Amesos 2.0 Reference Guide

Total Page:16

File Type:pdf, Size:1020Kb

Amesos 2.0 Reference Guide SANDIA REPORT SAND2004-4820 Unlimited Release Printed September 2004 Amesos 2.0 Reference Guide Marzio Sala Prepared by Sandia National Laboratories Albuquerque, New Mexico 87185 and Livermore, California 94550 Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy's National Nuclear Security Administration under Contract DE-AC04-94-AL85000. Approved for public release; further dissemination unlimited. Issued by Sandia National Laboratories, operated for the United States Department of Energy by Sandia Corporation. NOTICE: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government, nor any agency thereof, nor any of their employees, nor any of their contractors, subcontractors, or their employees, make any warranty, express or implied, or assume any legal liability or re- sponsibility for the accuracy, completeness, or usefulness of any information, appara- tus, product, or process disclosed, or represent that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Govern- ment, any agency thereof, or any of their contractors or subcontractors. The views and opinions expressed herein do not necessarily state or reflect those of the United States Government, any agency thereof, or any of their contractors. Printed in the United States of America. This report has been reproduced directly from the best available copy. Available to DOE and DOE contractors from U.S. Department of Energy Office of Scientific and Technical Information P.O. Box 62 Oak Ridge, TN 37831 Telephone: (865) 576-8401 Facsimile: (865) 576-5728 E-Mail: [email protected] Online ordering: http://www.doe.gov/bridge Available to the public from U.S. Department of Commerce National Technical Information Service 5285 Port Royal Rd Springfield, VA 22161 Telephone: (800) 553-6847 Facsimile: (703) 605-6900 E-Mail: [email protected] Online ordering: http://www.ntis.gov/help/ordermethods.asp?loc=7-4-0#online NT O E F E TM N R E A R P G E Y D • • U N A I C T I E R D E S M T A ATES OF ¢¡¤£¦¥¨§ © © § © ¦ !" # § © © Printed %$&(')%* Amesos 2.0 Reference Guide Marzio Sala Computational Math & Algorithms Sandia National Laboratories P.O. Box 5800, MS 1110 Albuquerque, NM 87185-1110 Abstract This document describes the main functionalities of the version 2.0 of the AMESOS pack- age. AMESOS provides an object-oriented interface to several serial and parallel sparse direct solvers libraries for the solution of the linear system of equations +-,/.1032 (1) , where + is a real sparse, distributed matrix, defined as an Epetra RowMatrix object, and 0 and are defined as Epetra MultiVector objects. AMESOS provides a common look-and- feel for all interfaces, and insulates the user from each solver’s details, such as matrix and vector formats, and data distribution. Currently supported libraries are: LAPACK, KLU, UMFPACK, PARDISO, TAUCS, SuperLU, SuperLU DIST, MUMPS, DSCPACK. This document is organized as follows. First, Section 1 introduces the design of AMESOS. Section 2 presents the basic usage of the AMESOS package. Section 3 details how to configure and compile AMESOS. Section 4 describes the interfaces of AMESOSto the supported direct solvers. A brief note on the examples included in the distribution is reported in Section 5. 3 Acknowledgments The author would like to acknowledge the support of the ASCI and LDRD programs that funded development of AMESOS. 4 Amesos 2.0 Reference Guide Contents 1 The Design of AMESOS . 6 2 Basic Usage . 7 2.1 Supported Matrix Formats . 9 2.2 Parameters for All AMESOS Solvers . 9 3 Configuring and Installing AMESOS . 12 4 Supported Solvers . 14 4.1 Interface to LAPACK . 14 4.2 Interface to KLU . 14 4.3 Interface to UMFPACK 4.3 . 16 4.4 Interface to PARDISO 1.2.3 . 16 4.5 Interface to TAUCS 2.2 . 17 4.6 Interface to SuperLU 3.0 . 17 4.7 Interface to SuperLU DIST 2.0 . 17 4.8 Interface to MUMPS 4.3.1 . 19 4.9 Interface to DSCPACK 1.0 . 22 5 Guide to the Examples . 22 5 1 The Design of AMESOS The AMESOS package, developed by (in alphabetical order) T. Davis, M. Heroux, R. Hoekstra, M. Sala, and K. Stanley, is an effort to define a set of object-oriented, abstract interfaces for the usage of serial and parallel sparse direct solvers. Although one serial direct solver, KLU, is distributed within AMESOS, the goal of the AMESOS project is to make it easier to interface a code that makes use of EPETRA objects with direct solver libraries developed outside TRILINOS. AMESOS is written in C++, and has been designed with the following requirements: Simplicity of usage: Solving linear system (1) in a language like MATLAB is very easy, just write X = A \ b. It should not be much more difficult in a C++, production code. Flexibility: More than one algorithm must be available, to