Real-time operating systems generally depend on some form of priority information for making scheduling decisions. Priorities may take the form of small integers or deadline times, for example, and the priorities indicate the preferred order for execution of the jobs. Unfortunately, most systems suffer from some degree of priority inversion where a high priority job must wait for a lower priority job to execute. The authors consider the nature of the nonpreemptible code sections, called critical sections or critical regions, which give rise to this priority inversion in the context of a soft real-time operating system where average response time for different priority classes is the primary performance metric. An analytical model is described which is used to illustrate how critical regions may affect the time-constrained jobs in a multimedia (soft real-time) task set.>

Workstations for digital audio and video applications must support predictable processing, presentation, and transmission of continuous media data while making efficient use of limited hardware resources. The authors present techniques for managing network resources and processing messages in a predictable fashion. Their solutions use a novel structure for the protocol processing software, and they examine several alternatives and discuss the engineering tradeoffs associated with each. Using simulation, they evaluate their protocol processing structures and show that the more sophisticated structures can improve the response time and predictability of time-constrained messages. The ideas presented in this paper can be applied to resource management in the operating system and to the structuring of application-level servers. The designers of gateway machines (for internetwork routing) which handle both time-constrained traffic and non-time-constrained traffic may also benefit from this study.>

An extension of the uniprocessor Hartstone benchmark for the distributed real-time environment, called the Distributed Hartstone benchmark, is described. The Distributed Hartstone measures system performance in the critical areas of communication latency and bandwidth, protocol preemptability, and priority queueing at the protocol and media access levels. Areas of the system which are particularly important for distributed, real-time computing are described. On the basis of the requirements that specify various areas of the system that a distributed real-time benchmark must stress, a series of task sets in the style of the Hartstone benchmarks are given. The benchmark results from a distributed real-time operating system (ARTS testbed) are given.>

The authors present the motivation for using an object model for real-time operating systems, and describe the object model used in the ARTS kernel. A discussion is presented of the novel object classification and the priority inheritance properties which arise from this taxonomy. Various methods for implementing critical regions are described, and some guidelines as to the use of each are given.>

The priority-inversion problems in real-time communication are addressed, and solutions developed for the ARTS distributed real-time operating system are presented. The performance results of the multi-thread-based protocol implementation are compared with those of other implementation schemes, and the schedulability is analyzed. Experimental results indicate that the multi-thread-based protocol implementation could eliminate potential priority-inversion problems and also demonstrate the same schedulability as the softint implementation scheme in spite of about 10% additional implementation overhead.>