OLCF Experts Teach and Learn at Cray User Group
HPC leaders attend four-day supercomputing conference
Members of the OLCF staff joined high-performance computing professionals from around the U.S. for the 2011 Cray User Group (CUG) meeting held in Fairbanks, AK, May 23-26. The technical conference provided an opportunity for users of Cray supercomputers, such as OLCF’s Cray XT5 “Jaguar,” to exchange problem-solving information via presentations, tutorials, meetings and workshops, as well as informal “Birds of a Feather” discussions, which allowed attendees to gather based on a shared interests. The theme of this year’s conference was “Golden Nuggets of Discovery,” a nod to both the discovery of gold shortly after the founding of Fairbanks and the rich scientific discoveries made possible through high-performance computing.
“The Cray Users Group is an excellent forum for sharing our successes and issues with other Cray sites and with Cray management,” said OLCF Project Director Buddy Bland of the importance of OLCF’s participation at CUG. “Since most Cray systems are located at government research laboratories, we get to share experiences with our peers from around the world, to the benefit of all parties.”
OLCF staff participated in both General and Parallel Technical Sessions. General Technical Sessions provided topics of interest to workshop attendees, including up-to-date corporate and technical information from Cray executives. The Parallel Technical Sessions give participants the opportunity to focus on specific knowledge domains such as data management, code optimization, and I/O [data input/out-put] performance.
During Monday’s sessions, OLCF User Assistance and Outreach group leader Ashley Barker and user support specialist Robert Whitten gave a presentation on building an electronic knowledge base, a valuable resource for Jaguar’s users. Bland gave a presentation on Tuesday titled, “Titan: ORNL’s New System for Scientific Computing,” which outlined the facility’s plan for the installation of a 10- to 20-petaflop hybrid computer system containing both CPUs (general processing units) and GPUs (graphic processing units).
Other OLCF staff presentations included a nuclear physics code optimization by computational scientist Rebecca Hartman-Baker, a talk by Linux-based code developer David Dillow on how to avoid I/O congestion, and a discussion on the importance of hierarchical parallelism in the hybrid multicore era by Bronson Messer, OLCF interim director of science. Messer’s presentation included case studies from the OLCF Center for Application Acceleration Readiness, in which committee members have worked for nearly two years to optimize half a dozen of the most highly utilized HPC codes for use on hybrid architectures. — by Caitlin Rockett