Often Knowledge Rep- Resentation Support for Data Archaeology. in Redeened Aggregation Adaptation Using Original View

it is easier to adapt a view if certain additional information is kept in the view. Such additional information includes keys of base relations, attributes involved in selection conditions, counts of the number of derivations of each tuple, additional aggregate functions beyond those requested, and identiiers indicating which subquery in a union each tuple came from. Depending on the type of anticipated change, the view can be deened to contain the appropriate additional information. Additionally , it can be beneecial to reserve some physical space in each record to allow in-place adaptation involving addition of attributes. We have derived tables of adaptation techniques (see GMR95] for a complete list) that can be used in three important ways. Firstly, the query optimizer can use the tables to nd the adaptation technique (and compute its cost estimate) given the properties of the current schema visa -vis the assumptions stated in the table. Secondly, a database administrator or user can use the tables to see what assumptions would need to be satissed in order to make view adaptation possible at the most eecient level, and deene the view accordingly. Thirdly, the database administrator can interact with the query optimizer to build appropriate access methods and indexes on the base relations and on the materialized views, in order to facilitate eecient adaptation. The main contributions of this paper are (a) the derivation of a comprehensive set of view adaptation techniques, (b) the smooth integration of such techniques into the framework of current relational database systems using existing optimization technology , and (c) the identiication of guidelines that can be provided to users and database administrators in order to facilitate view adaptation. Acknowledgments We thank Arun Netravali for pointing out the importance of redeenition to data visualization, and Shaul Dar and Tom Funkhouser for discussions of the relationship between view maintenance and data visualization. eeciently using the rst observation above. In that case, the new view V 0 is V EXCEPT V 2 and we can make use of the old extension of V. In the general case, there is another possibility that the optimizer can consider for computing V 0. Suppose that V 2 changes with both V + 2 and V ? 2 nonempty. The new answer is V EXCEPT V + 2 UNION U where U is V 1 \ V ? 2. While we probably have not materialized V 1 , …