Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 11 Next »

Modules required

  1. lang/Perl/5.28.1-GCCcore-6.3.0
  2. data/netCDF-Fortran/4.4.5-intel-2018.5.274
  3. data/netCDF/4.6.2-intel-2018.5.274
  4. toolchain/intel/2018.5.274
  5. devel/CMake/3.12.1-intel-2018.5.274

Prerequisite steps

Install Required Perl Module 

[user@login001 ~]$ srun -p sandbox --mem=6G -c 2 -t 60 --pty /bin/bash
[user@node-0005 ~]$ module load lang/Perl/5.28.1-GCCcore-6.3.0
[user@node-0005 ~]$ mkdir .perl5
[user@node-0005 ~]$ ln -s /mnt/home/noac/${USER}/.perl5 perl5  
[user@node-0005 ~]$ cpan install  XML::LibXML

Add Personal Perl Library Paths to .bash_profile

#!/bin/bash
echo -e "\nPATH=\"${HOME}/perl5/bin/"\${PATH:+:\${PATH}}"\"" >> ~/.bash_profile
echo -e "\nPERL5LIB=\"${HOME}/perl5/lib/perl5"\${PERL5LIB:+:\${PERL5LIB}}"\"" >> ~/.bash_profile
echo -e "\nPERL_LOCAL_LIB_ROOT=\"${HOME}/perl5"\${PERL_LOCAL_LIB_ROOT:+:\${PERL_LOCAL_LIB_ROOT}}"\"" >> ~/.bash_profile
echo -e "\nPERL_MB_OPT=\"--install_base \"${HOME}/perl5\"\"" >> ~/.bash_profile
echo -e "\nPERL_MM_OPT=\"INSTALL_BASE=${HOME}/perl5\"" >> ~/.bash_profile
echo -e "\n" >> ~/.bash_profile
echo -e 'export ${PATH}' >> ~/.bash_profile
echo -e 'export ${PERL5LIB}' >> ~/.bash_profile
echo -e 'export ${PERL_LOCAL_LIB_ROOT}' >> ~/.bash_profile
echo -e 'export ${PERL_MB_OPT}' >> ~/.bash_profile
echo -e 'export ${PERL_MM_OPT}' >> ~/.bash_profile



Setup a joined directory of a specific version of NetCDF

Older models, such as CESM 1.x, require that all parts of NetCDF be in a single directory.  Unfortunately, the modules on the cluster do not provide this, so we need to create our own joined directory on the version of NetCDF we want to use.  The script below will; create a  directory called netcdf4 in your home directory that will then be populated with symlinks to the NetCDF libraries.


netcdf4_single_dir.sh 
#!/bin/bash

mkdir -p ~/netcdf4/bin ~/netcdf4/include ~/netcdf4/lib ~/netcdf4/lib/pkg-config

cd ~/netcdf4/bin
find /opt/apps/software/data/PnetCDF/1.9.0-intel-2018.5.274/bin/ -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF-Fortran/4.4.5-intel-2018.5.274/bin -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF/4.6.2-intel-2018.5.274/bin/ -maxdepth 1 -type f -exec ln -s {} . \;

cd ~/netcdf4/include
find /opt/apps/software/data/PnetCDF/1.9.0-intel-2018.5.274/include -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF-Fortran/4.4.5-intel-2018.5.274/include -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF/4.6.2-intel-2018.5.274/include -maxdepth 1 -type f -exec ln -s {} . \;

cd ~/netcdf4/lib
find /opt/apps/software/data/PnetCDF/1.9.0-intel-2018.5.274/lib -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF-Fortran/4.4.5-intel-2018.5.274/lib -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF/4.6.2-intel-2018.5.274/lib64 -maxdepth 1 -type f -exec ln -s {} . \;

cd ~/netcdf4/lib/pkg-config
find /opt/apps/software/data/PnetCDF/1.9.0-intel-2018.5.274/lib/pkg-config -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF-Fortran/4.4.5-intel-2018.5.274/lib/pkg-config -maxdepth 1 -type f -exec ln -s {} . \;
find /opt/apps/software/data/netCDF/4.6.2-intel-2018.5.274/lib64/pkg-config -maxdepth 1 -type f -exec ln -s {} . \;


