|
Size: 1710
Comment:
|
Size: 6037
Comment: LD_LIBRARY_PATH problem
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 2: | Line 2: |
| <<TableOfContents>> | |
| Line 3: | Line 4: |
| At Zeuthen a cluster of 16 dual opteron machines is available. It is integrated into the SGE batch system. The documentation in ["Batch System Usage"] applies to it. | == Introduction == There are 8 dedicated parallel clusters (blade centers) in testing mode, but you can also run parallel MPI jobs in the SGE farm. The documentation in [[Batch_System_Usage]] applies there. For discussions and information regarding the usage of the PAX cluster a mailing list has been introduced: <<MailTo(zn-cluster AT desy DOT de)>>. To get subscribed to that list, send an email to <<MailTo(sympa AT desy DOT de)>> with the subject: '''subscribe zn-cluster''' |
| Line 6: | Line 10: |
| === Openmpi === Since SL5, all batch worker nodes have the openmpi implementation of the MPI standard installed. Recently the machines were upgraded to the default SL5.5 packages of openmpi. For 64 bit applications use the installation in /usr/lib64/openmpi/1.4-gcc/bin, for 32 bit use the binaries from /usr/lib/openmpi/1.4-gcc/bin . |
|
| Line 7: | Line 13: |
| 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: | Additional openmpi versions are installed to support the Intel and PGI compilers: |
| Line 9: | Line 15: |
| /usr/local/ibgd/mpi/osu/gcc/mvapich-0.9.5/bin/mpicc | {{{ /usr/lib64/openmpi/1.4-icc/bin /usr/lib64/openmpi-1.3.2-pgi/bin }}} If you don't want to specify the full path to your preferred MPI implementation, configure a default by using the ini command or running mpi-selector-menu on a build machine. |
| Line 11: | Line 21: |
| /usr/local/ibgd/mpi/osu/intel/mvapich-0.9.5/bin/mpicc | ==== Building applications ==== 64 bit MPI Applications can be compiled on any 64 bit SL5 machine, e.g. sl5-64.ifh.de. |
| Line 13: | Line 24: |
| /usr/local/ibgd/mpi/osu/pgi/mvapich-0.9.5/bin/mpicc | ==== Running your application ==== To run an MPI program outside the batch system, you must specify a machinefile listing all the machines and the number of cores your application should run on. A typical machine file looks like this: |
| Line 15: | Line 27: |
| {{{ pax8c slots=8 pax8d slots=8 pax8e slots=8 pax8f slots=8 }}} The command line would look like this: |
|
| Line 16: | Line 35: |
| Compilers for C++ and Fortran are available as well. | {{{ /usr/lib64/openmpi/1.4-gcc/bin/mpirun -np 32 -machinefile ./machinefile ./program }}} More information on openmpi is in the openmpi FAQ: http://www.open-mpi.org/faq/ |
| Line 18: | Line 40: |
| === Mvapich2 === Two additional mpi implementations are installed on all pax machines: |
|
| Line 19: | Line 43: |
| {{{ /usr/lib64/mvapich2/1.7-gcc/bin /usr/lib64/mvapich2/1.7-intel/bin }}} To use mvapich2, add one of those versions to your path and compile your application with that mpi compiler. To run it outside the batch system, follow these instructions: http://mvapich.cse.ohio-state.edu/overview/mvapich2/ Applications built with mvapich2 will only run on machines with Infiniband hardware, so they will work on the pax machines but not on desktops, workgroup servers or the farm. The machine file format is different from the one for openmpi, you must list the host name for every core you want to use, e.g. if you want to run four processes, two processes on each of pax09 and pax08: {{{ pax88 pax89 pax88 pax89 }}} The preferred way to run a application with mvapich2 is mpiexec. If you want to use the deprecated mpd startup method as opposed to mpirun_rsh, you must also first create the file ~/.mpd.conf that contains of one line like this: {{{ MPD_SECRETWORD=password }}} |
|
| Line 20: | Line 67: |
| /!\ '''ATTENTION''': The PAX cluster was split off the normal Zeuthen batch. To access the PAX batch system you will need to call `ini pax`. | |
| Line 21: | Line 69: |
| 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: | Alternatively source a script: |
| Line 23: | Line 71: |
