Using Tough2-MP on a cluster—optimization methodology and study of scalability

Evaluating the impacts of heat and gas production in a deep geological repository for radioactive waste necessitates a large number of numerical simulations. The task is best approached by running a massively parallel code such as Tough2-MP on a cluster. One key issue then becomes the optimal utilization of the computer resources. The newly developed methodology facilitates the efficient distribution of the workload on the cluster by guiding the selection of the partitioning settings for the model and the repartitioning of the computation load among the nodes and cores. Its application involves a series of numerical test routines which are performed prior to the realization of the repository simulations. w methodology has been applied on two actual simulation cases. One is based on the Couplex-Gaz 1b exercise with variable mesh size, the other is the larger MAVL 3D model concerning the storage of intermediate-level long-lived radioactive waste. The results from testing show a good scalability of the Tough2-MP code under certain conditions. ractices for domain partitioning and processor load repartitioning have been derived. The results may be applicable, too, for efficiency enhancements of other domain-decomposition based parallel simulation software.