Data warehouses are databases that are used for analytical tasks. The common assumption is that data warehouses separate data input out of operational operational databases from output to so-called data marts, which provide data for particular on-line analytical processes (OLAP). In this article it is assumed that the data warehouse is distributed of a network and that not all OLAP tasks are supported at all network locations. This implies that only some fagments need to be available locally. A heuristic cost optimisation approach is suggested for setting up local data warehouse fragments. The underlying cost model integrates query and maintenance costs.