Distributed recursive sets: programmability and effectiveness for data intensive applications

This paper presents a concurrent object model based on distributed recursive sets for data intensive applications that use complex, recursive data layouts. The set abstraction is used to represent irregular (recursive) data layouts. The distributed set abstraction is used to transparently distribute large data across multiple address spaces. We effectively map data to processors by using information about data accesses to preserve locality. This results in a consistency scheme with small communication requirements. The programming model is easy to use since the system automatically does the data distribution, mapping and consistency.