Workload Characterization and Performance Assessment of Yellowstone using XDMoD and Exploratory Data Analysis (EDA)

08/01/2014 - 9:25am
Mesa Lab Main Seminar Room

Ying Yang portrait

Ying Yang, SIParCS Intern
(University of Buffalo)


Open XDMoD is an open source tool designed to audit and facilitate the utilization of supercomputers by providing a wide range of metrics on resources, including resource utilization, resource performance, and impact on scholarship and research. The framework includes a computationally lightweight application kernel auditing system that utilizes performance kernels chosen from both low-level benchmarks and actual scientific and engineering applications to measure overall system performance from the user’s perspective. We shred and ingest the workload and performance monitoring data of Yellowstone into XDMoD and create superMoD to monitor Yellowstone. We developed two new metrics that are important to monitor Yellowstone performance and contribute back to the open source. We have also conducted exploratory data analysis with ingested data using R. Methods such as Kernel density estimation, regression, K-nearest neighbor, clustering, Bayesian network and other data mining methods has been utilized to visualize and explore patterns in the data.

Video Presentation