Scienti c Computation With

JavaSpaces provides a simple yet expressive mechanism for distributed computing with commodity technology. We discuss the suit-ability of JavaSpaces for implementing diierent classes of concurrent computations based on low-level metrics (null messaging and array I/O), and present performance results for several parametric algorithms. We found that although ineecient for communication intensive problems, JavaSpaces yields good speedups for parametric experiments, relative to both sequential Java and C. We also outline a dynamic native compilation technique, which for short, compute-intensive codes further boosts performance without compromising Java portability or extensive algorithm recoding. Discussion and empirical results are presented in the context of our public benchmark suite.