Skip to main content

Installation

PALADIN is distributed by Fraunhofer IST. A site that received the source builds it on the machine or cluster it will run on; there is no package manager entry and no container image.

Requirements​

ComponentVersion / note
Linux64-bit, x86-64
C++ compilerC++14, used through mpic++
MPIOpenMPI 5 by default, any MPI with mpic++, mpiexec and MPMD support otherwise
GNU Makefor the build targets below
Python3.x with development headers (Python.h) and a matching python3-config

MPMD support matters: a run starts one MPI rank as the Python server process and the remaining ranks as start_client. Slurm handles that natively; a bare mpiexec needs the colon syntax shown on the scenario page.

Note: the Python that runs the scenario is not the Python that was linked at build time. The extension is loaded at run time from libs/, so the interpreter only has to match the ABI of the headers used for the build.

Build​

All settings are Make variables, so nothing has to be edited for a normal build:

make MPIPATH=/opt/openmpi-5 PYTHON_CONFIG_PATH=/usr/bin/python3-config -j
VariableDefaultMeaning
MPIPATH/opt/openmpi-5Root of the MPI installation; mpic++ is taken from $MPIPATH/bin
PYTHON_CONFIG_PATH/usr/bin/python3.11-configScript that supplies the Python include flags
DESTINATION.Install root. With a path, the tree is created there and the Python part is copied into it
DEBUGunsetDEBUG=1 builds with -O0 -g instead of -O3

Targets:

TargetResult
makeLibraries, start_client, signal_helper, and the Python part in the install root
make fullThe same plus the example cases
make cleanRemoves objects, libraries and executables
make clean_examplesRemoves the example directory of a full build
make clean_fullBoth of the above

What lands in the install root:

PathContent
libs/pyPALADIN.soThe Python extension module: all classes, all constants
start_clientThe MPI client binary that advances the particles
signal_helperSends control commands (pause, unpause, dump, shutdown, kill) to a running job
python_scripts/pyPALADINServer.py (the server class), constants.py (units and physical constants), console_operations.py
submit_pySlurm submit wrapper
examples/The four example cases, only after make full

Environment​

Three variables have to be set in the shell, or inside the job script. PALADINPATH is the install root, the directory that contains libs/ and python_scripts/:

export PALADINPATH=/path/to/paladin
export PYTHONPATH=$PALADINPATH/libs:$PALADINPATH/python_scripts:$PYTHONPATH
export LD_LIBRARY_PATH=$PALADINPATH/libs:$LD_LIBRARY_PATH

PYTHONPATH makes import pyPALADIN and import pyPALADINServer work from anywhere; LD_LIBRARY_PATH is needed by start_client for the shared libraries in libs/. On a cluster, submit_py sets all three itself for the batch job from PALADINPATH and PYTHONPATH, so the interactive shell does not have to be prepared.

Warning: submit_py has the MPI root compiled in as /opt/openmpi-5 and starts mpiexec from there. If your MPI lives elsewhere, edit the MPIPATH line at the top of the script — the environment variable does not override it.

Verify the installation​

python3 -c "import pyPALADIN, pyPALADINServer; print('import ok')"

If the import fails with a message about a missing shared object, LD_LIBRARY_PATH is not set. If it fails with an undefined symbol from MPI, the extension was built against a different MPI than the one on the path at run time.

A full check is one of the example cases: run it, and look at the header of the run output — it prints the PALADIN version and the run name, and the run directory appears next to the script.

Updates​

There is no in-place update mechanism. A new version replaces libs/, start_client, python_scripts/ and submit_py; scenarios that only use documented parameters keep working, but check the parameters page for renamed or added keys before reusing an old script.