569 lines
22 KiB
Bash
Executable File
569 lines
22 KiB
Bash
Executable File
#!/bin/bash
|
||
#
|
||
# Measure transfer rates and CPU usage with various different options,
|
||
# multiple times, then calculate the mean and standard deviation for each
|
||
# set of measurements.
|
||
#
|
||
# The report is written to stdout as tab-separated values, each line
|
||
# prefixed with an opaque system ID (based on "uname -a"), the PV version
|
||
# expressed as an integer, and a run ID based on the start date and time.
|
||
#
|
||
# Takes a path to a pv binary as an argument.
|
||
|
||
# Defaults.
|
||
pv='pv' # pv executable to run the measurements with
|
||
rounds='10' # how many rounds of measurements to take
|
||
testFileMB='256' # max size of each of the test files, in MiB
|
||
testZeroesMB='1024' # amount of /dev/zero data to use, in MiB
|
||
|
||
# Script information for --help and --version.
|
||
programName='benchmark-pv-transfers'
|
||
programVersion='0.1.0'
|
||
bugReportsTo='https://codeberg.org/ivarch/pv/issues'
|
||
copyrightYear='2026'
|
||
copyrightHolder='Andrew Wood'
|
||
|
||
# Constants.
|
||
fieldsPerRecord='4' # measurements taken: rate, time - real, user, sys.
|
||
|
||
# Define the measurements. Each measurement definition contains:
|
||
# - The ID of the measurement
|
||
# - The name of the measurement
|
||
# - A space-separated list of single-letter options that PV must support
|
||
# - How many testFileMB multiples are transferred, or Z for testZeroesMB
|
||
# - The command to run
|
||
# In the command to run, {PV} is replaced with the path to PV, {FILE1} and
|
||
# {FILE2} are replaced by the random data filenames, {ZSIZE} is replaced by
|
||
# testZeroesMB, and {OUTPUT1} and {OUTPUT2} are replaced by a temporary
|
||
# output filenames.
|
||
# Each definition is stored as single line, with the above parts separated
|
||
# by "!".
|
||
measurementDefinitions=''
|
||
# Add a definition (name, options list, data size, command) to the
|
||
# definitions array.
|
||
addDefinition () {
|
||
measurementDefinitions="$(
|
||
printf '%s\n%s!%s!%s!%s!%s\n' \
|
||
"${measurementDefinitions}" \
|
||
"$(printf '%s\n' "$1" | md5sum | cut -b1-7)" \
|
||
"$1" "$2" "$3" "$4" \
|
||
| grep .
|
||
)"
|
||
}
|
||
# Add a set of standard definitions for measurements, naming them with a
|
||
# prefix "$1: ", each with extra options "$2".
|
||
addStandardMeasurements () {
|
||
# From file via stdin to file via stdout.
|
||
addDefinition "$1: stdin file to file" "$2" 1 "{PV} $2 < {FILE1} > {OUTPUT1}"
|
||
# From file to file via stdout.
|
||
addDefinition "$1: file to file" "$2" 1 "{PV} $2 {FILE1} > {OUTPUT1}"
|
||
# From two files to file via stdout.
|
||
addDefinition "$1: two files to file" "$2" 2 "{PV} $2 {FILE1} {FILE2} > {OUTPUT1}"
|
||
# From pipe to file via stdout.
|
||
addDefinition "$1: pipe to file" "$2" 1 "cat {FILE1} | {PV} $2 > {OUTPUT1}"
|
||
# From file via stdin to pipe.
|
||
addDefinition "$1: stdin file to pipe" "$2" 1 "{PV} $2 < {FILE1} | cat > {OUTPUT1}"
|
||
# From file to pipe.
|
||
addDefinition "$1: file to pipe" "$2" 1 "{PV} $2 {FILE1} | cat > {OUTPUT1}"
