of thesis entitled “Distributed Java Virtual Machine with Thread Migration” submitted by Wenzhang Zhu for the degree of Doctor of Philosophy at the University of Hong Kong in August 2004 Recent advances in better Java class libraries and Just-in-Time (JIT) compilation techniques have greatly improved the performance of Java to match that of C/C++. To fully exploit Java’s multithreading feature on clusters, an attractive goal is to extend the Java Virtual Machine (JVM) to be “cluster-aware” so that a group of JVMs running on distributed nodes can work together as a single, more powerful JVM to support true parallel execution of a multithreaded Java application. In a cluster-aware JVM, the Java threads created within one program can run on different cluster nodes to achieve a higher degree of execution parallelism. The distributed system would provide all the virtual machine services to Java programs, and should be fully compliant with the Java language specification. We refer to such a distributed system as a Distributed JVM (DJVM). In our study, we solve the problem of transparent Java thread migration in a JIT-enabled DJVM. The problem of thread migration is to suspend a running thread in a multithreaded application on one node, and resume its execution on a target node, such that the thread continues to execute and communicate with other threads of the same application on different nodes. In this act, the challenges lie in dealing with different types of pointers associated with the context of the migratory thread. The JIT compilation in JVM has made the native code of Java methods be dynamically placed, thus the method pointer relocation becomes difficult. Shared Java objects