Difference between revisions of "Testsuite Machines"

From Einstein Toolkit Documentation
Jump to: navigation, search
(→‎General)
(use pushable URL for testsuite_results repo)
 
(6 intermediate revisions by 5 users not shown)
Line 3: Line 3:
 
=General=
 
=General=
  
−
FIXME: this is out of date as it refers to simfactory version 1, which is obsolete!
+
To check out the ET, follow the instructions at http://einsteintoolkit.org/download/ under "Current development version".
  
−
To check out the ET:
+
Now configure simfactory:
−
 
 
−
  mkdir etrelease
 
−
  cd etrelease
 
−
  curl -O https://github.com/gridaphobe/CRL/raw/master/GetComponents
 
−
  chmod a+x GetComponents
 
−
  ./GetComponents --root=. -a https://svn.einsteintoolkit.org/manifest/trunk/einsteintoolkit.th
 
  
 +
  cd Cactus
 
   cp simfactory/etc/defs.local.ini.simple simfactory/etc/defs.local.ini
 
   cp simfactory/etc/defs.local.ini.simple simfactory/etc/defs.local.ini
 
   nano simfactory/etc/defs.local.ini
 
   nano simfactory/etc/defs.local.ini
Line 28: Line 23:
 
   sim login <machine>
 
   sim login <machine>
 
   sim build
 
   sim build
−
   sim create-submit ettests_1proc --testsuite --procs 16 --num-threads 16 --walltime 3:00:00
+
   sim create-submit ettests_1proc --testsuite --procs <N> --num-threads <N> --walltime 3:00:00
−
   sim create-submit ettests_2proc --testsuite --procs 16 --num-threads 8 --walltime 3:00:00
+
   sim create-submit ettests_2proc --testsuite --procs <N> --num-threads <N/2> --walltime 3:00:00
  
−
Replace 16 and 8 with the number of cores on each node, and half of this, respectively, for the machine you are using.  Remember that "procs" here means "cores" and "num-threads" means "number of threads per process". The idea is to use a full node, i.e. all the cores, and then either one or two MPI processes.
+
Replace <N> and <N/2> with the number of cores on each node, and half of this, respectively, for the machine you are using.  Remember that "procs" here means "cores" and "num-threads" means "number of threads per process". The idea is to use a full node, i.e. all the cores, and then either one or two MPI processes.
  
 
When the jobs have finished, you should have the summary.log files in
 
When the jobs have finished, you should have the summary.log files in
Line 40: Line 35:
 
Update the testsuite status page by adding the log files to the release-info repository:
 
Update the testsuite status page by adding the log files to the release-info repository:
  
−
   svn checkout https://svn.einsteintoolkit.org/www/release-info
+
   git clone git@bitbucket.org:einsteintoolkit/testsuite_results
 
   cd release-info
 
   cd release-info
−
   scp machine:<simulations>/ettests_1proc/output-0000/TEST/sim/summary.log <machine>__1_16.log
+
   scp machine:<simulations>/ettests_1proc/output-0000/TEST/sim/summary.log <machine>__1_<N>.log
−
   scp machine:<simulations>/ettests_2proc/output-0000/TEST/sim/summary.log <machine>__2_8.log
+
   scp machine:<simulations>/ettests_2proc/output-0000/TEST/sim/summary.log <machine>__2_<N/2>.log
−
   svn commit <machine>*.log
+
   git add <machine>*.log
 +
  git commit
 +
  git push
  
 
To re-run the tests with an updated checkout, run the GetComponents command above with the --update flag, rebuild, delete the "ettests_*" simulations, and resubmit the simulations.
 
To re-run the tests with an updated checkout, run the GetComponents command above with the --update flag, rebuild, delete the "ettests_*" simulations, and resubmit the simulations.
−
 
−
=Machines=
 
−
 
−
==Kraken==
 
−
 
−
Edit simfactory/mdb.pm
 
−
 
−
  -        myproxy-logon -p 7514 -s myproxy.teragrid.org -T -l @USER@ -o @SOURCEDIR@/.globus/proxy-teragrid
 
−
  +        myproxy-logon -p 7514 -s myproxy.teragrid.org -T -l <username> -o @SOURCEDIR@/.globus/proxy-teragrid
 
−
 
−
where <username> is your user name on Kraken.  This is related to [https://trac.einsteintoolkit.org/ticket/381 Ticket #381]
 
−
 
−
Change the sourcebasedir in simfactory/udb.pm to be under /lustre/scratch.  This is necessary because the Cactus directory must be visible from the compute node when running the tests (to see the parameter files, testsuite reference output and test.ccl files), and the home directory on Kraken is not visible from the compute nodes.
 
