Hi all!
How good is HP/UX MirrorDisk striping?
Has anybody measured it in real world? Maybe compared to some hardware striping...and to JBOD's and so on...
I'm just thinking of increasing Baan/Oracle performance.
Maybe $BSE/tmp and Oracle areas.
-Jukka

Hi Jukka,

Striping is not really the right term in connection with LVM mirroring.

Striping is normally done on block level (lets say 2k or something). Mirroring with LVM is done on extent level (most of the time 4 MB). They can not be combined!

What can be done is the so called "distributed allocation" with Mirroring. Here the LVM will just allocate the extents dirtributed over the physical disks.

Yo could call it striping with a stripe size of 4 MB. This obviously is not nearly as good as striping with a much smaller stripesize.


I can not give you concreate measurements. But we use for all our installations distributed allocation with mirroring for the database files! I would not place the redo- or archive logs in the same stripeset as the database files or use the mechanism at all.
Also I would not expect Baan istself or $BSE/tmp to gain much from distributed allocation.

Even without measurements, I have a very good feeling about using distributed allocation. Some HP folks even say, that a FC10 with distributed allocation is as fast or even faster than the raid products (VA7100/VA7400) of HP and might even be able to compete with an XP!

Try it!

Markus

Uh, Mirror/UX does strip smaller sizes that PE (4MB). That is you can choose strips of 4, 8, 16, 32, or 64K blocks.

I have not done any performance tests though. It would be faster than not stripping and putting the same contents on the volumes.

Still not clear for me. Can I stripe and mirror with extent size smaller than 1M?