NEWS
Parallelization of the FDTD Method Based on the Open Computing Language and the Message Passing Interface
20/01/2012
Parallelization of the FDTD Method Based on the Open Computing Language and the Message Passing Interface
Tomasz P. Stefanski, Nicolas Chavannes, and Niels Kuster, in Microwave and Optical Technology Letters, Volume 54, Issue 3, pp. 785–789, March 2012, online January 20
A novel approach for parallelization of the finite-difference time-domain (FDTD) method based on open computing language (OpenCL) and the Message Passing Interface (MPI) is presented. Because of the portability of OpenCL, the developed code targets not only distributed shared memory computer clusters based on multicore central processing units but also clusters accelerated by graphics processing units. Subsequently, this article shows the results of numerical tests to evaluate the hybrid MPI-OpenCL FDTD solver in electromagnetic simulations.
The scientific and technical impact of the study can be summarized as:
- A parallel programming model for the FDTD method was developed based on the Message Passing Interface (MPI) and the Open Computing Language (OpenCL).
- Due to the portability of OpenCL, the developed code targets not only distributed shared memory computer clusters based on multi-core central processing units (CPUs), but also clusters accelerated by graphics processing units (GPUs).
- Results of numerical tests are presented showing efficiency and portability of the hybrid MPI-OpenCL FDTD solver in electromagnetic simulations.
- A single cluster node accelerated by four GPUs delivered almost three times the speed in comparison to four nodes executing the code on multi-core CPUs. This shows the advantage of multi-GPU acceleration of the FDTD method in comparison to the code execution on small CPU based clusters.