Additions to the config_machines.xml

 

As Mana has two different Infiniband networks (QDR and HDR), two different machine entries are created.  This also means that subsequent files will also come in duplicate, with only changes in selecting the right network to use.


config_machines.xml 
<machine MACH="uhhpc_qdr">
<DESC>User Defined Machine</DESC> <!-- can be anything -->
<OS>LINUX</OS> <!-- LINUX,Darwin,CNL,AIX,BGL,BGP -->
<COMPILERS>intel</COMPILERS> <!-- intel,ibm,pgi,pathscale,gnu,cray,lahey -->
<MPILIBS>impi</MPILIBS> <!-- openmpi, mpich, ibm, mpi-serial -->
<CESMSCRATCHROOT>~/lus_scratch/cesm/case</CESMSCRATCHROOT> <!-- complete path to the 'scratch' directory -->
<RUNDIR>$CASEROOT/run</RUNDIR> <!-- complete path to the run directory -->
<EXEROOT>$CASEROOT/bld</EXEROOT> <!-- complete path to the build directory -->
<DIN_LOC_ROOT>~/cesm/input</DIN_LOC_ROOT> <!-- complete path to the inputdata directory -->
<DIN_LOC_ROOT_CLMFORC>USERDEFINED_optional_build</DIN_LOC_ROOT_CLMFORC> <!-- path to the optional forcing data for CLM (for CRUNCEP forcing) -->
<DOUT_S>TRUE</DOUT_S> <!-- logical for short term archiving -->
<DOUT_S_ROOT>$CASEROOT/output</DOUT_S_ROOT> <!-- complete path to a short term archiving directory -->
<DOUT_L_MSROOT>USERDEFINED_optional_run</DOUT_L_MSROOT> <!-- complete path to a long term archiving directory -->
<CCSM_BASELINE>USERDEFINED_optional_run</CCSM_BASELINE> <!-- where the cesm testing scripts write and read baseline results -->
<CCSM_CPRNC>USERDEFINED_optional_test</CCSM_CPRNC> <!-- path to the cprnc tool used to compare netcdf history files in testing -->
<BATCHQUERY>squeue -a</BATCHQUERY>
<BATCHSUBMIT>sbatch</BATCHSUBMIT>
<SUPPORTED_BY>uh</SUPPORTED_BY>
<GMAKE_J>8</GMAKE_J>
<MAX_TASKS_PER_NODE>19</MAX_TASKS_PER_NODE>
</machine>


<machine MACH="uhhpc_hdr"> 
<DESC>User Defined Machine</DESC> <!-- can be anything --> 
<OS>LINUX</OS> <!-- LINUX,Darwin,CNL,AIX,BGL,BGP --> 
<COMPILERS>intel</COMPILERS> <!-- intel,ibm,pgi,pathscale,gnu,cray,lahey --> 
<MPILIBS>impi</MPILIBS> <!-- openmpi, mpich, ibm, mpi-serial -->
<CESMSCRATCHROOT>~/lus_scratch/cesm/case</CESMSCRATCHROOT> <!-- complete path to the 'scratch' directory -->
<RUNDIR>$CASEROOT/run</RUNDIR> <!-- complete path to the run directory --> 
<EXEROOT>$CASEROOT/bld</EXEROOT> <!-- complete path to the build directory --> 
<DIN_LOC_ROOT>~/cesm/input</DIN_LOC_ROOT> <!-- complete path to the inputdata directory --> 
<DIN_LOC_ROOT_CLMFORC>USERDEFINED_optional_build</DIN_LOC_ROOT_CLMFORC> <!-- path to the optional forcing data for CLM (for CRUNCEP forcing) -->
<DOUT_S>TRUE</DOUT_S> <!-- logical for short term archiving --> 
<DOUT_S_ROOT>$CASEROOT/output</DOUT_S_ROOT> <!-- complete path to a short term archiving directory -->
<DOUT_L_MSROOT>USERDEFINED_optional_run</DOUT_L_MSROOT> <!-- complete path to a long term archiving directory -->
<CCSM_BASELINE>USERDEFINED_optional_run</CCSM_BASELINE> <!-- where the cesm testing scripts write and read baseline results -->
<CCSM_CPRNC>USERDEFINED_optional_test</CCSM_CPRNC> <!-- path to the cprnc tool used to compare netcdf history files in testing -->
<BATCHQUERY>squeue -a</BATCHQUERY>
<BATCHSUBMIT>sbatch</BATCHSUBMIT>
<SUPPORTED_BY>uh</SUPPORTED_BY>
<GMAKE_J>8</GMAKE_J>
<MAX_TASKS_PER_NODE>39</MAX_TASKS_PER_NODE> 
</machine>

