Modern Operating System Scheduling Strategy

The classical all-purpose OS has not satisfied the demand of co-existence of hard real-time applications, soft real-time applications and general applications. So devising scheduling policy which provides universal subsistence for all types of applications is the hot spot in the research of modern OS.This paper studies the scheduling mechanisms of Linux, Unix and Windows NT OS and points their shortcomings. At the same time, the article analyses classical policy and scheme which support real-time application in or out of china. RT-Linux provides hard real-time for preferable support .The article analyzes it and figures out an outline of a scheduling mechanism. It differentiates hard real-time, soft real-time and best-effort applications and supplies different applications for CPU's bandwidth according to their own policy. Hard real-time applications are executed with a priority because of its processing in real-time kernel .The control of time ratio shared by CPU for soft real-time tasks can keep general time-sharing task not to starve. Further, the access control of soft real-time tasks and the modification of CPU's time-sharing ratio according to the dropped deadline guarantees the QoS demand of soft real-time task. Therefore, it ensures that the tasks of hard real-time do not drop deadline, guarantees the QoS demand of soft real-time tasks and does not starve common tasks.In order to verify the validity of scheduling policy, we design a virtual simulation system for the process scheduling of OS by the use of UML method. In this article, we devise the simulation system taking virtual time for driver, taking task for the underlying scheduling object and taking random producer for producer of system load. Simultaneously we test the RMA and round robin with multiple feedback in the underlying real-time scheduling algorithm. We test the RMA and round robin with multiple feedback in the simulation system. When the system load over 0.69, there are deadline dropping in the RMA scheduling. With the time going by, the tasks in the round robin with multiple feedback get approximate time. We test scheduling policy in the 3.3, the result show that the scheduling policy in 3.3 ensures that the task of hard real-time does not drop deadline, guarantees the QoS demand of soft real-time tasks and does not starve common tasks.