Profiling and Bottlenecks Analysis of an Agent-based Dengue Fever Simulation Model

Agent-Based Model simulations plays a critical role in computational epidemiology, usually employing GPU aceleration to account for large scale scenarios. In this work, we profile and analyze an agent based model for dengue fever disease spreading simulation as part of a bachelor thesis work, to find bottlenecks and performance issues. We have discovered a bottleneck in two of the model kernels, and inefficient host-to-device memory transfers. The use of a unified memory architecture as also running more kernels in parallel using ISO C++17 Parallelism are proposed as future solutions to these challenges.