| At present, virtualization technology has been widely applied in the data center, but the I/O performance of VM is always a bottleneck restricting the development of virtualization. Even in SR-IOV, the I/O performance of VM is still poorer than the original system. Optimizing the VCPU scheduler and interrupt handling are the mainly method to improve the I/O performance of VM.VCPU scheduling has a greater impact on VM's I/O performance, but the VCPU scheduling strategy still meets intolerable I/O delay, especially for I/O-latency sensitive VMs which suffer significant performance degradation when competing with CPU-intensive VMs. In this paper we present a flexible and efficient VCPU scheduling algorithm(FLMS) which utilizes VM classification, VCPU binding and flexible slicer to reduce the VM response delay. We have also redesigned the load balancing strategy to ensure optimal VCPU migration. FLMS is suitable for the current mainstream virtualization solutions and ensures the fairness of the whole system. Besides, CPU resource is a key factor of network performance in high-performance networks, the I/O processing usually cannot get enough resources, FLMS design the lending mechanism to solve the problem.Interrupt handling in virtualization environment lead to a context switch and,damage the VM performance.This paper proposes a optimization method based on multiprocessor architecture to solve the problem. Reserving CPUs for the VMM to hand interrupts and bounding the VM to the rest of the CPUs to eliminate the influence of Interrupt injection. Besdies, optimizing load balancing for particular interrupt, bounding it to the CPU with lightest load.Experimental results show that FLMS can improve VM's I/O performance, It has a 30% improvement comparing with the latest strategy. With hardware assisted, FLMS makes it possible for VMs to reach nearly bare-metal performance?The optimized interrupt handing can also reduce the I/O latency and improve the bandwidth. |
PDF Full Text Request