IOStat doesn't work?

Message

Syster · Post by **Syster** » Thu May 19, 2011 12:56 pm

Hi all!

I need to get iostat running on my server. So, i've instaled sysstats. But it seems that iostat doesn't working anyway. I've done some simple test:
Firstly, i've checked stats via iostats:

Server linux # iostat -k
Linux 2.6.37-gentoo-r4 (Server) 05/19/11 _i586_ (1 CPU)

avg-cpu: %user %nice %system %iowait %steal %idle
6.66 0.00 1.65 2.02 0.00 89.67

Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
sda 0.00 0.00 0.00 160 0
sdb 1.95 16.40 17.57 5139802 5505604

Then i started to copying 100MB file from /dev/zero to file on disk.
While system was copying the file, i've checked what iostat returns:

Server linux # iostat -k
Linux 2.6.37-gentoo-r4 (Server) 05/19/11 _i586_ (1 CPU)

avg-cpu: %user %nice %system %iowait %steal %idle
6.66 0.00 1.65 2.02 0.00 89.67

Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
sda 0.00 0.00 0.00 160 0
sdb 1.95 16.40 17.91 5139814 5614576

Almost no difference!

Does iostat some special kernel config?
I've found that kernel should be compiled with CONFIG_BLK_STATS=y but it seems to be deprecated in 2.6 kernel line

PS: /dev/sdb is connected to server via USB to ATA converter

Sysa · Post by **Sysa** » Thu May 19, 2011 5:27 pm

Before:

Code: Select all

sdb ... 5505604

After:

Code: Select all

sdb ... 5614576

difference is 100Mb - it means that write operation was finished.

What you want to see?

Try to send more (*Gbs) to see running process...

Syster · Post by **Syster** » Thu May 19, 2011 5:36 pm

So why kB_read/s and kB_write/s is always at the same level?
I don't get it.

Sysa · Post by **Sysa** » Thu May 19, 2011 6:02 pm

Syster wrote:So why kB_read/s and kB_write/s is always at the same level?
I don't get it.

It is because system in the same state.
Look at iostat during long I/O, e.g.

Code: Select all

emerge --sync

in one console and

Code: Select all

watch -n1 iostat

in another and you'll see the difference...

py-ro · Post by **py-ro** » Thu May 19, 2011 9:46 pm

And if u copy from /dev/zero, you will probably get a sparse file.

py

Post by Hu » Fri May 20, 2011 3:44 am

py-ro wrote:And if u copy from /dev/zero, you will probably get a sparse file.

Not necessarily. Reading /dev/zero provides an infinite stream of nulls. It is up to the tool reading from /dev/zero whether this results in a sparse file. Simple tools, such as cat or dd, will default to creating non-sparse files.