Abstract

In real-time systems, deadlines are imposed on the response time. Not meeting a deadline in a hard real-time system is equivalent to its failure, while in soft real-time systems occasional minor delays in responding to events are acceptable. Only when the delays are frequent or considerable, performance degradation up to system malfunction can be observed. The developers for real-time systems must therefore pay special attention to the performance of the system. The scheduling of tasks in the system becomes critically important as it directly affects the system performance. While, traditionally, real-time system developers have used low-level software programming paradigms, the rising complexity of real-time software is creating a demand for CASE tools that allow for development using a combination of visual modeling and design, augmented with code-segments. One such tool is the ObjecTime Developer based on ROOM (Real-Time Object Oriented Modeling) development methodology. In this thesis we study usability and effectiveness of this tool for building a soft real time system with special attention on the performance and behaviour of the system under overload conditions. As a working example we develop a radar simulator system, which observes air targets whose speeds, number and distances are constantly changing; thus, creating a constantly varying load of the entire system, with dynamically appearing objects. The major contribution of this thesis is developing custom overload handling policies to improve performance and illustrating how they may be implemented within the framework of the tool. The native scheduling policy of ObjecTime, the classical priority-based policy and our own policy based on the period-of-execution are studied