A system of recommendations is a specific type of filter of information that helps the user to select such articles of his (her, your) interest as movies, musical, web pages, magazines, books, etc. Nowadays, the web sites that give these services need that the great quantity of information got for all the implicit or explicit actions of million users on million articles, is tried in a rapid way and with the minor possible infrastructure, this in order to obtain rapid and better indexes of useful preferences and to minor cost The Present work has as aim to present two alternatives of processing recommendation of musical articles based on the implicit preferences of the users and using a model of massive and scalable programming inside Hadoop's framework as a system of the execution of tasks in parallel and tolerantly to failures.