An Efficient Block Sampling Strategy for Online Aggregation in the Cloud

Abstract

As the development of social network, mobile Internet, etc., an increasing amount of data are being generated, which beyond the processing ability of traditional data management tools. In many real-life applications, users can accept approximate answers accompanied by accuracy guarantees. One of the most commonly used approaches is online aggregation. Online aggregation responds aggregation queries against the random samples and refines the result as more samples are received. In the era of big data, more and more data analysis applications are migrated to the cloud, so online aggregation in the cloud has also attracted more attention. There can be a huge difference between the number of tuples in each group when dealing with group-by queries. As a result, answers of online aggregation based on uniform random sampling can result in poor accuracy for groups with very few tuples. Data in the cloud are usually organized into blocks and this data organization makes sampling more complex. In this paper, we propose an efficient block sampling which can exactly reflect the importance of different blocks for answering group-by queries. We implement our methods in a cloud online aggregation system called COLA and the experimental results demonstrate our method can get results with higher accuracy.