Abstract:
Data warehouses (DW) that store enormous quantities of data put a major challenge in what concerns performance and scalability, as users request instant answers to their queries. Traditional solutions rely on very expensive architectures and structures for speedup and scale-up. The Summary Warehouse (SW) is an inexpensive solution that has the potential to deliver very fast approximate answers to aggregate queries using only general-purpose sampling summaries. Although summaries are expected to be extremely fast, some analysis requires larger summaries to estimate individual group results, compromising the speedup advantage. This is the accuracy/speedup (A/S) tradeoff. In this paper we propose the " Distributed Set-of-Summaries for Quality of Service " (DSQoS) that solves the A/S issue by optimizing the accuracy and response time for each query pattern in order to guarantee a desired Quality of Service (QoS). This QoS is defined in terms of response time and accuracy bounds. The strategy determines the required summary size to guarantee the accuracy targets and then dynamically select a set of summaries, distributed in various nodes, which can ensure the QoS constraints (time and accuracy). The strategy presents enormous possibilities since each node can contain summaries with different sizes, depending on the node characteristics, and can dynamically be added and removed from the system. We discuss the design of the approach and the strategies used to process queries. In the experimental section we show how the approach is able to deliver almost instant and accurate answers without employing expensive architectures, which would be impossible using other strategies.
