Modelling grid architecture.

This thesis evaluates software engineering methods, especially event modelling of distributed systems architecture, by applying them to specific data-grid projects. Other methods evaluated include requirements' analysis, formal architectural definition and discrete event simulation. A novel technique for matching architectural styles to requirements is introduced. Data-grids are a new class of networked information systems arising from e-science, itself an emergent method for computer-based collaborative research in the physical sciences. The tools used in general grid systems, which federate distributed resources, are reviewed, showing that they do not clearly guide architecture. The data-grid projects, which join heterogeneous data stores specifically, put required qualities at risk. Such risk of failure is mitigated in the EGSO and AstroGrid solar physics data-grid projects' designs by modelling. Design errors are trapped by rapidly encoding and evaluating informal concepts, architecture, component interaction and objects. The success of software engineering modelling techniques depends on the models' accuracy, ability to demonstrate the required properties, and clarity (so project managers and developers can act on findings). The novel formal event modelling language chosen, FSP, meets these criteria at the diverse early lifecycle stages (unlike some techniques trialled). Models permit very early testing, finding hidden complexity, gaps in designed protocols and risks of unreliability. However, simulation is shown to be more suitable for evaluating qualities like scalability, which emerge when there are many component instances. Design patterns (which may be reused in other data-grids to resolve commonly encountered challenges) are exposed in these models. A method for generating useful models rapidly, introducing the strength of iterative lifecycles to sequential projects, also arises. Despite reported resistance to innovation in industry, the software engineering techniques demonstrated may benefit commercial information systems too