How to measure the transfer rate over the production?

[Boonchu Ngampairoijpibu_1]

I found a problem to find the real measurement rate of disk transfer. How do I have a cool procedure to measure the transfer disk rate? What if I have two Fibre Channel FC60 and full split-type ULTRA LVDs with two disk SC10 enclosures? I read the previous article in regards of using dd command and output to /dev/null. But, I am quite not sure how to get the real measurement rate.

[Vincenzo Restuccia]

Try iostat.

[Stefan Farrelly]

if you do an ioscan -fknCdisk it lists all the disk devices (or Luns - which are groups of disks) visible to your HP server.

eg. ioscan -fknCdisk

Class I H/W Path Driver S/W State H/W Type Description
======================================================================
disk 22 8/0.4.0 sdisk CLAIMED DEVICE SEAGATE ST39173WC
/dev/dsk/c0t4d0 /dev/rdsk/c0t4d0
disk 23 8/0.12.0 sdisk CLAIMED DEVICE SEAGATE ST39173WC
/dev/dsk/c0t12d0 /dev/rdsk/c0t12d0
disk 24 8/0.13.0 sdisk CLAIMED DEVICE SEAGATE ST39173WC
/dev/dsk/c0t13d0 /dev/rdsk/c0t13d0
disk 25 8/0.14.0 sdisk CLAIMED DEVICE SEAGATE ST39173WC
/dev/dsk/c0t14d0 /dev/rdsk/c0t14d0
disk 26 8/0.15.0 sdisk CLAIMED DEVICE SEAGATE ST34371W
/dev/dsk/c0t15d0 /dev/rdsk/c0t15d0
disk 1 8/4.0.0 sdisk CLAIMED DEVICE EMC SYMMETRIX
/dev/dsk/c1t0d0 /dev/rdsk/c1t0d0

You can see from this there are 5 disk devices on the SCSI controller 8/0 (H/W Path) and 1 on 8/4 (The EMC device/lun).

To test each in turn do;

time dd if=/dev/rdsk/cXXXXX of=/dev/null bs=1024k count=50

This will read 50 MB off the disk device (you can set this to whatever value you want but 50 is sufficient for a good read test) and give you a time in seconds to do it.
eg.
50+0 records in
50+0 records out

real 3.5
user 0.0
sys 0.0

Now, divide 50 (MB) by 3.5 seconds gives 14.2 MB/s for that disk device. Pretty quick - considering the HP FWD SCSI controller will only run as fast as around 15 MB/s.

When I do the same test on the EMC device (which is a Lun - combined over multiple disks inside the EMC frame - you will need to talk to EMC to find out exactly how, but that doesnt really concern us, just its speed!)
time dd gives me 4.6 secs, so 10.8 MB/s. But, the EMC has tons of cache so if I repeat the test on the EMC I now get 2.9s, for 17.2 MB/s (Which is great - the HP scsi controller wont really go any faster). The first disk I tried it on was an internal drive (no cache - or very small cache so no improvement) so multiple dd tests give the same result - it doesnt get faster like the EMC does.

This is how to test your disk device speeds.

[Boonchu Ngampairoijpibu_1]

Thanks for response, especially
Stefan guide me to the real life.
I got FC60 with cache size 256 Mbytes.

I try with RAID 1 device. It returns
as you say. First time, it tooks about
2.55 seconds per 50 Mbytes. Without RAID
, it responses with 3.32 seconds per
50 Mbytes.

But, it tooks better response when I use
them twice. It responses back 0.733 seconds
per 50 Mbytes.

So it writes to cache to acheive the better
rate.

Stefan, If I want to test with lvm device,
does it return with the same rate?

[David Hixson]

Boonchu,

That test will work exactly the same if you give it an lvol instead of a raw disk device.

[Roberto Gallis]

I think is more usefull to test the speed using file systems.
The tests above are directly to the disk...
Probabli You use a filesystem to read/write to your disks lvol.
So, try a write test

time prealloc file

Try a dimension bigger then your cache and try to do the test 2-3 times with different file names

Then try to read using dd

dd if=file of=/dev/null
with the right block size

Regards

Roberto

[Roberto Gallis]

Sorry,
the right command is

time prealloc file