The fast growth of scientific and business data has resulted in the evolution of the cloud computing. The MapReduce parallel programming model is a new framework favoring conception of algorithms for cloud computing. Such framework favors processing problems across huge datasets using a large number of computers. Hadoop is an implementation of MapReduce framework that becomes one of the most interesting approaches for cloud computing. Our contribution consists in investigating how the MapReduce framework can create new trend in conception of operational research algorithms and could define a new methodology to adapt algorithms to Hadoop. Our investigations are directed on the shortest path problem on large-scale real-road networks. The proposed algorithm is tested on a graph modeling French road network using the OpenStreetMap data. The computational results push us into considering that Hadoop offers a promising approach for problems where data are so large that numerous memory problems and excessive computational time could arise using a classical resolution scheme.