HP-UX LVM Change allocation policy

[Harold Gutierrez Calder]

Hi,
I would like to know how I/O performace in an HP-UX system could be affected if the allocation policy in a VG is changed from distributed to not distributed, with all the LUNs provided by SAN.

[Torsten.]

Hi Harold,

the man page says:

"When the distributed allocation policy is turned on, only one free extent is allocated from the first available physical volume. The next free extent is allocated from the next available physical volume. Allocation of free extents proceeds in round-robin order on the list of available physical volumes.

When the distributed allocation policy is turned off, all available free extents are allocated from each available physical volume before proceeding to the next available physical volume."


Now there are a lot of questions, like

- what is your "SAN"?
- why do you want to change this?
- want do you want to do after the change?


As long as you don't modify your VG config, nothing will happen.

[James R. Ferguson]

HI:

You'll probably need to measure --- your milage may vary.

You don't offer the SAN type nor the LVM layout details you now use, nor for that matter the kind of access you do (sequential or random I/O). SANs provide data cacheing and can store data in various RAID formats.

Empirical measurements will probably offer you the best advice for your environment.

Regards!

...JRF...

[Torsten.]

Talking about performance ... there is almost never a general solution. This depends always on soooo, soooo many different things.

Next to me is a XP24k disk array and an old L-class server. The array will only laugh about the traffic my good old server is producing.

The server is just about to glow, but this is almost not even noticeable on the array.

So I don't need to think about to optimize on the array side ... ;-)

What I can do is to stripe across different PVs to utilize both FC HBAs at the same time - but as said before, performance depends on so many different things.

[Ulrich Sander]

Hi Harold,
I acutally found a big differnce between distributed and non distributed:
We have two HPVM Guest System on the same hardware one distributed and one not. Creating a tablespace of 20GB needs 5 minutes on the distributed onw and over 8 minutes on the other one! We have a FAS3170 from NetApp as SAN. I think NetApp distributes a LUN always quasi optimal on its disks. So I guess looking from the disks you can't decide whether the data is accessed via distributed LVM layout or not.
I think HPUX have also a limitation how much IOs are allowed on one LUN