Automated Programmatic Performance Analysis of Parallel Programs

Abstract

'Katherine E. Isaacs' 'Abhinav Bhatelé'] A variety of performance measurement tools exist, including profilers and tracing tools that can generate performance data [1, 7, 17, 28]. However, the data generated can be extremely large, making it challenging to sift through this data to identify performance issues. Several performance measurement tools also provide visual analytics counterparts to facilitate performance analysis. Typically, the analysis support is in the form of a graphical user interface (GUI) to visualize and manipulate performance data [3, 14, 16, 22] although some tools also provide a scripting interface [9, 26]. The GUIs help in visualizing performance data and in many cases, the user can connect such data to source code (file and line numbers).