By Rob Mitchum // November 14, 2017
A share of a prestigious award, a demo of a new research data portal, and a plenary talk are among the highlights of CI research at this week’s SC17 Supercomputing conference in Denver, CO.
The HPCWire Editor’s Choice Award for Best Use of HPC in Physical Sciences, announced Monday night, went to the National Center for Supercomputing Applications for their work with the ArcticDEM mapping project. The award also acknowledged the use of CI’s Swift parallel scripting language, which allowed ArcticDEM researchers to analyze their images on the Blue Waters supercomputer.
“A notable milestone of ArcticDEM was to productively use the Blue Waters supercomputer as a huge many-task engine” said Swift team leader and CI Senior Fellow Mike Wilde. “It demonstrates that supercomputers are not just large single-program engines. We can use a supercomputer in a more dynamic and resilient fashion to address important problems – like geospatial ones – that are many-task in nature.”
Swift, in a new Python-friendly flavor called “Parsl” (Python Parallel Scripting Library), also plays a role in an exciting new research data portal which will make its premiere Thursday at the Department of Energy booth. CI Senior Fellow and Globus co-founder Ian Foster will demo the innovative pipeline, which pulls data from the Materials Data Facility with Globus, sends it to the Argonne Leadership Computing Facility for machine learning and analysis, and documents the process and results using the open-source notebook Jupyter.
The demo’s use case will take materials science simulation data generated at ALCF by UIUC’s Andre Schleife and couple standard analyses with machine learning algorithms to extrapolate new data points, said postdoctoral researcher Logan Ward, a member of the project team. Ben Blaiszik, software developer at Globus, added that the pipeline will also be useful for other data-heavy fields such as quantum chemistry, cosmology, and climate science, and applications such as nuclear reactor safety and proton therapy for cancer.
“ALCF is interested in finding ways to capture the data that’s generated at ALCF from various projects and index them in a way that makes them discoverable and reusable, and brings them together with machine learning and analysis tools,” Blaiszik said. “It’s bringing together and extending, at a high level, a lot of modular pieces that several teams have developed separately.”
Another Computation Institute center, the Urban Center for Computation and Data, was featured in Monday night’s plenary session on “The Century of the City.” Moderated by UrbanCCD Director and CI Senior Fellow Charlie Catlett, the panel also included CI Senior Fellow Pete Beckmanand featured the Array of Things urban sensing project as a leading example of how high-performance computing can help cities.
To promote the plenary session, the conference’s #HPCConnects campaign visited Argonne National Laboratory for a video with Catlett and Beckman, which you can watch below.