Perfromance on 11.23

[Trng]

Hi,

i have HPUX Machine 11.23 with 5 VG and 22 Logical volume on it (Total number of Luns assigned from Hitachi NSC55 to this machine is 23.Now i am planning to scrap this server with new database/Application.I want to create 20 New filesystem on it.

1-is there any performance issue will come ,if i combine this 23 luns of 20GB size os each lun from storage and use it a single LUN OF 460GB on HPUX 11.23 box ? .







regards
Trng

[Kapil Jha]

460Gb is not too much!! but still, what I personally perfer is to have 100GB Lun (if file system is not too big)

SO you can go with 200 or even 460GB Lun there would not be any performance issue if you use it as one LUN of 460GB

But is always better to have small LUNs for better performance.

BR,
Kapil+

[gb karki]

This is depending on RAID configuration in storage.

Rgds
Karki

[Suraj K Sankari]

Hi,

Always create small size of file system yes you can go for 100GB to 200GB
if you want to get more performance then make it strip also.
for more information about RAID see the below link
http://www.aaa-datarecovery.com/raid_tutorial.htm

Suraj

[Michal Kapalka (mikap)]

hi,

i think its depend on the storage which level of RAID , is used and SAN infrastructure, on our old DMX we hade 600 x 12 GB lun size with 4 alternate paths, and when we migrated it to the new Hitachi USP-V and increased the space on it, we made aroun 69 GB lun size to decrease numbers of luns on HP-UX 11.23.

100/200 GB is ok.

mikap

[Hein van den Heuvel]

From a SAME (Stripe and Mirror Everything) perspective I like a a single big lun. Typically it will provide the best spread possible.

Do you want all volume to have the SAME storage attributes, or do you want some extra safe and extra fast (RAID 0+1) (OS? DATA) and some cheaper (RAID 5 : backups, historical data) and some fast & looose ( RAID 0 : temp, scratch, re-generatable reports? )
If you want some storage to be be different it will need to be a different LUN.

The most important input to decide this has to come from your Storage folks, either in-house, or the vendor. They will know whether a single lun will give the concurrency and the spread over all the disks you need.

I do NOT know the NSC55, but maybe it has a limitation as to how many physical disks can be grouped under a LUN? Maybe it has CACHE rules which make it more or less desirable to have more or fewer luns?

From the HPUX angle I like 'a few' LUNs to help me visualize the IO load generated by the system for the purpose of SAR and IOSTAT or such. But the I would NOT generate those luns all at a the same size, but each at a size reasonable for the purpose. Continuing on the simplistic examples above: a LUN for the OS, one for the data, one for the development environment and for scratch?

Finally, you want to keep several LV's, agains to stats (glance), and minor performance impact like file creation/allocation which is somewhat serialized per LV / file-system.

Regards,
Hein van den Heuvel
HvdH Performance COnsulting.