|
||
# From two files to pipe.
|
||
addDefinition "$1: two files to pipe" "$2" 2 "{PV} $2 {FILE1} {FILE2} | cat > {OUTPUT1}"
|
||
# From pipe to pipe.
|
||
addDefinition "$1: pipe to pipe" "$2" 1 "cat {FILE1} | {PV} $2 | cat > {OUTPUT1}"
|
||
}
|
||
# Generate the measurement definitions, most of which are based on repeated
|
||
# blocks of the above standard measurements, with various different options.
|
||
defineMeasurements () {
|
||
addStandardMeasurements 'Default' ''
|
||
addStandardMeasurements 'No-splice' '-C'
|
||
addStandardMeasurements 'Pipe buffer 1M' '-J 1M'
|
||
addStandardMeasurements 'Transfer buffer 1M' '-B 1M'
|
||
|
||
addStandardMeasurements 'Directed output' '-o {OUTPUT2}'
|
||
addStandardMeasurements 'Directed output with no-splice' '-o {OUTPUT2} -C'
|
||
addStandardMeasurements 'Directed output with pipe buffer 1M' '-o {OUTPUT2} -J 1M'
|
||
addStandardMeasurements 'Directed output with transfer buffer 1M' '-o {OUTPUT2} -B 1M'
|
||
|
||
addStandardMeasurements 'Discard' '-X'
|
||
addStandardMeasurements 'Discard with no-splice' '-X -C'
|
||
addStandardMeasurements 'Discard with pipe buffer 1M' '-X -J 1M'
|
||
addStandardMeasurements 'Discard with transfer buffer 1M' '-X -B 1M'
|
||
|
||
addDefinition "Zeroes: stdout to /dev/null" '-S -s' Z "{PV} -S -s {ZSIZE}M /dev/zero > /dev/null"
|
||
addDefinition "Zeroes: stdout to pipe" '-S -s' Z "{PV} -S -s {ZSIZE}M /dev/zero | cat > /dev/null"
|
||
addDefinition "Zeroes: discarded" '-X -S -s' Z "{PV} -X -S -s {ZSIZE}M /dev/zero"
|
||
addDefinition "Zeroes: stdout to /dev/null with no-splice" '-C -S -s' Z "{PV} -C -S -s {ZSIZE}M /dev/zero > /dev/null"
|
||
addDefinition "Zeroes: stdout to pipe with no-splice" '-C -S -s' Z "{PV} -C -S -s {ZSIZE}M /dev/zero | cat > /dev/null"
|
||
addDefinition "Zeroes: discarded with no-splice" '-C -X -S -s' Z "{PV} -C -X -S -s {ZSIZE}M /dev/zero"
|
||
}
|
||
|
||
# Write an error message $1 to standard error, prefixed with the program
|
||
# name.
|
||
error () {
|
||
printf '%s: %s\n' "${programName}" "$1" >&2
|
||
}
|
||
|
||
# Output an error message $1, and exit with a failure status.
|
||
die () {
|
||
error "$1"
|
||
exit 1
|
||
}
|
||
|
||
# Show all measurement definitions.
|
||
showMeasurementDefinitions () {
|
||
printf '%s\n%s\n' 'ID!Name!Opts!Size!Command' "${measurementDefinitions}" \
|
||
| awk -F '!' '{printf "%s\t%-60s\t%s\n", $1, $2, $5}'
