ABSTRACT

In this paper, we have analysed i) the relationship between the checkpoint cost and the optimal checkpoint interval and ii) the relationship between the checkpoint cost and the number of processors (processes) and we have also determined the optimal number of processors (processes) required for executing the fault tolerant MPI applications. We have presented an experimental study in which, we have used an optimal checkpoint restart model [1] with Weibull�s and Exponential distributions to determine the optimal checkpoint interval. We have observed that, the optimal checkpoint intervals obtained using Weibull�s distribution, produce minimal average completion time; as compared with the optimal checkpoint intervals obtained using Exponential distribution. The optimal checkpoint interval is approximately directly proportional to the checkpoint cost and inversely proportional to shape parameter. The study indicates that, the checkpoint cost of MPI applications increases with the number of processors (processes) used for execution. We have determined the optimal number of processes (processors) required to execute the MPI applications considered in this paper, as 4.