Commits

Peter Lichtner  committed 9269db3

Updated QG for parallel io.

  • Participants
  • Parent commits 13ee7e1

Comments (0)

Files changed (1)

File docs/user_manual/installation.tex

 
 Besides \href{http://www.mcs.anl.gov/petsc/}{PETSc}, third party libraries commonly installed are
 \begin{itemize}
-\item MPI, message passing interface [\href{http://www.open-mpi.org/}{OpenMPI} or \href{http://www.mpich.org/}{MPICH}, note for parallel IO MPICH is required (see Section~\S\ref{parallelio})]
+\item MPI, message passing interface (\href{http://www.open-mpi.org/}{OpenMPI} or \href{http://www.mpich.org/}{MPICH})
 \item \href{http://www.hdfgroup.org/HDF5/}{HDF5}, required for parallel I/O and reading HDF5-formatted input files
 \item \href{http://glaros.dtc.umn.edu/gkhome/views/metis}{Metis} and \href{http://glaros.dtc.umn.edu/gkhome/metis/parmetis/overview}{ParMetis}, graph partitioning libraries required for 
       unstructured grids
 
 The installation of PFLOTRAN on MacOSX requires compiler versions 4.7 or later for gfortran and gcc to be compatible with Fortran 2003. 
 
+For MacOSX compilers can be obtained from several sources including \href{http://www.macports.org/}{MacPorts} and the web site \href{http://hpc.sourceforge.net/}{High Performance Computing for MacOSX}.
+
 \begin{comment}
 \subsection{MacOSX}
 
 \subsection{Parallel I/O using Sarat Sreepathi's PARALLELIO\_LIB with PFLOTRAN}
 \label{parallelio}
 
-The parallel I/O library enables a scalable general purpose parallel I/O capability for HPC by taking advantage of existing parallel I/O libraries, such as HDF5 which are being widely used by scientific applications, and modifying these algorithms to better scale to larger number of processors. The library has been tested with MPICH-3.0.2.
-
-{\em Caution:
-use of this library appears to require {\rm MPICH2} and it does not currently work with {\rm OPENMPI}}.
+The parallel I/O library enables a scalable general purpose parallel I/O capability for HPC by taking advantage of existing parallel I/O libraries, such as HDF5 which are being widely used by scientific applications, and modifying these algorithms to better scale to larger number of processors. The library has been tested with MPICH-3.0.2 and OpenMPI-1.6.
 
 \normalsize
-Values for the variables {\tt HDF5\_READ\_GROUP\_SIZE} and {\tt HDF5\_WRITE\_GROUP\_SIZE} must be set in the input file. 
+Values for the variables {\footnotesize\tt HDF5\_READ\_GROUP\_SIZE} and {\footnotesize\tt HDF5\_WRITE\_GROUP\_SIZE} must be set in the input file. 
 Typical values are to set the
 write group size equal to the number of processes on a compute node (typically 16 or 32). 
 %If that still results in a penalty, decrease it further. 
 A much higher read group size is preferred, e.g. use 512 when running on 512 cores so that one process reads the input file and broadcasts relevant sections.
-Put {\tt HDF5\_WRITE\_GROUP\_SIZE} under the {\tt OUTPUT} keyword:
+Put {\footnotesize\tt HDF5\_WRITE\_GROUP\_SIZE} under the {\footnotesize\tt OUTPUT} keyword:
 \footnotesize
 \begin{Verbatim}
 :=========================== output options ===================================
 HDF5_READ_GROUP_SIZE 1024
 \end{Verbatim}
 \normalsize
-Number of MPI tasks should be an exact multiple of {\tt HDF\_READ\_GROUP\_SIZE}.
+Number of MPI tasks should be an exact multiple of {\footnotesize\tt HDF\_READ\_GROUP\_SIZE}.
 
 For more details on the I/O library, please see Appendix A (Figs. A4, A5) in Sarat Sreepathi's <admin@sarats.com> \href{http://www.lib.ncsu.edu/resolver/1840.16/8317}{dissertation}.
 
 
 {\small\tt svn co http://ascem-io.secure-water.org/ascem-io DIRNAME}
 
-where {\tt DIRNAME} is the installation directory (Default: {\tt ascem-io}). 
+where {\footnotesize\tt DIRNAME} is the installation directory (Default: {\footnotesize\tt ascem-io}). 
 
 The username and password are:
 %\href{https://bitbucket.org/pflotran/pflotran-dev/wiki/Documentation/Strata}{Strata}
 
 \item Compile PARALLELIO library:
 
-To compile the parallel library first modify the Makefile by adding the lines below following the {\tt endif} statement:
+To compile the parallel library first modify the Makefile by commenting out
+{\footnotesize\tt MACHINE=cygnus} on line 61, and adding the lines below following the {\footnotesize\tt endif} statement (line 98):
 \footnotesize
 \begin{Verbatim}
 ifeq ($(MACHINE),jaguarpf)
 {\tt make} \hfill {\it (compile with {\tt mpicc})}
 \normalsize
 
-The will build the library {\footnotesize\tt libparallelio.a}.
+This will build the library {\footnotesize\tt libparallelio.a}.
 
 \item Compile PFLOTRAN: