Home / server / Questions / 146033
Accepted
GetFree
Asked: 2010-05-28 13:16:35 +0800 CST

Linux: How to find out whether the harddrive is causing a bottleneck?

  • 772

I'm not asking how to find out where a bottleneck is. I just what to know how to check if the harddrive is being used at its limit (and hence causing a bottleneck).

Something like top where I can see a usage percentage of CPU and RAM.
Is there something like that with which I can monitor harddrive activity?

Native linux tools preferred.

performance hard-drive linux monitoring performance-monitoring

9 Answers

  1. You want the sysstat utilities: iostat and sar are what you'll find most useful, I think.

    iostat will give you current stats. sar will grab and store and show you historical stats.

    Also possibly useful is iotop, which is like iostat but ties things to PIDs. It's newer though, and I don't know as much about it.

    • 15
  2. Best Answer

    Take a look at:

    • 12

  3. You already mentioned top. It's useful enough to give you a simple realtime figure.

    Check the %wa counter on the CPU line near the top of the screen. It tells you the percentage of time the CPU currently spends waiting on I/O. This should be close to zero.

    • 8

  4. I'm a huge fan of iostat for watching disk activity on the console. You'll be able to see the IOPs being service by the drive, throughput, it's utilization %, request wait time and more. If you are seeing high wait, service and utilization that'll be the give away.

    • 6

  5. Another tool that will allow you to see disk activity (among other things) on Linux (or AIX) is nmon. It will display either a table or a bar graph and shows both per-partition and per-device data.

    • 2

  6. sar is the answer. Its gathers statics on queue length, request response time, etc. You can run it interactively, or even better run it in the background and analyze the data later.

    • 1
    • 1

  8. iostat (or nmon which also displays the same stats in a different format) is good to get a system level diagnostic there are other tools which are more fine grained if you need it. I was recently introduced to blktrace/blkparse which came with the OS and will record and strace like level of detail about every IO request from every part of the system. You can record number and size of IO requests per process, including kernel processes like swap and filesystem journaling, as well as latency.

    • 0

  9. personally I find sar and iostat too verbose to be of value to me. they take multiple lines out output and are very difficult to view (my opinion). try collectl! same data and more, but much easier to read the output. -mark

    • 0