Accelerating The Distributed Simulations Of Agent-Based Models Using Community Detection

Abstract

The performance of parallel simulations of large agent-based models (ABMs) distributed across multiple CPUs is strongly dependent on the distribution of the agents among the processors. In this paper, we introduce an algorithm that uses community structure to distribute the agents among processors, thereby reducing the communication overhead. In particular, we compare three distribution schemes (initial-LPA, separate-LPA and integrated-LPA) for employing label propagation community detection (LPA) in the distributed simulations of self-organizing ABMs. In the novel third scheme, we provide completely seamless integration by exploiting the similar process flow patterns of LPA and ABMs. Using an example from computational epidemiology, we demonstrate that our algorithm accelerates the distributed simulation of ABMs on multiple processors up to eight times compared to a baseline distribution method.

Publication Date

12-27-2016

Publication Title

2016 IEEE RIVF International Conference on Computing and Communication Technologies: Research, Innovation, and Vision for the Future, RIVF 2016 - Proceedings

Number of Pages

25-30

Document Type

Article; Proceedings Paper

Personal Identifier

scopus

DOI Link

https://doi.org/10.1109/RIVF.2016.7800264

Socpus ID

85010699940 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/85010699940

This document is currently not available here.

Share

COinS