Petascale Programming Environments and Tools

MSI hosted two courses for graduate students and postdoctoral researchers on the Twin Cities campus during Summer 2010 in association with the Virtual School of Computational Science and Engineering (VSCSE). The VSCSE is a national virtual organization whose goal is to develop and deliver a computational science curriculum that accelerates the ability of faculty, staff, and students to use emerging computational resources to advance science and engineering.

During the week of July 5, the University of Minnesota Supercomputing Institute (MSI) was a participating site in the VSCSE's Petascale Programming Environments and Tools course. The week-long interactive workshop and lecture series commenced with a keynote from theoretical chemist Thom Dunning on July 6 and ran through July 9. Eleven students participated from MSI, joining 230 others from nine additional sites across the U.S. One of the course presentations was broadcast from MSI with graduate students Pei-Hung Lin and Jagan Jayaraj discussing their experience scaling a gas dynamics code up to 100,000 cores (top picture, below).

The course introduced students to the challenges, opportunities, techniques, and resources for scaling computational science codes to perform on petascale computing systems. Students learned techniques for coding and code optimization on highly parallel architectures (especially multicore architectures); how to debug code in a highly parallel environment; how to scale existing codes to highly parallel systems with thousands of cores; and how to develop parallel codes using numerical libraries, parallel I/O, and advanced compiler technologies. Pictures from the course are shown below.

The Petascale Programming course was powered by an H.323 videoconferencing service supported locally by the Office of Information Technology (OIT). Students in Minnesota received a high-definition video feed of the presenter through one projector, a video feed of the presenter’s slides on a second screen, and a third video feed that included the audiences from the remaining nine sites. All sites were linked with bidirectional audio enabling students in Minnesota to have face-to-face conversations with the course instructors as well as participants at other sites. Students at all sites were given accounts on the fastest supercomputers in the world for the duration of the course, and worked through exercises interactively with the instructors and other sites.

MSI hosted a second course entitled Big Data for Science, July 26-30. An article about the summer's record participation appeared in an article on HPC Wire.