offer optimal algorithms for small and large matrices, serial and parallel. Efficiency: The solution of (1) must be as efficient as possible, using state-of-the-art algo- rithms. Besides, the overhead due to the C++ design must be minimal. To fulfill these design requirements, we split the solution of linear system (1) into the following steps: 1. Definition of the sparsity pattern of the linear system matrix; 2. Computation of the symbolic factorization; 3. Definition of the values of the linear system matrix; 4. Computation of the numeric factorization; 5. Definition of the values of the right-hand side; 6. Solution of the linear system. Steps 1, 3 and 5 depend on the matrix and vector format. To increase flexibility, AMESOS requires the matrix to be derived from the Epetra_RowMatrix format, and the solution and right-hand side vector to the defined as Epetra_MultiVector’s. Steps 2, 4 and 6 correspond to three different methods in the AMESOS classes. From an ab- stract point of view, these steps do not depend on the direct solver of choice. However, their concrete implementations does, since different libraries may require different matrix and vector formats and distribution, may have different parameters or different ways of setting the same pa- rameter. To obtain flexibility, AMESOS insulates the user from the details specific to each solver, so that generic methods can be used to manipulate all the supported interface. This design goal is ac- complished using a pure virtual class, which defines methods SymbolicFactorization(), NumericFactorization() and Solve(), plus method SetParameters() which can be used to tune the interface. To increase efficiency, all AMESOS classes are defined as light containers. Each class simply converts the matrix A from the input Epetra_RowMatrix format into the solver’s required format, and sets the parameters are defined by the user. Therefore, AMESOS interfaces are as efficient as the underline solver library. 6 2 Basic Usage A fragment of code using AMESOS is as follows. Let us suppose that A is an Epetra_RowMatrix, and X and B are two Epetra_MultiVector’s. First, we need to include the header files for AMESOS: #include "Amesos.h" #include "Amesos_BaseSolver.h" Note that these header files will not include the header files for the supported libraries (which are of course needed to compile the AMESOS library itself). Then, we need to create an linear problem, as follows: Epetra_LinearProblem Problem(&A, &X, &B); At this point, we can create an AMESOS class using the factory class Amesos: Amesos_BaseSolver* Solver; Amesos Factory; char* SolverType = "Amesos_Klu"; // uses the KLU direct solver Solver = Factory.Create(SolverType, Problem); At this point, we can perform the symbolic factorization of the linear system matrix: AMESOS_CHK_ERR(Solver->SymbolicFactorization()); This phase does not require the numerical values of A, which can therefore be changed after the call to SymbolicFactorization(). However, the nonzero pattern of A cannot be changed. AMESOS_CHK_ERR is a macro (defined in Amesos_ConfigDefs.h) that checks the return code: if not zero, the macro prints out an error message, and returns. The numeric factorization is performed by AMESOS_CHK_ERR(Solver->NumericFactorization()); NumericFactorization() accesses the values of A, but does not consider the vectors X and B. Finally, to solve the linear system, we simply write AMESOS_CHK_ERR(Solver->Solve()); In the previous example, we showed how to use the KLU solver (see Section 4.2 for more details). Other interfaces can be created using the factory class by simply changing one parameter. Note that the supported solver can be serial or parallel, dense or sparse: the user code still remains the same (except for the name of the solver); AMESOS will take care of data redistribution if required by the selected solver. The list of supported solvers is reported in Table 1. Method Factory.Query() can be used to query the factory about the availability of a given solver: char* SolverType = "Amesos_Klu"; bool IsAvailable = Factory.Query(SolverType); 7 Class Communicator Matrix type Interface to Amesos Lapack serial general LAPACK Amesos Klu serial general KLU Amesos Umfpack serial general UMFPACK 4.3 Amesos Pardiso serial/OMP general PARDISO 1.2.3 Amesos Taucs serial symmetric TAUCS 2.2 Amesos Superlu serial general SuperLU 3.0 Amesos Superludist parallel general SuperLU DIST 2.0 Amesos Mumps parallel SPD, sym, general MUMPS 4.3.1 Amesos Dscpack parallel symmetric DSCPACK 1.0 Table 1. Supported interfaces. “serial” means that the supported direct solver is serial. In this case, when solving with more than one proces- sor, the linear problem is gathered to process 0, here solved, then the solution is broadcasted to the distributed solution vector.
