The usual approach to dealing with imperfections in data is to attempt to eliminate them. However, the nature of modern systems means this is often futile. This paper describes an approach which permits applications to operate notwithstanding inconsistent data. Instead of attempting to extract a single, correct view of the world from its data, a DataWarp application constructs a collection of interpretations. It adopts one of these and continues work. Since it acts on assumptions, the DataWarp application considers its recent work to be provisional, expecting eventually most of these actions will become definitive. Should the application decide to adopt an alternative data view, it may then need to void provisional actions before resuming work. We describe the DataWarp architecture, discuss its implementation and describe an experiment in which a DataWarp application in an environment containing inconsistent data achieves better results than its conventional counterpart.
展开▼
