X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre-iokit%2Fior-survey%2Fior-survey;h=b7c7421e78675de916b02044f07671a7b2bbe252;hp=764744ac7142b64bdd7eacae4c8e0e1c159bc8c0;hb=316113bbe60da637297a914f336c5f1f81257c6a;hpb=00fc7ec0e84c297275f3d746b18ea0310e69cf57 diff --git a/lustre-iokit/ior-survey/ior-survey b/lustre-iokit/ior-survey/ior-survey index 764744a..b7c7421 100644 --- a/lustre-iokit/ior-survey/ior-survey +++ b/lustre-iokit/ior-survey/ior-survey @@ -1,91 +1,69 @@ #!/bin/bash -cluster=adev -servers=(4 7) -server_disks=([4]=sdd [7]=sdd) -clients=(8-15 0-1) +# cluster name (all node names are this followed by the node number) +cluster=mdev -min_clients=1 -max_clients=10 +# client node numbers (individual numbers or inclusive ranges) +clients=(7-8) -per_client_size=4G -transfer_size=1M -tasks_per_client=1 +# numbers of clients to survey +clients_lo=1 +clients_hi=2 +clients_iterator="+=1" + +# numbers of tasks per client to survey +tasks_per_client_lo=1 +tasks_per_client_hi=8 +tasks_per_client_iterator="*=2" + +# record sizes to survey +rsize_lo=1M +rsize_hi=1M +rsize_iterator="*=2" + +## which tests to run (first must be write) +# remount) not really a test; just remount to uncache everything +# *write*) write +# *) read +#tests=(write rewrite read reread rewrite_again) +tests=(write rewrite remount read reread) + +# total # bytes written/read by any client node +min_per_client_size=4G +min_total_size=8G + +# should each task do I/O to its own file? file_per_task=1 +# the binaries IOR="/home/ericb/ior/src/C/IOR" -script="/home/ericb/eeb_ior_script" +llmount=/home/ericb/lustre/utils/llmount +pdsh=pdsh -testfile=/mnt/lustre/ior_survey_testfile +# the result file prefix (date/time + hostname makes unique) +#rslt=/home/ericb/ior_survey_`date +%F@%R`_`uname -n` +rslt=/home/ericb/ior_survey +# where lustre is mounted on the clients +lustre=/mnt/lustre -################################################################################ -cat > $script < -# blockSize= -b 8G -# transferSize= -t 1M -# filePerProc= -F -# testFile= -o /mnt/lustre/ior_testfile -# uniqueDir= -u -# verbose= -v -# reorderTasks= -C - -# unused options -# collective=0 (MPI only) -# individualDataSets=0 [not working] -# noFill=0 (HDF5 only) -# preallocate=0 (MPI only) -# useSharedFilePointer=0 [not working] -# useFileView= -# useStridedDataType=0(MPI only) -# showHints=0 -# showHelp=0 - -# setup - api=POSIX - fsync=1 - repetitions=1 - useO_DIRECT=0 - interTestDelay=10 - intraTestBarriers=1 - storeFileOffset=0 - quitOnError=1 - segmentCount=1 - singleXferAttempt=0 - -# write - readFile=0 - writeFile=1 - useExistingTestFile=0 - keepFile=1 - RUN - -# rewrite - useExistingTestFile=1 - RUN - -# read - readFile=1 - writeFile=0 - useExistingTestFile=1 - RUN - -# reread (bug in ior means it needs each run needs at least 1 directive) - readFile=1 - RUN - -# write again - readFile=0 - writeFile=1 - useExistingTestFile=1 - keepFile=0 - RUN -IOR STOP - -EOF +# basename of the test file(s) +testfile=${lustre}/ior_survey_testfile + +# how to unmount and remount the F/S on a client (to clear the cache) +# change this depending on lustre config (network type, MDS etc) +remount="umount $lustre && $llmount -o nettype=elan mdev6:/ll_mds/client $lustre" + +# pdsh args required to instantiate all instances of IOR in parallel +# the chosen module must support '-n ' +# -R, -f etc +pdsh_mpiargs="-Rmqsh" + +#don't spin for MPI completions +export LIBELAN_WAITTYPE=0 ################################################################################ +# dont change stuff below here unless you know what you're doing... count_range() { echo $1 | awk '{ nvals=split($1, vals, "-");\ @@ -98,9 +76,9 @@ base_range() { } idx2nodenum() { - n=$1; shift + local n=$1; shift while ((1)); do - range=$1; shift + local range=$1; shift if [ -z "$range" ]; then return fi @@ -115,18 +93,18 @@ idx2nodenum() { } n2noderange() { - n=$1; shift + local n=$1; shift sep="" nodes="[" while ((n > 0)); do - range=$1; shift + local range=$1; shift if [ -z "$range" ]; then return fi - base=`base_range $range` - chunk=`count_range $range` + local base=`base_range $range` + local chunk=`count_range $range` if ((chunk > n)); then chunk=n; fi - nodes="${nodes}${sep}${base}"; sep="," + local nodes="${nodes}${sep}${base}"; sep="," if ((chunk > 1)); then nodes="${nodes}-$((base+chunk-1))"; fi n=$((n-chunk)) done @@ -134,10 +112,10 @@ n2noderange() { } countnodes() { - radix=16384 - n=0 + local radix=16384 + local n=0 while ((radix > 