Merge pull request #7396 from GabyCT/topic/addghatensorflow

metrics: Enable Tensorflow metrics for kata CI
This commit is contained in:
GabyCT 2023-07-31 17:13:24 -06:00 committed by GitHub
commit b77d69aeee
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 181 additions and 64 deletions

View File

@ -71,3 +71,29 @@ checktype = "mean"
midval = 96939.0
minpercent = 20.0
maxpercent = 20.0
[[metric]]
name = "tensorflow"
type = "json"
description = "tensorflow resnet model"
# Min and Max values to set a 'range' that
# the median of the CSV Results data must fall
# within (inclusive)
checkvar = ".\"tensorflow\".Results | .[] | .resnet.Result"
checktype = "mean"
midval = 3566.0
minpercent = 20.0
maxpercent = 20.0
[[metric]]
name = "tensorflow"
type = "json"
description = "tensorflow alexnet model"
# Min and Max values to set a 'range' that
# the median of the CSV Results data must fall
# within (inclusive)
checkvar = ".\"tensorflow\".Results | .[] | .alexnet.Result"
checktype = "mean"
midval = 98.0
minpercent = 20.0
maxpercent = 20.0

View File

@ -71,3 +71,29 @@ checktype = "mean"
midval = 98687.0
minpercent = 20.0
maxpercent = 20.0
[[metric]]
name = "tensorflow"
type = "json"
description = "tensorflow resnet model"
# Min and Max values to set a 'range' that
# the median of the CSV Results data must fall
# within (inclusive)
checkvar = ".\"tensorflow\".Results | .[] | .resnet.Result"
checktype = "mean"
midval = 3546.0
minpercent = 20.0
maxpercent = 20.0
[[metric]]
name = "tensorflow"
type = "json"
description = "tensorflow alexnet model"
# Min and Max values to set a 'range' that
# the median of the CSV Results data must fall
# within (inclusive)
checkvar = ".\"tensorflow\".Results | .[] | .alexnet.Result"
checktype = "mean"
midval = 98.0
minpercent = 20.0
maxpercent = 20.0

View File

@ -75,16 +75,14 @@ function run_test_blogbench() {
info "Running Blogbench test using ${KATA_HYPERVISOR} hypervisor"
bash tests/metrics/storage/blogbench.sh
check_metrics
}
function run_test_tensorflow() {
info "Running TensorFlow test using ${KATA_HYPERVISOR} hypervisor"
# ToDo: remove the exit once the metrics workflow is stable
exit 0
bash tests/metrics/machine_learning/tensorflow.sh 1 20
check_metrics
}
function main() {

