Date Approved
12-15-2024
Graduate Degree Type
Project
Degree Name
Applied Computer Science (M.S.)
Degree Program
School of Computing and Information Systems
First Advisor
Christian Trefftz
Academic Year
2024/2025
Abstract
Community detection in complex networks is an essential process in the field of network science, offering insights into the underlying structure and functionality of interconnected systems. As the scale and complexity of networks grows, traditional CPU-based methods for community detection struggle to keep pace, leading to the exploration of GPU-accelerated solutions.
This project investigates the implementation of the Louvain algorithm for community detection using GPU-accelerated computing via the CuGraph library. By comparing it too traditional CPU based methods implemented with NetworkX, this study examines performance improvements, scalability, and applicability across real-world datasets. Two networks were used for experimentation: Zachary’s Karate Club graph, a widely used benchmark, and the larger email-EU Core dataset, representing real-world email communication networks.
The results indicate that GPU acceleration offers significant computational benefits. On the Email EU Core dataset, GPU-based execution demonstrated a 9.1x speedup compared to traditional methods, achieving similar modularity scores. This research highlights the potential of GPU accelerated computing in enhancing the performance and scalability of graph analytics, paving the way for more efficient analysis of large-scale networks in fields such as social media, biology, and citation networks.
ScholarWorks Citation
Pulaparthi, Venkata Satyanarayana, "GPU-Accelerated Community Detection: Performance Comparison of NetworkX and CuGraph" (2024). Culminating Experience Projects. 541.
https://scholarworks.gvsu.edu/gradprojects/541
