TAU Gateway: Performance Engineering for Scientific Gateways

In order to increase access to High Performance Computing HPC) systems, Science Gateways have been developed to provide web-based interfaces to scientific applications. Science Gateways, which have been de- ployed at Department of Energy Leadership Computing Facilties, through NSF XSEDE, and at academic and research institutions throughout the world, provide a simple interface which hides the technical details of compilation, data transfer, and job management on HPC systems. However, the resultant influx of users without HPC experience increases the burden on support staff and application developers. ParaTools, Inc. will address this problem by making improvements and developing new interfaces to the TAU Performance SystemQR , a performance tool developed in part with Department of Energy SciDAC funding, to improve usability when applied to software running in a Science Gateway environment. The key innovation is to provide performance data collection and analysis tools through the same interface as the domain-specific interface provided by Science Gateways. A user will collect performance with a single button press in the web interface and will share that data with Science Gateway support staff and application developers, who can in turn analyze and visualize the collected performance data through the Science Gateway interface. ParaTools, Inc. will develop TAU Gateway, a user-friendly interface to the TAU Performance Tools presented through the same interface as Science Gateway applications. TAU Gateway will package performance measure- ment, analysis, and visualization tools using container technologies in use at HPC centers for ready deployment across multiple Science Gateway software stacks. ParaTools will develop web interfaces for controlling instru- mentation which is compatible with multiple Science Gateway middleware packages; will develop technology for injecting the TAU Gateway environment into pre-existing container images without requiring that their build scripts be modified; will develop a collaborative environment enabling performance results to be easily shared between end users, support staff, and developers; and will develop interfaces for overlaying performance data on top of existing application-level visualizations. The software developed as part of the Phase I activities will be integrated and tested on a production Science Gateway environment. Usage of Science Gateways is rapidly increasing. An analysis of usage patterns on the NSF XSEDE eXtreme Science and Engineering Discovery Environment) systems found a five-fold increase in the number of active Science Gateway users over the period 2011-2017, and the number of active Science Gateway users now exceeds the number of traditional HPC users on XSEDE. However, the development of tools to support Science Gateway use, including performance tools, has not kept up with growth in usage. If this project is successful and is carried over into Phase II and III, performance improvements facilitated by TAU Gateway will provide benefits to support staff and application developers by aiding them in making efficient use of computational resources, will provide benefits to domain scientists by reducing time to solution, thereby accelerating the pace of scientific discovery.