Adaptive Process Scheduling For Real-Time Systems

With the development of new computing sciences and technologies, such as multimedia technology, embedded system, mobile computing and pervasive computing, the environments real-time systems face become more and more dynamic and unpredictable. How to try to perform tasks' processing in prescriptive time and be able to control all real-time devices and tasks' coordinated running in dynamic and unpredictable environments bring new challenge to real-time systems. In this dissertation, some issues of adaptive scheduling in dynamic environments are discussed to enhance real-time systems' performance.The previous process scheduling approaches for real-time systems in dynamic environment are studied in this dissertation. Based on the closed-loop feedback dynamic process scheduling approaches, a framework of the adaptive process scheduling is proposed. Three adaptive process-scheduling approaches based on the framework are also developed in the dissertation. The main contributions of this dissertation are described as follows:(1) It proposes a formal definition of the adaptive process scheduling and an adaptive process-scheduling framework.Some shortcomings of the closed-loop feedback based process scheduling approaches are analyzed at first. Then necessity of adopting the adaptive process scheduling approaches is proposed. In order to strictly distinguish the differences between the adaptive and the closed-loop feedback process scheduling approaches, the formal definitions of them are proposed. A system framework of the adaptive process scheduling based on the formal definition is also proposed.(2) It proposes an adaptive process scheduling approach based on fuzzy inference and genetic algorithm.In the Fuzzy Inference and Genetic Algorithm Based Adaptive Process Scheduling (FuGAPS) approach, CPU resources are allocated through closed loops by PID (Proportional-Integral-Derivative) controllers according to system scheduling errors. Fuzzy controller decides the parameters of the PID controllers according to the current system state. The fuzzy rules of the fuzzy controller are decided by a genetic algorithm. The FuGAPS approach is suitable for the real-time system whose mathematic model is difficult to be created. But it needs an offline genetic searching environment.(3) It proposes an adaptive process scheduling approach based on linear regression.In the Linear Regression Based Adaptive Process Scheduling (LiRAPS) approach, the system resources are allocated through the closed loop by the feedback controllers according to the system scheduling errors. The parameters of the feedback controllers are decided by a multiple linear regression model. The LiRAPS approach is suitable for linear model system. The parameters of the scheduler are changed at once after online adaptation. This makes the real-time systems have a quicker adaptive response.(4) It proposes an adaptive process scheduling approach based on the curvilinear regression that can be transformed to a linear regression.In the Linear Regression Transform Enabled Curvilinear Regression Based Adaptive Process Scheduling (LiCAPS) approach, the system resources are allocated through the closed loop by the feedback controllers according to the system scheduling errors. The parameters of the feedback controllers are decided by a multiple curvilinear regression model that can be transformed to a multiple linear regression model. The LiCAPS approach is suitable for the systems whose model is a curvilinear model that can be transformed to a linear model. The parameters of the scheduler are changed at once after online adaptation. This enables the real-time systems to have a quicker adaptive response.As to the FuGAPS approach, a case study, i.e. web service response delay control, is given in this dissertation. Web service processes are divided into several process class queues according to the demand of the response delay. These processes in those class queues are scheduled according to the FuGAPS approach. Experimental data show that the FuGAPS approach can effectively guara...