Skip Navigation CECS Home | ANU Home | Search ANU | Search FEIT | Feedback

Towards High-performance and Fault-tolerant Distributed Java Implementations.

Approach

Our approach is to use the newly-released Jikes RVM as a basis for a distributed JVM. This is due to its use of advanced IT (Just-In-Time) translation for high performance. Its implementation, largely based on Java itself, makes it also suitable for such an extension.

As an alternative to using a a general Distributed Shared Memory implementation, a more specialized scheme for distribution, which can better take advantage of Java's semantics, will be investigated.

The target platform is the ANU Beowulf Cluster. This contemporary cluster has 96 dual Pentium III nodes connected using a Fast Ethernet switch; it is thus ideal to demonstrate scalability of a dJVM on a large, COTS cluster.