Performance of a Massively Parallel Method of Moment Solver and Its Application

─ A massively parallel Method of Moment (MoM) solver able to run on 200,000 CPU cores and solve matrices larger than 1.3 million unknowns is presented. The solver implements a novel LU decomposition algorithm based on the Communication Avoiding LU (CALU) scheme. By using a new pivoting policy, the communication between processes is improved enhancing the parallel speed up of the algorithm. Solver effectiveness and performance are demonstrated comparing the results with two of the most important math libraries used by direct dense solvers: the commercial MKL and the open source ScaLapack. Results show how simulation time is reduced significantly thanks to this novel LU decomposition algorithm making possible the simulation of incredibility electrically large problems using MoM. Index Terms ─ Communication avoiding, high performance, LU decomposition, massively parallel, method of moments.

[1]  Allen Taflove,et al.  Computational Electrodynamics the Finite-Difference Time-Domain Method , 1995 .

[2]  James Demmel,et al.  ScaLAPACK: A Portable Linear Algebra Library for Distributed Memory Computers - Design Issues and Performance , 1995, Proceedings of the 1996 ACM/IEEE Conference on Supercomputing.

[3]  Jiming Song,et al.  Multilevel fast multipole algorithm for electromagnetic scattering by large complex objects , 1997 .

[4]  Carretera de Valencia,et al.  The finite element method in electromagnetics , 2000 .

[5]  Barbara Chapman,et al.  Using OpenMP - portable shared memory parallel programming , 2007, Scientific and engineering computation.

[6]  T.K. Sarkar,et al.  Solving MoM problems with million level unknowns using a parallel out-of-core solver on a high performance cluster , 2009, 2009 IEEE Antennas and Propagation Society International Symposium.

[7]  Robert A. van de Geijn,et al.  Parallel Solution of Integral Equation-Based EM Problems in the Frequency Domain , 2009 .

[8]  T.K. Sarkar,et al.  Parallel MoM Using Higher-Order Basis Functions and PLAPACK In-Core and Out-of-Core Solvers for Challenging EM Simulations , 2009, IEEE Antennas and Propagation Magazine.

[9]  Fernando Gustavo Tinetti,et al.  Using OpenMP: Portable Shared Memory Parallel Programming , 2010 .

[10]  H. Fangjing,et al.  An Efficient Parallel Multilevel Fast Multipole Algorithm for Large-scale Scattering Problems , 2010 .

[11]  James Demmel,et al.  CALU: A Communication Optimal LU Factorization Algorithm , 2011, SIAM J. Matrix Anal. Appl..

[12]  Tapan K. Sarkar.  Higher Order Basis Based Integral Equation Solver - HOBBIES , 2012 .

[13]  James Demmel,et al.  LU Factorization with Panel Rank Revealing Pivoting and Its Communication Avoiding Version , 2012, SIAM J. Matrix Anal. Appl..

[14]  Tapan K. Sarkar,et al.  Performance of a Massively Parallel Higher-Order Method of Moments Code Using Thousands of CPUs and Its Applications , 2014, IEEE Transactions on Antennas and Propagation.

[15]  Torsten Hoefler,et al.  Using Advanced MPI: Modern Features of the Message-Passing Interface , 2014 .

[16]  M. Zhu,et al.  A Hybrid MoM-PO Method Combining ACA Technique for Electromagnetic Scattering from Target above a Rough Surface , 2014 .

[17]  Ji-Hoon Park,et al.  A Hybrid UTD-ACGF Technique for DOA Finding of Receiving Antenna Array on Complex Environment , 2015, IEEE Transactions on Antennas and Propagation.