SC Conference - Activity Details

Methods and Challenges Scaling FLASH to Petascale Computation

Katherine Riley  (Argonne National Laboratory)
Dean M. Townsley  (University of Chicago)
Raymond A. Bair  (Argonne National Laboratory)
Anshu Dubey  (University of Chicago)
Robert T. Fisher  (University of Chicago)
Don Q. Lamb  (University of Chicago)
Nathan Hearn  (University of Chicago)
Posters Session
Tuesday,  05:15PM - 07:00PM
Room Rotunda Lobby
Using the Argonne Leadership Computing Facility's (ALCF) Intrepid 163,840-core Blue Gene/P, the FLASH code, and computational resources provided by the DOE Office of Science under the INCITE program, members of the University of Chicago DOE/NNSA ASC FLASH Flash Center, working with ALCF staff at Argonne National Laboratory, have obtained preliminary results for a fundamental physical process in the modeling of thermonuclear supernovae: the degree to which buoyancy-driven turbulence enhances the burning rate of the nuclear flame during the initial deflagration phase of the explosion. We describe the challenges and successes in optimizing this large-scale adaptive mesh simulation. We focus on the challenges presented by the memory needs and load balancing of this application, and the steps we took to improve single-cpu performance. We present scaling results up to 131,072 cores and an analysis of the scaling properties, and discuss how these results are guiding future efforts to enhance performance.
   IEEE Computer Society  /  ACM     2 0   Y E A R S   -   U N L E A S H I N G   T H E   P O W E R   O F   H P C