XP12000 - I/O size is varying

[Jean-Baptiste Broccard]

Hi all,
I extracted data from XP Disk array and made some calculations. I observe that for a given type of LUN and a given type of test, the IO size is varying.
Example:
I have a Oracle DB in FS mode
=> redovg is on lun1,2...4
=> datavg is on lun5,6...20
I run a huge Sqlloader, the tests report a total of 100Million IO
If I divide the total IO on redo luns by the total MB I have like 8KG IO size.
If I do the same for data luns, IO size is roughly 32KB.
If I have log rotation (mass seq read on redo), I reach ~80KB IO size.

I don't understand why and how these IO sizes vary. Does it come from the server (I use vxfs and 8KB block size) ? From Oracle ?

[Peter Mattei]

Yes sure, the IO size is determined by the Server (OS, HBA/driver, Oracle etc)

Cheers
XP-Pete

[Jean-Baptiste Broccard]

Thanks for the confirmation, but having information on these black boxes XP are so tough even working in HP !

But there's something I want to be sure. Assume that we are in FS mode, buffer-cache flush to disk will make the XP IO Size, won't it ?

Assume that we are in Oracle Raw device, Oracle is 8KB block size. Oracle flush to disk will make the IO Size from XP side ?

I wonder how Oracle flush cache to disk, but i should post on an Oracle forum...

Thanks anyway!

[Amar_Joshi]

Hi,

An another point to consider in calculating IO & Transfer rate (Mbps) is that there is a bunch of IOs which are ACKNOWLEDGMENT. These IO do NOT carry any good amount of data but they are IO. So considering the fact of these ACK IOs, you will never be able to get the formula working in this way.

My 2 cents worth...

[Jean-Baptiste Broccard]

From what I observed, I don't think ACK IO are counted when monitoring XP, neither monitoring server (w/ sar for instance). Can you confirm ?
How can I determine maximum IO size that HBA can send ?

JB