Create env_mach_specific files

A files for the uhhpc_qdr and the uhhpc_hdr machines needs to be made to define the build environment.  These two files should be identical as building for the QDR or the HDR network are identical on Mana.


env_mach_specific.uhhpc_qdr 
#! /bin/csh -f

# -------------------------------------------------------------------------
# UHHPC_QDR build specific settings
# -------------------------------------------------------------------------


source /etc/profile.d/lmod.csh
module purge

module load devel/CMake/3.12.1-intel-2018.5.274
module load lang/Perl/5.28.1-GCCcore-6.3.0
module load data/netCDF-Fortran/4.4.5-intel-2018.5.274
module load data/netCDF/4.6.2-intel-2018.5.274
module load toolchain/intel/2018.5.274

setenv NETCDF ${HOME}/netcdf4
setenv LD_LIBRARY_PATH ${HOME}/netcdf4/lib/:$LD_LIBRARY_PATH
setenv LIBRARY_PATH ${HOME}/netcdf4/lib/:$LIBRARY_PATH
setenv PATH ${HOME}/netcdf4/bin/:$PATH
setenv CPATH ${HOME}/netcdf4/include/:$CPATH

# -------------------------------------------------------------------------
# Build and runtime environment variables - edit before the initial build
# -------------------------------------------------------------------------
limit stacksize unlimited
limit datasize unlimited


env_mach_specific.uhhpc_hdr 
#! /bin/csh -f

# -------------------------------------------------------------------------
# UHHPC_QDR build specific settings
# -------------------------------------------------------------------------


source /etc/profile.d/lmod.csh
module purge

module load devel/CMake/3.12.1-intel-2018.5.274
module load lang/Perl/5.28.1-GCCcore-6.3.0
module load data/netCDF-Fortran/4.4.5-intel-2018.5.274
module load data/netCDF/4.6.2-intel-2018.5.274
module load toolchain/intel/2018.5.274

setenv NETCDF ${HOME}/netcdf4
setenv LD_LIBRARY_PATH ${HOME}/netcdf4/lib/:$LD_LIBRARY_PATH
setenv LIBRARY_PATH ${HOME}/netcdf4/lib/:$LIBRARY_PATH
setenv PATH ${HOME}/netcdf4/bin/:$PATH
setenv CPATH ${HOME}/netcdf4/include/:$CPATH

# -------------------------------------------------------------------------
# Build and runtime environment variables - edit before the initial build
# -------------------------------------------------------------------------
limit stacksize unlimited
limit datasize unlimited


