Using online worst-case execution time analysis and alternative tasks in real time systems

Abstract

As embedded hardware becomes more powerful, it allows for more complex realtime systems running tasks with highly dynamic execution times. This dynamicitymakes the already formidable task of producing accurate WCET analysis evenmore di?cult. Since the variation in execution time depends on task input andthe state of the system, it is postulated that a more accurate estimate for theWCET can be found online with knowledge about the task parameters.This thesis will explore the concept of online execution time analysis and itspotential utilization. Line detection in images through Hough line transform isfound to be a relevant application whose execution time can be estimated bythe contrast of the input image. A system for scheduling tasks utilizing theironline WCET estimate is then discussed. It dynamically checks for potentialdeadline misses and degrades tasks, either by running a more e?cient alternativetask instead or by aborting the task, until timely execution is guaranteed. Anexperiment is presented, demonstrating a higher throughput of tasks with onlineWCET estimation. Finally, the work on a framework for more precise simulationsand experiments is presented.