A Parallel Many-core CUDA-based Graph Labeling Computation

Stefano Quer

2020

Abstract

When working on graphs, reachability is among the most common problems to address, since it is the base for many other algorithms. As with the advent of distributed systems, which process large amounts of data, many applications must quickly explore graphs with millions of vertices, scalable solutions have become of paramount importance. Modern GPUs provide highly parallel systems based on many-core architectures and have gained popularity in parallelizing algorithms that run on large data sets. In this paper, we extend a very efficient state-of-the-art graph-labeling method, namely the GRAIL algorithm, to architectures which exhibit a great amount of data parallelism, i.e., many-core CUDA-based GPUs. GRAIL creates a scalable index for answering reachability queries, and it heavily relies on depth-first searches. As depth-first visits are intrinsically recursive and they cannot be efficiently implemented on parallel systems, we devise an alternative approach based on a sequence of breadth-first visits. The paper explores our efforts in this direction, and it analyzes the difficulties encountered and the solutions chosen to overcome them. It also presents a comparison (in terms of times to create the index and to use it for reachability queries) between the CPU and the GPU-based versions.

Download


Paper Citation


in EndNote Style

TY - CONF

JO - Proceedings of the 15th International Conference on Software Technologies - Volume 1: ICSOFT,
TI - A Parallel Many-core CUDA-based Graph Labeling Computation
SN - 978-989-758-443-5
AU - Quer S.
PY - 2020
SP - 597
EP - 605
DO - 10.5220/0009780205970605


in Harvard Style

Quer S. (2020). A Parallel Many-core CUDA-based Graph Labeling Computation.In Proceedings of the 15th International Conference on Software Technologies - Volume 1: ICSOFT, ISBN 978-989-758-443-5, pages 597-605. DOI: 10.5220/0009780205970605


in Bibtex Style

@conference{icsoft20,
author={Stefano Quer},
title={A Parallel Many-core CUDA-based Graph Labeling Computation},
booktitle={Proceedings of the 15th International Conference on Software Technologies - Volume 1: ICSOFT,},
year={2020},
pages={597-605},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0009780205970605},
isbn={978-989-758-443-5},
}