next up previous
Next: Implementation Notes Up: CLUSTEREASY: The Parallel Computing Previous: CLUSTEREASY: The Parallel Computing


Obtaining, Compiling, and Running CLUSTEREASY

The file structure of CLUSTEREASY is the same as it is for LATTICEEASY, except for the addition of a file called mpiutil.cpp with functions for communicating between processors. All of the files have been modified in the parallel version, however, so you need to download the CLUSTEREASY files from the LATTICEEASY download page.

As noted above, CLUSTEREASY requires a cluster using MPI. On such a cluster there should typically be a script called mpiCC for compiling MPI C++ programs. The makefile that comes with CLUSTEREASY uses this command, so if your cluster is set up differently you may need to modify it accordingly.

You will also need a freely available set of Fourier Transform routines called FFTW. You can contact your system administrator if this is not already installed on your cluster. CLUSTEREASY calls FFTW for Fourier Transforms in 2D and 3D. FFTW doesn't have a 1D parallel FFT, so for 1D CLUSTEREASY uses a parallelized version of the FFTEASY routines that come with LATTICEEASY. Note that the parallel FFTEASY routine does not scale well, so large runs in 1D may become very slow.

Note that FFTW can be installed in three different ways: single-precision, double-precision, or both. When it is installed for both, all compiler flags for FFTW must include a letter ``s'' or ``d'' that specifies whether single or double precision is being used. By default CLUSTEREASY assumes that FFTW is installed for single and double precision. If it is installed for single precision only on your system you should remove the letter ``s'' from all the FFTW compiler flags in the makefile and FFTW header names in latticeeasy.h.

If FFTW is installed for both precisions and you wish to do runs with double precision, you need to make the following changes

  1. Add the line #define float double to latticeeasy.h and ffteasy.cpp, just below the list of header files.
  2. Change the letter ``s'' to ``d'' in the two FFTW header file names in latticeeasy.h.
  3. Change the letter ``s'' to ``d'' in the FFTW compiler flags in makefile.

If you have mpiCC and FFTW set up properly on your system then you should be able to compile CLUSTEREASY simply by typing ``make.'' Note that you do not specify the number of processors when you compile the program, but rather in the command line when you run it. Your cluster should have instructions for executing MPI programs, but on most systems the command is
mpirun -np <number of processors> latticeeasy

Note that more processors doesn't always mean faster performance. If you use too many processors the program will spend more time communicating between processors than evolving the fields. You may have to do some trial and error for your particular problems, but a good rule of thumb is that you probably won't get much benefit from using more processors than $N/4$, where $N$ is the number of gridpoints along an edge. Also, you will get slightly better performance per processor if the number of processors is a factor of $N$ so that the processors can divide the lattice up evenly.


next up previous
Next: Implementation Notes Up: CLUSTEREASY: The Parallel Computing Previous: CLUSTEREASY: The Parallel Computing

Go to The LATTICEEASY Home Page
Go to Gary Felder's Home Page
Send email to Gary Felder at gfelder@email.smith.edu
Send email to Igor Tkachev at Igor.Tkachev@cern.ch

This documentation was generated on 2008-01-21