Abstract Machines for Dynamic Computation

Abstract

In this thesis we address the challenges associated with the provision of dynamic software architectures. These are systems in which programs are constructed from separately compiled units with a facility for the replacement of these units at runtime. Typical examples of applications which will benefit from this dynamic approach are long-lived systems in which downtime is highly undesirable, for example, web-servers, database engines, and equipment controllers. In addition, dynamic software architectures are also gaining popularity with the recent advent of wide-area Internet applications, where it is often impractical to compile a program in its entirety or begin execution in a single step.

Our approach to dynamic software architectures differs from earlier attempts in that we guarantee the safety of the replacement operation. This is done by founding our techniques on the rigour of strong typing. In the first half of the thesis we take an existing static software architecture with strong typing facilities and modular program construction, namely the Standard ML platform, and equip it with facilities for separate-compilation and code-replacement of modules. The resulting dynamic software architecture, which we call Dynamic ML, ensures the safety of replacement through an effective use of state-of-the-art advances in the fields of types in compilation and abstract machines.

In the latter half of the thesis we extend Dynamic ML with a facility for distributed execution and adapt our code-replacement model accordingly. This will permit the construction of larger dynamic architectures, for example, across a distributed network of workstations. We also perform a mechanical verification of the distributed algorithm by model checking, to gain further confidence in the correctness of our approach. At the end of the thesis we outline an implementation of our techniques for the Java language, demonstrating the portability of our approach.