MPI HyperQuicksort

https://www.overleaf.com/2843638255vbbtgrffwdbn

https://docs.google.com/spreadsheets/d/1eT6IGapXR3tc1kAAuBgW7sdR-1guILFOPUcmq0dxb7E/edit

9 pages double column, single spacing, 10 point font with more performance graphs and more references

Commands

read generator

./generator.out 0 10000 1000000 normal /gpfs/u/home/PCPA/PCPAgjnn/scratch/datafile.txt
ls ~/scratch/datafile.txt
hexdump ~/scratch/datafile.txt

ssh dcsfen02
module load xl_r spectrum-mpi cuda/10.2
make generator project project-cuda
salloc -N 1 --partition=rpi --gres=gpu:4 -t 60

Make directories

mkdir SS WS

for weak scaling

sbatch -N 1 --ntasks-per-node=1 --partition=dcs --gres=gpu:6 -t 10 ./benchmark1.sh
sbatch -N 1 --ntasks-per-node=2 --partition=dcs --gres=gpu:6 -t 10 ./benchmark2.sh
sbatch -N 1 --ntasks-per-node=4 --partition=dcs --gres=gpu:6 -t 10 ./benchmark4.sh
sbatch -N 1 --ntasks-per-node=8 --partition=dcs --gres=gpu:6 -t 10 ./benchmark8.sh
sbatch -N 1 --ntasks-per-node=16 --partition=dcs --gres=gpu:6 -t 10 ./benchmark16.sh

for strong scaling

sbatch -N 1 --ntasks-per-node=1 --partition=dcs --gres=gpu:6 -t 10 ./benchmark.sh
sbatch -N 1 --ntasks-per-node=2 --partition=dcs --gres=gpu:6 -t 10 ./benchmark.sh
sbatch -N 1 --ntasks-per-node=4 --partition=dcs --gres=gpu:6 -t 10 ./benchmark.sh
sbatch -N 1 --ntasks-per-node=8 --partition=dcs --gres=gpu:6 -t 10 ./benchmark.sh
sbatch -N 1 --ntasks-per-node=16 --partition=dcs --gres=gpu:6 -t 10 ./benchmark.sh