Dynamical Time Slice Scheduling For Virtual Machine Monitor

Many of the traditional computer technology are used in cloud com-puting, including virtualization which plays an important role. The perfor-mance of I/O request processing will affect the performance of the entire system. The traditional VMM scheduler will treat I/O tasks and CPU tasks equally, which leads to poor efficiency and performance of the entire sys-tem. Some existing research did not consider the allocation of sufficient CPU resources to handle I/O requests and therefore affect the efficiency of I/O tasks. Some existing research will mark the type of VMs only when the system is started up. when a VM that is not marked as I/O-intensive receive a lot of I/O requests, the scheduler can not change the scheduling policy, leading to the delay of I/O request.In this paper, we prove the existence of I/O response latency issues of Xen through experiments. When I/O requests arrive, I/O-intensive virtual machines can only wait for its turn to be scheduled to process I/O requests. If the system load is relatively high, I/O latency will become very large. We also prove that if a virtual machine doesn’t get enough time to process I/O request, it will not be able to handle large amounts of packets arriving in a short time, resulting in performance degradation.This paper presents a dynamic time slice scheduling strategy and the implement the system prototype in Xen. This scheduling strategy monitors interrupt frequency and CPU usage, distinguishes different types of VMs depending on the interrupt frequency, and assign the appropriate time slice to VM based on its CPU usage. We improve the performance of I/O request with ensuring the fairness of scheduling.Finally, we verify that the performance of network I/O, and disk I/O response time and throughput of or dynamic time slice scheduler are bet-ter than that of Xen credit scheduler. And for the CPU-intensive VMs, the performance difference between the two is very little. We verify that our scheduler improve the performance of I/O request with ensuring the fairness of scheduling and introducing a small overhead.