An Optimized Scheduling Strategy Based on Task Type in Xen

Abstract

Along with the development of virtualization technology, the computing tasks’ type worked in virtual machine is more diverse and GPU intensive task has become a trend. But most current VMM scheduling algorithms pay little attention to the task types of diversification, but abide by fair principle and distribute CPU resources equally, which would be unable to meet the requirement for latency-sensitive tasks. Through theoretical analysis and a series of experiments, this paper confirme that the CPU resources sensitivity of different task type is different and put forward an optimized scheduling strategy based on task type. The VMM calculates the real-time scheduling priority according to CPU and GPU resource utilization of all active guest domains, and then redistributes the physical CPU resources. The experiments show that the optimized scheduling strategy can promote the performance of the GPU intensive tasks and improve user experience.