A basic prerequisite for parallel programming is a good communication API. The recent interest in using Java for scienti c and engineering application has led to several international e orts to produce a message passing interface to support parallel computation. In this paper we describe and then discuss the syntax, functionality and performance of one such interface, mpiJava, an object-oriented Java interface to MPI. We first discuss the design of the mpiJava API and the issues associated with its development. We then move on to briefly outline the steps necessary to 'port' mpiJava onto a range of operating systems, including Windows NT, Linux and Solaris. In the second part of the paper we present and then discuss some performance measurements made of communications bandwidth and latency to compare mpiJava on these systems. Finally, we summarise our experiences and then briefly mention work that we plan to undertake.

Several Java bindings of the Message Passing Interface standard, MPI, have been developed recently. Message buffers have usually been restricted to arrays with elements of primitive type. We discuss use of the Java object serialization model for marshalling general communication data in MPI. This approach is compared with a Java transcription of the standard MPI derived datatype mechanism. We describe an implementation of the mpiJava interface to MPI incorporating automatic object serialization. Benchmark results show that the current JDK implementation of serialization is (not unexpectedly) probably not fast enough for high performance applications. Means of solving this problem are discussed.

We sketch a proposed reference implementation for MPJ, the Java Grande Forum's MPI-like message-passing API [9, 3]. The proposal relies heavily on RMI and Jini for nding computational resources, creating slave processes, and handling failures. User-level communication is implemented eciently directly on top of Java sockets. Current address: University of Portsmouth, UK Contents 1 Introduction 3 2 Some design decisions 4 3 Overview of the Architecture 7 4 Process creation and monitoring 8 4.1 The MPJ daemon . . . . . . . . . . . . . . . . . . . . . . . . . . 9 4.2 The MPJ slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.3 The MPJ client . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 4.4 Handling MPJ aborts|Jini events . . . . . . . . . . . . . . . . . 12 4.5 Other failures|Jini leasing . . . . . . . . . . . . . . . . . . . . . 12 5 Sketch of a \Device-Level" API for MPJ 13 5.1 Minimal API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...

In this paper we sketch out a proposed reference implementation for message passing in Java (MPJ), an MPI-like API from the Message-Passing Working Group of the Java Grande Forum [1,2]. The proposal relies heavily on RMI and Jini for finding computational resources, creating slave processes, and handling failures. User-level communication is implemented efficiently directly on top of Java sockets.

this paper is organized as follows. In Section 2, we describe the basic support for concurrent computing/programming provided by Java, as well as some other features that are relevant to understand the proposals described here. Readers that are familiar with Java's concurrency features can skip this section. In Section 3, we describe the parameters chosen to classify each of the selected projects. In Section 4, we describe the Java environments and mechanisms for supporting high-performance network-based computing that were included in this survey. Section 5 presents a classi cation of these systems, based on the parameters described in Section 3. Section 6 concludes this work