Researches On Real-Time Issues In Open Systems

Currently, open system technology is widespreadly used in the area of information technology. Applying open system technology can develop, integrate, upgrade and maintain application system conveniently. Therefore, the cost of building such system is reduced and the productivity is improved. However, existing open system technology can not support applications with non-functional requirements, such as real-time performance, which are no less important than the functional requirements.At the same time, the coverage of real-time application is increasingly broadening. From traditional industrial control systems to newly emerged networked multimedia systems, the operations with time limitation are needed in them. These systems are required to be open in order to shorten the development cycle and to reduce the cost. Therefore, it is necessary to study how to integrate real-time application with open system architecture. The dissertation is devoted to show the research on this subject. The main contribution is as follows:According to the features of real-time systems and characteristics of open system, the influence that real-time behaviors impact on system's openness is discussed in details. Based on the research, a concept model built on middleware is proposed and the features of middleware for real-time application are presented. In terms of the model,the main stream of current middleware------CORBA, which is selected as researchobject, is analyzed in depth. These analysises specify the direction to improve real-time CORBA to deal with real-tune problem in open system.Real-time CORBA is confined to fixed-priority real-time system. In order to build such a real-time application system, schedulability analysis must be done firstly to verify its distributed real-time performance. However, current scheduling solutions provided by many real-time CORBA systems, pay attention to server-end and depend on single CPU rather than the distributed scheduling. Then an end-to-end scheduling approach for real-time CORBA is proposed. According to this approach, each remote call in a client is mapped to a task that executes on server host. The schedulability of each real-time CORBA application can be verified and the worst-case response time of real-time CORBA application can be obtained through this approach.The scheduling service specified in real-time CORBA helps to simplify development of real-time applications. But usually it uses fixed priority schedulingVIIImethods, which can only be applied to a "closed" CORBA application system. A dynamic scheduling service model is proposed, which can be applied to a dynamic, open real-time CORBA system. A dynamically varying client is permitted to execute on a particular processing node with simple admission test. The admission test is accomplished by cooperation of client scheduler and server scheduler. The scheduling service can support both static scheduling and dynamic scheduling. Based this model, a dynamic CORBA system is presented.In distributed system, interactions between given objects is only possible if a binding has been established between them. The binding model for real-time object should be powerful enough to enhance prediction of interaction. However, the binding model specified in real-time CORBA is too simple to manage and control binding. Based on the binding model suggested by RM-ODP, a binding model for real-time CORBA is proposed, which can be used to establish explicit binding, and manage and control the established one. At the same time, the binding model supports pluggable protocols.A real-time CORBA prototype called rtORB is described at the end. Between ORB and operating system, a platform dependent layer is introduced to decouple the dependence between ORB and operating system. So the prototype is easily ported to different operating systems. Several methods are used to optimize its performance. A series of test show that rtORB is capable enough to support real-time application.