Home / server / Questions / 342273
Accepted
rush
Asked: 2011-12-20 01:29:26 +0800 CST

Disk IOPS count in GNU/Linux

  • 772

I can't find information about ways to count disk iops in GNU/Linux for some time period. I saw iostat, but I'm not sure it is right way. Also i need tool, that can count iops for all time since system is started, but iostat doesn't do this. I mean disk iops for whole systems and for any logical volume.

May be there are some tools to limit disk iops for logical volume or smth like that.

If it matters, I use Debian. Also I use hardware MegaRaid Raid Controller. May be it influences on smth.

linux iops

3 Answers

  1. Best Answer

    iostat is the right way to get IO count over time.

    If you want totals since boot, you can read those from /proc/disktats. Description of the file format is in your kernel's Documentation/iostats.txt (or here).

    This information is also available per-device or per-partition in /sys/block/${DEVICE}/stat and /sys/block/${DEVICE}/${DEVPART}/stat (substitute ${DEVICE} with your device name eg. sda and ${DEVPART} with your partition name eg. sda5).

    • 7

  2. I prefer to use cd /dev; iostat -xk 3 sd? fio? to watch disk IO. Take a look at this sample excerpt:

    avg-cpu:  %user   %nice %system %iowait  %steal   %idle
           1.20    0.00    4.58    0.00    0.00   94.22

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await  svctm  %util
sdg               0.00     0.00    6.67  238.00  3413.33 39774.67   353.04     0.25    1.02   0.37   9.17
sda               0.00     0.00    5.33 3570.67  2730.67 42230.50    25.15     0.44    0.12   0.07  25.20
sdc               0.00     0.00   10.33  795.00  3089.33 44510.00   118.21     0.40    0.47   0.16  12.83
sdf               0.00     0.00    6.67  254.67  3413.33 40318.67   334.68     0.24    0.93   0.35   9.07
sdh               0.00     0.00   14.33  338.00  3444.00 43286.67   265.26     0.27    0.78   0.29  10.23
sdi               0.00     0.00    8.67  906.33  4437.33 44533.17   107.04     0.36    0.40   0.15  14.17
sdb               0.00     0.00    4.67 2355.33  2389.33 44427.50    39.68     0.51    0.21   0.08  18.87
sdd               0.00     0.00    7.00  256.00  3414.67 40434.67   333.46     0.32    1.22   0.37   9.60
sde               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
fioa              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00

    Your average IOPS for this interval is the sum of r/s and w/s and your avgrq-sz (average request size in sectors) gives you an idea of whether the workload is random or sequential.

    Take a look at sdg vs. sda in the example above. Both are writing around 40MBps to disk but the request size is much lower for sda (random workload) resulting in a higher IOPS.

    If you want to track IOPS (and other performance) for an extended period of time I strongly suggest using nmon to collect the data and generate pretty graphs.

    • 3

  3. Does iotop do what you want ? It's maybe a little more real-time than you wanted.

    else, try iostat 5 5

    • 0