An approach to providing small-waiting time during debugging message-passing programs

Abstract

Cyclic debugging, where a program is executed repeatedly, is a popular methodology for tracking down and eliminating bugs. Breakpointing is used in cyclic debugging to stop the execution at arbitrary points and inspect the program's state. These techniques are well understood for sequential programs but they require additional efforts when applied to parallel programs. For example, record&replay mechanisms are required due to nondeterminism. A problem is the cost associated with restarting the program's execution every time from the beginning until arriving at the breakpoints. A corresponding solution is offered by combining checkpointing and debugging, which allows restarting an execution at an intermediate state. However, minimizing the replay time is still a challenge. Previous methods either cannot ensure that the replay time has an upper bound or accept the probe effect, where the program's behavior changes due to the overhead of additional code. Small waiting time is the key that allows to develop debugging tools, in which some degree of interactivity for the user's investigations is required. This paper introduces the MRT method to limit the waiting time with low logging overhead and the four-phase-replay method to avoid the probe effect. The resulting techniques are able to reduce the waiting time and the costs of cyclic debugging.