Date of Award


Degree Name

Master of Science


Computer Science

First Advisor

Dr. Elise de Doncker

Second Advisor

Dr. John Kapenga

Third Advisor

Dr. Diana Prieto


Epidemiology, influenza, gpu, parallel, modeling

Access Setting

Masters Thesis-Open Access


Simulations of influenza spread are useful for decision-making during public-health emergencies. Policy-makers use models to predict disease spread and estimate the effects of various intervention strategies. Effective modeling of targeted intervention strategies requires accurate modeling of individual-level behavior and transmission. However, this greatly increases the computational costs of these agent-based models. In addition, if the models are used as an outbreak progresses, some operational decisions must occur rapidly in order to contain the spread of the disease.

Graphics Processing Units (GPUs) are a type of specialized processor used to drive graphical displays. Many recent devices also allow users to write and execute programs using the GPU's processor. This processor is designed with functional units that perform the same operation on many pieces of data at once, providing significant increases in processing power. However, good performance depends on effective usage of the GPU architecture, and not all algorithms can be accelerated using this approach. To date, most influenza simulations have had very limited gains from GPU acceleration.

This thesis investigates the performance of influenza simulations when accelerated by GPU devices. We implement a simulation using an agent-based model of low computational complexity, as well as a framework designed to support creation, processing, and analysis of simulations. We further investigate several approaches for accelerating the processing of this simulation using commonly-available hardware. We discuss strategies for adapting such models to GPU computation, and demonstrate an implementation of our model running entirely in GPU memory. This accelerated implementation supports simulation of very large populations and produces excellent performance., reaching speedups of approximately 95x. Additionally, we implement an OpenMP parallelization of the baseline model and compare results.