Starting conditions for post-mortem debugging using deterministic replay of real-time systems

Abstract

Repeatable executions are required in order to successfully debug a computer system. However, for real-time systems, interactions with the environment and race conditions in the execution of multitasking real-time systems software make reproducible behavior difficult to achieve. Earlier work on debugging of real-time software has established the use of a deterministic replay, a record/replay solution, as a viable approach to reproduce executions. When combining the deterministic replay approach with infinite loop recorders (similar to black-box recorders in airplanes) for post-mortem debugging, it is essential that the recordings are sufficiently long and detailed in order to be able to re-execute the system. Basis problems however, are how to find a well-defined starting point within the recording, and how to find a reachable state in the rebooted/restarted system to match that instance? Previous work has not presented solutions to these fundamental problems, in this paper we do. We also present some implementation details from an industrial case study.