cudadl/run_timing_experiment.sh

#!/bin/bash
#
if [ "${1}" == "dlrho" ] && [ "${#}" != "2" ] ; then
    echo "Usage (dlrho): ${0} dlrho mem_per_computer"
    exit
elif [ "${1}" == "controller" ] && [ "${#}" != "4" ] ; then
    echo "Usage (controller): ${0} controller mem_per_computer num_dpnodes num_workers"
    exit
elif [ "${#}" != "2" ] && [ "${#}" != "4" ] ; then
    echo "Usage: ${0} {dlrho, controller} mem_per_computer [num_dpnodes] [num_workers]"
    exit
fi
#
DATA_PATH="/home/sengler/data"
PORT=42000
PROBLEMS=(44 46 48 50 52 54 56 58 60 62 64 66 68 70 74 78 82 84 72 76 80 86 88)
#
MEM_PER_COMPUTER=${2} # ex: 200
#
NUM_DPNODES=${3} # ex: 1
NUM_WORKERS=${4} # ex: 8
#
MEM_PER_DPNODE=$((MEM_PER_COMPUTER/NUM_DPNODES))
#
# Make sure to run 'gen_all_N.sh' first to generate the problems.
#
for i in ${PROBLEMS[@]} ; do
    echo "Starting ${i}..."
    #
    if [ ! -f "${DATA_PATH}/N_${i}" ]; then
        echo "File ${DATA_PATH}/N_${i} was not found. Skipping..."
        continue
    fi
    #
    if [ "${1}" == "dlrho" ] ; then
        ./dlrho $MEM_PER_COMPUTER < "${DATA_PATH}/N_${i}" > "${DATA_PATH}/dlrho_${i}" 2>&1
    elif [ "${1}" == "controller" ] ; then
        ./controller -p $PORT -n $NUM_DPNODES -m $MEM_PER_DPNODE "${DATA_PATH}/N_${i}" 1 > "${DATA_PATH}/controller_${i}" 2>&1 &
        CONTROLLER_PID=$!
        trap "kill $CONTROLLER_PID ; exit" SIGINT
        #
        sleep 5
        # let the controller have time to start listening
        #
        for (( j=0; j<${NUM_DPNODES}; j++)); do
            ./dpnode 127.0.1.1 $PORT > "${DATA_PATH}/dpnode_${i}_${j}" 2>&1 &
        done
        #
        for (( j=0; j<${NUM_WORKERS}; j++)); do
            ./worker 127.0.1.1 $PORT $j > "${DATA_PATH}/worker_${i}_${j}" 2>&1 &
        done
        #
        wait $CONTROLLER_PID
    else
        echo "Option ${1} not recognized."
    fi
done