Parallelization strategies for high-performance and energy-efficient epidemic spread simulations

Simulation analysis of epidemic disease spread is crucial for a proper social and governmental response. Certain susceptible–infected–recovered (SIR) models based on cellular automata (CA) have proven to be effective tools for this purpose. Despite the growing interest in these simulation models, few studies have addressed computational efficiency. Many models are not parallelized and, as a result, are computationally inefficient. Moreover, computational efficiency is often solely associated with runtime, with limited attention given to energy consumption and energy-efficient software implementations.

This paper presents various parallel implementations of a successful Covid-19 cellular automaton SIR model on multiprocessors and Graphics Processing Units (GPUs), significantly improving the performance of existing codes while substantially reducing energy consumption. The performance analysis of these parallel implementations demonstrates that simulations can be reduced from hours to under a second, with energy consumption reduced by more than three orders of magnitude. Additionally, the results reveal that in cases where multiple parallel multiprocessor alternatives are available, there is not always a direct correlation between the shortest execution time and the lowest energy consumption in CA simulations.

This work aims to support practitioners interested in implementing or utilizing parallel, energy-efficient SIR model simulations for future epidemic outbreaks, green computing initiatives, and efficient cellular automata simulations in general.