The Research And Implementation Of Real-Time Middleware Based On Priority

With the development of the network, distributed computing has become the mainstream of the computing technology undoubtedly. As one of the key techniques, although the current state-of-the-art distributed middlewares could make information and computational resource shared more transparently and effectively by providing fine communication transparency and well designed development environment, the serious problem behind these systems is the lack of time-constraint support for real-time applications. Consequently, while more and more mission critical application domains have adopted distributed object-oriented architecture, real-time middlewares with time-constraint support become the focus on such preliminary research to fulfil the increasing requirement for it.The paper is mainly devoted to the concept of real-time middleware, architecture composition, validity verification and real-time infrastructure. Moreover, an implementation of highly network manageable, self adaptive priority tuning real-time middleware system is introduced finally.The dissertation aims at making such contributions:Concepts about real-time middleware are identified firstly, with some ambiguities to be clarified. The paper defines a real-time middleware architecture as the role of real-time foundation infrastructure. The paper also gives the point of view that distributed real-time middleware should provide the predictability of response for such applications. Furthermore, real-time system, real-time OS and real-time middleware are distinguished in detail, with the conclusion that real-time middleware should provide the mechanism of time-constraints presentation and resource control.Based on queuing simulation, a new FP queuing model for real-time middleware is proposed to evaluate the real-time middleware architecture. Factors that impact request time constraints are collected, of which the comprehensive analysis is made to verify the new architecture and demonstrate its effectiveness.With the regard to the compatibility of the real-time middleware for heterogeneous platform, this paper brings forward the real-time abstract layer (RT-AL) different from the traditional Ad-hoc scheme solution. The RT-AL provides consistent thread interface and identical thread scheduling model from the semantic aspect. What's the most important, priority inversion could be avoided by the RT-AL. And the RT-AL can also be optimized for network and I/O operations. The way to build the real-time middleware based on the RT-AL completely follows the modularity in software engineering and consequently greatly simplifies the work of development and maintenance.To deal with the difference on operating system level, this paper designs a thread-level algorithm based upon priority inheritance protocol to prevent priority inversion. A dynamicpriority mapping and scheduling algorithm to achieve strict order is also presented. Through analysis of multi-threaded execution order under fixed priority scheduling model, a multithreaded programming model with totally predicable execution order is proposed.Combined with soft real-time scheduler and multi-threaded library, a customized self-scheduled thread library is presented. Such self-scheduled thread library can bring priority expansion and schedule convenience for operating system. The key ideas of design and implementation of such library are introduced in the paper.Conclusively, the process to develop a priority-based real-time ORB (RTORB) is illustrated for verification and test. The advantages of RTORB is demonstrated by the practical system, including the well network manageability, the support of multi-level thread pool and priority enabled queue, the ability of adaptive priority tuning and the resource planning. Through experimental evaluations, the RTORB shows far more effectiveness than the ordinary ORB for distributed real-time applications. Thus the real-time middleware architecture presented in this paper is proved to be effective. More improvement could be expected in the future.