Modeling & Simulation as a Service with the Massive Multi-Agent System MARS

Abstract

There is an increasing demand for very large-scale agent-based models. High numbers of individual entities and complex interactions between them require new ways of modeling and simulation. The creation of a distributed simulation model imposes a major challenge in the fields of network communication and coordination to the developer. Integrating multi-scale GIS and time-series data into such a model is another challenge altogether. We introduce the massive Multi-Agent Research and Simulation system (MARS). It is designed to provide a Modeling and Simulation as a Service (MSaaS) solution to end users. MARS allows domain experts to integrate their data and models through a user-friendly web interface. The complexity of distributed and scalable simulation is handled in the background by a mechanism we call Agent Shadowing. Finally a layer-based segmentation of the model is proposed. It allows domain specialists to focus on one aspect at a time while developing their simulation model. A large-scale model from the ecological modeling domain is showcased. The model integrates various GIS data formats with collected time-series datasets and simulates a scalable amount of agents. Results from this simulation demonstrate the capabilities of MARS to support the workflow as required by the development of large-scale agent-based simulation models.