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 # and max # threads one instance will spawn
  41 SG_MAX_QUEUE=16
  42
  43 # map given device names into SG device names
  44 i=0
  45 devs=()
  46 if [ "$scsidevs" ]; then
  47         # we will test for a LUN, the test for a partition
  48         # if the partition number is > 9 this will fail
  49     for d in $scsidevs; do
  50         devs[$i]=`sg_map | awk "{if ($ 2 == \"$d\") print $ 1}"`
  51         if [ -z "${devs[i]}" ]; then
  52             echo "Can't find SG device for $d, testing for partition"
  53             pt=`echo $d | sed 's/[0-9]$//'`
  54             # Try again
  55             devs[$i]=`sg_map | awk "{if ($ 2 == \"$pt\") print $ 1}"`
  56             if [ -z "${devs[i]}" ]; then
  57                 echo "Can't find SG device $pt"
  58                 exit 1
  59            fi
  60         fi
  61         i=$((i+1))
  62     done
  63 elif [ "$rawdevs" ]; then
  64     for r in $rawdevs; do
  65         RES=`raw -q $r`
  66         if [ $? -eq 0 ];then
  67             devs[$i]=$r
  68             i=$((i+1))
  69         else
  70             echo "Raw device $r not set up"
  71             exit 1
  72         fi
  73     done
  74 else
  75     echo "Must specify scsidevs or rawdevs"
  76     exit 1
  77 fi
  78
  79 ndevs=${#devs[@]}
  80
  81 # determine block size. This should also work for raw devices
  82 # If it fails, set to 512
  83 bs=$((`sg_readcap -b ${devs[0]}  |awk '{print $2}'`))
  84 if [ $bs == 0  ];then
  85         echo "sg_readcap failed, setting block size to 512"
  86         bs=512
  87 fi
  88 rsltf=${rslt}.summary
  89 workf=${rslt}.detail
  90 echo -n > $rsltf
  91 echo -n > $workf
  92
  93 print_summary () {
  94     if [ "$1" = "-n" ]; then
  95         minusn=$1; shift
  96     else
  97         minusn=""
  98     fi
  99     echo $minusn "$*" >> $rsltf
 100     echo $minusn "$*"
 101 }
 102
 103 print_summary "$(date) sgpdd-survey on $rawdevs$scsidevs from $(hostname)"
 104
 105 for ((rsz=$rszlo;rsz<=$rszhi;rsz*=2)); do
 106     for ((crg=$crglo;crg<=$crghi;crg*=2)); do
 107         for ((thr=$thrlo;thr<=$thrhi;thr*=2)); do
 108             if ((thr < crg || thr/crg > SG_MAX_QUEUE)); then
 109                 continue
 110             fi
 111             # compute parameters
 112             bpt=$((rsz*1024/bs))
 113             blocks=$((size*((1024*1024)/bs)/crg))
 114             count=$blocks
 115             # show computed parameters
 116             actual_rsz=$((bpt*bs/1024))
 117             actual_size=$((bs*count*crg/1024))
 118             str=`printf 'total_size %8dK rsz %4d crg %5d thr %5d ' \
 119                          $((actual_size*ndevs)) $actual_rsz $((crg*ndevs)) $((thr*ndevs))`
 120             echo "==============> $str" >> $workf
 121             print_summary -n "$str"
 122             freemem=`awk < /proc/meminfo '/^MemTotal:/ {printf "%d\n", $2}'`
 123             if (((actual_rsz*thr/crg + 64)*crg*ndevs > freemem)); then
 124                 print_summary "ENOMEM"
 125                 continue
 126             fi
 127             # run tests
 128             for action in $actions; do
 129                 print_summary -n "$action "
 130                 echo "=====> $action" >> $workf
 131                 tmpf=${workf}_tmp
 132                 # start test
 133                 t0=`date +%s.%N`
 134                 for ((i=0;i<ndevs;i++)); do
 135                     dev=${devs[i]}
 136                     if [ $action = read ]; then
 137                         inf="if=$dev"
 138                         outf="of=/dev/null"
 139                         skip=skip
 140                     else
 141                         inf="if=/dev/zero"
 142                         outf="of=$dev"
 143                         skip=seek
 144                     fi
 145                     for ((j=0;j<crg;j++)); do
 146                         sgp_dd 2> ${tmpf}_${i}_${j} \
 147                             $inf $outf ${skip}=$((1024+j*blocks)) \
 148                             thr=$((thr/crg)) count=$count bs=$bs bpt=$bpt time=1&
 149                     done
 150                 done
 151                 wait
 152                 t1=`date +%s.%N`
 153                 # collect/check individual stats
 154                 echo > $tmpf
 155                 ok=0
 156                 for ((i=0;i<ndevs;i++)); do
 157                     for ((j=0;j<crg;j++)); do
 158                         rtmp=${tmpf}_${i}_${j}
 159                         if grep 'error' $rtmp > /dev/null 2>&1; then
 160                                 echo "Error found in $rtmp"
 161                         elif 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