| In this dissertation, we study the problem of supporting schema evolution in snapshot databases and historical databases. Schema evolution requires several error-prone and time-consuming tasks, including migration of stored data, rewriting of applications, and in historical databases, management of heterogeneous data. Thus, our primary goal is to relieve users from the burden of performing the tasks manually by having the system to perform them automatically.;The outline and the main contributions of this dissertation are as follows. First, we design schema modification operators (SMOs), for high-level specification of schema changes. Then, based on SMOs, we develop a workbench system called PRISM , where users can evaluate the effect of schema changes ("what-if" scenario) and (semi-)automatically execute data migration and application query translation. For query translation, the system computes for the user possible inverse schema mappings and performs automatic query translation according to the inverse selected by the user.;Evolution of schemas also affect historical data management: repeated schema changes lead to the accumulation of successive schema versions in history, since historical data must be archived according to the schema version under which they were originally created, as to assure faithful preservation of the database history. To write a historical query on such databases, users have to refer to every schema version containing relevant parts of the historical database. Since this is too cumbersome for typical application, we let users write a single historical query on the current schema version, as if all historical data were stored under that version. Thus our PRIMA system uses efficient rewriting techniques to map the queries formulated against the current schema onto the relevant schema versions.;We also study the problem of efficient support for transaction-time databases with evolving schemas and propose novel optimization techniques for temporal queries. In particular, we propose an efficient temporal coalescing technique called CNesT, which outperforms the best coalescing algorithm to date by almost two orders of magnitude. |