Create the mkbatch files

mkbatch.uhhpc_qdr 
#! /bin/csh -f

#################################################################################
if ($PHASE == set_batch) then
#################################################################################

source ./Tools/ccsm_getenv || exit -1

module load lang/Perl/5.28.1-GCCcore-6.3.0
set ntasks = `${CASEROOT}/Tools/taskmaker.pl -sumonly`
set maxthrds = `${CASEROOT}/Tools/taskmaker.pl -maxthrds`
module purge
@ nodes = $ntasks / ${MAX_TASKS_PER_NODE}
if ( $ntasks % ${MAX_TASKS_PER_NODE} > 0) then
@ nodes = $nodes + 1
@ ntasks = $nodes * ${MAX_TASKS_PER_NODE}
endif
@ taskpernode = ${MAX_TASKS_PER_NODE} / ${maxthrds}
set qname = batch
set tlimit = "3-00:00:00"

if ($?TESTMODE) then
set file = $CASEROOT/${CASE}.test
else
set file = $CASEROOT/${CASE}.run
endif

cat >! $file << EOF1
#!/bin/csh
#SBATCH --job-name=${CASE}
#SBATCH --constraint="ib_qdr"
#SBATCH --distribution="*:*:*"
#SBATCH --partition=exclusive
#SBATCH --time=$tlimit
#SBATCH --job-name=${CASE}
#SBATCH --ntasks=$ntasks
#SBATCH --cpus-per-task=$maxthrds
#SBATCH --output=${CASE}.%A.out


# Configure the Intel MPI parameters
setenv I_MPI_FABRICS "shm:ofi"
setenv I_MPI_PMI_LIBRARY "/lib64/libpmi.so"
# ### FOR QDR NETWORK #####
setenv FI_PROVIDER "psm"
setenv FI_PSM_TAGGED_RMA 0
setenv FI_PSM_AM_MSG 1
setenv FI_PSM_UUID \`uuidgen\`
# # ###### ######## ###### ##
module purge

EOF1

#################################################################################
else if ($PHASE == set_exe) then
#################################################################################
module load lang/Perl/5.28.1-GCCcore-6.3.0
set maxthrds = `${CASEROOT}/Tools/taskmaker.pl -maxthrds`
set maxtasks = `${CASEROOT}/Tools/taskmaker.pl -sumtasks`
module purge


cat >> ${CASEROOT}/${CASE}.run << EOF1
# -------------------------------------------------------------------------
# Run the model
# -------------------------------------------------------------------------

sleep 25
cd \$RUNDIR
echo "\`date\` -- CSM EXECUTION BEGINS HERE"
setenv OMP_NUM_THREADS ${maxthrds}
module load data/netCDF-Fortran/4.4.5-intel-2018.5.274
module load data/netCDF/4.6.2-intel-2018.5.274
module load toolchain/intel/2018.5.274
srun --ntasks=${maxtasks} --cpu_bind=sockets --cpu_bind=verbose --kill-on-bad-exit \$EXEROOT/cesm.exe >&! cesm.log.\$LID
wait
echo "\`date\` -- CSM EXECUTION HAS FINISHED"

EOF1


#################################################################################
else if ($PHASE == set_larch) then
#################################################################################

#This is a place holder for a long-term archiving script

#################################################################################
else
#################################################################################

echo " PHASE setting of $PHASE is not an accepted value"
echo " accepted values are set_batch, set_exe and set_larch"
exit 1

#################################################################################
endif
#################################################################################


mkbatch.uhhpc_hdr 
#! /bin/csh -f

#################################################################################
if ($PHASE == set_batch) then
#################################################################################

source ./Tools/ccsm_getenv || exit -1
maxtasks
module load lang/Perl/5.28.1-GCCcore-6.3.0
set ntasks = `${CASEROOT}/Tools/taskmaker.pl -sumonly`
set maxthrds = `${CASEROOT}/Tools/taskmaker.pl -maxthrds`
module purge
@ nodes = $ntasks / ${MAX_TASKS_PER_NODE}
if ( $ntasks % ${MAX_TASKS_PER_NODE} > 0) then
@ nodes = $nodes + 1
@ ntasks = $nodes * ${MAX_TASKS_PER_NODE}
endif
@ taskpernode = ${MAX_TASKS_PER_NODE} / ${maxthrds}
set qname = batch
set tlimit = "3-00:00:00"