Recommended publications
  • MUMPS Users' Guide
    MUltifrontal Massively Parallel Solver (MUMPS 5.4.1) Users’ guide * August 3, 2021 Abstract This document describes the Fortran 95 and C user interfaces to MUMPS5.4.1, a software package to solve sparse systems of linear equations, with many features. For some classes of problems, the complexity of the factorization and the memory footprint of MUMPS can be reduced thanks to the Block Low-Rank (BLR) feature. We describe in detail the data structures, parameters, calling sequences, and error diagnostics. Basic example programs using MUMPS are also provided. Users are allowed to save to disk MUMPS internal data before or after any of the main steps of MUMPS (analysis, factorization, solve) and then restart. *Information on how to obtain updated copies of MUMPS can be obtained from the Web page http://mumps-solver.org/ 1 Contents 1 Introduction 5 2 Notes for users of other versions of MUMPS 6 2.1 ChangeLog........................................7 2.2 Binary compatibility....................................7 2.3 Upgrading between minor releases............................7 2.4 Upgrading from MUMPS 5.3.5 to MUMPS 5.4.1 ...................8 2.5 Upgrading from MUMPS 5.2.1 to MUMPS 5.3.5 ...................8 2.6 Upgrading from MUMPS 5.1.2 to MUMPS 5.2.1 ...................8 2.7 Upgrading from MUMPS 5.0.2 to MUMPS 5.1.2 ...................9 2.7.1 Changes on installation issues...........................9 2.7.2 Selective 64-bit integer feature..........................9 2.8 Upgrading from MUMPS 4.10.0 to MUMPS 5.0.2 .................. 10 2.8.1 Interface with the Metis and ParMetis orderings................
    [Show full text]
  • Arxiv:2105.11220V1 [Cs.DC] 24 May 2021 the Motion of Particles Is Described by a Set of Convection-Diffusion-Reaction Equations
    Towards Parallel CFD computation for the ADAPT framework Imad Kissamiy∗, Christophe Ceriny, Fayssal Benkhaldoun∗, and Gilles Scarella∗ Universit´ede Paris 13 y LIPN, ∗ LAGA, 99, avenue Jean-Baptiste Cl´ement, 93430 Villetaneuse, France [email protected] Abstract. In order to run Computational Fluid Dynamics (CFD) codes on large scale infrastructures, parallel computing has to be used because of the computational intensive nature of the problems. In this paper we investigate the ADAPT platform where we couple flow Partial Differ- ential Equations and a Poisson equation. This leads to a linear system which we solve using direct methods. The implementation deals with the MUMPS parallel multi-frontal direct solver and mesh partitioning meth- ods using METIS to improve the performance of the framework. We also investigate, in this paper, how the mesh partitioning methods are able to optimize the mesh cell distribution for the ADAPT solver. The ex- perience gained in this paper facilitates the move to the 3D version of ADAPT and the move to a Service Oriented view of ADAPT as future work. Keywords: Unstructured mesh, Mesh partitioning, Parallel direct solver, Multi-frontal method, MUMPS, METIS, Multi-physics, Multi-scale and Multilevel Algorithms. 1 Introduction and context of the work In the last recent decades CFD (Computational Fluid Dynamics) used to play an important role in industrial designs, environmental impact assessments and academic studies. The aim of our work is to fully parallelize an unstructured adaptive code for the simulation of 3D streamer propagation in cold plasmas. As a matter of fact, the initial sequential version of the Streamer code needs up to one month for running typical benchmarks.
