JaCaL : an implementation of Linda in Java

Abstract

Java is an object-oriented programming language with built-in features for creating distributed programs. A key feature-set that is missing, however, is'an easy-to-use, reliable, and scaleable tool for writing truly parallel programs. The Linda para\lel programming model defmes a client-server approach where concurrent execution requests are serviced and results are stored to a shared data repository called a Tuple Space. Tuples consist of heterogeneous coIIections of data of various types. An interpretation of the model aIIows the programmer to create active tuples, in which user-defined functions are automatically and transparently launched in concurrently executing processes distributed on remote workstations. This paper presents an implementation of Linda in Java called JaCaL, a library of classes and interfaces easily integrated into a client application. The Tuple Space has been implemented as a data store residing on a single machine, whilst being accessible by clients distributed on other machines. Clients can create active tuples, which cause processes on distributed machines to be transparently invoked to execute user-defined functions, employing a load-balanced worker process model. No pre-processor is required to parse the client program before compilation. The implementation allows a Java application developer to create more efficient coarse grained parallel programs with minimal effort. Performance measurements have been made and compared to those of another, similar implementation.