Infi: Difference between revisions

From Wiki
Jump to navigation Jump to search
← Older edit
No edit summary
m INFI: moved to Infi
 
(16 intermediate revisions by 7 users not shown)
Line 1: Line 1:
11 Dual Intel Xeon processor at 3.06GHz
go back to [[Main Page]], [[Computational Resources]], [[Clusters]], [[Local Clusters]]




11 Dual Intel Xeon processor at 3.06GHz


== Access ==
== Access ==


ssh -X 10.3.30.254
ssh -X 10.3.30.254
== User details ==
/home  is on infi (not in sirius as kimik's and tekla's clusters).
Backups are for files size lower than 10 Mb.
There is a quota limit
  20 Gb


== Queues ==
== Queues ==


There is a single queues called "n" (without quotes). This queue has 12 nodes (except when some nodes have hardware problems) with 2 processors each. So you ask for slots (NCPUS) and system will allocate for each node 2 slots, p.e. you ask for 8 slots and system will take 4 nodes as it uses 2 slots by node. (can someone write this clearly?, because I can even barely understand it... :))
There is a single queues called "n" (without quotes). This queue has 12 nodes (except when some nodes have hardware problems) with 2 processors each. Because each node has two processors, you should ideally ask for an even number of CPU to optimize the performance. :-)
So you should send jobs asking for even slots (NCPUS).


To send jobs use  
To send jobs use  
       qsub name_of_your_script
       qsub name_of_your_script


below you can see examples of script for ADF, NWChem, DL_POLY, etc.






We run SGE 6.1, that has many new features, like boolean expressions for hosts, queues, etc.
[http://docs.sun.com/app/docs/doc/820-0699?l=en&q=sge+6.1 a manual here]
e.g. to see your jobs
  qstat -u my_username
to see all jobs
  qstat -u *
to see how many slots (quota) you are using and how many you have available
  qquota -u my_username
=='''IMPORTANT:'''==
As we still have some problems, to kill or delete jobs please use "qd":
  '''qd job_id program'''
  while "program" could be: adf nwchem dlpoly
for example, to delete job_id 4000 that runs a adf process (adf.exe) use:
  '''qd 4000 adf'''
for DLPOLY:
  '''qd 4000 dlpoly'''
 
 
Below you can see examples of script for ADF, NWChem, DL_POLY, etc.


We run SGE 6.1, that has many new features, like boolean expressions for hosts, queues, etc.
[http://docs.sun.com/app/docs/doc/820-0699?l=en&q=sge+6.1 a manual here]


----
----
Line 31: Line 68:
[http://aliga.iciq.es/wiki/index.php/NWCHEM NWChem]
[http://aliga.iciq.es/wiki/index.php/NWCHEM NWChem]
   
   
[http://aliga.iciq.es/wiki/index.php/ADF_2006 ADF 2007]
[http://aliga.iciq.es/wiki/index.php/ADF_2006 ADF 2006]
[http://aliga.iciq.es/wiki/index.php/ADF_2006 ADF 2006]


Line 38: Line 77:
  # queue system setup:
  # queue system setup:
  # pe request
  # pe request
  #$ -pe n0 6
  #$ -pe n 4
  #
  #ADF environment
  #MPI stuff
export ADFHOME=/opt/adf2006.01
export ADFBIN=$ADFHOME/bin
export ADFRESOURCES=$ADFHOME/atomicdata
export SCMLICENSE=$ADFHOME/license
export SCM_TMPDIR=$TMPDIR
export SCM_USETMPDIR=yes
  #MPI
  export MPIDIR=/opt/mpi
  export MPIDIR=/opt/mpi
  export PATH=$MPIDIR:$MPIDIR/bin:$PATH
  export PATH=$MPIDIR:$MPIDIR/bin:$PATH
Line 55: Line 86:
  export SCM_MACHINEFILE=$TMPDIR/machines
  export SCM_MACHINEFILE=$TMPDIR/machines
  export SCMWISH=""
  export SCMWISH=""
  export NSCM=6
  export NSCM=4
  export P4_GLOBMEMSIZE=32000000
  export P4_GLOBMEMSIZE=16000000
  export GFORTRAN_UNBUFFERED_ALL=y
  export GFORTRAN_UNBUFFERED_ALL=y
   
  export TMPDIR=/scratch
  cd /home/pmiro/Water/
#
   
# ADF Stuff
  $ADFBIN/adf -n $NSCM < aigua.in > aigua.out
export TMPDIR=/scratch
export ADFHOME=/opt/adf2006.01
export ADFBIN=$ADFHOME/bin
export ADFRESOURCES=$ADFHOME/atomicdata
export SCMLICENSE=$ADFHOME/license
export SCM_TMPDIR=$TMPDIR
export SCM_USETMPDIR=yes
#
  cd /home/ezuidema/test
  #
  $ADFBIN/adf -n $NSCM < test.in > test.out
#
mv TAPE21 test.t21


== ADF 2006 Scaling ==
== ADF 2006 Scaling ==
Line 111: Line 154:
  # queue system setup:
  # queue system setup:
  # pe request
  # pe request
  #$ -pe n0 4
  #$ -pe n 4
   