| #$ -pe mpich-ppn2 4 | * zsh users: {{{ [oreade38] ~ % . /usr/gridengine/pax/common/settings.sh }}} * tcsh users: {{{ [oreade38] ~ $ source /usr/gridengine/pax/common/settings.csh }}} Switching back to use the standard farm works similarly: * zsh users: {{{ [oreade38] ~ % . /usr/gridengine/default/common/settings.sh }}} * tcsh users: {{{ [oreade38] ~ $ source /usr/gridengine/default/common/settings.csh }}} |
| Line 25: | Line 89: |
| It is important to request the right limit for memory with the parameter h_vmem. The machines have 3673204k of RAM and by default two jobs are executed on one node, so the maximal amount of memory is 1650M per process. | 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 up to 8 slots for 8 MPI processes on a single node: |
| Line 27: | Line 91: |
| {{{ #$ -pe multicore-mpi 8 }}} For more MPI processes that have no big communication overhead, use -pe mpi. Be sure to call the right mpirun version for your architecture. If you application was compiled for 64 bit, use {{{ /usr/lib64/openmpi/1.4-gcc/bin/mpirun -np $NSLOTS yourapp }}} The MPI runtime will automatically select the right network type. For more demanding MPI jobs you can select one of the pax blade centers like this in your job script. You can request up to 1024 slots, as a blade center contains 128 CPU cores and the batch system contains 8 blade centers: {{{ #$ -pe pax 128 /usr/lib64/openmpi/1.4-gcc/bin/mpirun -np $NSLOTS yourapp }}} Finally, here's a list of common pitfalls when using the pax batch system: * Please be aware that all requested resources (via the '''-l''' qsub switch) are meant '''per job slot'''. As the pax nodes only provide 24GB (8 core systems -> 3GB per job slot), you cannot request more than 3GB h_vmem in your job scripts. Otherwise your job won't start! * /!\ If your MPI application relies on LD_LIBRARY_PATH to load its shared libraries or modules, this will fail on remote notes, as the batch system will remove this variable from the environment. In that case you'll have to wrap ''yourapp'' in a shell script that sets up the environment and calls your binary application. === Mvapich2 === With mvapich2 1.7, there is working integration into the SGE batch system. Just use a command like this: {{{ #$ -pe pax 128 /usr/lib64/mvapich2/1.7-intel/bin/mpiexec -n $NSLOTS yourapp }}} |
|
| Line 28: | Line 124: |
| Line 31: | Line 126: |
| 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: | == BLAS library == Both ATLAS and GotoBLAS are available. |
| Line 33: | Line 129: |
| 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. |
* ATLAS is in /opt/products/atlas * libgoto is in /usr/lib or /usr/lib64 respectively. == Further documentation == [[http://www-zeuthen.desy.de/technisches_seminar/texte/waschk_20100427.pdf|Paralleles Rechnen in Zeuthen - die neuen Cluster]] , 04/27/10, technical seminar [[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
Contents
Introduction
There are 8 dedicated parallel clusters (blade centers) in testing mode, but you can also run parallel MPI jobs in the SGE farm. The documentation in Batch_System_Usage applies there.
For discussions and information regarding the usage of the PAX cluster a mailing list has been introduced: <zn-cluster AT desy DOT de>. To get subscribed to that list, send an email to <sympa AT desy DOT de> with the subject: subscribe zn-cluster
Building Applications
Openmpi
Since SL5, all batch worker nodes have the openmpi implementation of the MPI standard installed. Recently the machines were upgraded to the default SL5.5 packages of openmpi. For 64 bit applications use the installation in /usr/lib64/openmpi/1.4-gcc/bin, for 32 bit use the binaries from /usr/lib/openmpi/1.4-gcc/bin .
Additional openmpi versions are installed to support the Intel and PGI compilers:
/usr/lib64/openmpi/1.4-icc/bin /usr/lib64/openmpi-1.3.2-pgi/bin
If you don't want to specify the full path to your preferred MPI implementation, configure a default by using the ini command or running mpi-selector-menu on a build machine.
