Commit 2b73d387 authored by Marc Vef's avatar Marc Vef
Browse files

Adding dynamic IOR slurm script

parent 61237c58

ifs/scripts/slurm/mogon1_ior_ramdisk.sh

0 → 100755
+162 −0
Original line number Diff line number Diff line 
#!/bin/bash

# Slurm stuff



#SBATCH -J adafs_ior

#SBATCH -p nodeshort

#SBATCH -t 300

#SBATCH -A zdvresearch

#SBATCH --gres=ramdisk:20G



usage_short() {

        echo "

usage: adafs_ior.sh [-h] [-n <MD_PROC_N>] [-i <MD_ITER>] [-I <NUM_ITEMS>] [-u]

                    benchmark_dir

        "

}



help_msg() {



        usage_short

    echo "

This slurm batch script is for mdtesting adafs



positional arguments:

        benchmark_dir           path where the dependency downloads are put





optional arguments:

        -h, --help

                                shows this help message and exits

        -n <MD_PROC_N>

                                number of processes used in mdtest

                                defaults to '1'

        -i <MD_ITER>

                                number of iterations done in mdtest

                                defaults to '1'

        -I <NUM_ITEMS>

                                number of files per process in mdtest

                                defaults to '500000'

        -u, --unique

                                use if files should be placed in a unique directory per-process in mdtest

        "

}



MD_PROC_N=16

MD_ITER=1

MD_ITEMS="500000"

MD_UNIQUE=""



POSITIONAL=()

while [[ $# -gt 0 ]]

do

key="$1"



case ${key} in

    -n)

    MD_PROC_N="$2"

    shift # past argument

    shift # past value

    ;;

    -i)

    MD_ITER="$2"

    shift # past argument

    shift # past value

    ;;

    -I)

    MD_ITEMS="$2"

    shift # past argument

    shift # past value

    ;;

    -u|--unique)

    MD_UNIQUE="-u"

    shift # past argument

    ;;

    -h|--help)

    help_msg

    exit

    #shift # past argument

    ;;

    *)    # unknown option

    POSITIONAL+=("$1") # save it in an array for later

    shift # past argument

    ;;

esac

done

set -- "${POSITIONAL[@]}" # restore positional parameters



# positional arguments

if [[ -z ${1+x} ]]; then

    echo "Positional arguments missing."

    usage_short

    exit

fi



VEF_HOME="/home/vef"

HOSTFILE="${VEF_HOME}/jobdir/hostfile_${SLURM_JOB_ID}"

MD_DIR=$1

ROOTDIR="/localscratch/${SLURM_JOB_ID}/ramdisk"



# Load modules and set environment variables

PATH=$PATH:/home/vef/adafs/install/bin:/home/vef/.local/bin

C_INCLUDE_PATH=$C_INCLUDE_PATH:/home/vef/adafs/install/include

CPATH=$CPATH:/home/vef/adafs/install/include

CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/home/vef/adafs/install/include

LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/vef/adafs/install/lib

LIBRARY_PATH=$LIBRARY_PATH:/home/vef/adafs/install/lib

export PATH

export CPATH

export C_INCLUDE_PATH

export CPLUS_INCLUDE_PATH

export LD_LIBRARY_PATH

export LIBRARY_PATH

export LDFLAGS='-L/home/vef/adafs/install/lib/'

export CPPFLAGS='-I/home/vef/adafs/install/include/'

module load devel/CMake/3.8.0

module load mpi/OpenMPI/2.0.2-GCC-6.3.0

module load devel/Boost/1.63.0-foss-2017a

export CC=$(which gcc)

export CXX=$(which g++)



# printing stuff

echo "files per process: ${MD_ITEMS}"



# create a proper hostfile to run

srun -n ${SLURM_NNODES} hostname -s | sort -u > ${HOSTFILE} && sed -e 's/$/ max_slots=32/' -i ${HOSTFILE}



echo "Generated hostfile no of nodes:"

cat ${HOSTFILE} | wc -l



NONODES=$(cat ${HOSTFILE} | wc -l)

let MD_PROC_N=${NONODES}*16



echo "

############################################################################

############################### DAEMON START ############################### ############################################################################

"

# start adafs daemon on the nodes

python2 ${VEF_HOME}/ifs/scripts/startup_adafs.py -c -J ${SLURM_JOB_ID} --numactl "--cpunodebind=0,1 --membind=0,1" ${VEF_HOME}/ifs/build/bin/adafs_daemon ${ROOTDIR} ${MD_DIR} ${HOSTFILE}



#echo "logfiles:"

#cat /tmp/adafs_daemon.log

# pssh to get logfiles. hostfile is created by startup script

${VEF_HOME}/.local/bin/pssh -O StrictHostKeyChecking=no -i -h /tmp/hostfile_pssh_${SLURM_JOB_ID} "tail /tmp/adafs_daemon.log"



echo "

