Agent-Based Modelling as a Service on Amazon EC2: Opportunities and Challenges

Abstract

Modelling and Simulation is heavily influenced by availability of computational power and resources, to successfully complete simulation tasks. In this paper, we investigate deploying the FLAME framework, the only supercomputing framework that automatically produces parallelisable code on different parallel hardware architectures, on cloud infrastructures. The framework focuses on agent-based modelling (ABM) technique which has presented various challenges in the high-performance computing fields and how these reflect in Cloud environments. Computationally these simulations are extremely complex to program with interconnected software, using massive amount of computational power and architectural challenges. High-performance computing grids have provided solutions to some of these issues, but are still not capable enough to solve most of the issues faced by the modelers. This paper discusses the computational problems of executing these simulations and open challenges. Presenting ABM-as-a-service with a possible framework on how this can be implemented with a platform as a service backend. Computational problems such as memory, processing and time are discussed highlighting the issues for enabling these services for non-computing scientists.