Page Rank Performance Evaluation of Cluster Computing Frameworks on Cray Urika-GX Supercomputer
Robert W. Techentin, Matthew W. Markland, Ruth J. Poole, David R. Holmes, Clifton R. Haider, Barry K. Gilbert · 2016
Modern “big data” and analytics software platforms offer a variety of algorithms to the analytics practitioner. It is often possible to choose from several available implementations to solve a given set of problems, even on the same set of hardware, within the same software ecosystem, or leveraging completely different software environments. Choices of techniques should be informed by the relative performance and scalability of the implementations, whether or not they run on the same hardware platform. This paper presents a relative performance comparison of three implementations of the popular PageRank graph analytic algorithm running on the Cray Urika-GX high performance analytics appliance. Relative performance and scaling presented here, along with additional information about the problem set and resources available to the analyst, could be used to make an informed decision about which PageRank implementation to choose.