Abstract | The development of Java has seen increasing attention as the most popular platform for distributed computing. However, despite Java's advantages in the area of portability and rapid prototyping, its efficiency is unavoidably compromised through its commitment to portability. In this paper we present performance analysis and comparisons of evaluation results for both Java and C/Fortran on three different message-passing parallel platforms-a shared memory multi-processor (Sun E4000), a Linux cluster, and a distributed memory computer (IBM SP-2). The NAS Embarrassingly Parallel and Integer Sort benchmarks were selected for this evaluation. Both the original Fortran/C codes and Java versions of these two kernels were used for obtaining the performance measurements as part of our project. The evaluation results demonstrate the feasibility of message-passing computing with Java on a wide range of computer platforms. Depending on the system and the software components installed, significant impact on the message-passing performance will have the efficiency of the native MPI library and the version of the Java platform. |
---|