The BofS solution to limitations of approximate summaries

Data warehouses are of crucial importance to decision-making in competitive organizations. The fact that they store enormous quantities of data is a challenge in what concerns performance and scalability, as users request instant answers. None of the traditional performance strategies is sufficiently good to make complex aggregation queries take only minutes or seconds. The summary warehouse (SW) achieves such a speedup by storing only general-purpose sampling summaries well-fit for aggregated exploration analysis. The major limitation of SWs results from the tradeoff between accuracy and speed: smaller, faster summaries cannot answer less-aggregated queries. In this paper we present the Bag-of-Summaries approach (BofS) designed to deal effectively with this problem: BofS maintains a set of summaries with varied sizes and chooses the right one to answer a query with the desired accuracy and best possible speedup, based on query granularity considerations. We also present experimental results that show the advantage of BofS.