NIIF Systems

From Nano Group Budapest
Revision as of 10:25, 6 September 2016 by Thiering Gergo (talk | contribs)
Jump to navigation Jump to search

NIIF systems are maintained by NIIF.

There are useful guides about it in their wikipedia, but mostly in hungarian.

Legacy information: Computing Service Legacy

Debrecen

This system maintained by NIIF.

Specification

Type SGI ICE8400EX
# of CPUs / node 2
# of cores / CPU 6
Memory / node 47 GB
Memory / core 3.9 GB
CPU Intel Xeon X5680 @ 3.33 GHz SMT on
Architecture x86_64 / intel64 / em64t little-endian
Scheduler Slurm
MPI SGI MPT (mpt)
ESZR local

Shell Environment

Precompiled Environment

From Skynet sync up the precompiled environment:

 cd /share/niif/debrecen
 sshput

The precompiled environment does not contain the program binaries and data libraries. To transfer a PROGRAM:

 cd /share/niif/debrecen/pkg
 sshtx put debrecen PROGRAM

Shell Framework

Login to debrecen:

 sshin debrecen

On debrecen:

 cd $HOME
 git clone git://github.com/thieringgergo/shf3.git

Source and setup the Shell Framework in $HOME/.profile:

 source $HOME/shf3/bin/shfrc
 # set the prompt
 shf3/ps1 DEBRECEN[\\h]
 # set framework features
 shf3/alias yes
 shf3/screen yes
 shf3/mc/color yes
 # screen workaround
 if shf3/is/screen ; then
   source "/etc/profile.d/modules.sh"
 fi
 # tab complete
 source $HOME/shf3/bin/complete

Parallel Compressor

Enable the parallel compressor for the framework:

 cd $HOME
 echo "sys_zip_xg_gz=pigz" > shf3/lib/sys/zip/xg/config.$USER

Module Environment

ESZR is our unified computing environment. Enable ESZR system modules in $HOME/.profile :

 # reset
 module purge
 module use ${HOME}/site/eszr/mod/common
 module load eszr/local
 module load eszr/env/local
 module load eszr/sys/niif/debrecen
 module load eszr/sys/niif/debrecen.mpt
 module use ${HOME}/site/eszr/mod/local
 module load sgi/2011
 source ${ESZR_ROOT}/env/alias

Logout and login again.

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -g c
Queue Allowed Groups Purpose
test.q ALL Test queue (2 nodes)
debrecen.q ALL Production

General information on jobs:

 qstat -u "*"

Job Setup

Setup the Queue file and edit the parameters:

 cd $HOME/shf3/mid/que
 cp templates/niif/debrecen .
 mcedit debrecen

Job template is in $HOME/shf3/mid/que/templates/niif/debrecen.job

Job Monitoring

Average node utilization of a job:

 jobmon JOBID

Per node utilization:

 pcpview -j JOBID

Check the last 3 columns of cpu:

 us - user load
 sy - system load
 id - idle

The user load should be around the maximum and the other two around 0. Node utilization chart:

 pcpview -c -j JOBID

Maximum utilization is 50% since SMT is enabled, in the chart it is 12 (# of cores per node).

Special Options

If you need to allocate a full node but want to start arbitrary number of MPI processes set in the job file:

 SLTPN=12

which will specify the total number of SGE slots per node and the total number of slots will be NODES*SLTPN .

Pécs

The system is maintained by NIIF.

