Periodic Task Mining in Embedded System Traces

Modern systems are growing in complexity beyond deep comprehension of developers. Increasing difficulties of keeping software projects on schedule and increasing recall rates are symptoms of this development.  Consequently, developers need new methods and tools to build embedded systems, such as tools that dynamically analyze systems and recover comprehensible specifications of particular aspects. In this paper, we address the problem of discovering temporal behavior of real-time systems by mining periodic task sets and their temporal characteristics from system execution traces. We leverage the periodic nature of real-time systems to achieve this goal in an automatic way.
We propose PeTaMi (PEriodic TAsk MIner) - a novel approach and a tool to mine periodic tasks along with information on their periods and response time profiles from execution traces of real-time systems. PeTaMi embraces an important observation we make about operation of periodic tasks: their individual jobs are usually followed by intervals of task inactivity of a considerable duration. We evaluated PeTaMi on two case studies (unmanned aerial vehicle and a commercial car in operation) using traces containing tens of thousands of recorded execution events.