A Periodic Task-Oriented Scheduling Architecture in Cloud Computing

Peng Zhang, Yan Li, Hailun Lin, Jianwu Wang, Chuang Zhang

Abstract: With the rapid development of cloud computing, more and more applications are deployed in the cloud. This trend brings cloud infrastructure providers to build distributed data centers to provide physical resources for multi-domain-deployed jobs of these applications. However, the heterogeneity of resources and the different execution periodicity of tasks in jobs pose great challenges to the efficiency of large scale job executions. To solve this problem, we propose a periodic task-oriented scheduling architecture in cloud computing. This architecture provides resource domains for the scheduler to shield the details of the underlying heterogeneous resources. Moreover, it implements job-level scheduling and task-level scheduling by our proposed PTS algorithm and RMDU4T algorithm to improve the execution efficiency and resource utilization. Experimental results show the proposed architecture can reduce the scheduling times to less than 10% and reduce the average job completion time by 10%~40% under different distribution of tasks and resources, and the CPU resource utilization is improved by about 15%.