DESC0024832

In the effort to propagate the use of ASCR-funded research codes into more general computing communities, NexGen Analytics proposes to develop and support an open-source visual user support tool, the ôWork Visualizerö, in order to address the challenges of performance analysis and optimization of application codes that run on heterogeneous architectures at scale. While there exist tools that support parallel debugging or trace-based analysis of CPU or GPU runtimes, NexGen Analytics proposes to develop and support a tool that would integrate informative visualization with the dual goal to 1) coarsen up a level of abstraction over the fine granularity of trace-based profiling, e.g. like the Projections tool, and 2) support users who run at large scale. A guiding idea would be to make this tool as light-weight as possible so that it could run along with an application (with some acceptable overhead) to provide visual insight of performance in real time. The Work Visualizer would be structured according to the core capabilities of 1) collection, 2) analysis and 3) visualization. Communication data, systems level properties of the compute platform, as well as host and device event-based sampling would all be ingested in order to then be analyzed and displayed to the application user. NexGen Analyticsĺ team would leverage its experience with machine learning and VTK-based visualization to make an informative, well-designed, and ultimately, helpful and useful tool. As more and more general computing needs are being met in the Cloud, with an increasing use of Cloud-based heterogeneous compute platforms, NexGen Analytics observes an opportunity to spread the use of HPC tooling into the commercial technology sector. NexGen Analyticsĺ private sector market focus would particularly target potential clients in the growing AI domain, especially for its reliance on the use of accelerators.