April 4-6

Log in

Abstract Details

files Add files

status:file name:submitted:by:
approveddebs_poster_sherwood2016.pdf2016-04-08 10:54:49Brian Cornille


Author: Brian S. Cornille
Requested Type: Poster
Submitted: 2016-02-12 14:11:45

Co-authors: C. R. Sovinec

Contact Info:
University of Wisconsin-Madison
1500 Engineering Dr
Madison, WI   53706
United States

Abstract Text:
The legacy DEBS resistive MHD code [D. D. Schnack et al., JCP 70, 330 (1987)] is frequently used for modeling the Madison Symmetric Torus RFP experiment. Recent modifications for parallel computation allow for faster turnaround and greater numerical resolution in RFP simulations. Since the semi-implicit time advance of DEBS requires solving a narrow-banded matrix equation, the parallel implementation is designed around the use of the parallel linear algebra library ScaLAPACK. These matrix equations lie across the radial mesh of a cylindrical domain, and that mesh needs to be distributed across multiple processors. The parallel implementation for DEBS is a balance between minimizing the changes to the code and maximizing effectiveness of the parallel computations. Results from a strong scaling study show that the number of radial points per processor is best kept near or greater than fifty. A weak scaling study indicated that for typical radial mesh sizes the limiting factor of parallel execution time is due to latency associated with inter-processor communication. Thus, in production level calculations, best performance is often obtained on a single shared memory compute node. A production-scale RFP application is used to exercise the parallel implementation of DEBS.