Graduate student Yuguang Wang, PhD in Computational Science and Engineering, will present a research proposal on Wednesday, April 10, 2024, from 10-11:30 am via Zoom online meeting.
The title of Wang’s proposal is, “GPU-based Out-of-Memory Graph Processing Design.”
Wang is advised by Professor Zhenlin Wang, Computer Science, and Junqiao Qiu.
Proposal Abstract
Graphics Processing Units (GPUs) have emerged as a popular platform for parallel computing due to their ability to perform massive parallelism and provide high memory bandwidth. Designing efficient GPU programs, however, faces challenges, particularly in optimizing thread utilization and facilitating fast data access. Our preliminary work has demonstrated the acceleration of Finite State Automata (FSA) processing on GPUs, a fundamental computational model employed in a variety of applications, including textual data analytics, malware detection, system verification, machine learning, and natural language processing. By leveraging chunk-level parallelism and employing speculation and recovery techniques, we have enabled parallel processing for both deterministic and non-deterministic finite automata.
Building on this success, we shift our focus to GPU-based graph processing, a domain that, like FSA, involves traversing nodes and edges but represents a broader spectrum of real-world entities and their interrelationships. As the size of graphs in real-world applications increasingly exceeds GPU memory limitations, many existing GPU-based graph processing frameworks fail to operate, necessitating the development of Out-of-Memory (OOM) graph processing solutions.
To address the performance challenges associated with OOM graph processing, we propose a GPU-based framework designed to efficiently process large-scale graphs.