The management of semantic multidimensional data models plays an important role during the phases of development and maintenance of data warehouse systems. Unfortunately, this is not done with the necessary stress by now. Reasons might be seen in the plethora of semantic notations or the insufficient tool support for multidimensional modeling. The paper on hand provides experiences gained within a project with an industry partner of the telecommunications industry. Their problem is a very huge data warehouse with more than 400 data cubes and several hundred key performance indicators. We developed a repository-based solution for managing the semantic data models. Our lessons learned show that especially for very large data models there has to be a repository based solution as well as a clear concept on how to break them up into their component pars. The aim of our principles is to increase the understandability as well as the maintainability of semantic multidimensional data models.