−
 
−
    set_option 'kraken', 'sourcebasedir', '/lustre/scratch/USERNAME';
 
−
 
−
Copy kraken.sh as kraken-testsuite.sh in simfactory/scriptfiles.  Replace the aprun command with the following:
 
−
 
−
<pre>
 
−
cd @SOURCEDIR@
 
−
 
−
export CCTK_TESTSUITE_RUN_COMMAND="aprun -n \$nprocs -d 1 \$exe \$parfile"
 
−
 
−
test_dir=TEST_$(date +%Y-%m-%d-%H%M%S)
 
−
mkdir $test_dir
 
−
 
−
CONFIGNAME=$(ls configs|tail) # SimFactory should provide the configuration name
 
−
 
−
for i in 1 2 3; do
 
−
  case $i in
 
−
    1 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=1
 
−
      export OMP_NUM_THREADS=1;;
 
−
    2 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=1;;
 
−
    3 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=2;;
 
−
  esac
 
−
  make $CONFIGNAME-testsuite PROMPT=no
 
−
  cp TEST/$CONFIGNAME/summary.log kraken__${CCTK_TESTSUITE_RUN_PROCESSORS}_${OMP_NUM_THREADS}.log
 
−
  mv TEST $test_dir/TEST.$i
 
−
done
 
−
</pre>
 
−
 
−
==Datura==
 
−
 
−
Copy datura.sh as datura-testsuite.sh in simfactory/scriptfiles.  Replace the mpirun command with the following:
 
−
 
−
<pre>
 
−
cd @SOURCEDIR@
 
−
 
−
test_dir=TEST_$(date +%Y-%m-%d-%H%M%S)
 
−
mkdir $test_dir
 
−
 
−
CONFIGNAME=$(ls configs|tail) # SimFactory should provide the configuration name
 
−
 
−
export CACTUS_STARTTIME=$(date +%s)
 
−
 
−
for i in 1 2 3; do
 
−
  case $i in
 
−
    1 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=1
 
−
      export OMP_NUM_THREADS=1;;
 
−
    2 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=1;;
 
−
    3 )
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=2;;
 
−
  esac
 
−
  export CCTK_TESTSUITE_RUN_COMMAND="${MPIDIR}/bin/mpirun -v --mca btl openib,self --mca mpi_leave_pinned 0 -np \$nprocs -npernode $CCTK_TESTSUITE_RUN_PROCESSORS \$exe -L 3 \$parfile"
 
−
  make $CONFIGNAME-testsuite PROMPT=no
 
−
  cp TEST/$CONFIGNAME/summary.log datura__${CCTK_TESTSUITE_RUN_PROCESSORS}_${OMP_NUM_THREADS}.log
 
−
  mv TEST $test_dir/TEST.$i
 
−
done
 
−
</pre>
 
−
 
−
==BlueDrop==
 
−
 
−
The following script is a slightly modified version of run_tests adapted to run on
 
−
the Power7 machine BlueDrop:
 
−
 
−
set -e
 
−
 
−
cd Cactus
 
−
 
−
#Blue drop (Note that we need to specify the absolute path for the bluedrop_run*.ll files):
 
−
 
−
for i in 1 2 3; do
 
−
  case $i in
 
−
    1 )
 
−
    hostname | grep ^bd && export CCTK_TESTSUITE_RUN_COMMAND="poe \$exe \$parfile -retry -1 -llfile /home/bcmundim/bluedrop_run_1.ll"
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=1
 
−
      export OMP_NUM_THREADS=1;;
 
−
    2 )
 
−
      hostname | grep ^bd && export CCTK_TESTSUITE_RUN_COMMAND="poe \$exe \$parfile -retry -1 -llfile /home/bcmundim/bluedrop_run_2.ll"
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=1;;
 
−
    3 )
 
−
      hostname | grep ^bd && export CCTK_TESTSUITE_RUN_COMMAND="poe \$exe \$parfile -retry -1 -llfile /home/bcmundim/bluedrop_run_2.ll"
 
