Scheduling Model of Virtual Machine Base on Task Type in Multi-core System

Abstract

The traditional virtual machine scheduling algorithm does not fully consider the execution efficiency of parallel applications. When multiple virtual machines cooperate to execute the parallel computing tasks, the virtual machine monitor still allocates the physical CPUs by the time-division multiplexing method. That will lead the parallel tasks to be serialized and the efficiency degraded greatly. The modern chip multiprocessors platform involves several available computing cores, to meet the need of the concurrent execution of multiple virtual machines. In this paper, we proposed a dynamic scheduling strategy –CON-Credit scheduler, which helps to speed up the parallel applications in virtual environment with multi-cores or many cores system. The main feature of CON-Credit is to map the virtual CPU to the physical CPU directly, so the virtual machines involves parallel tasks can take fully advantage of the underlying hardware resources. More precisely, the CON-Credit algorithm dynamically allocated processor cores to the virtual domains according to the type of the application. For the parallel applications, CON-Credit chooses to schedule a bulk of physical CPUs at the same time to avoid the extra makespan of discrete dispatch in traditional virtual machine scheduling algorithm. The experimental results show that the CON-Credit algorithm improved the execution efficiency of the parallel application and optimized the overall performance of the virtual machine system.