A Picture Is Worth a Thousand Numbers—Enhancing Cube’s Analysis Capabilities with Plugins

In the last couple of years, supercomputers became increasingly large and more and more complex. Performance analysis tools need to adapt to the system complexity in order to be used effectively at large scale. Thus, we introduced a plugin infrastructure in Cube 4, the performance report explorer for Score-P and Scalasca, which allows to extend Cube’s analysis features without modifying the source code of the GUI. In this paper we describe the Cube plugin infrastructure and show how it makes Cube a more versatile and powerful tool. We present different plugins provided by JSC that extend and enhance the CubeGUI’s analysis capabilities. These add new types of system-tree visualizations, help create reasonable filter files for Score-P and visualize simple OTF2 trace files. We also present a plugin which provides a high-level overview of the efficiency of the application and its kernels. We further discuss context-free plugins, which are used to integrate command-line Cube algebra utilities, like cube_diff and similar commands, in the GUI