Computing latency of a real-time system modeled by Synchronous Dataflow Graph

Abstract

Mixed applications that gather real-time tasks and best effort jobs require a research effort in order to be effectively modeled and executed. Therefore, in this study we define a general and intuitive communication model between multi-periodic real-time tasks. We first demonstrate that the communications between real-time tasks can be directly expressed as a "Synchronous Data-flow Graph". This modeling allows precise definition of the system latency. Accordingly, we develop an exact evaluation method to calculate the worst case latency of a system from a given input to a connected outcome. Then, we frame this value using two algorithms that compute its upper and lower bounds. Finally, we show that these bounds can be computed using a polynomial amount of computation time, while the time required to compute the exact value increases linearly according to the average repetition factor. Furthermore, the gap between the exact result and its upper (resp. lower) bound is evaluated between 10 and 15 % (resp. 20 and 30%).