A Customizable Implementation of RMI for High Performance Computing

This paper describes an implementation of Java’s Remote Method Invocation (RMI) that is designed to run on top of the Globus high performance computing protocol. The primary contribution of this work is to illustrate how the object serialization mechanism used by RMI can be extended so that it becomes more configurable. This allows the implementation of object serialization protocols that are more efficient than the defaults or that are compatible with other distributed object models like HPC++, which is based on C++. Both issues are important when RMI is to be used in scientific computing.