X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fscripts%2Fbdev-io-survey.sh;h=016bd65bb3beed50a160ee18b24c81011f8910b2;hb=9fd45a236b6fb7afb559903f05eba0c668ff3c61;hp=9fbd772c7a112b931ef3606cc4731f5f8b992c79;hpb=4240f48c73f70aaaf69a387c161bbece52a80982;p=fs%2Flustre-release.git diff --git a/lustre/scripts/bdev-io-survey.sh b/lustre/scripts/bdev-io-survey.sh index 9fbd772..016bd65 100755 --- a/lustre/scripts/bdev-io-survey.sh +++ b/lustre/scripts/bdev-io-survey.sh @@ -20,12 +20,18 @@ declare -a cleanup_pids declare -a cleanup_mounts # global for completing the table. XXX this is a wart that could go cur_y="0" +# a global which funcs use to get at the blocks[] array +last_block=-1 +# prefix to run oprofile or readprofile +oprofile="" +readprofile="" # defaults for some options: min_threads=1 max_threads=4 possible_tests="sgp_dd ext2_iozone echo_filter" run_tests="$possible_tests" +echo_module="" # optional output directory output_dir="" @@ -223,7 +229,7 @@ sgp_dd_start() { local bdev=${blocks[$i]}; case "$wor" in - w) ifof="if=/dev/zero of=$bdev" ;; + [wo]) ifof="if=/dev/zero of=$bdev" ;; r) ifof="if=$bdev of=/dev/null" ;; *) die "asked to do io with $wor?" esac @@ -265,6 +271,10 @@ ext2_iozone_prepare() { echo "iozone binary not found in PATH" return 1 fi + if ! iozone -i 0 -w -+o -s 1k -r 1k -f /dev/null > /dev/null; then + echo "iozone doesn't support -+o" + return 1 + fi if ! which mke2fs; then echo "mke2fs binary not found in PATH" return 1 @@ -295,7 +305,7 @@ ext2_iozone_setup() { case "$wor" in w) rm -f $f ;; - r) ;; + [or]) ;; *) die "asked to do io with $wor?" esac } @@ -308,30 +318,36 @@ ext2_iozone_start() { local f="$tmpdir/mount_$id/iozone" case "$wor" in - w) args="-i 0 -w" ;; - r) args="-i 1 -w" ;; + [wo]) args="-i 0 -w" ;; + r) args="-i 1" ;; *) die "asked to do io with $wor?" esac - echo iozone "$args -r ${iosize}k -s ${io_len}k -I -f $f" + echo iozone "$args -r ${iosize}k -s $(($io_len / $threads))k \ + -t $threads -+o -x -I -f $f" } ext2_iozone_result() { local output=$1 - - kps=`awk '($2 == "reclen"){results=NR+1}(results == NR){print $3}' \ - < $output` - do_bc_scale 0 "$kps / 1024" -} -ext2_iozone_cleanup() { - local id=$1 local wor=$2 - local f="$tmpdir/mount_$id/iozone" + local string + local field case "$wor" in - w) ;; - r) rm -f $f ;; + [wo]) string="writers" + field=7 + ;; + r) string="readers" + field=6 + ;; *) die "asked to do io with $wor?" esac + + do_bc_scale 1 `awk '($1 == "Parent" && $'$field' == "'$string'") \ + {print $'$(($field + 2))'}' $output` / 1024 +} +ext2_iozone_cleanup() { + # the final read w/o -w removed the file + local nothing=0 } ext2_iozone_finish() { local index=$1 @@ -349,7 +365,7 @@ ext2_iozone_teardown() { # the echo_client setup is nutty enough to warrant its own clenaup running_config="" -running_modules="" +running_module="" declare -a running_names declare -a running_oids @@ -372,12 +388,12 @@ detach quit EOF done - running_names="" + unset running_names - for m in $running_modules; do + for m in $running_module; do rmmod $m done - running_modules="" + running_module="" [ ! -z "$running_config" ] && lconf --cleanup $running_config