Memory management strategies for decision support systems

Memory is a particularly vital resource for decision support systems. This is because effective decision support frequently requires the analysis of large amounts of data and results in the production of equally large (and often larger) query results. In most situations, the amount of main memory available in the system is not enough to meet the response time goals of interactive analysis. This thesis deals with ways of managing memory more effectively so that we can better meet the demands imposed on decision support systems.; The first part of the thesis concerns memory management for relational databases running complex decision support queries. We discuss several algorithms for distributing available memory among concurrently running operators of a query in ways that satisfy operator requirements and scheduling constraints. One of these algorithms, based on linear programming, is also optimal in that it is guaranteed to produce the best query response time.; The second part of the thesis discusses memory management for data mining applications. Most current data mining algorithms have rather long response times that render them unsuitable for interactive use. This thesis incorporates the first proposal for using caching in the data mining context. We show how memory can be used to build a knowledge cache that can cut down the time required for association rule mining by several orders of magnitude.; The final part of this thesis aims at broadening the scope of data mining queries and extending the benefits of caching to this more powerful query model. We view data mining as a special kind of aggregation operation that can be combined with selections and group-bys to yield a richer set of analysis tools. To make this system interactive, we describe the design of a chunked cache that employs a specialized semantic caching scheme to store the results of association mining queries in a uniform and scalable fashion. We believe that this idea will make a significant contribution towards bringing OLAP and various data mining techniques together under a common, unifying framework that can be used to build extensible and scalable decision support systems.