The Design And Implementation Of A Unikernel CPU Scheduler Based On Thread Activity

As a cutting-edge technology in cloud computing,Unikernel combines the isolation of virtual machines and the portability of containers,and has high research value.At the same time,a CPU scheduler plays a key role in operating systems,especially for unikernels which are constructed by library operating systems.However,existing researches focus on applying Unikernel to specific scenarios,and their CPU schedulers are either derived from the complex scheduling mechanism of existing operating systems,which is contrary to the minimalism of unikernels,or designed simply and are difficult to adapt to cloud computing with variable types of applications.The fundamental reason is that these CPU schedulers are not specifically designed according to the characteristics of unikernels.To solve this problem,combined with the single address space feature of unikernels,i.e.a thread is the basic scheduling unit of unikernel CPU scheduler,a thread activity(the measure of thread recent runtime)based CPU scheduler is designed and implemented.The definition formula of thread activity is given according to decaying average mathematical model,then the calculation formula of thread activity is deduced from the definition formula.When computing thread activity,approximate calculation is applied to reduce overhead,and renormalization is used to avoid overflow.Dynamic scheduling timers are designed to wake up the CPU scheduler based on thread activity,and thread hysteresis is designed to cope with the performance loss problem caused by thread competition.Based on OSv,an open-source unikernel project,a prototype of thread activity based CPU scheduler is implemented.Compared with the original CPU scheduler of OSv,testing results show that thread switching time can be reduced by 60.66% at most and program runtime can be reduced by a maximum of 5.7%.