OpenCL for Large-Scale Agent-Based Simulations

Abstract

NetLogo is a Java-based multi-agent programmable modeling environment. Our aim is to improve the execution speed of NetLogo models with large number of agents by means of heterogeneous computing. Firstly, we describe OpenCL as a suitable computing platform. Then we propose a new NetLogo-to-OpenCL extension (NL2OCL) which encapsulates functionality of OpenCL and enables NetLogo to undertake agents’ computations simultaneously on graphic processor units. The architecture of our extension is presented. An experimental flocking model with 40,000 agents is used for evaluation of NL2OCL functioning. When using NL2OCL the simulation runs more than 300-times faster than the original model which was created in NetLogo solely. It means that with NL2OLC, drawbacks in maximum size of the NetLogo model and the simulation speed are tackled. Our approach allows using standard PC configurations with suitable graphical cards for large agent-based simulations while preserving advantages of NetLogo. It is a good alternative for researchers who cannot afford high performance computational systems.