Comparing communication strategies for Impact-T's Poisson solver
February 1, 2011
Brian Austin
NERSC Petascale Postdoc
The process of designing particle accelerators increasingly relies on computer simulations codes to predict and tailor details of the particle phase space. Impact-T is an example of an accelerator simulation code that uses an FFT-based Poisson solver to compute space-charge forces between particles in the accelerator. At high concurrency, the performance of this solver is limited primarily by network latency during the transpose step of the 3-d FFT. I will describe two different strategies I have used to mitigate the effects of communication on the overall performance of Impact-T's Poisson solver. The first strategy uses non-blocking MPI functions to overlap the transpose (communication) of a Green's function with the 1-d FFT of the particle density. The second approach packs elements of the Green's function and density into a single message buffer to minimize the number of messages sent. The message consolidation strategy performs better in nearly all tests and increases the speed of the solver by up to 2.5x relative to original code.
About NERSC and Berkeley Lab
The National Energy Research Scientific Computing Center (NERSC) is the primary high-performance computing facility for scientific research sponsored by the U.S. Department of Energy's Office of Science. Located at Lawrence Berkeley National Laboratory, the NERSC Center serves more than 4,000 scientists at national laboratories and universities researching a wide range of problems in combustion, climate modeling, fusion energy, materials science, physics, chemistry, computational biology, and other disciplines. Berkeley Lab is a U.S. Department of Energy national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California for the U.S. DOE Office of Science. For more information about computing sciences at Berkeley Lab, please visit www.lbl.gov/cs.
