Closed connoraird closed 2 months ago
Results of running benchmarks CONQUEST 276 parallel speedup.pdf
To generate this data, from within a directory containing these files
submit-kathleen.sh - Script to submit jobs to kathleen (UCL cluster)
#!/bin/bash -l
#$ -l h_rt=1:00:00
#$ -N conquest_vtune_hotspot
#$ -pe mpi 41
#$ -l mem=3G
#$ -cwd
APP=../../Conquest
export OMP_NUM_THREADS=1
export OMP_STACKSIZE=100M
export MKL_NUM_THREADS=1i
export LD_LIBRARY_PATH=/shared/ucl/apps/libxc/6.2.2/intel-2022/lib:$LD_LIBRARY_PATH
mpiexec -np 1 $APP
test_*
benchmark directoriesI ran the following script.
for d in test_*/ ; do
echo "$d"
cd $d
for i in 1 2 4 8 16; do
echo "np-1-nt-$i"
mkdir np-1-nt-$i
cp C* H* ../submit-kathleen.sh np-1-nt-$i
cd np-1-nt-$i
sed -i "s/export OMP_NUM_THREADS=.*/export OMP_NUM_THREADS=${i}/g" "submit-kathleen.sh"
qsub submit-kathleen.sh
cd ../
done
cd ../
done
Once each job had completed, I ran the following script to gather the data into a format I could work with.
for d in */ ; do
for i in 1 2 4 8 16; do
echo "$d, $i, " | tr '\n' ' ' >> data.csv
cat $d/np-1-nt-$i/Conquest_out | grep "Total run time" | grep -Eo '[+-]?[0-9]+([.][0-9]+)' | tr '\n' ' ' >> data.csv
echo ', ' | tr '\n' ' ' >> data.csv
cat $d/np-1-nt-$i/time.001 | grep "Exact" | grep -Eo '[+-]?[0-9]+([.][0-9]+)' | tr '\n' ' ' >> data.csv
echo ', ' | tr '\n' ' ' >> data.csv
cat $d/np-1-nt-$i/exx_timers.001 | grep "nsup" | grep -Eo '[+-]?[0-9]+([.][0-9]+)' >> data.csv
done
echo ' ' >> data.csv
done
Dear Connor, results are impressive, that's a major achievement ! Thank you.
Description
Speedup plot
This plot shows the performance of test
test_004_isol_C2H4_4proc_PBE0CRI
for 1 mpi process