An Optimized Scheduling Strategy In Xen Based On Task Type

With the rapid development of computer hardware, researching on virtualization technology has become a hot point again in the area of computers. Virtualization technology can maximize the performance of the hardware and build the cloud computing platform rapidly for enterprise, then improve operational efficiency and save the economic cost to avoid wasted space. Virtualization software is in the intermediate layer between the underlying hardware and the operating system, which abstract the underlying hardware resources into another form of virtual hardware resources, such as virtual CPU, virtual memory, etc., providing to the guest operating system of the top layer.Virtual machine provided services for mainframe systems in the early. With the support for virtualization of the hardware and the performance improvements of personal computer, the virtual machine technology has been turned to the general user community from the original server private. Today's computation, especially in client side, has already gone into the transition from computer-centric to user-centric. People concern more on what kind of interface and what kind of service could be provided to adapt user's various requirements. The types of the tasks on virtual machines have been more diverse, there are two tasks of type in the virtual domain:the CPU intensive task and the GPU intensive task.This paper takes Xen platform as the study platform. Firstly, we describe the virtualization research status especially about the current Xen scheduling algorithm and in-depth analysis the operating principle of three core subsystems in Xen hypervisor. Then we do three sets of experiments and analysis the programming model of the GPU-intensive task. The conclusion is that:the CPU resource sensitivity of GPU tasks is higher than that of CPU tasks. Based on the above conclusion, we analysis the weakness of the Credit scheduling algorithm, then we proposed an optimized scheduling strategy in Xen based on task type. This strategy monitors the real-time CPU and GPU resources utilization of the domains to calculate the customer's real-time scheduling priority. Xen scheduler re-allocates the physical CPU resources based on the real-time scheduling priority. When the physical CPU resources cannot meet the requirements of the tasks, this strategy will increase the real-time scheduling priority of the GPU-intensive domain to improve the performance of GPU-intensive tasks. This paper designs the adjustable CPU and GPU intensive tasks scene based CSSVM architecture. These two scenes has played a significant role in the final experiment.They are the effective solution to the problem that the benchmark of the virtual machine cannot work effectively.Finally, the paper sets up a virtual experiment platform and does three groups of targeted experiments. The experimental results verify the correctness and the feasibility of the optimized strategy. The optimized scheduling strategy in Xen based on task type effectively promotes the performance of the GPU-intensive tasks and improve the user's experience when the CPU resource cannot meet the requirements of the GPU-intensive tasks.