The document describes using the MASON agent-based modeling framework together with the JPPF distributed computing framework to parallelize social simulation experiments across a computer grid. MASON models are combined with JPPF to distribute simulation runs across multiple machines. This allows exploring large parameter spaces more efficiently. An example application involves modeling agents interacting in multiple overlapping social networks representing different social contexts. Results show speedups close to linear scaling as processing units are added to the grid.

1. Parallel Execution Of Social Simulation Models In
A Grid Environment
Davide Nunes, Luis Antunes
GUESS / LabMAg / University of Lisbon, Portugal
{davide.nunes, xarax}@di.fc.ul.pt
June 5, 2012

2. Outline
1 Introduction
Objectives
Grid Computing
Simulation Platforms
2 MASON + JPPF - An Implementation
MASON
JPPF
Combining MASON with JPPF
Experiment Distribution Overview
Grid Performance Tests
3 A Model of Multiple Social Contexts
Multiple Social Contexts
Context Permeability, Switching and Segregation

3. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Objectives
Introduction
The Problem
Even for simple simulation models, exploring a big parameter
space is computationally expensive.
Davide Nunes, Luis Antunes (GUESS) MABS2012 3 / 16

4. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Objectives
Introduction
The Problem
Even for simple simulation models, exploring a big parameter
space is computationally expensive.
Configuring a High-performance cluster is time consuming and
requires technical expertise.
Davide Nunes, Luis Antunes (GUESS) MABS2012 3 / 16

5. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Objectives
Introduction
The Problem
Even for simple simulation models, exploring a big parameter
space is computationally expensive.
Configuring a High-performance cluster is time consuming and
requires technical expertise.
Objective
To provide an implementation where one can easily deploy
simple social simulation models in a distributed environment
with minimum effort.
Davide Nunes, Luis Antunes (GUESS) MABS2012 3 / 16

6. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Grid Computing
Grid Computing
A computer grid is simply a set of networked loosely coupled
computers acting together to perform very large tasks [6].
Davide Nunes, Luis Antunes (GUESS) MABS2012 4 / 16

7. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Grid Computing
Grid Computing
A computer grid is simply a set of networked loosely coupled
computers acting together to perform very large tasks [6].
A grid of computers executes working units called Jobs.
Davide Nunes, Luis Antunes (GUESS) MABS2012 4 / 16

8. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Grid Computing
Grid Computing
A computer grid is simply a set of networked loosely coupled
computers acting together to perform very large tasks [6].
A grid of computers executes working units called Jobs.
Jobs have multiple independent tasks that can be executed
separately.
Grid VS Cluster
Loosely Coupled VS Tightly Coupled Processing units.
Grids contain heterogeneous resources that can be
geographically distributed.
Davide Nunes, Luis Antunes (GUESS) MABS2012 4 / 16

9. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Simulation Platforms
Simulation Platforms
NetLogo [7] allows for the execution of multiple experiments
in a single machine, limited by the number of cores a machine
possesses.
Davide Nunes, Luis Antunes (GUESS) MABS2012 5 / 16

10. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Simulation Platforms
Simulation Platforms
NetLogo [7] allows for the execution of multiple experiments
in a single machine, limited by the number of cores a machine
possesses.
MASON [5] is a Java-based framework optimized for running
efficiently in a single thread ,we can however distribute
simulation runs across multiple processing units.
Davide Nunes, Luis Antunes (GUESS) MABS2012 5 / 16

11. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Simulation Platforms
Simulation Platforms
NetLogo [7] allows for the execution of multiple experiments
in a single machine, limited by the number of cores a machine
possesses.
MASON [5] is a Java-based framework optimized for running
efficiently in a single thread ,we can however distribute
simulation runs across multiple processing units.
Repast Simphony, a suit of platforms for agent-based
modelling and simulation.
Davide Nunes, Luis Antunes (GUESS) MABS2012 5 / 16

12. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Simulation Platforms
Simulation Platforms
NetLogo [7] allows for the execution of multiple experiments
in a single machine, limited by the number of cores a machine
possesses.
MASON [5] is a Java-based framework optimized for running
efficiently in a single thread ,we can however distribute
simulation runs across multiple processing units.
Repast Simphony, a suit of platforms for agent-based
modelling and simulation.
JADE is a Java based agent middleware that supports
distributed agent communication.
Davide Nunes, Luis Antunes (GUESS) MABS2012 5 / 16

13. MASON Self-contained model
Figure : UML diagram describing a self-contained model constructed
using the MASON framework.

14. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
JPPF
On JPPF
JPPF (Java Parallel Processing Framework) [4], is an
open-source, Java-based, framework for parallel computing.
Figure : Basic components of a JPPF grid [4]
Davide Nunes, Luis Antunes (GUESS) MABS2012 7 / 16

15. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
JPPF
On JPPF
JPPF (Java Parallel Processing Framework) [4], is an
open-source, Java-based, framework for parallel computing.
Figure : Basic components of a JPPF grid [4]
Node Candidates
Workstations
Computer Clusters
JPPF Grid Servers
Davide Nunes, Luis Antunes (GUESS) MABS2012 7 / 16

16. Combining MASON with JPPF
Figure : UML diagram describing the integration of a self-contained
MASON model with the JFFP framework.

17. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Experiment Distribution Overview
Experiment Distribution
Figure : Parameter space parallel exploration process overview
Davide Nunes, Luis Antunes (GUESS) MABS2012 9 / 16

18. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Grid Performance Tests
Job Execution Speedup
Figure : Speedup ratio observed for different numbers of tasks in a single
job submission to a grid with 30 processing units. These processing units
correspond to two 8-core and seven dual-core computers.
Davide Nunes, Luis Antunes (GUESS) MABS2012 10 / 16

19. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Multiple Social Contexts
Multiple Social Contexts
In real social world scenarios, agents interact in multiple
complex social relations with other agents and/or institutions.
Davide Nunes, Luis Antunes (GUESS) MABS2012 11 / 16

20. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Multiple Social Contexts
Multiple Social Contexts
In real social world scenarios, agents interact in multiple
complex social relations with other agents and/or institutions.
Social relations may be of different kind and quality,
possessing different topologies and social dynamics.
Davide Nunes, Luis Antunes (GUESS) MABS2012 11 / 16

21. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Multiple Social Contexts
Multiple Social Contexts
In real social world scenarios, agents interact in multiple
complex social relations with other agents and/or institutions.
Social relations may be of different kind and quality,
possessing different topologies and social dynamics.
We model social spaces with multiple concurrent social
networks [3, 1, 2].
Research Context
We focus on the study of dynamic consequences of the topological
structures underlying social simulation scenarios.
Davide Nunes, Luis Antunes (GUESS) MABS2012 11 / 16

22. A Model of Context Switching
Figure : Example of context switching [2] considering two contexts for
social agent denoted by the number 1. In this case, these contexts are
created by two distinct physical spaces. Common nodes in both
neighbourhoods (like agent 2) represent an acquaintance of actor 1 in
both of them. The dashed circle represents the scope of each context.

23. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
Summary
We have reduced the time needed to explore the parameter
exploration on simple social simulation experiments.
These grids are easy to construct and any computer can be
harvested for its processing power.
The most time consuming process is still the analysis of huge
amounts of data.
Automatic dynamic simulation model analysis methods are
still in need.
Working Example
http://labmag.di.fc.ul.pt/n/guess/resources/parallel
Davide Nunes, Luis Antunes (GUESS) MABS2012 13 / 16

24. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
References I
L Antunes, J Balsa, P Urbano, and H Coelho.
Exploring context permeability in multiple social networks.
In Proceedings of the World Congress on Social Simulation,
2008.
Luis Antunes, Davide Nunes, Helder Coelho, Jo˜o Balsa, and
a
Paulo Urbano.
Context Switching versus Context Permeability in Multiple
Social Networks.
In Progress in Artificial Intelligence, volume 5816 of Lecture
Notes in Computer Science, pages 547–559, 2009.
Luis Antunes, Paulo Urbano, and Helder Coelho.
Context Permeability.
Sciences-New York, 2006.
Davide Nunes, Luis Antunes (GUESS) MABS2012 14 / 16

25. Introduction MASON + JPPF - An Implementation A Model of Multiple Social Contexts Summary
References II
Laurent Cohen.
Jppf, 2005.
Sean Luke, Claudio Cioffi-Revilla, Liviu Panait, Keith Sullivan,
and Gabriel Balan.
MASON: A Multiagent Simulation Environment.
Simulation, 81(7):517–527, 2005.
Jarek Nabrzyski, Jennifer M. Schopf, and Jan Weglarz, editors.
Grid resource management: state of the art and future trends.
Kluwer Academic Publishers, Norwell, MA, USA, 2004.
U. Wilensky.
Netlogo.
1999.
Davide Nunes, Luis Antunes (GUESS) MABS2012 15 / 16

26. Thank you for your attention
Questions?