Gaspar Data-Centric Framework

Abstract

This paper presents the Gaspar data-centric framework to develop high performance parallel applications in Java based on a map pattern of computation. The framework provides an efficient data Application Programming Interface(API) that supports data tiling and flexible data layout. Data tiling and layout control enables the improvement of data locality, which is essential to foster application scalability in modern multi-core systems. This paper shows that the framework performance in modern multi-core systems is comparable with pure Java code.