    [Show full text]
  • Multifrontal Massively Parallel Solver (MUMPS 4.8.4) Users' Guide
    MUltifrontal Massively Parallel Solver (MUMPS 4.8.4) Users’ guide ∗ December 15, 2008 Abstract This document describes the Fortran 90 and C user interface to MUMPS 4.8.4 We describe in detail the data structures, parameters, calling sequences, and error diagnostics. Example programs using MUMPS are also given. A preliminary out-of-core functionality (with limited support) is included in this version. ∗Information on how to obtain updated copies of MUMPS can be obtained from the Web pages http://mumps.enseeiht.fr/ and http://graal.ens-lyon.fr/MUMPS/ 1 Contents 1 Introduction 4 2 Main functionalities of MUMPS 4.8.4 5 2.1 Inputmatrixstructure . ....... 5 2.2 Preprocessing ................................... .... 5 2.3 Post-processingfacilities . .......... 6 2.4 Solvingthetransposedsystem. ......... 6 2.5 Reduce/condense a problem on an interface (Schur complement and reduced/condensed right-handside) .................................... 7 2.6 Arithmeticversions .... ............. ............. ...... 8 2.7 Theworkinghostprocessor . ....... 8 2.8 Sequentialversion..... ............. ............. ...... 8 2.9 Sharedmemoryversion . ..... 8 2.10 Out-of-corefacility . ........ 8 3 Sequence in which routines are called 8 4 Input and output parameters 11 4.1 Versionnumber ................................... 11 4.2 Control of the three main phases: Analysis, Factorization,Solve ............. 11 4.3 Controlofparallelism . ....... 12 4.4 Matrixtype ...................................... 13 4.5 Centralized assembled matrix input: ICNTL(5)=0 and ICNTL(18)=0 .......... 13 4.6 Element matrix input: ICNTL(5)=1 and ICNTL(18)=0 . ............. 14 4.7 Distributed assembled matrix input: ICNTL(5)=0 and ICNTL(18)=0 .......... 14 4.8 Scaling.........................................6 . 15 4.9 Givenordering:ICNTL(7)=1 . ....... 15 4.10 Schur complement with reduced (or condensed) right-hand side: ICNTL(19) and ICNTL(26) .......................................
    [Show full text]
  • Python Interface Release 5.4.1
    Python Interface Release 5.4.1 Luis Saavedra Jul 22, 2021 Contents 1 Introduction 1 2 Installation 3 3 Preliminary 5 4 Python GetFEM interface7 4.1 Introduction........................................7 4.2 Parallel version.......................................7 4.3 Memory Management...................................8 4.4 Documentation.......................................8 4.5 Python GetFEM organization...............................8 5 Examples 11 5.1 A step-by-step basic example............................... 11 5.2 Another Laplacian with exact solution (source term)................... 14 5.3 Linear and non-linear elasticity.............................. 16 5.4 Avoiding the model framework.............................. 19 5.5 Other examples....................................... 20 6 How-tos 23 6.1 Import gmsh mesh..................................... 23 7 API reference 25 7.1 ContStruct......................................... 25 7.2 CvStruct.......................................... 27 7.3 Eltm............................................ 28 7.4 Fem............................................. 28 7.5 GeoTrans.......................................... 31 7.6 GlobalFunction....................................... 32 7.7 Integ............................................ 33 7.8 LevelSet.......................................... 35 7.9 Mesh............................................ 36 7.10 MeshFem.......................................... 44 7.11 MeshIm........................................... 50 7.12 MeshImData.......................................
    [Show full text]
  • DEPARTMENT of INFORMATICS Configuration of a Linear Solver For
    DEPARTMENT OF INFORMATICS TECHNISCHE UNIVERSITÄT MÜNCHEN Master’s Thesis in Computational Science and Engineering Configuration of a linear solver for linearly implicit time integration and efficient data transfer in parallel thermo-hydraulic computations Ravil Dorozhinskii DEPARTMENT OF INFORMATICS TECHNISCHE UNIVERSITÄT MÜNCHEN Master’s Thesis in Computational Science and Engineering Configuration of a linear solver for linearly implicit time integration and efficient data transfer in parallel thermo-hydraulic computations Konfiguration eines Lösers für linear-implizite Zeitintegration und effizienter Datentransfer in parallelen thermohydraulischen Berechnungen Author: Ravil Dorozhinskii Supervisor: Prof. Dr. Thomas Huckle Advisors: Dr. rer. nat. Tim Steinhoff, Dr. rer. nat. Tobias Neckel Submission Date: 07 March 2019 I confirm that this master’s thesis is my own work and I have documented all sources and material used. Munich, 07 March 2019 Ravil Dorozhinskii Abstract Key words. numerical time integration, direct sparse linear methods, multifrontal methods, MUMPS, BLAS, parallel performance, distributed-memory computations, multi-threading, MPI, OpenMP, non-blocking communication An application of linearly implicit methods for time integration of stiff systems of ODEs results in solving sparse systems of linear equations. An optimal selection and configuration of a parallel linear solver can considerably accelerate the time integration process. A comparison of iterative and direct sparse linear solvers has shown that direct ones are the most suitable for this purpose because of their natural robustness to ill-conditioned linear systems that can occur during numerical time integration. Testing of different direct sparse solvers applied to systems generated by ATHLET software has revealed that MUMPS, an implementation of the multifrontal method, performs better than the others in terms of the overall parallel execution time.
