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