Semester of Graduation

5

Degree

Master of Science (MS)

Department

Computer Science and Engineering

Document Type

Thesis

Abstract

NWGraph is a proposed C++ standard library for graph algorithms and interfaces. This research delves into NWGraph's two pivotal algorithms: triangle counting and breadth-first search, providing an in-depth analysis of their implementations and their unique characteristics while evaluating their parallel performance. While the original parallel graph implementations include TBB and standard C++ constructs, in this exploration, we adapt NWGraph to HPX, a third-party runtime known for its minimal overhead and compare its performance with the native implementations. We uncover performance bottlenecks that originate from the algorithms. This study represents a partial conversion of NWGraph's graph algorithms to HPX, demonstrating performance improvements in specific scenarios. We showcase how HPX facilities map straight to standard facilities and provide ease of programmability along with performance. We conclude with a holistic performance evaluation of the two algorithms in both HPX and the native implementations.

Date

10-27-2023

Committee Chair

Kaiser, Hartmut

Available for download on Friday, October 25, 2030

Share

COinS