Building applications
64 bit MPI Applications can be compiled on any 64 bit SL5 machine, e.g. sl5-64.ifh.de.
Running your application
To run an MPI program outside the batch system, you must specify a machinefile listing all the machines and the number of cores your application should run on. A typical machine file looks like this:
pax8c slots=8 pax8d slots=8 pax8e slots=8 pax8f slots=8
The command line would look like this:
/usr/lib64/openmpi/1.4-gcc/bin/mpirun -np 32 -machinefile ./machinefile ./program
More information on openmpi is in the openmpi FAQ: http://www.open-mpi.org/faq/
Mvapich2
Two additional mpi implementations are installed on all pax machines:
/usr/lib64/mvapich2/1.7-gcc/bin /usr/lib64/mvapich2/1.7-intel/bin
To use mvapich2, add one of those versions to your path and compile your application with that mpi compiler. To run it outside the batch system, follow these instructions: http://mvapich.cse.ohio-state.edu/overview/mvapich2/
Applications built with mvapich2 will only run on machines with Infiniband hardware, so they will work on the pax machines but not on desktops, workgroup servers or the farm.
The machine file format is different from the one for openmpi, you must list the host name for every core you want to use, e.g. if you want to run four processes, two processes on each of pax09 and pax08:
pax88 pax89 pax88 pax89
The preferred way to run a application with mvapich2 is mpiexec.
If you want to use the deprecated mpd startup method as opposed to mpirun_rsh, you must also first create the file ~/.mpd.conf that contains of one line like this:
MPD_SECRETWORD=password
Batch System Access
![/!\](/moin_static1911/sinorca4moin/img/alert.png "/!\"){kind=small}
ATTENTION: The PAX cluster was split off the normal Zeuthen batch. To access the PAX batch system you will need to call ini pax.
Alternatively source a script:
- zsh users:
[oreade38] ~ % . /usr/gridengine/pax/common/settings.sh
- tcsh users:
[oreade38] ~ $ source /usr/gridengine/pax/common/settings.csh
Switching back to use the standard farm works similarly: - zsh users:
[oreade38] ~ % . /usr/gridengine/default/common/settings.sh
- tcsh users:
[oreade38] ~ $ source /usr/gridengine/default/common/settings.csh
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 up to 8 slots for 8 MPI processes on a single node:
#$ -pe multicore-mpi 8
For more MPI processes that have no big communication overhead, use -pe mpi.
Be sure to call the right mpirun version for your architecture. If you application was compiled for 64 bit, use
/usr/lib64/openmpi/1.4-gcc/bin/mpirun -np $NSLOTS yourapp
The MPI runtime will automatically select the right network type.
For more demanding MPI jobs you can select one of the pax blade centers like this in your job script. You can request up to 1024 slots, as a blade center contains 128 CPU cores and the batch system contains 8 blade centers:
#$ -pe pax 128 /usr/lib64/openmpi/1.4-gcc/bin/mpirun -np $NSLOTS yourapp
Finally, here's a list of common pitfalls when using the pax batch system:
Please be aware that all requested resources (via the -l qsub switch) are meant per job slot. As the pax nodes only provide 24GB (8 core systems -> 3GB per job slot), you cannot request more than 3GB h_vmem in your job scripts. Otherwise your job won't start!
- If your MPI application relies on LD_LIBRARY_PATH to load its shared libraries or modules, this will fail on remote notes, as the batch system will remove this variable from the environment. In that case you'll have to wrap yourapp in a shell script that sets up the environment and calls your binary application.
Mvapich2
With mvapich2 1.7, there is working integration into the SGE batch system. Just use a command like this:
#$ -pe pax 128 /usr/lib64/mvapich2/1.7-intel/bin/mpiexec -n $NSLOTS yourapp
AFS Access
The application binary must be available to all nodes, that's why it should be placed in an AFS directory.
BLAS library
Both ATLAS and GotoBLAS are available.
- ATLAS is in /opt/products/atlas
- libgoto is in /usr/lib or /usr/lib64 respectively.
Further documentation
Paralleles Rechnen in Zeuthen - die neuen Cluster , 04/27/10, technical seminar
HPC-Clusters at DESY Zeuthen , 11/22/06, technical seminar