Abstract

In a typical database application, it is commonly assumed that user information requirements can only be satisfied by the most current data. The desirable attributes of infolmation often include being up-to-date, timeliness, and accuracy, with the implicit assumption that without these attributes a response to a query has little or no value. This assumption is challenged in this work. We consider the tradeoff between the cost of incomplete information, due to the use of stale data, and the incremental cost of providing a current answer. We propose that the database system be extended to include a data cache, in which copies of frequently needed data will be kept. Objects in the cache are not updated as the database changes, but rather are refreshed whenever the cost of using stale data exceeds some prespecified level. We also discuss alternative refresh policies and cache search schemes.

Share

COinS