Rate-monotonic scheduling in the presence of timing unpredictability

Abstract

Scheduling periodic hard real time tasks is of great importance to many real time applications, such as open architecture machine tool controllers and avionic systems. The rate monotonic scheduling algorithm has been proven to provide an optimal static priority assignment under idealized conditions. However some of these conditions are not met in a real computer system. In particular the release times of tasks can deviate from the specified time instants because of operating system software timer unpredictability. We investigate the timer behaviors in three commercial real time operating systems, VxWorks, QNX and pSOSystem. Based on our findings, we propose an empirical task schedulability model, called RMTU (Rate Monotonic in the presence of Timing Unpredictability), to augment the rate monotonic scheduling theory in order to handle timing unpredictability. We then design an approach to systematically derive the model parameters by measurements. With RMTU, task deadlines can be empirically guaranteed. The validity of RMTU is supported by our measurement data. These results are useful not only to control application developers, but also to real time practitioners at large.