Fig. 1

Object structure. Here the main object composition structure of d2o is shown. distributed data objects are composed objects, cf. "Composed object" section. All tasks that need information related to the distribution strategy are outsourced to a distributor. In this figure, three distributed data objects are shown where obj1 and obj2 share the same distributor. This is possible because they are essentially identical: they have the same global shape, datatype, and distribution strategy. Since it is expensive to instantiate new distributors, the distributed data objects get their instance from the distributor_factory that takes care of caching those distributors that have already been created. Furthermore, we illustrate the d2o_librarian that keeps weak references to the individual distributed data objects and assigns a unique cluster-wide identifier to them: the index