László Gulyás, Attila Szabó, Richárd Legéndi, Tamás Máhr, Rajmund Bocsi, and George Kampis

This paper discusses a collection of tools aimed at helping agent-based modelers to run large-scale computational experiments using agent-based social simulations. We address both the issue of running simulations with millions of agents (technically, distributing individual simulation runs over a cluster of computers) and the issue of handling enormous parameter spaces. For this latter, we address two approaches.The first is a purely computational one that helps distributing computational experiments over a cluster of computers. In connection, we also describe an on demand simulation service, where modelers can rent (potentially hundreds) of computers by the hour over the web, providing access to an „instant computer farm”. The second approach addressed is more statistical than computational. We briefly refer to and discuss the application of careful designs of experiments.

Download PDF