INTELLIGENT DATABASE CACHING THROUGH THE USE OF PAGE-ANSWERS AND PAGE-TRACES

In this paper a new method to improve the utilization of main memory systems is presented. The new method is based on prestoring in main memory a number of query answers, each evaluated out of a single memory page. To this end, the ideas of page-answers and page-traces are formally described and their properties analyzed. The query model used here allows for selection, projection, join, recursive queries as well as arbitrary combinations. We also show how to apply the approach under update traffic. This concept is especially useful in managing the main memories of an important class of applications. This class includes the evaluation of triggers and alerters, performance improvement of rule-based systems, integrity constraint checking, and materialized views. These applications are characterized by the existence at compile time of a predetermined set of queries, by a slow but persistent update traffic, and by their need to repetitively reevaluate the query set. The new approach represents a new type of intelligent database caching, which contrasts with traditional caching primarily in that the cache elements are derived data and as a consequence, they overlap arbitrarily and do not have a fixed length. The contents of the main memory cache are selected based on the data distribution within the database, the set of fixed queries to preprocess, and the paging characteristics. Page-answers and page-traces are used as the smallest indivisible units in the cache. An efficient heuristic. to select a near optimal set of page-answers and page-traces to populate the main memory has been developed, implemented, and tested. Finally, quantitative measurements of performance benefits are reported.