    [Show full text]
  • Combinatorial Problems in Solving Linear Systems Iain
    COMBINATORIAL PROBLEMS IN SOLVING LINEAR SYSTEMS IAIN S. DUFF and BORA UC¸AR Technical Report: No: TR/PA/09/60 CERFACS 42 av. Gaspard Coriolis, 31057 Toulouse, Cedex 1, France. Available at http://www.cerfacs.fr/algor/reports/ Date: August 7, 2009. COMBINATORIAL PROBLEMS IN SOLVING LINEAR SYSTEMS∗ IAIN S. DUFFy z AND BORA UC¸ARx Abstract. Numerical linear algebra and combinatorial optimization are vast subjects; as is their interaction. In virtually all cases there should be a notion of sparsity for a combinatorial problem to arise. Sparse matrices therefore form the basis of the interaction of these two seemingly disparate subjects. As the core of many of today's numerical linear algebra computations consists of the solution of sparse linear system by direct or iterative methods, we survey some combinatorial problems, ideas, and algorithms relating to these computations. On the direct methods side, we discuss issues such as matrix ordering; bipartite matching and matrix scaling for better pivoting; task assignment and scheduling for parallel multifrontal solvers. On the iterative method side, we discuss preconditioning techniques including incomplete factorization preconditioners, support graph preconditioners, and algebraic multigrid. In a separate part, we discuss the block triangular form of sparse matrices. Key words. Combinatorial scientific computing, graph theory, combinatorial optimization, sparse matrices, linear system solution 1. Introduction. In this short review paper, we examine the interplay between the solution of sparse linear systems and combinatorics. Most of this strong associa- tion comes from the identification of sparse matrices with graphs so that most algo- rithms dealing with sparse matrices have a close or exact analogue to an algorithm on a graph.
    [Show full text]
  • A Bibliography of Publications of Iain S. Duff
    A Bibliography of Publications of Iain S. Duff Iain S. Duff Rutherford Appleton Laboratory Chilton, Oxon UK Tel: +44-235-445803 FAX: +44-235-446626 E-mail: [email protected] (Internet) 19 March 2021 Version 2.61 Abstract [Duf91b]. 1991 [Ano91, FF93]. 1993 [Duf94a, GW94]. 1994 [MC95]. 1995 This bibliography records publications of Iain [Duf96a, PB96]. 1996 [DGDG97a]. 1997 S. Duff. [Duf98b]. 1999 [Duf00a]. 1st [DFT95, HPP88]. Title word cross-reference 2 [ADD89a, ADD92, DD87a, DD88, DD89, DD90a, DD90b, DD91, Duf88b, DV98a]. 200/VF [DD88, DD89]. 2000 − A 1 [ADLR15]. LU [DD92a, DD90b, DD91]. [ACM00, vdVDE+02]. 2001 QR [ADP94]. LDLT [ADGP07, DHL20]. [Ano01, Duf02a]. 2002 [vdVDE+03]. 2003 LU [DD94a, DD97b]. O(n1=2τ) [DW88]. P 4 [Duf04b]. 2008 [BBD+11, TCJ+10]. 2010 [DAGR87, ADGR90]. P 5 [TBC+11]. 2e [AG80]. 2nd [DAGR87, ADGR90]. QR [ADP96]. [AG80, BBD+11, DR85a]. 1 [DR82a, Duf82b, Duf82c, Duf84g]. 10P 3 [DDP94]. 3090 [ADD89a, ADD92, DD88, DD89]. 11th [ADD89a, ADD92, DD88, DD89, DD90a]. [Duf81i]. 14th [Ame94]. 15th [GW94]. 1978 3090-200 [DD88, DD89]. 3090-200/VF [DS79]. 1979 [GL80]. 1981 [DD88, DD89]. 31-September [ABD+99]. [Hen82, TH82, Wat82]. 1982 [PR83]. 1983 [ESY84, Kow84]. 1984 [DR85a]. 1986 5th [ABD+99, SMMG01, WDPW04]. [IP87, PA88]. 1987 [Duf88c]. 1988 [CRQR89, Duf89f, ES89]. 1989 70th [DW91]. [Ano89, DG89, DMSV90, EJP90]. 1990 1 2 8 [Duf87d, Duf89e, Duf90a]. 80 ALLIANT [DD90a, ADD89a, ADD92, [ADD89a, ADD92, DD90a, DD90b, DD91]. DD90b, DD91, Duf87d, Duf89e, Duf90a]. 818 [DV02a]. 837 [ADD04]. AMD [ADD04]. Analysis [ADLL00, ADLL01c, ADLL01d, ADGS10, ’90 [Sup90, Jap90]. 92-VAPP [BCRT92]. Duf72, Duf81i, Duf86a, Duf88c, Duf89f, ’94 [Ame94, DW94].