running_config="" @@ -405,7 +421,7 @@ echo_filter_config() { fi if [ $index = 0 ]; then - if ! lmc -m $config --add net \ + if ! lmc -m $config --add net \ --node localhost --nid localhost --nettype tcp; then echo "error adding localhost net node" return 1 @@ -434,12 +450,28 @@ echo_filter_prepare() { return 1; fi running_config="$config" + + echo 0 > /proc/sys/portals/debug + echo 0 > /proc/sys/portals/subsystem_debug + if ! grep -q '^obdecho\>' /proc/modules; then + local m if ! modprobe obdecho; then - echo "error running modprobe obdecho" - return 1; + if [ ! -z "$echo_module" ]; then + if ! insmod $echo_module; then + echo "err: insmod $echo_module" + return 1; + else + m="$echo_module" + fi + else + echo "err: modprobe $obdecho" + return 1; + fi + else + m=obdecho fi - running_modules="obdecho" + running_module=`basename $m | cut -d'.' -f 1` fi fi @@ -464,13 +496,13 @@ echo_filter_setup() { case "$wor" in w) ;; - r) return ;; + [or]) return ;; *) die "asked to do io with $wor?" esac running_threads=$threads oid=`lctl --device "\$"$name create $threads | \ - awk '/1 is object id/ { print $6 }'` + awk '/ #1 is object id/ { print $6 }'` # XXX need to deal with errors running_oids[$id]=$oid } @@ -479,17 +511,20 @@ echo_filter_start() { local iosize=$2 local wor=$3 local id=$4 + local rw + local name="echo_$id" - local pages=$(($io_len / 4)) + local len_pages=$(($io_len / $(($page_size / 1024)) / $threads )) + local size_pages=$(($iosize / $(($page_size / 1024)) )) case "$wor" in - w) args="-i 0 -w" ;; - r) args="-i 1 -w" ;; + [wo]) rw="w" ;; + r) rw="r" ;; *) die "asked to do io with $wor?" esac echo lctl --threads $threads v "\$"$name \ - test_brw 1 w v $pages ${running_oids[$i]} p$iosize + test_brw 1 $rw v $len_pages t${running_oids[$id]} p$size_pages } echo_filter_result() { local output=$1 @@ -499,7 +534,7 @@ echo_filter_result() { for mbs in `awk '($8=="MB/s):"){print substr($7,2)}' < $output`; do total=$(do_bc $total + $mbs) done - do_bc_scale $total / 1 + do_bc_scale 2 $total / 1 } echo_filter_cleanup() { local id=$1 @@ -508,13 +543,13 @@ echo_filter_cleanup() { local name="echo_$id" case "$wor" in - w) return ;; + [wo]) return ;; r) ;; *) die "asked to do io with $wor?" esac - lctl --device "\$"$name destroy ${running_oids[$i]} $threads - unset running_oids[$i] + lctl --device "\$"$name destroy ${running_oids[$id]} $threads + unset running_oids[$id] } echo_filter_finish() { local index=$1 @@ -551,6 +586,8 @@ test_one() { echo $test with $threads threads + $oprofile opcontrol --start + # start up vmstat and record its pid nice -19 vmstat 1 > $vmstat_log 2>&1 & [ $? = 0 ] || die "vmstat failed" @@ -567,10 +604,14 @@ test_one() { iostat_pids[$i]=$pid done + $oprofile opcontrol --reset + $readprofile -r + # start all the tests. each returns a pid to wait on pids="" for i in `seq 0 $last_block`; do local cmd=`${test}_start $threads $iosize $wor $i` + echo "$cmd" >> $tmpdir/commands $cmd > $tmpdir/$i 2>&1 & local pid=$! pids="$pids $pid" @@ -596,6 +637,16 @@ test_one() { pid_has_stopped $pid done + $readprofile | sort -rn > $tmpdir/readprofile + + $oprofile opcontrol --shutdown + $oprofile opreport > $tmpdir/oprofile + echo >> $tmpdir/oprofile + $oprofile opreport -c -l | head -20 >> $tmpdir/oprofile + + save_output $tmpdir/oprofile $opref.oprofile + save_output $tmpdir/readprofile $opref.readprofile + # collect the results of vmstat and iostat cpu=$(mean_stddev $(awk \ '(NR > 3 && NF == 16 && $16 != "id" ) \ @@ -620,7 +671,7 @@ test_one() { # record each index's test results and sum them thru=0 for i in `seq 0 $last_block`; do - local t=`${test}_result $tmpdir/$i` + local t=`${test}_result $tmpdir/$i $wor` save_output $tmpdir/$i $opref.$i echo test returned "$t" mb_s[$i]="$t" @@ -680,11 +731,11 @@ test_iterator() { done while [ -z "$cleanup" -a $thr -lt $(($max_threads + 1)) ]; do - for iosize in 64 128; do + for iosize in 128 512; do table_set $test 0 $cur_y $thr table_set $test 1 $cur_y $iosize - for wor in w r; do + for wor in w o r; do table_set $test 2 $cur_y $wor test_one $test 3 $thr $iosize $wor done @@ -705,8 +756,9 @@ test_iterator() { return $rc; } -while getopts ":d:b:l:t:T:r:" opt; do +while getopts ":d:b:l:t:T:r:e:" opt; do case $opt in + e) echo_module=$OPTARG ;; b) block=$OPTARG ;; d) output_dir=$OPTARG ;; l) io_len=$OPTARG ;; @@ -717,14 +769,21 @@ while getopts ":d:b:l:t:T:r:" opt; do esac done +page_size=`getconf PAGE_SIZE` || die '"getconf PAGE_SIZE" failed' + +[ ! -z "$echo_module" -a ! -f "$echo_module" ] && \ + die "obdecho module $echo_module is not a file" + if [ -z "$io_len" ]; then io_len=`awk '($1 == "MemTotal:"){print $2}' < /proc/meminfo` [ -z "$io_len" ] && die "couldn't determine the amount of memory" fi if [ ! -z "$output_dir" ]; then - [ ! -e "$output_dir" ] && "output dir $output_dir doesn't exist" - [ ! -d "$output_dir" ] && "output dir $output_dir isn't a directory" + if [ ! -e "$output_dir" ]; then + mkdir -p "$output_dir" || die "error creating $output_dir" + fi + [ ! -d "$output_dir" ] && die "$output_dir isn't a directory" fi block=`echo $block | sed -e 's/,/ /g'` @@ -738,10 +797,29 @@ for t in $run_tests; do fi done +if which opcontrol; then + echo generating oprofile results + oprofile="" +else + echo not using oprofile + oprofile=": " +fi + +if which readprofile; then + map="/boot/System.map-`uname -r`" + if [ -f /proc/profile -a -f "$map" ]; then + echo generating profiles with 'readprofile' + readprofile="readprofile -m $map" + fi +fi +if [ -z "$readprofile" ]; then + echo not using readprofile + readprofile=": " +fi + [ $min_threads -gt $max_threads ] && \ die "min threads $min_threads must be <= min_threads $min_threads" -last_block=-1 for b in $block; do [ ! -e $b ] && die "block device file $b doesn't exist" [ ! -b $b ] && die "$b isn't a block device" @@ -779,11 +857,14 @@ for t in $run_tests; do test_results="$test_results $t" done +save_output $tmpdir/commands commands + [ ! -z "$test_results" ] && ( echo echo "T = number of concurrent threads per device" echo "L = base io operation length, in KB" echo "m = IO method: read, write, or over-write" + echo "A = aggregate throughput from all devices" echo "C = percentage CPU used, both user and system" echo "MB/s = per-device throughput" echo "rR = read requests issued to the device per second"