Working on MESU computers from ISCD
Computers are managed by ISCD (Institut des Sciences du Calcul et des Données — Computing and Data Science Institute). They are on the DSI network of Sorbonne Université.
Informations
- The mesu wiki of ISCD
The mesu forum of ISCD- For any question mesu@upmc.fr
MESU computers, a summary
For more detailed information, please refer to the wiki of course.
Machines
There are 5 interactive machines, 3 standard ones and 2 for GPU, named from mesu0
to mesu4
. One should use alias to connect to them, respectively mesu.dsi.upmc.fr
and mesugpu.dsi.upmc.fr
.
Interactive standard machines are not designed for computation, pleaser use PBS to submit computed oriented jobs.
Physically, there are currently 2 computing machines (refer to wiki for updated informations) :
-
mesu-alpha
, 1024 cores, a 16 TB shared memory machine -
mesu-beta
, 2020 cores, a 21 TB distributed memory machine, which owns 4 (four) K5200 GPU boards.
Except for GPU oriented jobs, computing queues are available from any interactive front machine.
Tools
The machine is using modules for each user session configuration. To list available modules :
module avail
Practical use for our LPNHE applications\
Compilers
One can use GNU or Intel compilers. The GNU compiler is available without any particular configuration. For the Intel compiler, one should do for example :
module load intel-compilers-13
From this point, let's usefully define following variables :
#COMPILO=gcc4.3.4
COMPILO=icc13.0.1
SOFTDIR=$HOME/soft
Example of command line options for compiling (to be adapted according the needs of course) :
- OpenMP - GNU :
g++ `root-config --libs --cflags ` -I$LIBMEMTK/inc -L$LIBMEMTK/lib/$ARCH_TYPE
-lMEM -lEG -fopenmp go_openmp.cxx
- MPI - GNU :
module load openmpi
mpicxx `root-config --libs --cflags ` -I$LIBMEMTK/inc -L$LIBMEMTK/lib/$ARCH_T
YPE -lMEM -lEG go_mpi.cxx
- MPI - Intel :
module load intel-compilers-13
module load mpt
icpc -wd858 -O3 -Wall -fPIC -ipo -xavx `root-config --libs --cflags ` -I$LIB
MEMTK/inc -L$LIBMEMTK/lib/$ARCH_TYPE -lMEM -lEG -lmpi -L$HOME/soft/Cuba3.2-icc13
.0.1/lib -lcuba go_mpi.cxx
For the code profiling, one should use SGI tools like perfsuite
, available in modules.
CVMFS
The cvmfs
tool is installed on Mesu machines. On the other hand, nowadays, it does not seem to be fully functional.
(to be completed)
ROOT
The [root
](https://root# --- CUBA dep.cern.ch/) tool is not installed. It should be available through cvmfs
; if not, one will find a local version at /home/beau/soft/root5.34.10-gcc4.3.4
and root5.34.10-icc13.0.1
, respectively for gcc4.3.4
and icc13.0.1
compilers.
ROOTVERSION=root5.34.10
. $SOFTDIR/$ROOTVERSION-$COMPILO/bin/thisroot.sh
Other tools
LHAVERSION=lhapdf5.8.8
export LHAPDF=$SOFTDIR/$LHAVERSION-$COMPILO
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:`$SOFTDIR/$LHAVERSION-$COMPILO/bin/lhapdf-config --libdir`
export LHAPATH=$SOFTDIR/share/lhapdf/PDFsets
CUBAVERSION=Cuba3.2
export CUBALIB=$SOFTDIR/$CUBAVERSION-$COMPILO
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUBALIB/lib
export CUBACORES=0 # i.e. all cores
-
Task Spooler
: available binary at$SOFTDIR/bin
.