Study Of Control-performance-driven Scheduling Parameters Optimization In Real-time Systems

A real-time system is one that can perform calculations,process transactions and produce reactions within an explicit deterministic time.Control applications are often integrated into a real-time system for real-time control,formulating a real-time control system.A real-time system is often composed of multiple computing tasks,which share computing platform resources,so scheduling is needed for resource allocation.Scheduling may cause the fact that tasks cannot timely get access to required resources and delay is subsequently induced.In the control domain,delay caused by scheduling will make control tasks unable to execute strictly periodically and even jeopardize the control stability.Therefore,it is necessary to co-design the scheduling and control by taking issues from these two domains into account.Based on the schedulability analysis theory from computing domain,this thesis co-designs the scheduling and control by means of control-performance-driven scheduling parameters optimization.The problem of control-performance-driven scheduling parameters design is mathematically modeled as a constrained optimization problem to optimize the overall system control performance,with the constraints of system schedulability and control stability.Specifically,this thesis aims at studying the airborne real-time system.Nowadays,modern avionics systems adopt the integrated modular avionics(IMA)architecture,which is updated from the integrated modular avionics(IMA)architecture,and multi-core processors may be employed in the future.According to the chronological sequence,traditional embedded real-time systems,partitioned real-time systems and partitioned multiprocessor real-time systems are investigated in this thesis.The main contents and innovations are as follows:For traditional embedded real-time systems,the scheduling parameter optimization problem is studied for real-time systems adopting the Earliest Deadline First scheduling algorithm.The control performance function about task scheduling parameters is analyzed and the constrained optimization problem is specified,which takes the period and deadline of a control task as variables,control performance function as target function,system schedulability and control stability as constraints.The feasible space of the variables is complex due to the tight coupling between periods and deadlines of tasks caused by system schedulability constraints.An approximation algorithm is proposed to solve the optimization problem,which handles the parameter coupling by treating period and deadline as undifferentiated parameters when making adjustments.Experimental results show that the method proposed in this thesis handles the coupling problem better and improves the overall system control performance by selecting proper scheduling parameters.The schedulability test problem is studied for the ARINC 653 partitioned real-time system widely adopted in IMA.A schedulability criterion is derived by analyzing the relation between workload requirement and resource supply ability of the platform.The schedulability criterion,which takes the feature of the ARINC 653 system into consideration,is proven to be both sufficient and necessary.Meanwhile,to test the system schedulability by means of software execution simulation,a virtual IMA software test platform based on interface simulation was developed.The platform is able to provide partitioned scheduling execution environment with general computing devices,and can be used for software development,schedulability test and validation.It has been successfully applied to a major model project.Design of the Major Time Frame for the ARINC 653 partitioned real-time system is an opposite problem of the schedulability test.Based on the schedulability analysis,the period and utilization of a periodic resource partition are calculated to guarantee the schedulability of the task set integrated in a partition.Afterwards an algorithm referred to as MFBF is proposed to schedule partitions and generate the Major Time Frame.These two processes produce the Major Time Frame from several partition applications and experimental results show that the MFBF algorithm reduces the number of partition switches effectively.The problem of optimizing parameters for a partition containing control tasks is further analyzed.The period and utilization of the partition are selected as variables and the control performance function about these two parameters is formulated.Thus a parameter optimization problem is built and the method to solve this problem is also presented.Experimental results show the feasibility of the proposed model.Design of the Major Time Frame for the multi-core IMA adopting partitioned scheduling policy is investigated.In such a situation,a multi-core time resource partition is allowed to contain multiple parallel time windows,where the tasks are globally scheduled.A binary search algorithm is proposed to calculate the minimal resource parameters required by the interface model to guarantee that the task set in a partition is schedulable.Besides,a general method to optimize the system control performance by selecting interface parameter is discussed.Furthermore,to handle the problem of partition integration,the partition model is decomposed into a set of tasks to be scheduled afterwards.A Major Time Frame is generated by a case study of MPR partition model,showing the feasibility of the proposed method to generate the Major Time Frame.