|
||
}
|
||
|
||
# Write a line of results for the measurement with ID $1 and name $2, round
|
||
# ${thisRound}, reading the times from ${workDir}/times and deriving the
|
||
# rate from the elapsed time (the real time in the times file) and the size
|
||
# written in ${workDir}/size. Removes those two files in the process.
|
||
#
|
||
# The results are also spooled to ${workDir}/results for later analysis,
|
||
# prefixed with only the ID $1.
|
||
resultsLine () {
|
||
testTimeReal="$(awk '$1=="real" {print $2}' "${workDir}/times")"
|
||
testTimeUser="$(awk '$1=="user" {print $2}' "${workDir}/times")"
|
||
testTimeSystem="$(awk '$1=="sys" {print $2}' "${workDir}/times")"
|
||
testRate="$(awk -v t="${testTimeReal}" '{if (t>0) { printf "%.3f\n", $1/t } else { print "-" }}' < "${workDir}/size")"
|
||
test -n "${testRate}" || testRate='-'
|
||
rm -f "${workDir}/times" "${workDir}/size"
|
||
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' "${outputPrefix}" "${thisRound}" "$1" "${testRate}" "${testTimeReal}" "${testTimeUser}" "${testTimeSystem}" "$2"
|
||
printf '%s\t%s\t%s\t%s\t%s\n' "$1" "${testRate}" "${testTimeReal}" "${testTimeUser}" "${testTimeSystem}" >> "${workDir}/results"
|
||
}
|
||
|
||
# Run $2 in a shell under "time", writing $1 to ${workDir}/size and the
|
||
# times to ${workDir}/times.
|
||
captureTimes () {
|
||
printf '%s\n' "$1" > "${workDir}/size"
|
||
(
|
||
TIMEFORMAT="real %3R
|
||
user %3U
|
||
sys %3S"
|
||
time sh -c "{ $2; } 2>&3"
|
||
) 3>&2 2>"${workDir}/times"
|
||
rm -f "${workDir}/elapsed"
|
||
}
|
||
|
||
# Run all defined measurements for which the required options are available.
|
||
#
|
||
# If "${workDir}/permitted-measurements" is not empty, only measurements
|
||
# with IDs listed in that file will be taken.
|
||
gatherMeasurements () {
|
||
printf '%s\n' "${measurementDefinitions}" \
|
||
| {
|
||
while read -r definitionLine; do
|
||
measurementId="${definitionLine%%!*}"
|
||
measurementName="${definitionLine#*!}"
|
||
measurementName="${measurementName%%!*}"
|
||
requiredOptions="${definitionLine#*!}"
|
||
requiredOptions="${requiredOptions#*!}"
|
||
requiredOptions="${requiredOptions%%!*}"
|
||
dataSize="${definitionLine#*!}"
|
||
dataSize="${dataSize#*!}"
|
||
dataSize="${dataSize#*!}"
|
||
dataSize="${dataSize%%!*}"
|
||
templateCommand="${definitionLine##*!}"
|
||
|
||
test -n "${templateCommand}" || continue
|
||
|
||
if test -s "${workDir}/permitted-measurements"; then
|
||
grep -Fqx "${measurementId}" "${workDir}/permitted-measurements" || continue
|
||
fi
|
||
|
||
if test -n "${requiredOptions}"; then
|
||
optionLetters="$(printf 'x%s\n' "${requiredOptions}" | sed 's/-/\n-/g' | grep '^-' | cut -b 2)"
|
||
optionsPresent='true'
|
||
for pvOption in ${optionLetters}; do
|
||
grep -Fq " -${pvOption}" "${workDir}/help" || optionsPresent='false'
|
||
done
|
||
${optionsPresent} || continue
|
||
fi
|
||
|
||
activeCommand="$(
|
||
printf '%s\n' "${templateCommand}" | sed \
|
||
-e "s!{PV}!${pv}!g" \
|
||
-e "s!{FILE1}!${workDir}/file1!g" \
|
||
-e "s!{FILE2}!${workDir}/file2!g" \
|
||
-e "s!{ZSIZE}!${testZeroesMB}!g" \
|
||
-e "s!{OUTPUT1}!${workDir}/output1!g" \
|
||
-e "s!{OUTPUT2}!${workDir}/output2!g"
|
||
)"
|
||
|
||
if test "${dataSize}" = "Z"; then
|
||
dataSize="${testZeroesMB}"
|
||
else
|
||
dataSize=$((dataSize*testFileMB))
|
||
fi
|
||
|
||
captureTimes "${dataSize}" "${activeCommand}"
|
||
resultsLine "${measurementId}" "${measurementName}"
