Implementation Aspects of a Persistent DSM Operating System in Java

摘要

The Java trademark encompasses the Java Virtual Machine (JVM), the Java language itself, and a large continuously growing class library. Beyond the development of applets in the context of the World Wide Web, more and more people use Java for large scaled standalone applications. This user group points out the performance disadvantage of the Java environment in comparison to sophisticated C/C++ compilers. As a result of this highly optimizing Java compilers begin to emerge which compile Java to machine code and abandon JVM-platform independence. The Plurix project goes one step further developing a native PC Operating System (OS) in Java. The central abstraction of the Plurix OS is a persistent Distributed Shared Memory (DSM). Our Plurix Java Compiler (PJC) translates Java sources into Intel protected mode code. It is itself written in Java and after boostrapping it will become an integral part of the Plurix OS. In this paper we shortly review the persistent DSM environment of Plurix and give an overview on the architecture of PJC. We present how the basic runtime structures are modeled in Java and how PJC is built on top of it. The implementation of the Java language in the persistent Plurix DSM reveals interesting semantic issues. We suggest extending initialization rules for classes and an additional attribute clarifying semantic ambiguities of static variables. Finally, we present how a generalized type equivalence check scheme enhances flexibility in the Plurix DHS.