The current technology allows scientists of several domains to obtain more precise and large data. In the same time, computing architectures evolve too, by providing even more computing resources, with more powerful machines and the pooling of them. In this thesis, in a first time we propose to study the problem of the mapping of asynchronous iterative applications tasks into heterogeneous and volatile environments. Our solution allows also to overcome the heterogeneity of host machines while offering an easier implementation of policies for fault tolerance. The experiments we have conducted are encouraging ad show that there is real potential for the use of such a platform for running scientific applications.