|
||
done
|
||
}
|
||
}
|
||
|
||
# Run several rounds of benchmark measurements using $1 as the pv
|
||
# executable, producing tab-separated data, including a final section
|
||
# containing the means and standard deviations for each measurement type.
|
||
#
|
||
# If $2 is not blank, only run the measurements whose IDs are listed in it.
|
||
#
|
||
runBenchmarks () {
|
||
pv="$1"
|
||
restrictTo="$2"
|
||
|
||
# Check there's enough room for the test files - make them smaller,
|
||
# if not.
|
||
tmpSpaceMB="$(df -kP "${TMPDIR:-/tmp}" | awk 'FNR==2 {print int($4/1024)}')"
|
||
while test "${testFileMB}" -gt 4; do
|
||
test "${tmpSpaceMB}" -gt $((2+3*testFileMB)) && break
|
||
testFileMB=$((testFileMB/2))
|
||
done
|
||
|
||
# Capture the help text so that capabilities can be checked.
|
||
${pv} -h > "${workDir}/help"
|
||
|
||
# Write a list of permitted measurement IDs, one per line. An empty
|
||
# file means no restriction.
|
||
printf '%s\n' "${restrictTo}" \
|
||
| tr ',; \t' '\n\n\n\n' \
|
||
| sort -u \
|
||
| grep . \
|
||
> "${workDir}/permitted-measurements"
|
||
|
||
# Define identifiers for the system this is running on, the pv
|
||
# version being benchmarked, and this specific benchmark run.
|
||
sysId="$(uname -a | md5sum | cut -b1-7)"
|
||
pvId="$(${pv} --version | awk 'FNR==1{print $2}' | awk -F . '{print 1000000*$1+1000*$2+$3}')"
|
||
runId="$(date '+%Y%m%d%H%M%S')"
|
||
outputPrefix="$(printf '%s\t%s\t%s\n' "${sysId}" "${pvId}" "${runId}")"
|
||
|
||
# Basic system information, and a header line for the test results.
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'System hostname' "$(uname -n)"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'System load' "$(uptime | awk '{printf "%.2f\n",$(NF-2)}')"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'System kernel type' "$(uname -s)"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'System kernel release' "$(uname -r)"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'System OS' "$(uname -o)"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'PV path' "${pv}"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'PV version' "$(${pv} -V | awk 'FNR==1 {print $2}')"
|
||
printf '%s\t%s\t%s\n' "${outputPrefix}" 'Test file size (MB)' "${testFileMB}"
|
||
|
||
# Generate two files of random data.
|
||
dd if='/dev/urandom' of="${workDir}/file1" bs=1048576 count="${testFileMB}" 2>/dev/null
|
||
dd if='/dev/urandom' of="${workDir}/file2" bs=1048576 count="${testFileMB}" 2>/dev/null
|
||
|
||
# Run several rounds of measurements.
|
||
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' "${outputPrefix}" '#' 'ID' 'MiB/sec' 'Real time' 'User CPU time' 'System CPU time' 'Raw measurement'
|
||
thisRound=0
|
||
while test ${thisRound} -lt "${rounds}"; do
|
||
thisRound=$((1+thisRound))
|
||
gatherMeasurements
|
||
done
|
||
|
||
# For each of the types of measurement, report the mean and standard
|
||
# deviation of each field.
|
||
awk -F "\t" '{print $1}' < "${workDir}/results" > "${workDir}/measurement-ids"
|
||
true > "${workDir}/measurement-ids-used"
|
||
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' "${outputPrefix}" 'μ/σ' 'ID' 'MiB/sec' 'Real time' 'User CPU time' 'System CPU time' 'Aggregated measurement'