if ($?TESTMODE) then
set file = $CASEROOT/${CASE}.test
else
set file = $CASEROOT/${CASE}.run
endif

cat >! $file << EOF1
#!/bin/csh
#SBATCH --job-name=${CASE}
#SBATCH --constraint="ib_hdr"
#SBATCH --distribution="*:*:*"
#SBATCH --partition=exclusive
#SBATCH --time=$tlimit
#SBATCH --job-name=${CASE}
#SBATCH --ntasks=$ntasks
#SBATCH --cpus-per-task=$maxthrds
#SBATCH --output=${CASE}.%A.out


###### ######## ###### ##
# Libfabric method
###### ######## ###### ##
# Configure the Intel MPI parameters
setenv I_MPI_FABRICS "shm:ofi"
setenv I_MPI_PMI_LIBRARY "/lib64/libpmi.so"
 setenv I_MPI_HYDRA_TOPOLIB "ipl" # May be required if newer libfabric and intel MPI is used
### FOR HDR NETWORK #####
# https://ofiwg.github.io/libfabric/master/man/
# https://ofiwg.github.io/libfabric/v1.9.1/man/
setenv FI_PROVIDER "shm,verbs;ofi_rxm"
setenv FI_MR_CACHE_MONITOR "disabled" # currently a bug exists that a segfault could happen
setenv FI_VERBS_MR_CACHE_ENABLE "0" # currently a bug exists that a segfault could happen
setenv  FI_VERBS_INLINE_SIZE "256"
setenv FI_UNIVERSE_SIZE "${maxtasks}" # should equal at least the max number of tasks one task will communicate with
setenv FI_VERBS_IFACE "i"
###### ######## ###### ##

###### ######## ###### ##
# DAPL method (deprecated but not gone in Intel 2018)
###### ######## ###### ##
# Configure the Intel MPI parameters

#setenv I_MPI_FABRICS "shm:dapl"
#setenv I_MPI_PMI_LIBRARY "/lib64/libpmi.so"

###### ######## ###### ##

module purge

EOF1

#################################################################################
else if ($PHASE == set_exe) then
#################################################################################
module load lang/Perl/5.28.1-GCCcore-6.3.0
set maxthrds = `${CASEROOT}/Tools/taskmaker.pl -maxthrds`
set maxtasks = `${CASEROOT}/Tools/taskmaker.pl -sumtasks`
module purge


cat >> ${CASEROOT}/${CASE}.run << EOF1
# -------------------------------------------------------------------------
# Run the model
# -------------------------------------------------------------------------

sleep 25
cd \$RUNDIR
setenv OMP_NUM_THREADS ${maxthrds}
module load data/netCDF-Fortran/4.4.5-intel-2018.5.274
module load data/netCDF/4.6.2-intel-2018.5.274
module load toolchain/intel/2018.5.274
echo "\`date\` -- CSM EXECUTION BEGINS HERE" 
srun --ntasks=${maxtasks} --cpu_bind=sockets --cpu_bind=verbose --kill-on-bad-exit \$EXEROOT/cesm.exe >&! cesm.log.\$LID
wait
echo "\`date\` -- CSM EXECUTION HAS FINISHED"

EOF1


#################################################################################
else if ($PHASE == set_larch) then
#################################################################################

#This is a place holder for a long-term archiving script

#################################################################################
else
#################################################################################

echo " PHASE setting of $PHASE is not an accepted value"
echo " accepted values are set_batch, set_exe and set_larch"
exit 1

#################################################################################
endif
#################################################################################






Need to create and use machine scripts specific for Mana:

/cesm1_2_2/scripts/ccsm_utils/Machines/


env_mach_specific.uhhpc

mkbatch.uhhpc


  • No labels