EURACE

NW-GRID Project Report

EURACE

EURACE is a major European attempt to construct an agent based model of the European economy with a very large population of autonomous, purposive agents interacting in a complicated economic environment. To create it, major advances are needed, in particular in terms of economic modeling and software engineering. See http://www.eurace.org/index.php?TopMenuId=3&section=42&webPage=138 .

In the paper below, we describe the general structure of the economic model developed for EURACE and present the Flexible Large-scale Agent Modeling Environment (FLAME) that will be used to describe the agents and run the model on massively parallel supercomputers. Illustrative simulations with a simplified model based on EURACE's labour market module are presented in a paper in 2008 (see below).

Project Investigators

Dr David Worth, Prof Chris Greenough, Mr L-Shawn Chin

Scientific/ Technical Objectives

The basis for this software platform is the FLAME framework for agent modelling developed by Simon Coakley from the Department of Computer Science at the University of Sheffield. It has primarily been used in modelling biological systems but it is now being extended to model interacting agents in computational economics in the EURACE project. The agent memory and messages sent between agents are defined in XML and the behaviour of the agents is definied with functions written in C.

The role of the Software Engineering Group is to improve the parallel version for FLAME in the following areas:

• Optimise serial portions of the code where possible.
• Optimise MPI message passing.
• Increase flexibility in partitioning agents over compute nodes. At present this is done via domain decompostion (works well with biological models) but the economic agents will interact based on more than physical proximity.
• Implement load balancing for hetergoeneous agents connected by many levels of dynamic communications networks.

We will also be advising those writing the C code that implements agent behaviour on how to avoid serial bottlenecks in the algortihms they choose to implement.

Role of NW-GRID

One of our development and testing platforms for changes in parallel implementation. NW-GRID will also allow us to consider grid enabling the application at a later stage. FLAME has now been run in parallel on up to 100 procs on NW-GRID using the Daresbury SUN cluster, the Daresbury HP cluster and HPCx.

Progress to Date and Publications

2008

Paper published: C. Dessenberg, S. van der Hoog and H. Dawid "EURACE: a Massively Parallel Agent Based Model of the European Economy" Applied Mathematics and Computation 204 (2008) 541-52 http://econpapers.repec.org/paper/halwpaper/halshs-00339756_5fv1.htm

Technical Report RAL-TR-2008-022 (STFC, 2008): C. Greenough, L.S. Chin, D.J. Worth, M. Holcome and S. Coakley "The Exploitation of Parallel High Performance Systems in the FLAME Agent Based Simulation Framework" http://www.sesp.cse.clrc.ac.uk/Publications/FLAME/FLAME.pdf

30 March 2007

We have done some optimisation work on the serial parts of the code and on MPI message passing. We have also ported existing biological models and a simple economic model to parallel platforms.