Interactive in situ visualization and analysis using Ascent and Jupyter

There is a gap in interactive in situ solutions for HPC simulations. While we have access to fully-featured visualization UIs through tools like ParaViewCatalyst and VisItLibSim,we lack in situ infrastructure to access more general interactive environments like Jupyter. The Jupyter ecosystem of tools provides a rich paradigm for interactive data analysis and is well-suited to help expand interactive use of in situ. Both the complexity of the Jupyter ecosystem and the security requirements of HPC centers pose challenges to developing software infrastructure that enables direct use of Jupyter in simulation codes. With this work, we developed a system enabling simulations instrumented with Ascent to connect to Jupyter. This system allows simulation users to interact with their data in situ using Jupyter Notebooks. The system combines Ascent's embedded Python filter infrastructure with a Client/Server Jupyter Bridge Kernel to simplify both deployment and security considerations on HPC systems. We describe the design of this system, demonstrate basic usage, and describe a prototype of an Ascent rendering UI built on top of this system.