|
||
{
|
||
while read -r measurementId; do
|
||
# Skip this type of measurement if already processed.
|
||
grep -Fqx "${measurementId}" "${workDir}/measurement-ids-used" && continue
|
||
printf '%s\n' "${measurementId}" >> "${workDir}/measurement-ids-used"
|
||
# Get the associated measurement name.
|
||
measurementName="$(printf '%s\n' "${measurementDefinitions}" | awk -F '!' -v "x=${measurementId}" '$1==x{print $2}')"
|
||
# Separate out this measurement type's results.
|
||
awk -F "\t" -v "mId=${measurementId}" '$1==mId {print}' < "${workDir}/results" \
|
||
> "${workDir}/measurements"
|
||
# Calculate the mean of each field.
|
||
awk -F "\t" -v "mId=${measurementId}" -v "mName=${measurementName}" -v "fieldcount=${fieldsPerRecord}" \
|
||
'BEGIN { samples=0 }
|
||
{ samples++; for (field=1; field<=fieldcount; field++) { total[field] += $(1+field) } }
|
||
END { printf "%s\t%s", "μ", mId; for (field=1; field<=fieldcount; field++) { printf "\t%.3f", total[field]/samples }; printf "\t%s\n", mName }' \
|
||
< "${workDir}/measurements" > "${workDir}/mean"
|
||
# Calculate the standard deviation of each field.
|
||
cat "${workDir}/mean" "${workDir}/measurements" \
|
||
| awk -F "\t" -v "mId=${measurementId}" -v "mName=${measurementName}" -v "fieldcount=${fieldsPerRecord}" \
|
||
'BEGIN { samples=0 }
|
||
FNR==1 { for (field=1; field<=fieldcount; field++) { mean[field] += $(2+field) } }
|
||
FNR>1 { samples++; for (field=1; field<=fieldcount; field++) { variance=$(1+field)-mean[field]; sum_variance_squared[field] += (variance*variance) } }
|
||
END { printf "%s\t%s", "σ", mId; for (field=1; field<=fieldcount; field++) { printf "\t%.3f", sqrt(sum_variance_squared[field]/samples) }; printf "\t%s\n", mName }' \
|
||
> "${workDir}/stddev"
|
||
sed "s!^!${outputPrefix}\t!" "${workDir}/mean" "${workDir}/stddev"
|
||
done
|
||
} < "${workDir}/measurement-ids"
|
||
}
|
||
|
||
# Reformat data on stdin so that each (whitespace-separated) column is
|
||
# space-padded and right-aligned to a consistent width for all rows.
|
||
#
|
||
# Note that this can only handle ASCII, not UTF-8.
|
||
lineUpColumns () {
|
||
tee "${workDir}/lineup-temp" \
|
||
| awk 'BEGIN {cols=0}
|
||
/./ {
|
||
if (NF > cols)
|
||
cols=NF
|
||
for (col=1; col<=NF; col++) {
|
||
if (length($col) >= max[col])
|
||
max[col]=length($col)
|
||
}
|
||
}
|
||
END {
|
||
for (col=1; col<=cols; col++) {
|
||
printf "%d ", max[col]
|
||
}
|
||
printf "\n"
|
||
}
|
||
' > "${workDir}/lineup-colwidths"
|
||
cat "${workDir}/lineup-colwidths" "${workDir}/lineup-temp" \
|
||
| awk '
|
||
FNR==1 { cols=NF; for (col=1; col<=NF; col++) { max[col]=$col } }
|
||
FNR>1 {
|
||
for (col=1; col<=cols; col++) {
|
||
if (col>1)
|
||
printf " "
|
||
val=(col<=NF ? $col : "")
|
||
printf "%" max[col] "s", val
|
||
}
|
||
printf "\n"
|
||
}'
|
||
}
|
||
|
||
# Read a stream of benchmark data on stdin containing runs from a single
|
||
# system, and report how the measurements changed across the different PV
|
||
# versions.
|
||
compareVersionResults () {
|
||
cat > "${workDir}/raw-system-data"