−
      export CCTK_TESTSUITE_RUN_PROCESSORS=2
 
−
      export OMP_NUM_THREADS=2;;
 
−
  esac
 
−
  make sim-testsuite PROMPT=no
 
−
  mv TEST/sim/summary.log ../${HOSTNAME}__${CCTK_TESTSUITE_RUN_PROCESSORS}_${OMP_NUM_THREADS}.log
 
−
done
 
−
 
−
Remember to change the absolute path (/home/bcmundim in the script above) to the appropriate location
 
−
of your LoadLeveler batch job script files, the bluedrop_run_*.ll. The following batch job scripts
 
−
were used to run the test suites; bluedrop_run_1.ll and bluedrop_run_2.ll, respectively:
 
−
 
−
## Comment the class line to run on short queue:
 
−
#@ class = debug
 
−
#@ job_type = parallel
 
−
##@ node_usage = not_shared
 
−
#@ node_usage = shared
 
−
#@ environment = COPY_ALL
 
−
#@ tasks_per_node = 1
 
−
#@ node = 1
 
−
#@ wall_clock_limit = 0:30:00
 
−
### uncomment below for a normal batch job
 
−
# #@ output = $(host).$(jobid).$(stepid).out
 
−
# #@ error = $(host).$(jobid).$(stepid).err
 
−
 
−
#@ queue
 
−
## uncomment for a normal batch job
 
−
# $HOME/a.out
 
−
 
−
## Comment the class line to run on short queue:
 
−
#@ class = debug
 
−
#@ job_type = parallel
 
−
##@ node_usage = not_shared
 
−
#@ node_usage = shared
 
−
#@ environment = COPY_ALL
 
−
#@ tasks_per_node = 2
 
−
#@ node = 1
 
−
#@ wall_clock_limit = 0:30:00
 
−
### uncomment below for a normal batch job
 
−
# #@ output = $(host).$(jobid).$(stepid).out
 
−
# #@ error = $(host).$(jobid).$(stepid).err
 
−
 
−
#@ queue
 
−
## uncomment for a normal batch job
 
−
# $HOME/a.out
 

Latest revision as of 18:29, 3 November 2022

This page contains notes and instructions for people running the ET testsuites on various different machines. If you have experience running testsuites on a machine which is not listed here, please consider adding some information which might help others (or yourself!) in future.

General

To check out the ET, follow the instructions at http://einsteintoolkit.org/download/ under "Current development version".

Now configure simfactory: 

 cd Cactus
 cp simfactory/etc/defs.local.ini.simple simfactory/etc/defs.local.ini
 nano simfactory/etc/defs.local.ini

Edit defs.local.ini and replace 

 YOUR_LOGIN with your username
 YOUR@EMAIL.ADDRESS with your usual email address
 YOUR_ALLOCATION with your project allocation
 YOUR_THORNLIST with manifest/einsteintoolkit.th

See the machine-specific notes below for any additional steps for each machine. 

 sim sync <machine>
 sim login <machine>
 sim build
 sim create-submit ettests_1proc --testsuite --procs <N> --num-threads <N> --walltime 3:00:00
 sim create-submit ettests_2proc --testsuite --procs <N> --num-threads <N/2> --walltime 3:00:00

Replace <N> and <N/2> with the number of cores on each node, and half of this, respectively, for the machine you are using. Remember that "procs" here means "cores" and "num-threads" means "number of threads per process". The idea is to use a full node, i.e. all the cores, and then either one or two MPI processes.

When the jobs have finished, you should have the summary.log files in 

 <simulations>/ettests_1proc/output-0000/TEST/sim/summary.log
 <simulations>/ettests_2proc/output-0000/TEST/sim/summary.log

Update the testsuite status page by adding the log files to the release-info repository: 

 git clone git@bitbucket.org:einsteintoolkit/testsuite_results
 cd release-info
 scp machine:<simulations>/ettests_1proc/output-0000/TEST/sim/summary.log <machine>__1_<N>.log
 scp machine:<simulations>/ettests_2proc/output-0000/TEST/sim/summary.log <machine>__2_<N/2>.log
 git add <machine>*.log
 git commit
 git push

To re-run the tests with an updated checkout, run the GetComponents command above with the --update flag, rebuild, delete the "ettests_*" simulations, and resubmit the simulations.

Retrieved from "https://docs.einsteintoolkit.org/et-docs/index.php?title=Testsuite_Machines&oldid=7957"