Causal-Consistent Replay Debugging for Message Passing Programs.

Debugging of concurrent systems is a tedious and error-prone activity. A main issue is that there is no guarantee that a bug that appears in the original computation is replayed inside the debugger. This problem is usually tackled by so-called replay debugging, which allows the user to record a program execution and replay it inside the debugger. In this paper, we present a novel technique for replay debugging that we call controlled causal-consistent replay. Controlled causal-consistent replay allows the user to record a program execution and, in contrast to traditional replay debuggers, to reproduce a visible misbehavior inside the debugger including all and only its causes. In this way, the user is not distracted by the actions of other, unrelated processes.