A mixed-language programming methodology for high performance Java']Java computing

Java is quickly becoming the most popular platform for distributed computing. However, its performance is still subject to concerns in comparison to other programming languages such as C and Fortran. As a consequence, programmers of high-performance applications are usually reluctant to embrace Java as an alternative language in their work. This article introduces the Java-to-C Interface (JCI) tool which generates automatically the wrapper code interfacing existing scientific libraries to Java. Thus, facilitating rapid development and software reuse, the JCI tool provides application programmers with immediate accessibility to existing scientific libraries from Java. While beneficial to the software developer, the additional advantages of mixed-language programming in terms of application performance are addressed in detail within the context of this work. We also present analysis and comparisons of evaluation results for mixed-language codes in Java and C/Fortran on a high-performance distributed memory computer (IBM SP-2). The NAS Embarrassingly Parallel and Integer Sort benchmarks as well as the Matrix Multiplication kernel from the PARKBENCH suite were selected for our experiments. The evaluation results demonstrate the feasibility and efficiency of our mixed-language programming methodology with Java.