A scalable spatial skyline evaluation system utilizing parallel independent region groups

Abstract

This research presents two parallel solutions to efficiently address spatial skyline queries. First, we propose a novel concept called independent regions for parallelizing the process of spatial skyline evaluation. Spatial skyline candidates in an independent region do not depend on any data point in other independent regions. Then, we propose a GPU-based solution. We use multi-level independent region group-based parallel filter to support efficient multi-threading spatial skyline non-candidate elimination. Beyond that, we propose comparable region to accelerate non-candidate elimination in each independent region. Secondly, we propose a MapReduce-based solution. We generate the convex hull of query points in the first MapReduce phase. In the second phase, we calculate independent regions based on the input data points and the convex hull of the query points. With the independent regions, spatial skylines are evaluated in parallel in the third phase, in which data points are partitioned by their associated independent regions in map functions, and spatial skyline candidates are calculated by reduce functions. The results of the spatial skyline queries are the union of outputs from the reduce functions. Our experimental results show that GPU multi-threading scheme is very efficient on small-scale input datasets. On the contrary, MapReduce scheme performs very well on large-scale input datasets.