############################################################################

############################ RUNNING BENCHMARK #############################

############################################################################

"

# Run benchmark

echo "Executing: mpiexec -np ${MD_PROC_N} --map-by node --hostfile ${HOSTFILE} -x LD_PRELOAD=/gpfs/fs2/project/zdvresearch/vef/fs/ifs/build/lib/libadafs_preload_client.so ${VEF_HOME}/benchmarks/mogon1/mdtest-1.9.3-modified/mdtest -z 0 -b 1 -i ${MD_ITER} -d ${MD_DIR} -F -I ${MD_ITEMS} -C -r -T -v 1 ${MD_UNIQUE}"



mpiexec -np ${MD_PROC_N} --map-by node --hostfile ${HOSTFILE} -x LD_PRELOAD=/gpfs/fs2/project/zdvresearch/vef/fs/ifs/build/lib/libadafs_preload_client.so ${VEF_HOME}/benchmarks/mogon1/mdtest-1.9.3-modified/mdtest -z 0 -b 1 -i ${MD_ITER} -d ${MD_DIR} -F -I ${MD_ITEMS} -C -r -T -v 1 ${MD_UNIQUE}



echo "

############################################################################

############################### DAEMON STOP ############################### ############################################################################

"

# shut down adafs daemon on the nodes

python2 ${VEF_HOME}/ifs/scripts/shutdown_adafs.py -J ${SLURM_JOB_ID} ${VEF_HOME}/ifs/build/bin/adafs_daemon ${HOSTFILE}



# cleanup

rm ${HOSTFILE}

ifs/scripts/slurm/mogon1_mdtest_ramdisk.sh

+73 −35
Original line number Diff line number Diff line
@@ -9,8 +9,8 @@ 


usage_short() {

        echo "

usage: adafs_mdtest.sh [-h] [-n <MD_PROC_N>] [-i <MD_ITER>] [-I <NUM_ITEMS>] [-u]

                    benchmark_dir

usage: adafs_mdtest.sh [-h] [-n <PROC_PER_NODE>] [-b <BLOCKSIZE>] [-i <ITER>] [-Y]

                    benchmark_dir+file_prefix

        "

}
@@ -18,30 +18,34 @@ help_msg() { 


        usage_short

    echo "

This slurm batch script is for mdtesting adafs

This slurm batch script is for IOR testing adafs



positional arguments:

        benchmark_dir           path where the dependency downloads are put

        benchmark_dir           benchmark workdir





optional arguments:

        -h, --help              shows this help message and exits

        -n <MD_PROC_N>

                                number of processes used in mdtest

        -h, --help

                                shows this help message and exits



        -n <PROC_PER_NODE>

                                number of processes per node

                                defaults to '16'

        -b <BLOCKSIZE>

                                total number of data written and read (use 1k, 1m, 1g, etc...)

                                defaults to '1m'

        -i <ITER>

                                number of iterations done around IOR

                                defaults to '1'

        -i <MD_ITER>

                                number of iterations done in mdtest

                                defaults to '1'

        -I <NUM_ITEMS>          number of files per process in mdtest

                                defaults to '500000'

        -u, --unique            use if files should be placed in a unique directory per-process in mdtest

        -Y, --fsync             use fsync after writes

                                defaults to 'false'

        "

}



MD_PROC_N=16

MD_ITER=1

MD_ITEMS="500000"

MD_UNIQUE=""

PROC_PER_NODE=16

ITER=1

BLOCKSIZE="1m"

FSYNC=false



POSITIONAL=()

while [[ $# -gt 0 ]]
@@ -50,22 +54,22 @@ key="$1" 


case ${key} in

    -n)

    MD_PROC_N="$2"

    PROC_PER_NODE="$2"

    shift # past argument

    shift # past value

    ;;

    -i)

    MD_ITER="$2"

    -b)

    BLOCKSIZE="$2"

    shift # past argument

    shift # past value

    ;;

    -I)

    MD_ITEMS="$2"

    -i)

    ITER="$2"

    shift # past argument

    shift # past value

    ;;

    -u|--unique)

    MD_UNIQUE="-u"

    -Y|--fsync)

    FSYNC=true

    shift # past argument

    ;;

    -h|--help)
@@ -90,13 +94,13 @@ fi 


VEF_HOME="/home/vef"

HOSTFILE="${VEF_HOME}/jobdir/hostfile_${SLURM_JOB_ID}"

MD_DIR=$1

WORKDIR=$1

ROOTDIR="/localscratch/${SLURM_JOB_ID}/ramdisk"



# Load modules and set environment variables

PATH=$PATH:/home/vef/adafs/install/bin:/home/vef/.local/bin

C_INCLUDE_PATH=$C_INCLUDE_PATH:/home/vef/adafs/install/include

CPATH=$C_PATH:/home/vef/adafs/install/include

CPATH=$CPATH:/home/vef/adafs/install/include

CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/home/vef/adafs/install/include

LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/vef/adafs/install/lib

LIBRARY_PATH=$LIBRARY_PATH:/home/vef/adafs/install/lib
@@ -114,11 +118,8 @@ module load devel/Boost/1.63.0-foss-2017a 
export CC=$(which gcc)

export CXX=$(which g++)



# printing stuff

echo "files per process: ${MD_ITEMS}"



# create a proper hostfile to run

srun -n ${SLURM_NNODES} hostname -s | sort -u > ${HOSTFILE} && sed -e 's/$/ max_slots=32/' -i ${HOSTFILE}

srun -n ${SLURM_NNODES} hostname -s | sort -u > ${HOSTFILE} && sed -e 's/$/ max_slots=64/' -i ${HOSTFILE}



echo "Generated hostfile no of nodes:"

cat ${HOSTFILE} | wc -l
@@ -131,10 +132,8 @@ echo " 
############################### DAEMON START ############################### ############################################################################

"

# start adafs daemon on the nodes

python2 ${VEF_HOME}/ifs/scripts/startup_adafs.py -c -J ${SLURM_JOB_ID} --numactl "--cpunodebind=0,1 --membind=0,1" ${VEF_HOME}/ifs/build/bin/adafs_daemon ${ROOTDIR} ${MD_DIR} ${HOSTFILE}

python2 ${VEF_HOME}/ifs/scripts/startup_adafs.py -c -J ${SLURM_JOB_ID} --numactl "--cpunodebind=0,1 --membind=0,1" ${VEF_HOME}/ifs/build/bin/adafs_daemon ${ROOTDIR} ${WORKDIR} ${HOSTFILE}



#echo "logfiles:"

#cat /tmp/adafs_daemon.log

# pssh to get logfiles. hostfile is created by startup script

${VEF_HOME}/.local/bin/pssh -O StrictHostKeyChecking=no -i -h /tmp/hostfile_pssh_${SLURM_JOB_ID} "tail /tmp/adafs_daemon.log"
@@ -144,9 +143,48 @@ echo " 
############################################################################

"

# Run benchmark

echo "Executing: mpiexec -np ${MD_PROC_N} --map-by node --hostfile ${HOSTFILE} -x LD_PRELOAD=/gpfs/fs2/project/zdvresearch/vef/fs/ifs/build/lib/libadafs_preload_client.so ${VEF_HOME}/benchmarks/mogon1/mdtest-1.9.3-modified/mdtest -z 0 -b 1 -i ${MD_ITER} -d ${MD_DIR} -F -I ${MD_ITEMS} -C -r -T -v 1 ${MD_UNIQUE}"



mpiexec -np ${MD_PROC_N} --map-by node --hostfile ${HOSTFILE} -x LD_PRELOAD=/gpfs/fs2/project/zdvresearch/vef/fs/ifs/build/lib/libadafs_preload_client.so ${VEF_HOME}/benchmarks/mogon1/mdtest-1.9.3-modified/mdtest -z 0 -b 1 -i ${MD_ITER} -d ${MD_DIR} -F -I ${MD_ITEMS} -C -r -T -v 1 ${MD_UNIQUE}

BENCH_TMPL="mpiexec -np ${PROC_PER_NODE} --map-by node --hostfile ${HOSTFILE} -x LD_PRELOAD=/gpfs/fs2/project/zdvresearch/vef/fs/ifs/build/lib/libadafs_preload_client.so ior -a POSIX -i 1 -o ${WORKDIR} -b ${BLOCKSIZE} -F -w -r -W"



echo "#############"

echo "# 1. SEQUEL #"

echo "#############"

for TRANSFER in 4k 256k 512k 1m 2m 4m 8m 16m

do

    for i in {1..${ITER}}

    do

        CMD="${BENCH_TMPL} -t ${TRANSFER}"

        echo "## iteration $i"

        echo "## transfer size ${TRANSFER}"

        if [ "${FSYNC}" = true ] ; then

            CMD="${CMD} -Y"

            echo "## FSYNC on"

        fi

        echo "## Command ${CMD}"

        eval ${CMD}

    done

done



echo "#############"

echo "# 2. RANDOM #"

echo "#############"

for TRANSFER in 4k 256k 512k 1m 2m 4m 8m 16m

do

    for i in {1..${ITER}}

    do

        CMD="${BENCH_TMPL} -t ${TRANSFER} -z"

        echo "## iteration $i"

        echo "## transfer size ${TRANSFER}"

        if [ "${FSYNC}" = true ] ; then

            CMD="${CMD} -Y"

            echo "## FSYNC on"

        fi

        echo "## Command ${CMD}"

        eval ${CMD}

    done

done



# TODO 3. Striped later



echo "

############################################################################