Abstract
This paper presents a high-performance distributed shared memory system called VODCA, which supports a novel view-oriented parallel programming on cluster computers. One advantage of view-oriented parallel programming is that it allows the programmer to participate in performance optimization through wise partitioning of the shared data into views. Another advantage of this programming style is that it enables the underlying distributed shared memory system to optimize consistency maintenance. VODCA implements a view-based consistency model and uses an efficient view-oriented update protocol with Integrated Diff to maintain consistency of a view. Important implementation details of VODCA are described in this paper. Experimental results demonstrate that VODCA performs very well and its performance is comparable with MPI (message passing interface) systems
