Parallel k-dominant skyline queries in high-dimensional datasets

Abstract

The skyline operator has been used to select preference points in many applications. The previously proposed k-dominant skyline, which relaxes the idea of dominance, reduces the number of skyline points in high-dimensional datasets. However, retrieving both skyline and k-dominant skyline points in high-dimensional datasets are computationally expensive. In this paper, we aim to consider all attributes simultaneously when retrieving skylines and k-dominant skyline points using a new data representation. Moreover, we propose a parallel k-dominant skyline algorithm, which obtains efficiency by exploring data parallelism. In the proposed algorithm, we introduce a data representation that significantly reduces the number of verifications between points and the computation of verifications. We implement the proposed algorithm on GPU frameworks, which are designed to perform data-parallel computation. For skyline queries, the experimental results show that the proposed algorithm outperforms the state-of-the-art GPU-based algorithms. In high dimensional space, the proposed algorithm is up to 10 times faster than the state-of-the-art GPU-based algorithm for skyline queries. We also evaluate and discuss the performance of the proposed algorithm for k-dominant skyline queries. With the proposed data representation, each point is averagely checked with less than 5% of points in k-dominant skyline queries, and 95% of verifications take only two comparisons.