Evolution Transparency for Distributed Service Types

Name: Twittie Senivongse Degree: PhD in Computer Science Evolution Transparency for Distributed Service Types Abstract Large software systems are never static as they exist in an environment that is subject to constant changes in both functionality and technology. Managing this evolution is one of the major challenges in the engineering of large-scale software systems. When a distributed service evolves its interface, other parties in its environment who need to continue using the service will themselves have to evolve correspondingly if the evolved service, although functionally compatible, is not type-compatible with the original one. The autonomous and decentralised nature of distributed components makes such an assumption impractical since the service provider and client systems may not agree on the evolution. It may not even be possible to track down and alter all client programs which are distributed over the network. The best way to tackle this problem is to provide 'evolution transparency' to give the affected components the illusion that the service does not change. This thesis presents an RM-ODP-based model that hides from clients details of changes occuring to distributed service types over time. By allowing type versioning , the model supports program compatibility by enabling a client program of one type version to access a service instance of another functionally compatible type version, even though the versions are not considered compatible by the supporting type system. The model manages a cross-version binding and maintains semantic information which is used to transform the client's invocation to the format recognised by the accessed service object. With a prototype implementation on ANSAware together with an analysis of its mechanism, the evolution transparency support proves useful; clients are given the flexibility to defer their own evolution, and type substitutability is also extended from syntactic to functional compatibility.