|
||
# List the measurement IDs in the order they appear in the data.
|
||
awk -F "\t" '$4=="σ"{print $5}' < "${workDir}/raw-system-data" > "${workDir}/measurement-ids"
|
||
# List all PV versions for which any data is available.
|
||
awk -F "\t" '$4=="PV version"{print $2,$5}' "${workDir}/raw-system-data" | sort -nu > "${workDir}/pv-versions"
|
||
# Report on each measurement type in turn.
|
||
true > "${workDir}/measurement-ids-seen"
|
||
{
|
||
while read -r measurementId; do
|
||
# Skip if this measurement was already processed.
|
||
grep -Fqx "${measurementId}" "${workDir}/measurement-ids-seen" && continue
|
||
printf '%s\n' "${measurementId}" >> "${workDir}/measurement-ids-seen"
|
||
# Collect this measurement's mean and standard deviation
|
||
# records for each PV version. If there's more than one for
|
||
# a single version, average them.
|
||
# TODO: split out collection into a separate function for re-use later
|
||
{
|
||
while read -r pvId pvVersion; do
|
||
awk -F "\t" \
|
||
-v "pvId=${pvId}" -v "outPrefix=${pvVersion}" \
|
||
-v "mId=${measurementId}" \
|
||
-v "fieldcount=${fieldsPerRecord}" \
|
||
'BEGIN {samples=0}
|
||
$2==pvId && $5==mId && $4=="μ" { samples++; for (field=1; field<=fieldcount; field++) { mean[field] += $(5+field) } }
|
||
$2==pvId && $5==mId && $4=="σ" { for (field=1; field<=fieldcount; field++) { stddev[field] += $(5+field) } }
|
||
END {
|
||
if (samples > 0) {
|
||
printf "%s", outPrefix
|
||
for (field=1; field<=fieldcount; field++) {
|
||
printf "\t%.3f\t%.3f", mean[field]/samples, stddev[field]/samples
|
||
}
|
||
printf "\n"
|
||
}
|
||
}' \
|
||
< "${workDir}/raw-system-data"
|
||
done
|
||
} < "${workDir}/pv-versions" > "${workDir}/measurements-per-version"
|
||
# If there are not at least 2 PV versions for which this
|
||
# measurement was available, report nothing as no comparison
|
||
# can be made.
|
||
test "$(grep -c . "${workDir}/measurements-per-version")" -lt 2 && continue
|
||
# Show the measurement name and associated command.
|
||
# Take the measurement name from the input data.
|
||
measurementName="$(awk -F "\t" -v "mId=${measurementId}" '$4=="σ" && $5==mId {print $NF;exit}' "${workDir}/raw-system-data")"
|
||
# Take the command from the definitions.
|
||
templateCommand="$(printf '%s\n' "${measurementDefinitions}" | awk -F '!' -v "mId=${measurementId}" '$1==mId {print $5}')"
|
||
printf '\n%s\n' "${measurementName}"
|
||
test -n "${templateCommand}" && printf ' (%s)\n' "${templateCommand}"
