Grid computing is nowadays an established technology in elds such as High Energy Physics and Biomedicine, oering an alternative to traditional HPC for several problems; however, it is still an emerging discipline for the climate community and only a few climate applications have been adapted to the Grid to solve particular problems. In this paper we present an up-to-date description of the advantages and limitations of the Grid for climate applications (in particular global circulation models), analyzing the requirements and the new challenges posed to the Grid. In particular, we focus on production-like problems such as sensitivity analysis or ensemble prediction, where a single model is run several times with dierent parameters, forcing and/or initial conditions. As an illustrative example, we consider the Community Atmospheric Model (CAM) and analyse the advantages and shortcomings of the Grid to perform a sensitivity study of precipitation with SST perturbations in El Niño area, reporting the results obtained with traditional (local cluster) and Grid infrastructures.
We conclude that new specic middleware (execution work managers) are needed to meet the particular requirements of climate applications (long simulations, check-pointing, etc.). This requires the side-by-side collaboration of IT and climate groups to deploy fully ported applications, such as the CAM for Grid (CAM4G) introduced in this paper.