Adaptive Data Management.

At present, the art of database design is conceptualized in two dimensions. The classes of objects of interest to an organization (entities) are typically portrayed in a spacial arrangement with the potential associations (relationships) among them represented as connecting edges. It is the principal thesis of this work that there exists a third dimension to the art and practice of database design--the occurrence dimension. In this dimension, the notion of occurrences of data objects and the valid relationships among them is explicitly and effectively recognized. The work presented here examines the occurrence dimension at three discrete levels of the design process. Initially, at the abstract data modeling level, two new database abstractions are introduced. The selector and adaptive selector abstractions are shown to be sufficient to capture and represent certain semantic information requirements which frequently arise in a database design. Without them, these important requirements could not be expressed in the design. The second level of the design process is concerned with representing the occurrence dimension in a generic data model. For purposes of demonstration, the relational model of data is used. The concept of artificial relations is introduced to serve as surrogates for selectors and adaptive selectors. A data definition language facility is presented and the relational algebra is extended to include new operators to manipulate these artificial relations. It is shown that the artificial relations in conjunction with the new operators can reduce the complexity of certain queries as well as reduce the amount of effort in responding to them. Lastly, the implications of the occurrence dimension at the level of internal data modeling are considered. Physical storage structures are recommended for the implementation of selectors and adaptive selectors. Several proposals are then made for the adaptive management of a stored database. Most notable among these is a proposal for a new database buffer management policy. It is shown through simulation experiments that this policy has the potential to significantly reduce the number of buffer faults experienced in an operational database environment, particularly when adaptive selectors are employed.PhDBusiness administrationUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/159766/1/8402310.pd