lustre-iokit/sgpdd-survey/sgpdd-survey

   1 #!/bin/bash
   2
   3 ######################################################################
   4 # customize per survey
   5
   6 # CHOOSE EITHER scsidevs or rawdevs
   7 # the SCSI devices to measure - WARNING: will be erased.
   8 # The raw devices to use
   9 # rawdevs=${rawdevs:-"/dev/raw/raw1"}
  10 # scsidevs=`ls /dev/sd[a-z] /dev/sd[a-z][a-z]` # all devices, if you use udev
  11
  12 # result file prefix.  date/time+hostname makes unique
  13 # NB ensure the path exists if it includes subdirs
  14 rslt=${rslt:-"/tmp/sgpdd_survey_`date +%F@%R`_`uname -n`"}
  15
  16 # what to do (read or write)
  17 actions=${actions:-"write read"}
  18
  19 # total size per device (MBytes)
  20 # NB bigger than device cache is good
  21 size=${size:-8192}
  22
  23 # record size (KBytes)
  24 rszlo=${rszlo:-1024}
  25 rszhi=${rszhi:-1024}
  26
  27 # Concurrent regions per device
  28 crglo=${crglo:-1}
  29 crghi=${crghi:-256}
  30
  31 # threads to share between concurrent regions per device
  32 # multiple threads per region simulates a deeper request queue
  33 # NB survey skips over #thr < #regions and #thr/#regions > SG_MAX_QUEUE
  34 thrlo=${thrlo:-1}
  35 thrhi=${thrhi:-4096}
  36
  37 #####################################################################
  38 # leave the rest of this alone unless you know what you're doing...
  39
  40 # sgp_dd's idea of disk sector size (Bytes)
  41 bs=512
  42 # and max # threads one instance will spawn
  43 SG_MAX_QUEUE=16
  44
  45 # map given device names into SG device names
  46 i=0
  47 devs=()
  48 if [ "$scsidevs" ]; then
  49         # we will test for a LUN, the test for a partition
  50         # if the partition number is > 9 this will fail
  51     for d in $scsidevs; do
  52         devs[$i]=`sg_map | awk "{if ($ 2 == \"$d\") print $ 1}"`
  53         if [ -z "${devs[i]}" ]; then
  54             echo "Can't find SG device for $d, testing for partition"
  55             pt=`echo $d | sed 's/[0-9]$//'`
  56             # Try again
  57             devs[$i]=`sg_map | awk "{if ($ 2 == \"$pt\") print $ 1}"`
  58             if [ -z "${devs[i]}" ]; then
  59                 echo "Can't find SG device $pt"
  60                 exit 1
  61            fi
  62         fi
  63         i=$((i+1))
  64     done
  65 elif [ "$rawdevs" ]; then
  66     for r in $rawdevs; do
  67         RES=`raw -q $r`
  68         if [ $? -eq 0 ];then
  69             devs[$i]=$r
  70             i=$((i+1))
  71         else
  72             echo "Raw device $r not set up"
  73             exit 1
  74         fi
  75     done
  76 else
  77     echo "Must specify scsidevs or rawdevs"
  78     exit 1
  79 fi
  80
  81 ndevs=${#devs[@]}
  82
  83 rsltf=${rslt}.summary
  84 workf=${rslt}.detail
  85 echo -n > $rsltf
  86 echo -n > $workf
  87
  88 print_summary () {
  89     if [ "$1" = "-n" ]; then
  90         minusn=$1; shift
  91     else
  92         minusn=""
  93     fi
  94     echo $minusn "$*" >> $rsltf
  95     echo $minusn "$*"
  96 }
  97
  98 time_v=`date`
  99 hostname=`hostname`
 100 if [ "$rawdevs" ]; then
 101         print_summary "$time_v sgpdd-survey on $rawdevs from $hostname"
 102 fi
 103 if [ "$scsidevs" ]; then
 104         print_summary "$time_v sgpdd-survey on $scsidevs from $hostname"
 105 fi
 106
 107 for ((rsz=$rszlo;rsz<=$rszhi;rsz*=2)); do
 108     for ((crg=$crglo;crg<=$crghi;crg*=2)); do
 109         for ((thr=$thrlo;thr<=$thrhi;thr*=2)); do
 110             if ((thr < crg || thr/crg > SG_MAX_QUEUE)); then
 111                 continue
 112             fi
 113             # compute parameters
 114             bpt=$((rsz*1024/bs))
 115             blocks=$((size*((1024*1024)/bs)/crg))
 116             count=$blocks
 117             # show computed parameters
 118             actual_rsz=$((bpt*bs/1024))
 119             actual_size=$((bs*count*crg/1024))
 120             str=`printf 'total_size %8dK rsz %4d crg %5d thr %5d ' \
 121                          $((actual_size*ndevs)) $actual_rsz $((crg*ndevs)) $((thr*ndevs))`
 122             echo "==============> $str" >> $workf
 123             print_summary -n "$str"
 124             freemem=`awk < /proc/meminfo '/^MemTotal:/ {printf "%d\n", $2}'`
 125             if (((actual_rsz*thr/crg + 64)*crg*ndevs > freemem)); then
 126                 print_summary "ENOMEM"
 127                 continue
 128             fi
 129             # run tests
 130             for action in $actions; do
 131                 print_summary -n "$action "
 132                 echo "=====> $action" >> $workf
 133                 tmpf=${workf}_tmp
 134                 # start test
 135                 t0=`date +%s.%N`
 136                 for ((i=0;i<ndevs;i++)); do
 137                     dev=${devs[i]}
 138                     if [ $action = read ]; then
 139                         inf="if=$dev"
 140                         outf="of=/dev/null"
 141                         skip=skip
 142                     else
 143                         inf="if=/dev/zero"
 144                         outf="of=$dev"
 145                         skip=seek
 146                     fi
 147                     for ((j=0;j<crg;j++)); do
 148                         sgp_dd 2> ${tmpf}_${i}_${j} \
 149                             $inf $outf ${skip}=$((1024+j*blocks)) \
 150                             thr=$((thr/crg)) count=$count bs=$bs bpt=$bpt time=1&
 151                     done
 152                 done
 153                 wait
 154                 t1=`date +%s.%N`
 155                 # collect/check individual stats
 156                 echo > $tmpf
 157                 ok=0
 158                 for ((i=0;i<ndevs;i++)); do
 159                     for ((j=0;j<crg;j++)); do
 160                         rtmp=${tmpf}_${i}_${j}
 161                         if grep 'time to transfer data' $rtmp > /dev/null 2>&1; then
 162                             ok=$((ok + 1))
 163                         fi
 164                         cat ${rtmp} >> $tmpf
 165                         cat ${rtmp} >> $workf
 166                         rm  ${rtmp}
 167                     done
 168                 done
 169                 if ((ok != ndevs*crg)); then
 170                     print_summary -n "$((ndevs*crg - ok)) failed "
 171                 else
 172                     # compute MB/sec from elapsed
 173                     bw=`awk "BEGIN {printf \"%7.2f MB/s\", $actual_size * $ndevs / (( $t1 - $t0 ) * 1024); exit}"`
 174                     # compute MB/sec from nregions*slowest
 175                     check=`awk < $tmpf \
 176                         '/time to transfer data/ {mb=$8/1.048576; if (n == 0 || mb < min) min = mb; n++}\
 177                         END {printf "%5d x %6.2f = %7.2f MB/s", n, min, min * n}'`
 178                     print_summary -n "$bw $check "
 179                 fi
 180                 rm $tmpf
 181             done
 182             print_summary ""
 183         done
 184     done
 185 done