Using sequential debugging techniques with massively parallel programs

Debugging is a crucial part of the software development process. Especially massively-parallel programs impose huge difficulties to program analyis and debugging due to their higher complexity compared to sequential programs. For debugging and analysing parallel programs there are several tools available, but many of these fail in case of massively-parallel programs with potentially thousands of processes. In this work we introduce the single process debugging strategy, a scalable debugging strategy for massively-parallel programs. The goal of this strategy is to make debugging large scale programs as simple and straight-forward as debugging sequential programs. This is achieved by adapting and combining several techniques which are well known from sequential debugging. In combination, these techniques give the user the possibility to execute and investigate small fractions of a possibly huge parallel program, without having to (re-)execute the entire program.