Patterns for Automatic Generation of Soft Real-time System Models

Abstract

Worst-case assumptions about the timing of systems are often too conservative when analyzing distributed soft real-time systems as they lead to over-dimensioned and expensive products. For these systems, a certain percentage of deadline misses is often affordable. Instead of a binary answer regarding the schedulability of such a system, a more interesting metric is the degree to which the system meets the timing requirements. For this, an appropriate model that realistically expresses the behavior of a soft real-time system when deployed on a multiprocessor platform should be built and analyzed. In this article, we present such a modeling approach based on the formal modeling language POOSL (parallel object-oriented specification language). Moreover, to alleviate the process of modeling, we present a pattern-based description language that allows an application, together with the multiprocessor platform and the deployment to be described in a concise way. Such a pattern-based description can be translated automatically into an executable POOSL model through which performance properties can be analyzed based on simulations. The suitability of our approach is demonstrated by exploring the design space of a distributed in-car radio navigation system.