0)); do - nodes=`n2noderange $((n+radix)) $@` + local nodes=`n2noderange $((n+radix)) $@` if [ -n "$nodes" ]; then n=$((n+radix)) fi @@ -147,7 +125,7 @@ countnodes() { } parse_number() { - str=$1 + local str=$1 case $str in *G|*g) n=`echo $str | sed 's/[gG]//'`; echo $((n*1024*1024*1024));; *M|*m) n=`echo $str | sed 's/[Mm]//'`; echo $((n*1024*1024));; @@ -157,10 +135,10 @@ parse_number() { } pp_number() { - n=$1 - G=$((1024*1024*1024)) - M=$((1024*1024)) - K=$((1024)) + local n=$1 + local G=$((1024*1024*1024)) + local M=$((1024*1024)) + local K=$((1024)) if ((n%G == 0 && n >= G)); then echo "$((n/G))G" elif ((n%M == 0 && n >= M)); then @@ -172,22 +150,145 @@ pp_number() { fi } -nservers=`countnodes ${servers[@]}` +if [ ${#tests[@]} -eq 0 -o "${tests[0]}" != "write" ]; then + echo "First test must be 'write'" 1>&2 + exit 1 +fi + +rsltf="${rslt}.summary" +workf="${rslt}.detail" +echo -n > $rsltf +echo -n > $workf + +print_summary () { + if [ "$1" = "-n" ]; then + minusn=$1; shift + else + minusn="" + fi + echo $minusn "$*" >> $rsltf + echo $minusn "$*" +} + +# convert params to actual numbers +min_per_client_size=`parse_number $min_per_client_size` +min_total_size=`parse_number $min_total_size` + +rsize_lo=`parse_number $rsize_lo` +rsize_hi=`parse_number $rsize_hi` + +# check on actual numbers of client nodes nclients=`countnodes ${clients[@]}` +if ((clients_hi > nclients)); then clients_hi=$nclients; fi + +for ((rsize=rsize_lo; rsize<=rsize_hi; rsize$rsize_iterator)); do + pp_rsize=`pp_number $rsize` + + for ((nclnt=clients_lo; nclnt<=clients_hi; nclnt$clients_iterator)); do + test_clients="${cluster}`n2noderange $nclnt ${clients[@]}`" + + per_client_size=$((min_total_size/nclnt)) + if ((per_client_size < min_per_client_size)); then + per_client_size=$min_per_client_size + fi + + for ((ntask=tasks_per_client_lo; ntask <= tasks_per_client_hi; ntask$tasks_per_client_iterator)); do + per_task_size=$((per_client_size/ntask)) + if ((per_task_size%rsize != 0)); then + per_task_size=$(((per_task_size/rsize + 1)*rsize)) + fi + total_size=`pp_number $((per_task_size*nclnt*ntask))` + + hdrstr=`printf "Total: %5sB rsize: %4sB clients: %4d tasks: %3d: " \ + $total_size $pp_rsize $nclnt $ntask` + print_summary -n "$hdrstr" + + for ((test_idx=0; test_idx < ${#tests[@]}; test_idx++)); do + test=${tests[$test_idx]} + + print_summary -n "$test " + echo "===========> ${hdrstr} on $test_clients doing $test" >> $workf + tmpf=${workf}_tmp + echo -n > $tmpf -if ((max_clients > nclients)); then max_clients=$nclients; fi -if ((file_per_task)); then minusFopt=-F; else minusFopt=""; fi + if [ "$test" = "remount" ]; then + echo "=> $remount" >> $tmpf + $pdsh -S -b -w "$test_clients" >> $tmpf 2>&1 \ + "$remount" + status=$? + echo "Completion Status: $status" >> $tmpf -for ((i=min_clients - 1;i Mount Check: $cmd" >> $tmpf + $pdsh -S -b -w "$test_clients" >> $tmpf 2>&1 \ + "$cmd" + status=$? + echo "Completion Status: $status" >> $tmpf + if ((status)); then + cat $tmpf >> $workf + rm $tmpf + print_summary "Lustre NOT mounted on $lustre somewhere" + exit 1 + fi - clients="${cluster}`n2noderange $((i+1)) ${clients[@]}`" - per_task_size=$((`parse_number $per_client_size`/tasks_per_client)) - per_task_size=`pp_number $per_task_size` + cmdline=( + $IOR # the command + -o${testfile} # test file prefix + -b${per_task_size} # bytes per task + -t${rsize} # record size + -e # fsync before close + -q # quit on error + ) - pdsh -b -Rmqsh -w "$clients" -n $tasks_per_client \ - $IOR -b${per_task_size} -t${transfer_size} $minusFopt -o $testfile -f $script + idx=${#cmdline[@]} + # keep the test file(s) unless this is the last test + ((test_idx < ${#tests[@]}-1)) && cmdline[$((idx++))]="-k" + + # use the existing test file(s) unless this is the first test + ((test_idx > 0)) && cmdline[$((idx++))]="-E" + + # file-per-task + ((file_per_task)) && cmdline[$((idx++))]="-F" + + case "$test" in + *write*) cmdline[$((idx++))]="-w" + awkstr="Max Write";; + *) cmdline[$((idx++))]="-r" + awkstr="Max Read";; + esac + + echo "=> ${cmdline[@]}" >> $tmpf + + $pdsh -S -b $pdsh_mpiargs -w "$test_clients" -n $ntask >> $tmpf 2>&1 \ + "${cmdline[@]}" + status=$? + + echo "Completion Status: $status" >> $tmpf + + if ((status)); then + result="ERROR" + else + result=`awk < $tmpf "/$awkstr/ {print $ 3; found=1; exit}\ + END {if (!found) print \"ERROR\"}"` + fi + fi + + cat $tmpf >> $workf + rm $tmpf + + str=`printf "%8s" "$result"` + print_summary -n "$str " + done + print_summary "" + done + done done