|
||
# Report each version's measurements and how they compare to
|
||
# the previous version.
|
||
#
|
||
# Since the column widths are set by an awk script which
|
||
# doesn't support UTF-8, ASCII headings are used initially,
|
||
# and adjusted after formatting.
|
||
{
|
||
printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
|
||
'PV' \
|
||
'M:Rate' 'S:Rate' 'C:Rate' \
|
||
'-' \
|
||
'M:tReal' 'S:tReal' 'C:tReal' \
|
||
'-' \
|
||
'M:tUser' 'S:tUser' 'C:tUser' \
|
||
'-' \
|
||
'M:tSys' 'S:tSys' 'C:tSys'
|
||
awk -F "\t" -v "fieldcount=${fieldsPerRecord}" \
|
||
'{
|
||
printf "%s", $1
|
||
for (field=0; field<fieldcount; field++) {
|
||
if (field > 0) {
|
||
printf "\t%s", "-";
|
||
}
|
||
mean=$(2+2*field)
|
||
stddev=$(3+2*field)
|
||
changescore=0
|
||
printf "\t%.3f\t%.3f", mean, stddev;
|
||
if (FNR>1) {
|
||
low=mean-stddev;
|
||
high=mean+stddev;
|
||
range=high-low;
|
||
prevlow=pmean[field]-pstddev[field];
|
||
prevhigh=pmean[field]+pstddev[field];
|
||
prevrange=prevhigh-prevlow;
|
||
if (prevrange < 1)
|
||
prevrange = 1;
|
||
if (high > prevhigh) {
|
||
changescore = (high - prevhigh) / prevrange;
|
||
} else if (low < prevlow) {
|
||
changescore = 0 - ((prevlow - low) / prevrange);
|
||
}
|
||
}
|
||
if (changescore < 0.01 && changescore > -0.01) {
|
||
printf "\t%d", 0
|
||
} else {
|
||
printf "\t%s%.2f", (changescore > 0.00 ? "+" : ""), changescore
|
||
}
|
||
pmean[field]=mean;
|
||
pstddev[field]=stddev;
|
||
}
|
||
printf "\n"
|
||
}' "${workDir}/measurements-per-version"
|
||
} \
|
||
| lineUpColumns \
|
||
| sed '1{s,M:,μ:,g;s,S:,σ:,g;s,C:,±:,g}'
|
||
done
|
||
} < "${workDir}/measurement-ids"
|
||
}
|
||
|
||
# Read a stream of benchmark data from stdin containing one or more runs
|
||
# from one or more systems and multiple PV versions, and for each individual
|
||
# system, report how the measurements changed across the different PV
|
||
# versions.
|
||
runVersionComparisons () {
|
||
cat > "${workDir}/raw-data"
|
||
# List the system IDs in the order they appear in the data.
|
||
awk -F "\t" '{print $1}' < "${workDir}/raw-data" | uniq > "${workDir}/sysids"
|
||
# Report on each system in turn.
|
||
true > "${workDir}/sysids-seen"
|
||
{
|
||
while read -r sysId; do
|
||
# Skip if this system was already processed.
|
||
grep -Fqx "${sysId}" "${workDir}/sysids-seen" && continue
|
||
printf '%s\n' "${sysId}" >> "${workDir}/sysids-seen"
|
||
# Extract the data for just this system.
|
||
awk -F "\t" -v "s=${sysId}" '$1==s {print}' < "${workDir}/raw-data" > "${workDir}/system-data"
|
||
# Report preamble.
|
||
printf '%79s\n' '' | tr ' ' '-'
|
||
printf '%s: %s - %s - %s %s\n' \
|
||
'System' \
|
||
"$(awk -F "\t" '$4=="System hostname"{print $5;exit}' "${workDir}/system-data")" \
|
||
"$(awk -F "\t" '$4=="System OS"{print $5;exit}' "${workDir}/system-data")" \
|
||
"$(awk -F "\t" '$4=="System kernel type"{print $5;exit}' "${workDir}/system-data")" \
|
||
"$(awk -F "\t" '$4=="System kernel release"{print $5;exit}' "${workDir}/system-data")"
|
||
cat <<EOF
|
||
|
||
For each measurement type, each of its aggregate benchmark results are
|
||
shown. Each field's mean and standard deviation are displayed along with a
|
||
change indicator showing how different this line's value is from the
|
||
previous line.
|
||
EOF
|
||
compareVersionResults < "${workDir}/system-data"
|
||
done
|
||
} < "${workDir}/sysids"
|
||
}
|
||
|
||
##############################################################################
|
||
# Main entry point.
|
||
|
||
# Process any command-line options.
|
||
action='benchmark'
|
||
restrictMeasurementIdList=''
|
||
while test -n "$1"; do
|
||
arg="$1"
|
||
shift
|
||
case "${arg}" in
|
||
'measurements'|'benchmark'|'analyse') action="${arg}" ;;
|
||
'-h'|'--help')
|
||
cat - <<EOF
|
||
Usage: ${programName} [OPTIONS] [ACTION]
|
||
Benchmark pv transfers.
