Study On The Real-time Scheduling Policy In Embedded Industrial Monitor And Control Systems

With the development and maturity of embedded industry, intelligent online monitor and error diagnostic systems have become one of the major fields. As a branch of embedded systems, real-time embedded industrial monitor and control system has not only resource constraints that are common to embedded systems, but also specific requirements on high reliability and real-time performance. To satisfy the timing requirements of such control systems, real system oriented predictable scheduling algorithms are needed in the operating system software layer.An abstracted model of a typical real-time embedded industrial monitor and control system is presented in this paper and the real-time characteristics of every module are analyzed. Overall system real-time requirements are given from the view of both embedded systems and industrial monitor and control applications. On the basis of the above analysis, we present some critical principles on selecting scheduling algorithms for such systems according to functionalities of the algorithms. The reasons why Deadline Monotonic algorithm is selected are also given.Ideal Deadline Monotonic algorithm has many assumptions, such as no resource sharing, no logical dependencies among tasks, no context switching overhead and so on. These assumptions may compromise the precision of the schedulability test in a real system.In this paper, Priority Ceiling Protocol, which is first presented by Lui Sha and his colleagues, is borrowed to extend the basic Deadline Monotonic algorithm with the ability to handle resource sharing problems. Then a process transition model is presented and how the scheduler can be activated is analyzed. Based on Katcher's integrated interrupt event-driven scheduling policy, a deep analysis into scheduling overhead and context switching overhead is made, and the schedulability test is further extended to incorporate these overheads.A task partitioning strategy is presented in this paper, which can partition a task into several sub-tasks. If there are tasks that may stay in a critical section for a long period in a system, the task partitioning strategy may improve/ the probability of a higher priority task passing the schedulability test. This strategy can also resolve the inconsistency in the mapping of real importance of a task to its assigned priority.With these revises, the extended Deadline Monotonic schedulability test can handle resource sharing and system overhead and still maintains the characteristics of the basic algorithm. It's apparent that the extended schedulability test has released some assumptions made in the basic algorithm, and it can guarantee the real-time requirements of complex systems.