal with the problem that multiple DAGs workflow have thesame priority, and gives the multi-priority multi-DAG mixed scheduling algorithm. Comparedwith E-Fairness algorithm, this algorithm can avoid additional data transfer overhead, shortenthe entire workflow execution Makespan, and improve resource utilization. Moreover,proposed a scheduling algorithm based on the dynamic of RANK-Hierarchical, which candeal with the problem that multiple DAGs workflow comes at different time and have variouskinds of structure. Compared with RANK-HYBD algorithm, the experiments show that theRANK-Hierarchical algorithm can not only improve the fairness of the scheduling of multipleDAGs, but also can shorten the average execution Makespan effectively.