Reducing virtual machine pre-emption in virtualized environment

摘要

A method for reducing virtual machine preemption in a virtualized environment is provided. The method includes dispatching a virtual central processing unit (CPU) to run in an emulation mode on a real CPU until the real CPU exits the emulation mode, determining whether the virtual CPU has loaded a wait state, determining whether a remaining time slice of the virtual CPU as a result of the dispatching is below a predefined threshold in an event that the virtual CPU has loaded the wait state and rescheduling the virtual CPU with a full time slice in an event the remaining time slice of the virtual CPU is below the predefined threshold.

主权项

1. A method for reducing virtual machine preemption in a virtualized environment, the method comprising:
dispatching a virtual central processing unit (CPU) to run in an emulation mode on a real CPU until the real CPU exits the emulation mode; determining whether the virtual CPU has loaded a wait state; in an event that the virtual CPU has loaded the wait state, determining whether a remaining time slice of the virtual CPU as a result of the dispatching is below a predefined threshold provided as a time slice percentage; in an event the remaining time slice of the virtual CPU is not below the predefined threshold, marking the virtual CPU as being idle with the remaining time slice being retained thereby; in an event the remaining time slice of the virtual CPU is below the predefined threshold, rescheduling the virtual CPU with a full time slice and marking the virtual CPU as being idle with the full time slice being retained thereby; increasing the predefined threshold in accordance with a determination that excessive virtual CPUs are exceeding respective time slices; and decreasing the predefined threshold in accordance with a determination that excessive hypervisor overhead is occurring.