Cluster Job Example: C/C++/FORTRAN Code

C/C++/FORTRAN (Traditional non-parallel code)

Programs compiled with the GNU compiler tools (gcc, g++, g77) will only run on a single CPU by default. Programs compiled with the Portland Group's compilers (pgcc, pgCC, pgf77, pgf90, pgf95, and pghpf) can run on multiple processors if the -mautopar flag was used during the compilation and the -np flag is used when running the executable, but only if those processors are all on a single multi-CPU system (see the Unix compiler documentation). C, C++, and FORTRAN programs would have to be modified to use parallel libraries such as MPI or OpenMP before they can spread across multiple systems. Therefore, use either "#PBS -l nodes=1:ppn=1" for a single-CPU job, "#PBS -l nodes=1:ppn=2" for a dual-CPU job, or "#PBS -l nodes=1:ppn=4" for a 4-CPU job.

Sample program file (fpi-serial.f)

c Copyright (c) 2001-2003 The Trustees of Indiana University.
c                         All rights reserved.
c Copyright (c) 1998-2001 University of Notre Dame.
c                         All rights reserved.
c Copyright (c) 1994-1998 The Ohio State University.
c                         All rights reserved.
c
c This file is part of the LAM/MPI software package.  For license
c information, see the LICENSE file in the top level directory of the
c LAM/MPI source distribution.
c
c $Id: fpi.f,v 1.4 2003/03/25 20:13:27 jsquyres Exp $
c
c Portions taken from the MPICH distribution example fpi.f.
c
c Example program to calculate the value of pi by integrating f(x) =
c 4 / (1 + x^2).
c
c Compile me with 'g77 -o fpi-serial fpi-serial.f'
      program main
      double precision  PI25DT
      parameter        (PI25DT = 3.141592653589793238462643d0)
      double precision  mypi, pi, h, sum, x, f, a
      integer*4 num_iters
      integer rank, size, i, rc
      character junk
c     Loop until finished
      num_iters = 200000
c     Calculate the interval size
      h = 1.0d0 / num_iters
      sum  = 0.0d0
      do 10 i = 1, num_iters
         x = h * (dble(i) - 0.5d0)
         sum = sum + 4.d0 / (1.d0 + x * x)
 10   continue
      mypi = h * sum
c     All finished
      write(6, 97) num_iters, mypi, abs(mypi - PI25DT)
 97   format(i20, ' points: pi is approximately: ', F18.16,
     +     ' error is: ', F18.16)
      stop
      end

Sample qsub command file (fpi-serial.sh)

#!/bin/sh
# Calculate pi by integrating f(x)=4/(1+x^2)

# Request 1 node with 1 free CPU core, since this code uses no parallel features
# by default. If Portland compilers are used with the -mautopar flag, change the
# ppn value to something between 2 and 8, and adjust the last line of this job file
# accordingly.
#PBS -l nodes=1:ppn=1

# Reserve 24 hours on selected cores
#PBS -l walltime=24:00:00

# Give the job a descriptive name for emails (name must start with a letter)
#PBS -N Finding_pi_serial

# Send mail to address given below when the job begins, ends normally, or aborts
#PBS -m bea
#PBS -M myusername@tntech.edu

cd $PBS_O_WORKDIR

# Single-CPU version
./fpi-serial
# Example 4-CPU version. Only works if code was compiled with Portland compilers
# and if you set the ppn value above to 4 -- commented out by default.
# ./fpi-serial -np 4

C/C++/FORTRAN (MPI-compatible parallel code)

If you've gone to the effort of converting your code into a parallel version using the MPI libraries, you can take advantage of multiple processors on several systems. Therefore, you won't be limited to a single node's capacity.

Sample program file (fpi.f)

c Copyright (c) 2001-2003 The Trustees of Indiana University.
c                         All rights reserved.
c Copyright (c) 1998-2001 University of Notre Dame.
c                         All rights reserved.
c Copyright (c) 1994-1998 The Ohio State University.
c                         All rights reserved.
c
c This file is part of the LAM/MPI software package.  For license
c information, see the LICENSE file in the top level directory of the
c LAM/MPI source distribution.
c
c $Id: fpi.f,v 1.4 2003/03/25 20:13:27 jsquyres Exp $
c
c Portions taken from the MPICH distribution example fpi.f.
c
c Example program to calculate the value of pi by integrating f(x) =
c 4 / (1 + x^2).
c
c Compile me with 'mpif77 -o fpi fpi.f'
      program main
      include 'mpif.h'
      double precision  PI25DT
      parameter        (PI25DT = 3.141592653589793238462643d0)
      double precision  mypi, pi, h, sum, x, f, a
      integer num_iters, rank, size, i, rc

c     Function to integrate
      f(a) = 4.d0 / (1.d0 + a * a)

c     Normal MPI startup
      call MPI_INIT(ierr)
      call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
      call MPI_COMM_SIZE(MPI_COMM_WORLD, size, ierr)
      print *, "Process ", rank, " of ", size, " is alive"

c     Loop until finished
      num_iters = 100000
      do 20 iter = 2, num_iters
c     Calculate the interval size
         h = 1.0d0 / iter
         sum  = 0.0d0
         do 10 i = rank + 1, iter, size
            x = h * (dble(i) - 0.5d0)
            sum = sum + f(x)
 10      continue
         mypi = h * sum
c     Collect all the partial sums
         call MPI_REDUCE(mypi, pi, 1, MPI_DOUBLE_PRECISION, MPI_SUM, 0,
     $        MPI_COMM_WORLD, ierr)
 20   continue

c     All finished
c     Node 0 prints the answer.
      if (rank .eq. 0) then
         write(6, 97) iter, pi, abs(pi - PI25DT)
 97      format(i3, ' points: pi is approximately: ', F18.16,
     +        ' error is: ', F18.16)
      endif
      call MPI_FINALIZE(rc)
      stop
      end

Sample qsub command file (fpi.sh)

#!/bin/sh
# Calculate pi by integrating f(x)=4/(1+x^2)

# Request up to 4 nodes with at least 1 free CPU core, since this code uses parallel features.
#PBS -l nodes=4:ppn=1

# Reserve 24 hours on selected cores
#PBS -l walltime=24:00:00

# Give the job a descriptive name for emails (name must start with a letter)
#PBS -N Finding_pi_parallel

# Send mail to address given below when the job begins, ends normally, or aborts
#PBS -m bea
#PBS -M myusername@tntech.edu

cd $PBS_O_WORKDIR
mpiexec -np 4 ./fpi

Sections

Personal tools