howto measure maximum throughput of harddisc?

[Bert Kamps]

I always use the following command to measure the maximum disc transfer rate. But the results are always different.

Do you know a good way to measure the disc speeds?

command:
time dd if=/dev/dsk/c0t0d0 of=/dev/null bs=1024k count=100

Do it copies 100 mb of data to /dev/null. Then you'll see the time it takes for that action.
You divide the 100 mb with the time, and then you'll have your maximum disk througput...

P.S. This thread has been moevd from Disk to HP-UX > sysadmin. - Hp forum moderator

[Bill Hassell · Accepted solution]

Be sure you use the raw disk device file and not the blocked one. Using the blocked device can result in buffered input through the system's buffer cache. The result is that the first and subsequent tests differ significantly in elapsed time.

Also, the disk (which may be an array controller) may have significant on-board buffering (disks aren't simple devices anymore) so it is a good idea to run several passes. And also vary the record size. Although the dd task is sequential reading, various features of the opsystem and hardware may create dips and spikes in performance.

[Bert Kamps]

Now i'm meausering only 1 'read'.
But i wanna know what happens when there's 10 processes or so that are reading data from the disc.

ps: the /dev/dsk/c0t0d0 file is a raw device file right?
(i always forget these kind of things :)

[Bert Kamps]

Oow yeah i'm sorry i put this post on the Networking forum. It's kinda late out here so i'm a bit goofy now :)))

[harry d brown jr]

For raw disk use /dev/rdsk/c0t0d0, as Bill said it won't be buffered. Also, your test doesn't adequately "stress" the disk. Try using 1, then 2, 4, 8, 16, ..., "dd's". That way you get a good sample of how many process will work together and what the max is.

live free or die
harry

[Bert Kamps]

aaahaa, that's the difference for /dev/dsk and /dev/rdsk. Gotta work on my unix theory i see... :)

Yeah i could run some dd's at the same time, but i thought there was some kinda cool program with nice graphs and stuff :) And that it would really stress your disc.

[Steven Sim Kok Leong]

Hi,

You can make use of the STKIO automated disk benchmark tool which is downloadable from the storagetek website and supports most Unixes.

http://www.storagetek.com/products/software/freeware/stkio/

It measures sequential read/write accesses as well as random read/write accesses under varying factors eg:
1) block size of each access
2) no. of threads used
3) file size used
4) etc.

You can also specify to execute the test on either a raw device or a filesystem.

It outputs results in bytes/sec, IO/s as well as time taken (msecs).

I have been using it to benchmark our XP256, other competitive raid arrays, HASS disks previously, and ultimately to qualify for purchase of XP256.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[harry d brown jr]

Bert,

If you want to "stress" your disks, threaten them with a copy of any MS product or MS os ;-))

live free or die
harry

[Santosh Nair_1]

You may also want to take a look at bonnie++, a public domain disk benchmarking suite:

http://www.coker.com.au/bonnie++/
Hope this helps.

-Santosh

[Roger Baptiste]

Bert,

I have used a free application tool called
IOTEST for disk throughput tests. You can download it from:

http://www.soliddata.com/products/iotest.html

It is a neat menu driven tool, gives you lot of options.


HTH
raj