Java Communications for Large-Scale Parallel Computing

Java has many features of interest to developers of large-scale parallel applications. At the same time, there are currently several barriers to the effective use of Java in this area. In this article we present part of the results and proposed solutions to these problems. In particular, we report about the current status of the organized collaborations within the Java Grande Forum in the area of Message Passing for Java (MPJ) and faster Remote Method Invocation (RMI). An outline of the current MPJ specification is given along with a discussion of several open issues and performance results on different platforms - Linux cluster, IBM SP-2, and Sun E4000. These “proof-of-concept” results are quite encouraging for future developments and efforts in this area. We also demonstrate that a much faster drop-in RMI and an efficient serialization can be designed and implemented in pure Java. Our benchmark results show that this better serialization and improved RMI design and implementation save more than 50% of the runtime in comparison to the standard implementations available at the moment. Our results demonstrate that fast parallel and distributed computing in Java is indeed possible.