2244
Comment:
|
2193
|
Deletions are marked like this. | Additions are marked like this. |
Line 6: | Line 6: |
Since the upgrade to SL5, both clusters use the openmpi implementation of the MPI standard. | |
Line 7: | Line 8: |
=== Infiniband === Applications for the cluster must be compiled on a 64 bit machine, at the moment, this means either lx64 or linfini. There are MPI versions for the GCC, Intel and PGI compilers installed: |
There are MPI versions for the GCC, Intel and PGI compilers installed: |
Line 10: | Line 10: |
/usr/local/ibgd/mpi/osu/gcc/mvapich-0.9.5/bin/mpicc | /opt/openmpi/gcc/bin/mpicc |
Line 12: | Line 12: |
/usr/local/ibgd/mpi/osu/intel/mvapich-0.9.5/bin/mpicc | /opt/openmpi/intel/bin/mpicc |
Line 14: | Line 14: |
/usr/local/ibgd/mpi/osu/pgi/mvapich-0.9.5/bin/mpicc |
/opt/openmpi/pgi/bin/mpicc |
Line 19: | Line 18: |
=== Infiniband === Applications for the plejade cluster must be compiled on a 64 bit SL5 machine, e.g. sl5-64.ifh.de. |
|
Line 20: | Line 23: |
Applications for this cluster can be compiled on and 32 bit machine. There are MPI versions for the GCC, Intel and PGI compilers installed: /opt/mpich/gcc/bin/mpicc /opt/mpich/intel/bin/mpicc /opt/mpich/pgi/bin/mpicc Compilers for C++ and Fortran are available as well. |
Applications for the geminide cluster must be compiled on a 32 bit SL5 machine, e.g. sl5.ifh.de. |
Line 42: | Line 37: |
The Opteron machines have 3.3G of RAM and by default two jobs are executed on one node, so the maximal amount of memory is 1650M per process. | The Opteron machines have 3.3G of RAM and by default two jobs are executed on one node, so the maximal amount of memory is 1650M per process: #$ -l h_vmem=1650M |
Line 45: | Line 42: |
If your application is using threads, it is recommended to set the value of h_stack (by default the same as h_vmem) to a sane value, e.g. 10M. Since SL5 is not the default OS on the batch farm yet, you must add this limit as well: #$ -l os=sl5 Be sure to call the right mpirun version for your compiler. If you application was compiled with GCC, use /opt/openmpi/gcc/bin/mpirun -np $NSLOTS yourapp |
|
Line 50: | Line 58: |
Be aware that the batch system renews the AFS token, but only on the node that starts the first process (node 0). That's why you should access the AFS from that node. An example scenario looks like this: | == Further documentation == |
Line 52: | Line 60: |
1. Copy data from AFS to node 0. 1. Copy it with scp to the nodes that need it to the directory $TMPDIR, the machine names are in $TMPDIR/machines 1. Run your MPI job. 1. Copy the results with scp from the local discs to node 0. 1. Copy the data from node 0 to AFS. |
[http://www-zeuthen.desy.de/technisches_seminar/texte/Technisches_Seminar_Waschk.pdf HPC-Clusters at DESY Zeuthen] , 11/22/06, technical seminar |
Usage of the Linux Clusters at DESY Zeuthen
At Zeuthen, two clusters are available, one with 16 dual Opteron machines connected by Infiniband and one with 8 dual Xeons and Myrinet. They are integrated into the SGE batch system. The documentation in ["Batch System Usage"] applies to them.
Building Applications
Since the upgrade to SL5, both clusters use the openmpi implementation of the MPI standard.
There are MPI versions for the GCC, Intel and PGI compilers installed:
/opt/openmpi/gcc/bin/mpicc
/opt/openmpi/intel/bin/mpicc
/opt/openmpi/pgi/bin/mpicc
Compilers for C++ and Fortran are available as well.
Infiniband
Applications for the plejade cluster must be compiled on a 64 bit SL5 machine, e.g. sl5-64.ifh.de.
Myrinet
Applications for the geminide cluster must be compiled on a 32 bit SL5 machine, e.g. sl5.ifh.de.
Batch System Access
A job script designated for a parallel job needs to specify the parallel environment and the number of required CPUs. The parameter looks like this for the Infiniband cluster:
#$ -pe mpich-ppn2 4
On the Myrinet cluster, it is similar:
#$ -pe mpichgm-ppn2 4
It is important to request the right limit for memory with the parameter h_vmem.
The Opteron machines have 3.3G of RAM and by default two jobs are executed on one node, so the maximal amount of memory is 1650M per process:
#$ -l h_vmem=1650M
The Xeons have 922.5M of RAM.
If your application is using threads, it is recommended to set the value of h_stack (by default the same as h_vmem) to a sane value, e.g. 10M.
Since SL5 is not the default OS on the batch farm yet, you must add this limit as well:
#$ -l os=sl5
Be sure to call the right mpirun version for your compiler. If you application was compiled with GCC, use
/opt/openmpi/gcc/bin/mpirun -np $NSLOTS yourapp
AFS Access
The application binary must be available to all nodes, that's why it should be placed in an AFS directory.
Further documentation
[http://www-zeuthen.desy.de/technisches_seminar/texte/Technisches_Seminar_Waschk.pdf HPC-Clusters at DESY Zeuthen] , 11/22/06, technical seminar