RCBW:the Reservation Of CPU Bandwidth Scheduler Mechanism In Non-realtime System

The cloud platform is one of the hot topics of commercial and scientific research and applications in domestic and abroad.With the wide use of cloud platforms,system administrators are paying increasingly attention to the Quality of Service(QoS)of the system resources,among which the CPU QoS serves as an extremely fundamental position.To improve the CPU QoS of the cloud platform,the underlying virtualization platform needs to provide the function of the virtual machine-level CPU QoS,and the operating system also needs to provide the function of the processlevel CPU QoS.However,current existing literature and enormous experimental results show that the Linux operating system does not have the complete capability of process-level CPU QoS.Based on Linux,this thesis researches process-level CPU QoS.It proposes RCBW,the nonreal-time reserving CPU bandwidth scheduling mechanism.It implements process-level control of CPU resource reservation in Linux kernel to improve the process-level CPU QoS.RCBW integrates a reservation-based scheduling algorithm based on Linux's existing non-real-time scheduler-the Completely Fair Scheduler,therefore,when a certain group of processes competes with other groups for CPU resources,the operating system can still accurately guarantee the minimum running time of a specific group with the support of RCBW.The RCBW scheduling mechanism primarily includes the design,modification and implementation of dynamic configuration module,scheduling module and bandwidth pool model.RCBW provides multiple interface files in the user space which allow system administrators to dynamically configure the minimum running time of a specific group.The experiment scheme we design in this paper is about the test and evaluation of the effectiveness and accuracy of the RCBW scheduling mechanism in both single-core and multi-core environments.The experimental results show that the RCBW scheduling mechanism can provide a guarantee of minimum runtime for a specific group in the CPU competition environment in an accurate level and an ability to dynamically modify this minimum runtime as needed.