Abstract:
Existing estimation approaches for multi-dimensional databases often rely on the assumption that data distribution in a small region is uniform, which seldom holds in practice. Moreover, their applicability is limited to specific estimation tasks under certain distance metric. This paper develops the Power-method, a comprehensive technique applicable to a wide range of query optimization problems under various metrics. The Power-method eliminates the local uniformity assumption and is accurate even in scenarios where existing approaches completely fail. Furthermore, it performs estimation by evaluating only one simple formula with minimal computational overhead. Extensive experiments confirm that the Power-method outperforms previous techniques in terms of accuracy and applicability to various optimization scenarios.
