Issue No. 02 - April (1996 vol. 8)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/69.494165
<p><b>Abstract</b>—The research in materialization of derived data elements has dealt so far with the <b>if</b> issue; that is, the question whether to physically store derived data elements. In the active database area, there has been some research on the <b>how</b> issue. In this paper, we deal with the <b>when</b> issue, devising an optimization model to determine the optimal materialization strategy. The decision problem confronted by the optimization model is more complex than "<it>to materialize or not to materialize</it>." The decision problem deals with devising the <it>materialization strategy</it> that consists of a set of interdependent decisions about each derived data element. Each decision relates to two issues:</p><ip2><lc><li><p>Should the value of a derived data element be persistent?</p></li><li><p>What is the required level of consistency of a derived value with respect to its derivers?</p></li></lc></ip2><ip2>For each derived data element, the decision is based on both its local properties (complexity of derivation, update and retrieval frequencies, etc.) and its interdependencies with other derived values. The optimization model is based on a heuristic algorithm that finds a local optimum (which is a global optimum in many cases) in O(N<super>2</super>) and a monitor that obtains feedback about the actual database performance. This optimization model is general and is not specific to any data model. Our experimental results show that a predictor for the optimal solution cannot be obtained in any intuitive or analytic way, due to the complexity of the involved considerations; thus, there is no obvious way to achieve these results without using the optimization model. This fact is a strong motivation for applying such an optimization model. Our experimental results further indicate that the optimization model is useful in the sense that the system performance (with respect to the applications' goal function) is substantially improved compared to any universal materialization policy.</ip2>
Database tuning, materialization strategies, derived data, active databases, database optimization.
O. Etzion and D. Botzer, "Optimization of Materialization Strategies for Derived Data Elements," in IEEE Transactions on Knowledge & Data Engineering, vol. 8, no. , pp. 260-272, 1996.