Parallel Generation of Massive Scale-Free Graphs

One of the biggest huddles faced by researchers studying algorithms for massive graphs is the lack of large input graphs that are essential for the development and test of the graph algorithms. This paper proposes two efficient and highly scalable parallel graph generation algorithms that can produc...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yoo, Andy, Henderson, Keith
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:One of the biggest huddles faced by researchers studying algorithms for massive graphs is the lack of large input graphs that are essential for the development and test of the graph algorithms. This paper proposes two efficient and highly scalable parallel graph generation algorithms that can produce massive realistic graphs to address this issue. The algorithms, designed to achieve high degree of parallelism by minimizing inter-processor communications, are two of the fastest graph generators which are capable of generating scale-free graphs with billions of vertices and edges. The synthetic graphs generated by the proposed methods possess the most common properties of real complex networks such as power-law degree distribution, small-worldness, and communities-within-communities. Scalability was tested on a large cluster at Lawrence Livermore National Laboratory. In the experiment, we were able to generate a graph with 1 billion vertices and 5 billion edges in less than 13 seconds. To the best of our knowledge, this is the largest synthetic scale-free graph reported in the literature.
DOI:10.48550/arxiv.1003.3684