Task Statistical Feature Based Resource Management For Cloud Computing Systems

Low resource utilization is a common issue in cloud computing systems.It increases the job completion time and the cost of users using cluster resources.Efficient resource allocation is a key issue in the design of cloud computing systems.The diversity of applications and the heterogeneous of cluster are the major challenge of resource management.Analysis of cluster workload trace provides a good understanding of cluster and job characteristics.The existing researches assumes that the resource consumption and completion time of the same type of tasks are the same.The similarities of tasks are further studied.The characteristics of the task can be well described by the probability distribution of resource consumption and completion time,and the probability distribution function can be obtained by Gaussian fitting.Based on the resource consumption probability distribution,a task statistical feature based resource allocation strategy is proposed.Through theoretical analysis,the resource allocation strategy can make the resource utilization theoretically optimal.The experimental results show that the strategy can achieve good performance,which can reduce the gap between resource allocation and resource consumption,effectively improve resource utilization to 85% and reduce job completion time over 25%.As the resources available on the node increase,the effect of reducing application execution time increases.Co-located latency-critical jobs and batch jobs is an effective way to increase data center utilization and reduce costs.Based on the completion time probability distribution,a transient resource management scheme based on task statistical features is proposed.The experimental results show the efficiency of the resource management scheme.Under different tail latency requirements,TLAA can improve the resource utilization and reduce the job completion time of batch jobs.CERR can reduce the impact of resource revocation on batch jobs and has better performance than existing resource revocation strategies.