Specification

Type SGI UV 1000
# of CPUs / node 2
# of cores / CPU 6
Memory 6 TB
CPU Intel Xeon X7542 @ 2.66 GHz SMT off
Architecture x86_64 / intel64 / em64t little-endian
Scheduler Slurm
MPI SGI MPT (mpt)
ESZR local

Shell Environment

Precompiled Environment

From Skynet sync up the precompiled environment:

 cd /share/niif/pecs
 sshput

The precompiled environment does not contain the program binaries and data libraries. To transfer a PROGRAM:

 cd /share/niif/pecs/pkg
 sshtx put pecs PROGRAM

Shell Framework

Login to pecs:

 sshin pecs

On pecs:

 cd $HOME
 git clone git://github.com/hornos/shf3.git

Source and setup the Shell Framework in $HOME/.profile:

 source $HOME/shf3/bin/shfrc
 # set the prompt
 shf3/ps1 PECS[\\h]
 # set framework features
 shf3/alias yes
 shf3/screen yes
 shf3/mc/color yes
 # screen workaround
 if shf3/is/screen ; then
   source "/etc/profile.d/modules.sh"
 fi
 # tab complete
 source $HOME/shf3/bin/complete

Parallel Compressor

Enable the parallel compressor for the framework:

 cd $HOME
 echo "sys_zip_xg_gz=pigz" > shf3/lib/sys/zip/xg/config.$USER

Module Environment

ESZR is our unified computing environment. Enable ESZR system modules in $HOME/.profile :

 # reset
 module purge
 module use ${HOME}/site/eszr/mod/common
 module load eszr/local
 module load eszr/env/local
 module load eszr/sys/niif/pecs
 module load eszr/sys/niif/pecs.mpt
 module use ${HOME}/site/eszr/mod/local
 module load sgi/2011
 source ${ESZR_ROOT}/env/alias

Logout and login again.

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -g c
Queue Allowed Groups Purpose
test.q ALL Test queue
pecs.q ALL Production

General information on jobs:

 qstat -u "*"

Job Setup

Setup the Queue file and edit the parameters:

 cd $HOME/shf3/mid/que
 cp templates/niif/pecs .
 mcedit debrecen

Job template is in $HOME/shf3/mid/que/templates/niif/pecs.job

Job Monitoring

Currently, monitoring is possible only by chart. You also have to enable the numainfo script in the queue file. In the job's submit directory:

 pcpview -c -j StdOut

Maximum utilization in the chart is 6 (# of cores per node).

Special Options

The UV is a ccNUMA SMP machine and you allocate CPU sockets and cores on one node. It is mandatory to set in the job file:

 NODES=1

The total number of SGE slots will be SCKTS*CORES .

Szeged

The system is maintained by NIIF.

Specification

Type HP CP4000BL
# of CPUs / node 4
# of cores / CPU 12
Memory / node 132 GB
Memory / core 2.75 GB
CPU AMD Opteron 6174 @ 2.2GHz
Architecture x86_64 / intel64 / em64t little-endian
Scheduler Slurm
MPI Intel (impi)
ESZR local

Shell Environment

Precompiled Environment

From Skynet sync up the precompiled environment:

 cd /share/niif/szeged
 sshput

The precompiled environment does not contain the program binaries and data libraries. To transfer a PROGRAM:

 cd /share/niif/szeged/pkg
 sshtx put szeged PROGRAM

Shell Framework

Login to szeged:

 sshin szeged

On szeged:

 cd $HOME
 git clone git://github.com/hornos/shf3.git

Source and setup the Shell Framework in $HOME/.bash_profile:

 source $HOME/shf3/bin/shfrc
 # set the prompt
 shf3/ps1 SZEGED[\\h]
 # set framework features
 shf3/alias yes
 shf3/screen yes
 shf3/mc/color yes
 # screen workaround
 if shf3/is/screen ; then
   source "/etc/profile.d/modules.sh"
 fi
 # tab complete
 source $HOME/shf3/bin/complete

Parallel Compressor

Enable the parallel compressor for the framework:

 cd $HOME
 echo "sys_zip_xg_gz=pigz" > shf3/lib/sys/zip/xg/config.$USER

Module Environment

ESZR is our unified computing environment. Enable ESZR system modules in $HOME/.bash_profile :

 # reset
 module purge
 module use ${HOME}/site/eszr/mod/common
 module load eszr/local
 module load eszr/env/local
 module load eszr/sys/niif/szeged
 module use ${HOME}/site/eszr/mod/local
 source ${ESZR_ROOT}/env/alias

