Fermilab Computing Division

Multi-core Optimizations for Synergia2

Full Title: Multi-core Optimizations for Synergia2
Date & Time: 30 Aug 2012 at 13:00
Event Location: DIR/ Racetrack
Event Moderator(s):
Event Info: Speaker: Qiming Lu, Scientific Computing Division at Fermilab

Abstract:

Synergia2 is a parallel 3D space charge PIC code that is being widely used by accelerator modeling community. The performance and strong scaling of the package on recent multi-socket multi-core architectures has been studied. Here we describe recent work to optimize Synergia2 for multicore architectures utilizing a hybrid MPI-OpenMP approach. We describe multiple hybridization options, including the communication avoidance scheme, multi-threading with OpenMP, customized FFT, and etc., each demonstrating much better scaling behavior than the pure-MPI approach. In all cases, we extend strong scaling by at least a factor of 8. We expect different optimization schemes to be optimal on different architectures.

Speaker Bio:

Qiming Lu

Scientific Software Infrastructure / Computational Physics for Accelerators group, Scientific Computing Division
Working on writing software and optimizations for large scale parallel simulations; building DAQ and other supporting softwares for experiments; etc.

Ph.D. in physics, Dept. of Physics, Rensselaer Polytechnic Institute, 2009
Studied the dynamic and long-term statistical behavior for large scale complex systems and interacting social networks.

M.S. in physics, University of Science and Technology of China, 2004

No talks in agenda


DocDB Home ]  [ Search ] [ Authors ] [ Events ] [ Topics ]

DocDB Version 8.7.23, contact Document Database Administrators
Execution time: 0 wallclock secs ( 0.20 usr + 0.04 sys = 0.24 CPU)