    [Show full text]
  • An Interview with IAIN S. DUFF Conducted by Thomas Haigh on 30
    An interview with IAIN S. DUFF Conducted by Thomas Haigh On 30 & 31st August, 2005 In East Hagbourne and Chilton, United Kingdom Interview conducted by the Society for Industrial and Applied Mathematics, as part of grant # DE-FG02-01ER25547 awarded by the US Department of Energy. Transcript and original tapes donated to the Computer History Museum by the Society for Industrial and Applied Mathematics © Computer History Museum Mountain View, California ABSTRACT Iain S. Duff explores the whole of his career to date. Duff grew up in Glasgow, attending The High School before earning a first class degree in Mathematics and Natural Philosophy from the University of Glasgow in 1969. He discusses the mathematical curriculum there, and recounts his early computing experiences as a summer fellow with IBM. Duff won a Carnegie Fellowship for postgraduate work at Oxford, culminating in a D.Phil. conferred in 1972. At Oxford he worked with the numerical analysis group and wrote his thesis on the analysis of sparse systems under the direction of Leslie Fox. While working toward his degree, Duff was already collaborating with Alan Curtis, Michael Powell, and John Reid at the nearby AERE Harwell nuclear power research center. His interest in sparse matrices was cemented by a 1971 visit to IBM’s Yorktown Heights laboratory for a conference, where he first met colleagues such as Alan George, Ralph Willoughby and Fred Gustavson. In 1972, Duff went to the United States on a Harkness postdoctoral fellowship, spending time at SUNY Stony Brook with Reggie Tewarson and Stanford, where he got to know Gene Golub.
    [Show full text]
  • High Performance Computing of Three-Dimensional Finite Element Codes on a 64-Bit Machine
    Journal of Applied Fluid Mechanics, Vol. 5, No. 2, pp. 123-132, 2012. Available online at www.jafmonline.net, ISSN 1735-3572, EISSN 1735-3645. High Performance Computing of Three-Dimensional Finite Element Codes on a 64-bit Machine M.P. Raju1† and S.K. Khaitan2 1 Department of Mechanical Engg., Case Western Reserve University, Cleveland, OH, 44106, USA 2 Department of Electrical Engg., Iowa State University, Ames, IA, 50014, USA †Corresponding Author Email: [email protected] (Received March 8, 2009; accepted March 28, 2009) ABSTRACT Three dimensional Navier-Stokes finite element formulations require huge computational power in terms of memory and CPU time. Recent developments in sparse direct solvers have significantly reduced the memory and computational time of direct solution methods. The objective of this study is twofold. First is to evaluate the performance of various state-of-the-art sequential sparse direct solvers in the context of finite element formulation of fluid flow problems. Second is to examine the merit in upgrading from 32 bit machine to a 64 bit machine with larger RAM capacity in terms of its capacity to solve larger problems. The choice of a direct solver is dependent on its computational time and its in-core memory requirements. Here four different solvers, UMFPACK, MUMPS, HSL_MA78 and PARDISO are compared. The performances of these solvers with respect to the computational time and memory requirements on a 64-bit windows server machine with 16GB RAM is evaluated. Keywords: Multifrontal, UMFPACK, MUMPS, HSL MA78, PARDISO, 64-bit. 1. INTRODUCTION iterative methods like SIMPLE algorithm used in finite volume formulations.