View File

@ -4,7 +4,7 @@
#
# SPDX-License-Identifier: Apache-2.0
set -e
set -o pipefail
# General env
SCRIPT_PATH=$(dirname "$(readlink -f "$0")")
@ -12,18 +12,30 @@ source "${SCRIPT_PATH}/../lib/common.bash"
IMAGE="docker.io/library/tensorflow:latest"
DOCKERFILE="${SCRIPT_PATH}/tensorflow_dockerfile/Dockerfile"
BATCH_SIZE="512"
NUM_BATCHES="300"
CMD_RESULT="cd benchmarks/scripts/tf_cnn_benchmarks/ && cat result"
CMD_FILE="cat benchmarks/scripts/tf_cnn_benchmarks/result | grep 'total images' | wc -l"
tensorflow_file=$(mktemp tensorflowresults.XXXXXXXXXX)
BATCH_SIZE="100"
NUM_BATCHES="100"
resnet_tensorflow_file=$(mktemp resnettensorflowresults.XXXXXXXXXX)
alexnet_tensorflow_file=$(mktemp alexnettensorflowresults.XXXXXXXXXX)
NUM_CONTAINERS="$1"
TIMEOUT="$2"
TEST_NAME="tensorflow"
PAYLOAD_ARGS="tail -f /dev/null"
# Options to control the start of the workload using a trigger-file
dst_dir="/host"
src_dir=$(mktemp --tmpdir -d tensorflow.XXXXXXXXXX)
MOUNT_OPTIONS="type=bind,src=$src_dir,dst=$dst_dir,options=rbind:ro"
# CMD points to the script that starts the workload
alexnet_start_script="alexnet_start.sh"
resnet_start_script="resnet_start.sh"
CMD_RESNET="$dst_dir/$resnet_start_script"
CMD_ALEXNET="$dst_dir/$alexnet_start_script"
timeout=600
INITIAL_NUM_PIDS=1
CMD_FILE="cat alexnet_results | grep 'total images' | wc -l"
RESNET_CMD_FILE="cat resnet_results | grep 'total images' | wc -l"
function remove_tmp_file() {
rm -rf "${tensorflow_file}"
rm -rf "${resnet_tensorflow_file}" "${alexnet_tensorflow_file}"
}
trap remove_tmp_file EXIT
@ -37,75 +49,110 @@ Usage: $0 <count> <timeout>
container.
Options:
<count> : Number of containers to run.
<timeout> : Timeout to launch the containers.
<timeout> : Timeout to launch the containers.
EOF
}
function resnet50_test() {
local CMD_RUN="cd benchmarks/scripts/tf_cnn_benchmarks/ && python tf_cnn_benchmarks.py -data_format=NHWC --device cpu --batch_size=${BATCH_SIZE} --num_batches=${NUM_BATCHES} > result"
info "Running Resnet50 Tensorflow test"
function create_resnet_start_script() {
local script="${src_dir}/${resnet_start_script}"
rm -rf "${script}"
cat <<EOF >>"${script}"
#!/bin/bash
python benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py -data_format=NHWC --device cpu --batch_size=${BATCH_SIZE} --num_batches=${NUM_BATCHES} > resnet_results
EOF
chmod +x "${script}"
}
function create_alexnet_start_script() {
local script="${src_dir}/${alexnet_start_script}"
rm -rf "${script}"
cat <<EOF >>"${script}"
#!/bin/bash
python benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py --num_batches=${NUM_BATCHES} --device=cpu --batch_size=${BATCH_SIZE} --forward_only=true --model=alexnet --data_format=NHWC > alexnet_results
EOF
chmod +x "${script}"
}
function tensorflow_test() {
info "Copy Resnet Tensorflow test"
local pids=()
local j=0
for i in "${containers[@]}"; do
sudo -E "${CTR_EXE}" t exec -d --exec-id "$(random_name)" "${i}" sh -c "${CMD_RUN}"
$(sudo -E "${CTR_EXE}" t exec -d --exec-id "$(random_name)" "${i}" sh -c "${CMD_RESNET}")&
pids["${j}"]=$!
((j++))
done
# wait for all pids
for pid in ${pids[*]}; do
wait "${pid}"
done
info "All containers are running the workload..."
for i in "${containers[@]}"; do
check_file=$(sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "${RESNET_CMD_FILE}")
retries="300"
for j in $(seq 1 "${retries}"); do
[ "${check_file}" -eq "1" ] && break
sleep 1
done
done
info "Copy Alexnet Tensorflow test"
local pids=()
local j=0
for i in "${containers[@]}"; do
$(sudo -E "${CTR_EXE}" t exec -d --exec-id "$(random_name)" "${i}" sh -c "${CMD_ALEXNET}")&
pids["${j}"]=$!
((j++))
done
# wait for all pids
for pid in ${pids[*]}; do
wait "${pid}"
done
for i in "${containers[@]}"; do
check_file=$(sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "${CMD_FILE}")
retries="200"
retries="300"
for j in $(seq 1 "${retries}"); do
[ "${check_file}" -eq 1 ] && break
[ "${check_file}" -eq "1" ] && break
sleep 1
done
done
for i in "${containers[@]}"; do
sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "${CMD_RESULT}" >> "${tensorflow_file}"
sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "cat resnet_results" >> "${resnet_tensorflow_file}"
done
local resnet50_results=$(cat "${tensorflow_file}" | grep "total images/sec" | cut -d ":" -f2 | sed -e 's/^[ \t]*//' | tr '\n' ',' | sed 's/.$//')
local average_resnet50=$(echo "${resnet50_results}" | sed "s/,/+/g;s/.*/(&)\/$NUM_CONTAINERS/g" | bc -l)
local res_results=$(cat "${resnet_tensorflow_file}" | grep "total images/sec" | cut -d ":" -f2 | sed -e 's/^[ \t]*//' | tr '\n' ',' | sed 's/.$//')
local resnet_results=$(printf "%.0f\n" "${res_results}")
local res_average=$(echo "${resnet_results}" | sed "s/,/+/g;s/.*/(&)\/${NUM_CONTAINERS}/g" | bc -l)
local average_resnet=$(printf "%.0f\n" "${res_average}")
for i in "${containers[@]}"; do
sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "cat alexnet_results" >> "${alexnet_tensorflow_file}"
done
local alex_results=$(cat "${alexnet_tensorflow_file}" | grep "total images/sec" | cut -d ":" -f2 | sed -e 's/^[ \t]*//' | tr '\n' ',' | sed 's/.$//')
local alexnet_results=$(printf "%.0f\n" "${alex_results}")
local alex_average=$(echo "${alexnet_results}" | sed "s/,/+/g;s/.*/(&)\/${NUM_CONTAINERS}/g" | bc -l)
local average_alexnet=$(printf "%.0f\n" "${alex_average}")
local json="$(cat << EOF
{
"Resnet50": {
"Result": "${resnet50_results}",
"Average": "${average_resnet50}",
"Units": "s"
}
}
EOF
)"
metrics_json_add_array_element "$json"
}
function axelnet_test() {
local CMD_RUN="cd benchmarks/scripts/tf_cnn_benchmarks/ && python tf_cnn_benchmarks.py --num_batches=${NUM_BATCHES} --device=cpu --batch_size=${BATCH_SIZE} --forward_only=true --model=alexnet --data_format=NHWC > result"
info "Running AxelNet Tensorflow test"
for i in "${containers[@]}"; do
sudo -E "${CTR_EXE}" t exec -d --exec-id "$(random_name)" "${i}" sh -c "${CMD_RUN}"
done
for i in "${containers[@]}"; do
check_file=$(sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "${CMD_FILE}")
retries="200"
for j in $(seq 1 "${retries}"); do
[ "${check_file}" -eq 1 ] && break
sleep 1
done
done
for i in "${containers[@]}"; do
sudo -E "${CTR_EXE}" t exec --exec-id "$(random_name)" "${i}" sh -c "${CMD_RESULT}" >> "${tensorflow_file}"
done
local axelnet_results=$(cat "${tensorflow_file}" | grep "total images/sec" | cut -d ":" -f2 | sed -e 's/^[ \t]*//' | tr '\n' ',' | sed 's/.$//')
local average_axelnet=$(echo "${axelnet_results}" | sed "s/,/+/g;s/.*/(&)\/$NUM_CONTAINERS/g" | bc -l)
local json="$(cat << EOF
{
"AxelNet": {
"Result": "${axelnet_results}",
"Average": "${average_axelnet}",
"Units": "s"
"resnet": {
"Result": ${resnet_results},
"Average": ${average_resnet},
"Units": "images/s"
},
"alexnet": {
"Result": ${alexnet_results},
"Average": ${average_alexnet},
"Units": "images/s"
}
}
EOF
@ -143,11 +190,14 @@ function main() {
check_ctr_images "${IMAGE}" "${DOCKERFILE}"
init_env
create_resnet_start_script
create_alexnet_start_script
info "Creating ${NUM_CONTAINERS} containers"
for ((i=1; i<= "${NUM_CONTAINERS}"; i++)); do
containers+=($(random_name))
sudo -E "${CTR_EXE}" run -d --runtime "${CTR_RUNTIME}" "${IMAGE}" "${containers[-1]}" sh -c "${PAYLOAD_ARGS}"
sudo -E "${CTR_EXE}" run -d --runtime "${CTR_RUNTIME}" --mount="${MOUNT_OPTIONS}" "${IMAGE}" "${containers[-1]}" sh -c "${PAYLOAD_ARGS}"
((not_started_count--))
info "$not_started_count remaining containers"
done
@ -158,12 +208,29 @@ function main() {
# Check that the requested number of containers are running
check_containers_are_up
resnet50_test
# Check that the requested number of containers are running
local timeout_launch="10"
check_containers_are_up & pid=$!
(sleep "${timeout_launch}" && kill -HUP "${pid}") 2>/dev/null & pid_tout=$!
axelnet_test
if wait "${pid}" 2>/dev/null; then
pkill -HUP -P "${pid_tout}"
wait "${pid_tout}"
else
warn "Time out exceeded"
return 1
fi
# Get the initial number of pids in a single container before the workload starts
INITIAL_NUM_PIDS=$(sudo -E "${CTR_EXE}" t metrics "${containers[-1]}" | grep pids.current | grep pids.current | xargs | cut -d ' ' -f 2)
((INITIAL_NUM_PIDS++))
tensorflow_test
metrics_json_save
rm -rf "${src_dir}"
clean_env_ctr
}
main "$@"