Ceci est une ancienne révision du document !
LAMMPS Molecular Dynamics Simulator
from wikipedia
LAMMPS ("Large-scale Atomic/Molecular Massively Parallel Simulator") is a molecular dynamics program from Sandia National Laboratories. LAMMPS makes use of MPI for parallel communication and is a free open-source code, distributed under the terms of the GNU General Public License.
LAMMPS was originally developed under a Cooperative Research and Development Agreement (CRADA) between two laboratories from United States Department of Energy and three other laboratories from private sector firms.[1] It is currently maintained and distributed by researchers at the Sandia National Laboratories.
For computational efficiency LAMMPS uses neighbor lists to keep track of nearby particles. The lists are optimized for systems with particles that are repulsive at short distances, so that the local density of particles never becomes too large.
On parallel computers, LAMMPS uses spatial-decomposition techniques to partition the simulation domain into small 3d sub-domains, one of which is assigned to each processor. Processors communicate and store "ghost" atom information for atoms that border their sub-domain. LAMMPS is most efficient (in a parallel computing sense) for systems whose particles fill a 3D rectangular box with approximately uniform density.
Versions installées
- Une ancienne version, version MPI, est installée sur le cluster Mesocomte
$ module avail lammps lammps/svn
- Une version OpenMP, GPU et Xeon PHI sont installées sur le cluster Lumière
$ module avail lammps lammps/20130930 lammps/20141024
Exécution LAMMPS
Version MPI
A exécuter sur le cluster Mésocomte (mesocluster.univ-fcomte.fr)
- LAMMPS_MPI.sge
#!/bin/bash -l #$ -q normal15d #$ -V #$ -N LAMMPS_MPI #$ -pe impi_tight 16 #$ -o $JOB_NAME.$JOB_ID.out module load lammps ## important : Toujours OMP_NUM_THREADS=1 export OMP_NUM_THREADS=1 ## lancement de l'application mpirun -np $NSLOTS lmp_meso -in inputData > output
Version OpenMP
A exécuter sur le cluster Lumière
- lammps_omp.sge
#!/bin/bash -l #$ -q all.q #$ -V #$ -N LAMMPS_OMP #$ -openmp 8 #$ -o $JOB_NAME.$JOB_ID.out #$ -cwd export OMP_NUM_THREAD=$NSLOTS module load lammps ## lancement de l'application lmp_meso -in inputData > output
Version CUDA
- laamps_cuda.sge
#!/bin/bash -l #$ -q tesla.q #$ -l gpu=1,k40 #$ -V #$ -N LAMMPS_MPI_CUDA #$ -pe impi_tight 2 #$ -o $JOB_NAME.$JOB_ID.out module load cuda module load lammps ## important : Toujours OMP_NUM_THREADS=1 export OMP_NUM_THREADS=1 ## lancement de l'application mpirun -np $NSLOTS lmp_meso-cuda -sf cuda -in inputData > output
Version Xeon Phi
- laamps_phi.sge
#!/bin/bash -l #$ -q xphi.q #$ -V #$ -N LAMMPS_MPI_PHI #$ -pe impi_tight 2 #$ -o $JOB_NAME.$JOB_ID.out module load intel/mic module load lammps ## important : Toujours OMP_NUM_THREADS=1 export OMP_NUM_THREADS=1 ## lancement de l'application ## lance un process MPI sur le noeud et 240 threads sur la carte mpirun -np $NSLOTS lmp_meso_phi -sf intel -pk intel 1 omp 1 -in inputData > output
Gains en performance sur GPU ou Xeon Phi
Toutes les opérations LAMMPS ne sont pas encore implémentées sur ces supports : La page http://lammps.sandia.gov/doc/Section_commands.html#comm propose une liste des commandes supportées par les différentes plates-formes d'exécution.
A titre indicatif, voici les accélérations obtenues sur deux exemples de modèles :
| Modèle | Support | Temps d'exécution | Gain |
|---|---|---|---|
| A | OpenMP (16 coeurs) | 4 h 43 | - |
| A | Xeon Phi | 2 h 13 | x2 |
| B | OpenMP (16 coeurs) | 0 h 43 | - |
| B | NVIDIA Kepler | 0 h 13 | x3 |