Performance analysis of dynamic soft real-time systems

Abstract

Soft real-time processing is real-time processing in which some or all applications are allowed to miss deadlines, particularly in situations of system overload. Resource allocation decisions in such systems are often based in part on specifications of application utility or benefit and directly affect which applications will miss deadlines and by how much. A feedback mechanism is often provided to inform the applications of their current resource allocations, allowing them to modify their processing to provide the best possible performance. These characteristics result in a situation in which traditional measures of performance, both general-purpose and real-time, can provide incomplete or inaccurate measures of system and application performance. We examine these issues and shortcomings, identify performance factors that need to be measured to address them, and propose specific metrics to quantify these performance factors. We conclude with results from the application of these metrics to our QoS level soft real-time system.