PCJ - New Approach for Parallel Computations in Java

摘要

In this paper we present PCJ - a new library for parallel computations in Java. The PCJ library implements partitioned global address space approach. It hides communication details and therefore it is easy to use and allows for fast development of parallel programs. With the PCJ user can focus on implementation of the algorithm rather than on thread or network programming. The design details with examples of usage for basic operations are described. We also present evaluation of the performance of the PCJ communication on the state of art hardware such as cluster with gigabit interconnect. The results show good performance and scalability when compared to native MPI implementations.