Updatable and evolvable transforms for virtual databases

Applications typically have some local understanding of a database schema, a virtual database that may differ significantly from the actual schema of the data where it is stored. Application engineers often support a virtual database using custom-built middleware because the available solutions, including updatable views, are unable to express necessary capabilities. We propose an alternative means of mapping a virtual database to a physical database that guarantees they remain synchronized under data or schema updates against the virtual schema. One constructs a mapping by composing channel transformations (CTs) that encapsulate atomic transformations --- including complex transformations such as pivoting --- with known updatability properties. Applications, query interfaces, and any other services can behave as if the virtual database is the implemented schema. We describe how CTs translate queries, DML, and DDL, and the properties that are necessary for such translation to be correct. We describe two example CTs in detail, and evaluate an implementation of channels for completeness and performance.