Communication Efficient Algorithms for Top-k Selection Problems
Data Structures and Algorithms
2015-10-20 v2 Distributed, Parallel, and Cluster Computing
Abstract
We present scalable parallel algorithms with sublinear per-processor communication volume and low latency for several fundamental problems related to finding the most relevant elements in a set, for various notions of relevance: We begin with the classical selection problem with unsorted input. We present generalizations with locally sorted inputs, dynamic content (bulk-parallel priority queues), and multiple criteria. Then we move on to finding frequent objects and top-k sum aggregation. Since it is unavoidable that the output of these algorithms might be unevenly distributed over the processors, we also explain how to redistribute this data with minimal communication.
Cite
@article{arxiv.1502.03942,
title = {Communication Efficient Algorithms for Top-k Selection Problems},
author = {Lorenz Hübschle-Schneider and Peter Sanders and Ingo Müller},
journal= {arXiv preprint arXiv:1502.03942},
year = {2015}
}