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -g c
Queue Allowed Groups Purpose
test.q ALL Test queue (2 nodes)
szeged.q ALL Production

General information on jobs:

 qstat -u "*"

Job Setup

Setup the Queue file and edit the parameters:

 cd $HOME/shf3/mid/que
 cp templates/niif/szeged .
 mcedit szeged

Job template is in $HOME/shf3/mid/que/templates/niif/szeged.job

Special Options

If you need to allocate a full node but want to start arbitrary number of MPI processes set in the job file:

 SLTPN=48

which will specify the total number of SGE slots per node and the total number of slots will be NODES*SLTPN . You can use the following combinations in MPI-OMP mode if you run out of memory. Set hybrid mode in the job file:

 MODE=mpiomp/impi

and the socket/core number accoring to your needs.

SCKTS (# of MPI proces / node) CORES (# of OMP threads / MPI proc) Memory / MPI proc
2 24 66 GB
4 12 33 GB
8 6 16.5 GB
12 4 8.3 GB
24 2 4.3 GB

Budapest

The system is maintained by NIIF.

Specification

Type HP CP4000BL
# of CPUs / node 2
# of cores / CPU 12
Memory / node 66 GB
Memory / core 2.75 GB
CPU AMD Opteron 6174 @ 2.2GHz
Architecture x86_64 / intel64 / em64t little-endian
Scheduler Slurm
MPI Intel (impi)
ESZR local

Shell Environment

Precompiled Environment

From Skynet sync up the precompiled environment:

 cd /share/niif/budapest
 sshput

The precompiled environment does not contain the program binaries and data libraries. To transfer a PROGRAM:

 cd /share/niif/budapest/pkg
 sshtx put budapest PROGRAM

Shell Framework

Login to budapest:

 sshin budapest

On budapest:

 cd $HOME
 git clone git://github.com/hornos/shf3.git

Source and setup the Shell Framework in $HOME/.bash_profile:

 source $HOME/shf3/bin/shfrc
 # set the prompt
 shf3/ps1 BUDAPEST[\\h]
 # set framework features
 shf3/alias yes
 shf3/screen yes
 shf3/mc/color yes
 # screen workaround
 if shf3/is/screen ; then
   source "/etc/profile.d/modules.sh"
 fi
 # tab complete
 source $HOME/shf3/bin/complete

Parallel Compressor

Enable the parallel compressor for the framework:

 cd $HOME
 echo "sys_zip_xg_gz=pigz" > shf3/lib/sys/zip/xg/config.$USER

Module Environment

ESZR is our unified computing environment. Enable ESZR system modules in $HOME/.bash_profile :

 # reset
 module purge
 module use ${HOME}/site/eszr/mod/common
 module load eszr/local
 module load eszr/env/local
 module load eszr/sys/niif/budapest
 module use ${HOME}/site/eszr/mod/local
 source ${ESZR_ROOT}/env/alias

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -g c
Queue Allowed Groups Purpose
test.q ALL Test queue (2 nodes)
budapest.q ALL Production

General information on jobs:

 qstat -u "*"

Job Setup

Setup the Queue file and edit the parameters:

 cd $HOME/shf3/mid/que
 cp templates/niif/budapest .
 mcedit budapest

Job template is in $HOME/shf3/mid/que/templates/niif/budapest.job

Special Options

If you need to allocate a full node but want to start arbitrary number of MPI processes set in the job file:

 SLTPN=24

which will specify the total number of SGE slots per node and the total number of slots will be NODES*SLTPN . You can use the following combinations in MPI-OMP mode if you run out of memory. Set hybrid mode in the job file:

 MODE=mpiomp/impi

and the socket/core number accoring to your needs.

SCKTS (# of MPI proc / node) CORES (# of OMP threads / MPI proc) Memory / MPI proc
2 12 33 GB
4 6 16.5 GB
8 3 8.3 GB