Supercomputer Simulation of Social Processes: New Technologies

Resumo

This article continues the work described in Vestnik RAN (no. 3, 2016) and in Herald of the RAS (no. 3, 2016). The previous articles analyzed the international experience in preparing and using agent-oriented models and technical groundwork for their implementation on supercomputers and described in detail the stages and methods of the efficient reflection of the computing core of a multiagent system on the architecture of a state-of-the-art supercomputer using the Supercomputer Technology for Agent-oRiented Simulation (STARS), developed by the authors. STARS was tested on two multiagent demographic models, built at the RAS Central Economics and Mathematics Institute, which differed in the level of detailization when simulating human reproductive behavior. This publication describes a technology of building multiagent simulations, making it possible to scale effectively models of this class up to 10⁹ agents and to apply it when creating a large-scale agent model for Eurasian countries. The objective of the model is to simulate the key migration processes and economic dynamics of these countries, as well as the aftermath of the implementation of large infrastructure projects because of the activity of multiple independent agents. The model was tested on various supercomputers, allowing conclusions on their technical characteristics.