12H Config with Dual Controllers

[Tracey]

When I first received my N box, I could have sworn SAM showed me both paths in the Disk Devices section (0/4 and 0/8) - now I can only see both of those hardware paths doing an ioscan - Sam just shows the 0/4 with two paths.

I want to do what has been suggested here and create two identical sizes LUN's and put one LUN on each controller, with alternate links to the other.

What I think I have gathered, is that I should make two equal sized luns 50 Gig each (I have 12 18 gig drives with hot spare on) Create a physical volume from the two LUNs and then create to logical volumns from this physical volume.

I've read most of the other threads about this, but I am confused since I can only see one controller via SAM. Is there something wrong with one controller? Will I have to do this via command line? While I dabble heavily into UNIX admin, I am a DBA by heart, any help with the commands will be greatly appreciated.

[Tracey]

I beleive I figured part of it out - when I went to create the new volume group, one of the options in Configure Disks (optional) was to choose hardware path for each LUN. The other option was to create it into a physical volume. Any comments are still appreciated!

[Robin Wakefield]

Tracey - this sounds OK. In SAM I see:

8/0.1 - controller
8/0.1.0 - LUN0
8/4.1.1 - LUN1
8/0.1.2 - LUN2
8/4.1.3 - LUN3

but ioscan -fuCdisk shows:

isk 5 8/0.1.0 sdisk CLAIMED DEVICE HP C5447A
disk 7 8/0.1.1 sdisk CLAIMED DEVICE HP C5447A
disk 8 8/0.1.2 sdisk CLAIMED DEVICE HP C5447A
disk 9 8/0.1.3 sdisk CLAIMED DEVICE HP C5447A
disk 6 8/4.1.0 sdisk CLAIMED DEVICE HP C5447A
disk 10 8/4.1.1 sdisk CLAIMED DEVICE HP C5447A
disk 11 8/4.1.2 sdisk CLAIMED DEVICE HP C5447A
disk 12 8/4.1.3 sdisk CLAIMED DEVICE HP C5447A

so sam only shows the primary controller/paths, ioscan shows them all.

Rgds, Robin.

[Robin Wakefield]

Tracey,

one other thing - rather than creating just 2 LUNs, I believe HP recommend 4 LUNs for optimum I/O, whilst giving you the ability to add up to 4 more if required.

Robin.

[Juha Eklund]

Yes,
and if You want maximum performance ( if that word can be used with 12H ) then You should also allocate less that ? of the maximum capacity of the AR .. Then all data can be allocated in RAID0+1 ( striped and mirrored ).

-- Juha --

[Tracey]

So that would be two LUNS on each controller?

With 12 18 gig drives (216 gig) - minus the hot swap option, left me with I beleive was 158 gig - which would roughly leave me 80 Gig.

I was looking for the option of striping, but didn't see it, but I did use 8K extents.

BTW: This array will be for an Oracle DB.

[Robin Wakefield]

Yes, create 4 LUNs, then set the primary link for two of them down the 2nd controller using pvchange -s /dev/......

Robin.

[Steve Lewis]

Tracy,

You can use diskinfo on each of the disk device files in /dev/rdsk to double-check the size if they are not listed in SAM. (Also SAM is diabolically slow at displaying the info). If they haven't been created by default (they sometimes aren't), use insf/mksf or even mknod to create two sets of device files for each LUN on each controller. To balance all i/o on both controllers, vgcreate the VG onto the cXtTdD disk, then vgextend the VG onto the alternate path using cYtTdD. For the other LUN/LV, initially add the cY.... disk then vgextend onto cX.... to create the pvlink.

You will end up with something like:
vgNN c0t1d0 pvlink=c4t1d0 50Gb
c4t0d1 pvlink=c0t0d1 50Gb

This will mean you use both controllers on your machine and on the 12H.

Either stripe LVs over both LUNs or lvextend each LV onto specific paths. Balance the io as best you can for your database over them both.

Use arraydsp -i and arraydsp -a [ID] to get quicker info on your 12H, than SAM.


Hope this helps,
Steve

[A. Clay Stephenson]

Hi Tracey,

Yes, it is perfectly normal not to be able to see the alternate paths.

You actually can set up striping in an AutoRAID but it does not stripe across disks in the conventional sense. It alternates between the two external controller to maximize bandwidth. The idea
is to compose each volume group of two equally sized LUN's. LUN0's primary path should be through Controller X (alternate Y) and LUN1's path should be through Controller Y (alternate X). Each logical volume can then be stripped across both LUN's. If you are running vxfs, I suggest a stripe size of 64K. You will then see both controllers on the 12H working.

In reality, more than 2 LUN's per volume group is of no benefit. Things APPEAR better to Glance, sar, ... but only because the statistics are spread across what the host sees as more devices. In reality, actual performance remains the same. I've never be able to measure actual throughput gains beyond 2 LUNs/vg. I'd rather keep the LUN's available since 8 isn't all that many. But in any event, the real answer is configure and then measure.

If you stripe like this and allocate no more than 60% (a little less is better) as LUN's, you will be surprised how well the old 12H's perform.

Regards, Clay