Hi Stefan,

that's what IBM told us.

--
Mit freundlichen Grüßen

Michael Prix

On 02/11/2013 01:46 PM, Stefan Folkerts wrote:

Thanks Michael, so the use of the filepool storagepool type does not set
the O_SYNC flag (and therefor uses the cache on the raid controller) but a
normal diskpool does (and therefor doesn't use the cache)?



On Mon, Feb 11, 2013 at 11:45 AM, Michael Prix <mp AT rs6000.darktech DOT 
org>wrote:

Hi Stefan,

I assume with diskpool you mean primary pools devicetype disk.
Create a filepool on the internal disks and measure performance against it.
If this is as expected, the internal RAID-controller honors the
O_SYNC-writes TSM uses for diskpools and in this case the cache of the
RAID-controller is not used.

--
Mit freundlichen Grüßen / kind regards

Michael Prix


On 02/11/2013 11:08 AM, Stefan Folkerts wrote:

Hi Chavdar,

If it would be the raidcontroller I would expect a CIFS copy to be slow as
well but it is not, a LAN based CIFS copy to the same disk the diskpool is
on is fast, the disk is only slow when using it with TSM.

Regards,
     Stefan


On Sun, Feb 10, 2013 at 9:24 PM, Chavdar Cholev <chavdar.cholev AT gmail DOT com

**wrote:

  Hi Stefan,

if it is HP server check to you have cache battery on RAID (if any) I
had simmilar issue, when I do backup form disk to LTO it was ~80-90
MB/s, but when nodes baked up to this disk stg it was ~8-10 MB/s even
I have etherchannel 2x1Gbps ...
Regards
Chavdar

On Sun, Feb 10, 2013 at 4:55 PM, Stefan Folkerts
<stefan.folkerts AT gmail DOT com> wrote:

Hi all,

I am running into a strange performance issue at a small TSM site.
They have an new intel based TSM server running Windows 2008 R2 running

TSM

5.5 (don't ask) with enough CPU and memory to run the server 4 times

over.

It has 2 disks in raid 1 for the TSM log, 4 disks in raid 10 for the TSM
database and 5 disks (all 10k) for the diskpool in raid 5.
The server has 2 1Gb/s ethernet ports in a 2Gb/s LACP channel.

A normal CIFS copy to the server raid 5 filesystem loads the interface
up
to 25%.
A TSM backup to LTO (I believe LTO4) loads the interface up to about the
same load.
However a TSM backup to the diskpool only get the load up to 5-6%.
I have tried a default dsmserv.opt and dsm.opt and 'tuned' ones.
Multiple clients or just one, MSSQL or fileserver data, nothing matters,

as

soon as I go to the diskpool the performance is gone.

Even a local backup to 127.0.0.1 is slow to the diskpool but fast to

tape.

I did filesystem checks, recreated the filesystem, swapped the
raidcontroller (that was done before performance checks and seems a bit
silly now) but I can't find the issue.
There are no errors in Windows or TSM, everything is just fine but very
very slow.
I recreated the diskpool volumes one by one to make sure there is no

weird

fragmentation going on, that didn't change anything, even with a single
1Gb/s connection the speed is still many times faster to tape than it is

to

the diskpool but a filecopy via CIFS to the same disk is fast.

Has anybody ever seen this before?

Regards,
     Stefan