Contact Details

Postal address

Innovative Manufacturing Research Centre
Cambridge University Engineering Department
Unit 26a Cambridge Science Park
Cambridge, CB4 0FP, UK

Location maps

Bus connections

Position

Research Associate

Affiliation

Job submission to the compute server via sun gridengine

• Do not write your own hostfiles as you will usually not know which nodes gridengine will choose and how many processors will be available. SGE will do it for you.
• Do not use email notification when you have a restricted account. The emails will never arrive (until I figure our a way around it) and it will trigger a security alert in cairngorm linked to your userid.
• For LAM/MPI you do not need to boot the lam. SGE will boot the lam automatically with the assigned nodes.

qsub mpi_job.sge 

#!/bin/bash # # UNIVERSITY OF CAMBRIDGE # Institute for Manufacturing # Computational Simulation for Manufacturing (CSfM) group # # September 2005 # # Example startup script for a MPICH2 run # # Filename: mpich2_job.sge # Created : 2005-09-06 19:34 by M Gross # # lines starting with a single hash symbol (#) are comment lines # lines starting with a hash-dollar (#$) are grid-engine parameters # # setting the shell used for execution of this job #$ -S /bin/bash # switch on email notification # REMEMBER: only use it if you have a full accont i.e. full # login access to cairngorm # # please fill in your camid below ##$ -M camid@cam.ac.uk # email notification at (b)egin and (e)nd of the job ##$ -m be # Work from submission directory #$ -cwd # parallel environment (pe) request #$ -pe mpich 12 # enables $TMPDIR/rsh to catch rsh calls if available set path=($TMPDIR $path) . ~/.bashrc # just copy hosts and machines file to stdout for the record more $TMPDIR/hosts more $TMPDIR/machines # count the hosts ... we could beautyfy this by expanding startmpi a bit ... but it works for now HOSTS=`cat $TMPDIR/machines | wc -l - | sed 's/-//'` # copy mpdboot line to stdout for the record echo mpdboot -n $HOSTS -f $TMPDIR/hosts -v --ifhn=mpi-`hostname` echo '================================================' mpdboot -n $HOSTS -f $TMPDIR/hosts -v --ifhn=mpi-`hostname` echo '================================================' # since using the local harddisk is faster, copy the necessary files # hardcoded node list at the moment, would be nicer to parse host/machine file for host in imrc-csfm-01 imrc-csfm-02 imrc-csfm-03 do echo scp cairngorm:ftLMPs.par $host:/home_local/markus/ scp cairngorm:ftLMPs.par $host:/home_local/markus/ echo scp cairngorm:ftLMPs.O3.x $host:/home_local/markus/ scp cairngorm:ftLMPs.O3.x $host:/home_local/markus/ done # change into the directory were we want to run cd /home_local/markus/ # start the job mpiexec -genv LD_LIBRARY_PATH `echo $LD_LIBRARY_PATH` -machinefile $TMPDIR/machines \ -n $NSLOTS /home_local/markus/ftLMPs.O3.x # close mpd mpdallexit #copy data back to the home directory mv ftLMPs.log ~ #done. # # EOF mpich2_job.sge

#!/bin/bash # # UNIVERSITY OF CAMBRIDGE # Institute for Manufacturing # Computational Simulation for Manufacturing (CSfM) group # # November 2004 # # Example startup script for a MPI run # # Filename: job1.sge # Created : 2004-12-04 19:34 by M Gross # # lines starting with a single hash symbol (#) are comment lines # lines starting with a hash-dollar (#$) are grid-engine parameters # # setting the shell used for execution of this job #$ -S /bin/bash # switch on email notification # REMEMBER: only use it if you have a full accont i.e. full # login access to cairngorm # # please fill in your camid below ##$ -M camid@cam.ac.uk # email notification at (b)egin and (e)nd of the job ##$ -m be # Work from submission directory #$ -cwd # change into the directory were the work should be done cd /home/markus/mpi_example # parallel environment (pe) request #$ -pe mpich 4 # enables $TMPDIR/rsh to catch rsh calls if available set path=($TMPDIR $path) . ~/.bashrc export P4_GLOBMEMSIZE=41943040 /opt/mpich-1.2.6/bin/mpirun -nolocal -np $NSLOTS \ -machinefile $TMPDIR/machines cpi.x # # EOF mpi_job.sge

