LUN Setup on VA7400

[Bob Ferro]

I have a VA7400 with 1GB dual cache controllers and 10 - 73GB Fibre Drive running on RP5470 with HP 11i. I've read a lot of the messages on this forum but I have one question. I want to setup the LUNs for Oracle 9i. The following is the setup I was going to use. I am mainly concerned with the oracle stuff. Everyone, take a shot, poke some holes.

VG01 armcfg â L 1 â a 50G â g 1 va7400
VG02 armcfg â L 2 â a 50G â g 2 va7400
VG03 armcfg â L 3 â a 5G â g 1 va7400
VG04 armcfg â L 4 â a 15G â g 2 va7400
VG05 armcfg â L 5 â a 20G â g 1 va7400
VG06 armcfg â L 6 â a 10G â g 2 va7400

VG01 = /oracle/data
VG02 = /oracle/index
VG03 = /oracle/system
VG04 = /or

P.S. This thread has been moved from General to Disk array. - Hp Forum Moderator

[Karsten_4 · Accepted solution]

Hi

looks ok.
You must(should) distribute the I/O Load evenly on both Red. Groups. And if possible, add more Disks. (10 Drives in an 7400 is the minimum)

[Bernhard Mueller]

Bob,

looks like there is something missing in your post and the minus characters are look a little akward. But basically, what's there looks fine to me.

i.e. alternating redundancy groups with each lun and have about the same sizes assigned in each RG.

Note that most likely you have dual paths to each lun and would want to alternate primary and alternate links in the vgcreate commands for a static load balancing.

Regards,
Bernhard

[Bob Ferro]

VG01 armcfg -L 1 -a 50G -g 1 va7400
VG02 armcfg -L 2 -a 50G -g 2 va7400
VG03 armcfg -L 3 -a 5G -g 1 va7400
VG04 armcfg -L 4 -a 15G -g 2 va7400
VG05 armcfg -L 5 -a 20G -g 1 va7400
VG06 armcfg -L 6 -a 10G -g 2 va7400

VG01 = /oracle/data
VG02 = /oracle/index
VG03 = /oracle/system
VG04 = /oracle/redo
VG05 = BaanIV application software
VG06 = junk stuff

Thanks for responding. The above is the correct layout. Is it better the above way or better to configure (2) LUNS in separate Raid Groups, put them both in the same Volume Group, then stripe across the controllers?

[James A. Donovan]

Where were you planning on putting your archived redo logs? On VG04, with the online redo? I would recommend placing archived redo on another lun. In your case since you have redo on VG04 (redundancy group 2), I would put archived redo (/oracle/arch) on VG07 (redundancy group 1).

Other than that, I think your setup looks fine.

[Bernhard Mueller]

Bob,

you are correct, since you have _only_ 10 disks, I would (for performance reasons) actually recommend to do what you suggest;

i.e. create 2 LUNs (one in each RG) for each VG and do LVM striping across the two LUNs.

Note: you should not use BBR, i.e.
lvcreate -L <#MB> -n -i 2 -I 64 -r N vg0#


Another option you _might_ want to consider is using RAID1+0 instead of AUTORAID, which does cost available LUN space, but prevents the controllers from performing "dynamic data migration".
(Note: you cannnot go back from RAID1+0 to AUTORAID)

Regards,
Bernhard

[Bernhard Mueller]

Sorry, for the mistake in the previous message, of course it is vice versa. You cannot go from AUTORAID to RAID1+0 (it would require a reformatting of the array) However, it is possbible to move - online - from RAID1+0 to AUTORAID.

Apologies for giving the wrong information.

Regards,
Bernhard

[Bob Ferro]

Bernhard,

Thanks for the info but I have one question. To do LVM striping when creating the logical volumes, what should the stripe size be? I am using the server for Oracle/Baan and the Oracle db block size is 16K. Should I set the stripe size 16k?

Thanks again,

Bob

P.S. I had a HP CE in to upgrade the firmware and install the latest version of SDM and he gives you an "A" for your VA setup. Great job.

[Michael Schulte zur Sur]

Hi Bob,

I found something to consider. If not 16kb, you should at least take a multiple of 16kb, maybe 64kb. You have a terrific hardware, something I would love to have in my hands, but why do you want to use software striping, when you can have it on the hardware level? Am I missing something? I am trying to get into this interesting subject.

Michael

RAID 1+0 striping of every LUN across all disks: In a very large redundancy group, striping across all disks reduces the number of LUNs required to achieve a balanced workload and eliminates "hot spots". This self-tuning performance insures consistency in meeting application Service Level Agreements.

[Bernhard Mueller]

Bob,

regarding the LVM stripe size on the VA wrt Baan w/Oracle I cannot give you a definite do or don't recommendation.

In general I have made good experiences with a LVM stripe size of 64K for LUNs on disk array systems. As I said I could not tell you whether a smaller size would give you more performance or would just add LVM overhead. My gut feeling is to select even a larger size if you have more cache.

Regards,
Bernhard