Virtual machine scheduling method, device and equipment and readable storage medium

Abstract

The invention discloses a virtual machine scheduling method, device and equipment, and a readable storage medium. The method comprises the steps of obtaining a pcpu utilization rate of each physical cpu; according to the pcpu utilization rate, determining a physical cpu to be adjusted in the physical cpu; determining a target sub-thread in the sub-threads according to the vcpu utilization rate of the sub-threads of the virtual machine on the physical cpu to be adjusted; and scheduling the target sub-thread from the source physical cpu to the target physical cpu; According to the method, device and equipment, the vcpu distribution and scheduling of the virtual machine are dynamically controlled by utilizing the pcpu utilization rate of each physical cpu and the vcpu utilization rate of the sub-thread of the virtual machine, so that the pcpu utilization rate of each physical cpu can be balanced, the synchronization delay of the virtual machine is reduced, and the performance of the virtual machine is improved; In addition, synchronous scheduling does not need to be carried out on all the vcpu of the virtual machine, and the situations of fragments used by the cpu and priority inversion are reduced.

Description

technical field [0001] The present invention relates to the technical field of virtualization, in particular to a virtual machine scheduling method, device, equipment and readable storage medium. Background technique [0002] With the development of science and technology in modern society, the application of virtualization technology is becoming more and more extensive. In the virtualization scenario, there will be synchronization delay problems that do not exist in the physical environment. Among them, the problem of spin locks (spinlocks) is particularly prominent. The spin locks are mainly used for synchronization between threads. The kernel thread will not be preempted, but in the virtualization environment, the child thread (vcpu) of the virtual machine (VM, Virtual Machine) holding the spin lock will be preempted by other threads, because the vcpu behaves as a user in the physical environment. State threads can be preempted by other high-priority threads. Only when t...

AI Technical Summary

Problems solved by technology

In the virtualization scenario, there will be synchronization delay problems that do not exist in the physical environment. Among them, the problem of spin locks (spinlocks) is particularly prominent. The spin locks are mainly used for synchronization between threads. The kernel thread will not be preempted, but in the virtualization environment, the child thread (vcpu) of the virtual machine (VM, Virtual Machine) holding the spin lock will be preempted by other threads, because the vcpu behaves as a user in the physical environment. State threads can be preempted by other high-priority threads. Only when the preempted vcpu gets the physical cpu (pcpu, cpu hyperthreading) running again can it complete the synchronization with other vcpus. The preempted time will cause synchronization delay. At the same time, it will also cause other vcpus waiting for the spin lock to be blocked, which will lead to a serious decline in the performance of the virtual machine; in the existing case, there have been 30% decline in database performance and the blue screen of the windows virtual machine

[0003] In the prior art, the co-scheduling (cooperative scheduling) algorithm is often used to solve the synchronization delay problem of the virtual machine, and the co-scheduling algorithm will schedule all the vcpus of the virtual machine at the same time, that is, all the vcpus of the virtual machine either obtain the pcpu at the same time, or Release the pcpu at the same time; and once a vcpu cannot get the pcpu, other vcpus that have already got the pcpu will also wait, which will cause the fragmentation problem of the cpu utilization, for example, the vcpu1 and vcpu2 of the virtual machine are in the pcpu1 and pcpu2 at the same time The running queue is ready to run, but at this time only pcpu1 is idle and pcpu2 is busy, then neither vcpu1 nor vcpu2 will be run, pcpu1 will wait until pcpu2 is also free, and then run vcpu1 and vcpu2 at the same time, pcpu1 waits This period of time is cpu fragmentation, which will reduce cpu utilization, because this time slice that could have been used to run other tasks is occupied by vcpu1 busy waiting; and the waiting time of vcpu will cause the running time of high-priority tasks to be delayed. Can be scheduled in time, leading to the problem of priority inversion

Images