Jet: Multilevel Graph Partitioning on Graphics Processing Units

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Gilbert, Michael S., Madduri, Kamesh, Boman, Erik G., Rajamanickam, Sivasankaran
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916081666359296
author Gilbert, Michael S.
Madduri, Kamesh
Boman, Erik G.
Rajamanickam, Sivasankaran
author_facet Gilbert, Michael S.
Madduri, Kamesh
Boman, Erik G.
Rajamanickam, Sivasankaran
contents The multilevel heuristic is the dominant strategy for high-quality sequential and parallel graph partitioning. Partition refinement is a key step of multilevel graph partitioning. In this work, we present Jet, a new parallel algorithm for partition refinement specifically designed for Graphics Processing Units (GPUs). We combine Jet with GPU-aware coarsening to develop a $k$-way graph partitioner, the Jet partitioner. The new partitioner achieves superior quality compared to state-of-the-art shared memory partitioners on a large collection of test graphs.
format Preprint
id arxiv_https___arxiv_org_abs_2304_13194
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Jet: Multilevel Graph Partitioning on Graphics Processing Units
Gilbert, Michael S.
Madduri, Kamesh
Boman, Erik G.
Rajamanickam, Sivasankaran
Distributed, Parallel, and Cluster Computing
Discrete Mathematics
The multilevel heuristic is the dominant strategy for high-quality sequential and parallel graph partitioning. Partition refinement is a key step of multilevel graph partitioning. In this work, we present Jet, a new parallel algorithm for partition refinement specifically designed for Graphics Processing Units (GPUs). We combine Jet with GPU-aware coarsening to develop a $k$-way graph partitioner, the Jet partitioner. The new partitioner achieves superior quality compared to state-of-the-art shared memory partitioners on a large collection of test graphs.
title Jet: Multilevel Graph Partitioning on Graphics Processing Units
topic Distributed, Parallel, and Cluster Computing
Discrete Mathematics
url https://arxiv.org/abs/2304.13194