#!/bin/bash # # UNIVERSITY OF CAMBRIDGE # Institute for Manufacturing # Computational Simulation for Manufacturing (CSfM) group # # November 2004 # # Example startup script for a LAM/MPI run # # Filename: job1.sge # Created : 2004-12-04 19:34 by M Gross # # lines starting with a single hash symbol (#) are comment lines # lines starting with a hash-dollar (#$) are grid-engine parameters # # setting the shell used for execution of this job #$ -S /bin/sh # switch on email notification # please fill in your camid below ##$ -M camid@cam.ac.uk # email notification at (b)egin and (e)nd of the job ##$ -m be # Work from submission directory #$ -cwd # change into the directory were the work should be done cd /home/markus/mpi_example # parallel environment (pe) request #$ -pe lam 4 # enables $TMPDIR/rsh to catch rsh calls if available set path=($TMPDIR /opt/lam/bin $path) /opt/lam/bin/mpirun /home/markus/lam_mandel/myapp # # EOF job1.sge

# We specify the full pathname to the "master" executable so that it # is sure to be found. It may not be necessary to specify the full # pathname in all cases; see mpirun(1) for more details. # h /home/markus/lam_mandel/master # # Run any number of slaves, but one per CPU is the most sensible thing. # Assuming the multicomputer is homogeneous, ship the executable # to each node. This is slower but more convenient than placing it # there yourself or relying upon NFS and your shell path to be right. # # We specify the full pathname to slave for the same reason as above. # C -s h /home/markus/lam_mandel/slave

#!/bin/bash # # UNIVERSITY OF CAMBRIDGE # Institute for Manufacturing # Computational Simulation for Manufacturing (CSfM) group # # November 2004 # # Example startup script for a Fluent run # # Filename: job1.sge # Created : Fri Nov 26 16:19:54 GMT 2004 by M Gross # # lines starting with a single hash symbol (#) are comment lines # lines starting with a hash-dollar (#$) are grid-engine parameters # # setting the shell used for execution of this job #$ -S /bin/bash # switch on email notification # please fill in your camid below #$ -M camid@cam.ac.uk # email notification at (b)egin and (e)nd of the job #$ -m be # Work from submission directory #$ -cwd # specifiy the queue to use. options are: node1,node2 and csfm.all # where csfm.all is a superset of node1 and node2 #$ -q csfm.all # change into the directory were the work should be done cd /home/markus/sge_example # some variables needed by FLUENT export FLUENT_ARCH=lnia64 export FLUENT_INC=/opt/Fluent.Inc # the command to execute /opt/Fluent.Inc/bin/fluent 2ddp -sge -g -i iterate.jou # # EOF fluent_job.sge

#!/bin/bash # # UNIVERSITY OF CAMBRIDGE # Institute for Manufacturing # Computational Simulation for Manufacturing (CSfM) group # # November 2004 # # Example startup script for compilation jobs # # Filename: make_job.sge # Created : Fri Nov 26 16:19:54 GMT 2004 by M Gross # # lines starting with a single hash symbol (#) are comment lines # lines starting with a hash-dollar (#$) are grid-engine parameters # # setting the shell used for execution of this job #$ -S /bin/bash # Work from submission directory #$ -cwd # specifiy the queue to use. options are: node1,node2 and csfm.all # where csfm.all is a superset of node1 and node2 #$ -q csfm.all # change into the directory were the make should be performed cd /home/user/my_code/repository/appxyz # the command to execute make # # EOF make_job.sge