   
  #MPI INTEL
  #MPI INTEL
  export MPIDIR=/opt/intel/mpi/
  export MPIDIR=/opt/mpi/
  export PATH=$MPIDIR:$MPIDIR/bin:$PATH
  export PATH=$MPIDIR:$MPIDIR/bin:$PATH
  export LD_LIBRARY_PATH=$MPIDIR/lib
  export LD_LIBRARY_PATH=$MPIDIR/lib
Line 120: Line 163:
  export MACHINEFILE=$TMPDIR/machines
  export MACHINEFILE=$TMPDIR/machines
  export NCPUS=4
  export NCPUS=4
export NWCHEM_BASIS_LIBRARY=/opt/nwchem/data/libraries/
   
   
  cd /home/abraga/TESTS
  cd /home/abraga/TESTS
   
   
  /opt/intel/mpi/bin/mpirun -np $NCPUS -machinefile $MACHINEFILE /opt/nwchem/bin/nwchem job.nw >& job.out
  /opt/intel/mpi/bin/mpirun -np $NCPUS -machinefile $MACHINEFILE /opt/nwchem/bin/nwchem job.nw >& job.out

Latest revision as of 10:33, 3 August 2009

go back to Main Page, Computational Resources, Clusters, Local Clusters


11 Dual Intel Xeon processor at 3.06GHz

Access[edit]

ssh -X 10.3.30.254

User details[edit]

/home is on infi (not in sirius as kimik's and tekla's clusters). Backups are for files size lower than 10 Mb.

There is a quota limit 

  20 Gb


Queues[edit]

There is a single queues called "n" (without quotes). This queue has 12 nodes (except when some nodes have hardware problems) with 2 processors each. Because each node has two processors, you should ideally ask for an even number of CPU to optimize the performance. :-)

To send jobs use 

     qsub name_of_your_script



We run SGE 6.1, that has many new features, like boolean expressions for hosts, queues, etc. a manual here e.g. to see your jobs 

  qstat -u my_username

to see all jobs 

  qstat -u *

to see how many slots (quota) you are using and how many you have available 

 qquota -u my_username

IMPORTANT:[edit]

As we still have some problems, to kill or delete jobs please use "qd": 

  qd job_id program
  while "program" could be: adf nwchem dlpoly

for example, to delete job_id 4000 that runs a adf process (adf.exe) use: 

  qd 4000 adf

for DLPOLY: 

  qd 4000 dlpoly



Below you can see examples of script for ADF, NWChem, DL_POLY, etc.


Avaliable Programs[edit]

DLPOLY 2

NWChem

ADF 2007

ADF 2006

ADF 2006 Script[edit]

#! /bin/bash
# queue system setup:
# pe request
#$ -pe n 4
#
#MPI stuff
export MPIDIR=/opt/mpi
export PATH=$MPIDIR:$MPIDIR/bin:$PATH
export LD_LIBRARY_PATH=$MPIDIR/lib
export P4_RSHCOMMAND=rsh
export SCM_MACHINEFILE=$TMPDIR/machines
export SCMWISH=""
export NSCM=4
export P4_GLOBMEMSIZE=16000000
export GFORTRAN_UNBUFFERED_ALL=y
export TMPDIR=/scratch
#
# ADF Stuff
export TMPDIR=/scratch
export ADFHOME=/opt/adf2006.01
export ADFBIN=$ADFHOME/bin
export ADFRESOURCES=$ADFHOME/atomicdata
export SCMLICENSE=$ADFHOME/license
export SCM_TMPDIR=$TMPDIR
export SCM_USETMPDIR=yes
#
cd /home/ezuidema/test
#
$ADFBIN/adf -n $NSCM <  test.in  > test.out
#
mv TAPE21 test.t21

ADF 2006 Scaling[edit]

DLPOLY Script[edit]

#! /bin/bash
# queue system setup:
# pe request
#$ -pe n0 8

export LD_LIBRARY_PATH=$MPIDIR/lib
export DLPOLYPATH=/home/pmiro/dlpoly_MPI/execute
export P4_RSHCOMMAND=rsh
export MACHINEFILE=$TMPDIR/machines
export NCPUS=8

cd /home/pmiro/Bola/SO4Simulation/DownTemp/02/

/opt/mvapich-0.9.9/bin/mpirun -np $NCPUS -machinefile MACHINEFILE $DLPOLYPATH/DLPOLY.X

DLPOLY Scaling[edit]

System with 27336 Atoms 

      Shared Nodes 
NCPUS    %     Time 1ns (Days)
  1     100         52
  2     203         26
  3     268         19
  4     369         14
  5     428         12
  6     465         11
  7     499         10
  8     557          9
  9     565          9
 10     702          8
 11     732          9

      Non-Shared Nodes 
NCPUS    %     Time 1ns (Days)
  1     100         52
  2     196         26
  4     368         14

NWChem Script[edit]

#! /bin/bash
# queue system setup:
# pe request
#$ -pe n 4

#MPI INTEL
export MPIDIR=/opt/mpi/
export PATH=$MPIDIR:$MPIDIR/bin:$PATH
export LD_LIBRARY_PATH=$MPIDIR/lib
export P4_RSHCOMMAND=rsh
export MACHINEFILE=$TMPDIR/machines
export NCPUS=4
export NWCHEM_BASIS_LIBRARY=/opt/nwchem/data/libraries/

cd /home/abraga/TESTS

/opt/intel/mpi/bin/mpirun -np $NCPUS -machinefile $MACHINEFILE /opt/nwchem/bin/nwchem job.nw >& job.out
Retrieved from "http://usrc.iciq.cat/wiki/index.php?title=Infi&oldid=2612"