This article deals with the optimization problem of the execution of massive analytical processing on distributed data warehouses (ED) where the number of simultaneous queries is counted by thousands. While taking as a starting point the techniques of optimization used in the undistributed context, we propose a new strategy of selection and storage of materialized views (MV) on distributed file system; then we handle the processing of the decisional queries workload by using the MV. Our approach plays a role of mediator between the users and the data warehouse to propose a better execution plan to their queries. The first results make us believe that in a distributed environment, our approach improves more than 50% the execution cost of a request compared to the system provided by default.