This paper presents a method for performance profiles development of scientific workflow. It addresses issues related to: workflows execution in a parameter sweep manner, collecting performance information about each workflow task, and analysis of the collected data with statistical learning methods. The main goal of this work is to increase the understanding about the performance of studied workflows in a systematic and predictable way. The evaluation of the presented approach is based on a real scientific workflow developed by the Spallation Neutron Source - a DOE research facility at the Oak Ridge National Laboratory. The workflow executes an ensemble of molecular dynamics and neutron scattering intensity calculations to optimize a model parameter value.
原文链接
分享
引用
