In this report, we tackle the problem of scheduling an Ocean-Atmosphere application in an heterogeneous environment. The application is used for long term climate forecast. In this context, we analyzed the execution of an experiment. An experiment is composed of several identical simulations composed of parallel tasks. On homogeneous platforms, we propose a heuristic and its optimizations, all based on the same idea: we divide the processors into disjoint sets, each group executing parallel tasks. On heterogeneous platforms the algorithm presented is applied on subsets of simulations. The computation of the subsets is done greedily and aims at minimizing the execution time by sending each subset on a cluster. We performed experiments on the french research grid \emph{Grid'5000} which exhibited some technical difficulties. We also present some modifications done to the heuristics to minimize the impact of these technical difficulties. Our simulations are then validated by experimentations.