The Design and Implementation of a Java Virtual Machine on a Cluster of Workstations

The Design and Implementation of a Java Virtual Machine on a Cluster of Workstations by Carlos Daniel Cavanna Master of Applied Science Edward S. Rogers Sr. Department of Electrical and Computer Engineering University of Toronto 2003 We present the design, implementation, and evaluation of a Java Virtual Machine (JVM) on a cluster of workstations, which supports shared memory in software. More specifically, we first extend the Jupiter JVM infrastructure to support multithreading on Symmetric Multiprocessors (SMPs). We then extend this multithreaded Jupiter to deal with memory consistency, private memory allocation, and the limitations on the use of some resources on a 16-processor cluster of PCs interconnected by a Myrinet network; which supports Shared Virtual Memory and a pthreads interface. Our experimental evaluation is performed using standard benchmarks. On a 4-processor SMP, it indicates that the performance of the multithreaded version of Jupiter is midway between a näıve and a state-of-the-art implementations of a Java interpreter. Our evaluation on the cluster demonstrates that good speedup is obtained.