APSkyline: Improved Skyline Computation for Multicore Architectures

Abstract

Abstract. The trend towards in-memory analytics and CPUs with an increasing number of cores calls for new algorithms that can efficiently utilize the available resources. This need is particularly evident in the case of CPU-intensive query operators. One example of such a query with applicability in data analytics is the skyline query. In this paper, we present APSkyline, a new approach for multicore skyline query processing, which adheres to the partition-execute-merge framework. Contrary to existing research, we focus on the partitioning phase to achieve significant performance gains, an issue largely overlooked in previous work in multicore processing. In particular, APSkyline employs an angle-based partitioning approach, which increases the degree of pruning that can be achieved in the execute phase, thus significantly reducing the number of candidate points that need to be checked in the final merging phase. APSkyline is extremely efficient for hard cases of skyline processing, as in the cases of datasets with large skyline result sets, where it is meaningful to exploit multicore processing.

High Parallel Skyline Computation over Low-Cardinality Domains

“…Afterwards the local Skyline results have to be merged. Liknes et al [7] present the APSkyline algorithm for efficient multicore computation of Skyline sets. They focus on the partitioning of the data and use the angle-based partitioning from [18] to reduce the number of candidate points that need to be checked in the final merging phase.…”

“…This saves memory and runtime. [13] suggests a grid-based partitioning, [18,7] uses an angle-based partitioning, and [19] uses hyperplane projections to divide the dataset into disjoint sets. The lattice algorithms are independent from the partitioning, because the dominance tests are done on the lattice structure instead of relying on a tuple-to-tuple comparison.…”

“…Since a large number of dominance tests can often be performed independently, Skyline computation has a good potential to exploit multicore architectures as described in [5][6][7]. In this paper we present algorithms for high-performance parallel Skyline computation which do not depend on tuple comparisons, but on the lattice structure constructed by a Skyline query over low-cardinality domains.…”

Scalable parallelization of skyline computation for multi-core processors

“…PSkyline [13] naively cuts the dataset, processes each partition on a separate core, and then merges the results together. APSkyline [16] follows the same pattern, but uses sophisticated angle-based partitioning of the dataset that does not scale with dimensionality (the reported experiments consider d = 5 at most). Additionally, PSFS [13], a weaker version of our Q-Flow, was introduced in [13] as a naive baseline and VSkyline [7] modifies PSkyline to utilize SIMD instructions.…”

“…This computational challenge has prompted the use of modern computing platforms, such as GPUs [3], [8] and multicore CPUs [13], [16], as well as distributed environments [12], including MapReduce [17], [19], [22], to accelerate the computation. Of these, multicore CPUs are a particularly attractive option, because the cost of shared data structures is much lower and parallel work need not be isolated.…”