Dynamically Evolvable Distributed Systems

Software systems must change over time. Changing business practices, the relentless advance of technology, and the demands of end users drive this evolution. The functionality required of applications inevitably changes in response to these factors. Consequently, in order to remain viable, applications must evolve to meet new requirements. Software component evolution is a major focus of effort in software engineering [6, 13]. The vast majority of commercial software is written in a few imperative languages, such as C++ or Java [1]. For these languages, software evolution is generally a slow, static process. Since any update requires stopping a program and overwriting all or part of it, incremental updates are often impractical, and major updates problematic. For a large class of critical applications, such as business transaction systems, telephone switching systems and emergency response systems, the interruption poses an unacceptable loss of availability. What is needed, then, is more support for applications that evolve during execution. Dynamic evolution provides a number of benefits in addition to easing upgrades to critical software. The Evolvable Systems Project (ESP) is developing techniques for safely and securely evolve distributed Java applications. Our current research involves (i) design and implementation of a type-safe model for dynamically modifying Java programs on a single host, (ii) application of the model for developing an adaptive security model, and (iii) extension of the single host model to a distributed environment. We briefly describe them below.