    [Show full text]
  • Short User Documentation Release 4.0
    Short User Documentation Release 4.0 Yves Renard, Julien Pommier March 22, 2010 CONTENTS 1 Introduction 1 2 How to install 3 2.1 Requirements...............................................3 2.2 Download sources............................................3 2.3 Compilling................................................4 3 Linear algebra procedures 7 4 Parallelization of GetFEM++ 9 5 Catch errors 11 6 Build a mesh 13 6.1 Add an element to a mesh........................................ 13 6.2 Remove an element from a mesh.................................... 15 6.3 Simple structured meshes........................................ 15 6.4 Mesh regions............................................... 16 6.5 Methods of the getfem::mesh object................................ 16 6.6 Using dal::bit_vector ....................................... 18 6.7 Face numbering............................................. 19 6.8 Save and load meshes.......................................... 19 7 Build a finite element method on a mesh 23 7.1 First level: manipulating fems on each elements............................ 24 7.2 Examples................................................. 25 7.3 Second level: the optional “vectorization”................................ 25 7.4 Third level: the optional linear transformation (or reduction)...................... 26 7.5 Obtaining generic mesh_fem‘s...................................... 27 7.6 The partial_mesh_fem object...................................... 27 8 Selecting integration methods 29 8.1 Methods of the mesh_im object....................................
    [Show full text]
  • Amesos 2.0 Reference Guide
    SANDIA REPORT SAND2004-XXXX Unlimited Release Printed October 2004 Amesos 2.0 Reference Guide Marzio Sala Prepared by Sandia National Laboratories Albuquerque, New Mexico 87185 and Livermore, California 94550 Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy's National Nuclear Security Administration under Contract DE-AC04-94-AL85000. Approved for public release; further dissemination unlimited. Issued by Sandia National Laboratories, operated for the United States Department of Energy by Sandia Corporation. NOTICE: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government, nor any agency thereof, nor any of their employees, nor any of their contractors, subcontractors, or their employees, make any warranty, express or implied, or assume any legal liability or re- sponsibility for the accuracy, completeness, or usefulness of any information, appara- tus, product, or process disclosed, or represent that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Govern- ment, any agency thereof, or any of their contractors or subcontractors. The views and opinions expressed herein do not necessarily state or reflect those of the United States Government, any agency thereof, or any of their contractors. Printed in the United States of America. This report has been reproduced directly from the best available copy. Available to DOE and DOE contractors from U.S.
    [Show full text]
  • Parallel Computation of Finite Element Navier-Stokes Codes Using MUMPS Solver
    IJCSI International Journal of Computer Science Issues, Vol. 4, No. 2, 2009 20 ISSN (Online): 1694-0784 ISSN (Print): 1694-0814 Parallel Computation of Finite Element Navier-Stokes codes using MUMPS Solver Mandhapati P. Raju 1 Mechanical Engineering, Case Western Reserve University, Cleveland, Ohio 44106 [email protected] Abstract is to use direct solvers in a distributed computing The study deals with the parallelization of 2D and 3D finite environment. element based Navier-Stokes codes using direct solvers. The system of non-linear equations obtained from the Development of sparse direct solvers using multifrontal solvers discretization of Navier-Stokes equations is usually solved has significantly reduced the computational time of direct using a Newton or a Picard algorithm. Newton algorithms solution methods. Although limited by its stringent memory are known for their quadratic convergence behavior. requirements, multifrontal solvers can be computationally efficient. First the performance of MUltifrontal Massively When the initial guess is close to the final solution, Parallel Solver (MUMPS) is evaluated for both 2D and 3D codes Newton achieves quadratic convergence. In this paper, in terms of memory requirements and CPU times. The scalability only the Newton algorithm is used. In using direct solvers, of both Newton and modified Newton algorithms is tested. factorization of the left hand side matrix is the most time Key words: finite element, MUMPS solver, distributed consuming step. To avoid factorization during every computing, Newton method. iteration, a modified Newton is used in which the factorization is done only during the first iteration. The left side matrix evaluated for the first iteration is retained and 1.
    [Show full text]