|
||
|
||
Actions:
|
||
|
||
measurements - list all benchmark measurement definitions
|
||
benchmark - take several rounds of measurements
|
||
analyse - analyse benchmark data on stdin from multiple pv versions
|
||
|
||
Options:
|
||
|
||
-p, --program FILE benchmark using FILE as the pv executable
|
||
-r, --rounds ROUNDS run ROUNDS sets of measurements (${rounds})
|
||
-m, --measurement ID only run this specific measurement
|
||
-s, --size SIZE attempt to use a test file of SIZE MiB (${testFileMB})
|
||
-z, --zeroes SIZE stop at SIZE MiB for /dev/zero measurements (${testZeroesMB})
|
||
|
||
-h, --help show this help
|
||
-V, --version show script version
|
||
|
||
Default values are shown in brackets.
|
||
|
||
Please report any bugs to: ${bugReportsTo}
|
||
EOF
|
||
exit 0
|
||
;;
|
||
'-V'|'--version')
|
||
cat - <<EOF
|
||
${programName} (pv) ${programVersion}
|
||
Copyright ${copyrightYear} ${copyrightHolder}
|
||
License: GPLv3+ <https://www.gnu.org/licenses/gpl-3.0.html>
|
||
This is free software: you are free to change and redistribute it.
|
||
There is NO WARRANTY, to the extent permitted by law.
|
||
EOF
|
||
exit 0
|
||
;;
|
||
'-p'|'--program'|'--pv') pv="$1"; test $# -gt 0 && shift ;;
|
||
'--program='*|'--pv='*) pv="${arg#*=}" ;;
|
||
'-r'|'--rounds') rounds="$1"; test $# -gt 0 && shift ;;
|
||
'--rounds='*) rounds="${arg#*=}" ;;
|
||
'-m'|'--measurement') restrictMeasurementIdList="${restrictMeasurementIdList} $1"; test $# -gt 0 && shift ;;
|
||
'--measurement='*) restrictMeasurementIdList="${restrictMeasurementIdList} ${arg#*=}" ;;
|
||
'-s'|'--size') testFileMB="$1"; test $# -gt 0 && shift ;;
|
||
'--size='*) testFileMB="${arg#*=}" ;;
|
||
'-z'|'--zeroes') testZeroesMB="$1"; test $# -gt 0 && shift ;;
|
||
'--zeroes='*) testZeroesMB="${arg#*=}" ;;
|
||
'-'*) die "${arg}: unknown option - try \`--help'" ;;
|
||
*) die "${arg}: unexpected argument - try \`--help'" ;;
|
||
esac
|
||
done
|
||
|
||
# Use /dev/shm for workspace if possible, to eliminate disk I/O as a factor.
|
||
if test -z "${TMPDIR}" && test -d '/dev/shm' && mountpoint -q '/dev/shm'; then
|
||
TMPDIR='/dev/shm'
|
||
export TMPDIR
|
||
fi
|
||
|
||
# Temporary working area, cleaned up on exit.
|
||
workDir="$(mktemp -d)" || exit 1
|
||
trap 'rm -rf "${workDir}"' EXIT
|
||
|
||
# Set everything to the "C" locale.
|
||
LANG=C
|
||
LC_ALL=C
|
||
export LANG LC_ALL
|
||
|
||
# Define the measurements.
|
||
defineMeasurements
|
||
|
||
# Run the selected action.
|
||
case "${action}" in
|
||
'measurements') showMeasurementDefinitions ;;
|
||
'benchmark') runBenchmarks "${pv}" "${restrictMeasurementIdList}" ;;
|
||
'analyse') runVersionComparisons ;;
|
||
esac
|