题目: Efficient Skyline Computation in MapReduce

时间:2015年12月15日上午10:00

地点:东五楼210

Abstract:

Skyline queries are useful for finding interesting tuples from a large data set according to multiple criteria. The sizes of data sets are constantly increasing and the architecture of back-ends are switching from single-node environments to non-conventional paradigms like MapReduce. Despite the usefulness of skyline queries, existing works on skyline computation in MapReduce do not take full advantage of parallelism but still run significant parts serially. In this paper, we propose a novel approach to compute skylines efficiently in MapReduce. We design a grid partitioning scheme to divide the data space into partitions, and employ a bitstring to represent the partitions. The bitstring is efficiently obtained in MapReduce, and it clearly helps prune partitions (and tuples) that cannot have skyline tuples. Based on the grid partitioning, we propose two MapReduce algorithms to compute skylines. Both algorithms utilize the bitstring and distribute the original tuples to multiple mappers and make use of them to compute local skylines in parallel. In particular, MapReduce Grid Partitioning based Single-Reducer Skyline Computation (MR-GPSRS) employs a single reducer to assemble the local skylines appropriately to compute the global skyline. In contrast, MapReduce Grid Partitioning based Multiple Reducer Skyline Computation (MR-GPMRS) further divides local skylines and distributes them to multiple reducers that compute the global skyline in an independent and parallel manner. The proposed algorithms are evaluated through extensive experiments, and the results show that MR-GPMRS significantly outperforms the alternatives in various settings.

Bio:

Yongluan Zhou is an Associate Professor in the Department of Mathematics and Computer Science at the University of Southern Denmark (SDU). Before joining SDU, he worked at EPFL as a postdoctoral researcher. He obtained his PhD in Computer Science at the National University of Singapore. His research interests span across database systems and distributed systems, especially in the area of query processing and optimization. Yongluan has published over 50 research articles in international journals and conference proceedings. He has also served on the program committees of various international conferences including ICDE, VLDB, CIKM, SSDBM and MDM.