On the Granularity of Events when Modeling Program Executions

In the context of parallel programming, program execution models are required to get better understanding of the behavior, properties, and characteristics of parallel programs. The main disadvantages of most existing models are the pretty stii deenition of the events and the generally not clearly deened semantic associated to the model. The goal of the paper is (1) to deene a program execution model that uniquely characterize a program execution in terms of program behavior, and (2) determine the suucient conditions on the granularity of the events that are required to achieve (1). One essential result of our analysis is that it deenes the theoretical framework for reasoning about execution replay, whose goal is to circumvent the non-determinism during the debugging phase. Another essential result is the possibility to model collective communication, which is part of the External User Interface (EUI) on the 1 announced IBM Scalable POWERparallel System 9076 SP1, and which will be part of the future Message Passing Interface (MPI) standard.