Though ATLAS has strong data servers, accessing files fon the local harddrive of the nodes is much faster, and given that you could read files in parallel accessing data from multiple other sources (nodes) via the network can be even faster.

Therefore we provide two different methods for distributing data over the nodes of ATLAS, depending on whether the targeted data set should be present on all the nodes to be accessed locally (DataClone) or should be accessed via the network from other nodes that only have a subset of the data each (DataSpray).