Oracle DB configuration with MC/SG OPS

[Pedro Sousa]

I need to configure an Oracle DB with OPS.

This DB will have, more or less, 100Gb, I've a 30*18Gb disk array, and I'm putting everything in RAID 1/0.

My idea is to create 2 VGs for the DB data, in order to take advantage of the 2 SPs from the disk array (each VG uses a different SP), plus 2 VGs for archive redologs, because they have to be on FS.

My questions are:
Is this correct?
If not:
How many VGs should I create?
How should they be configured?

thanks,
Pedro

[James R. Ferguson]

Hi:

In general, make sure that you do your vgcreate specifying max_pv greater than the default (16) and the max_pe to a size that also allows for future growth. See the man pages for vgcreate.

...JRF...

[Pedro Sousa]

Thanks for your hint...
But I should not be worried with, since each VG will only have 1 or 2 LUNs assigned. As each LUN is seen by the system as 1 Disk, that should be no problem, right?
My 2nd question refers to how should I access the VGs, in shared mode or what?

thanks again.

[Paul Torgerson]

Do you have multiple nodes for OPS? Are you running ServiceGuard/OPS? If so, then OPS demands all disk used for db's be raw and configured for shared access between the nodes. Then Oracle controls them via ogms/dlm within the constraints of ServiceGuard/OPS.

With ServiceGuard/OPS, you cannot utilize LVM mirroring between individual disks (Raid 1). Having your disks configured RAID 5 will protect the individual VG/LV sets in the case of an individual disk failure.

As for your archived logs, like you stated, they must be on FS space. This would take away from your space dedicated to oracle on your (I assume) model 30 array. Do you have any disk down another I/O channel you can put your logs onto? Perhaps to another HA device? This would help in your overall performance and allow all the M30 disk space to be utilized for Oracle.

Paul

[Pedro Sousa]

Regarding your questions:
We have two nodes for this cluster, and the DB will be configured in raw.
We have an EMC Clariion FC5500 disk array, it can take, more or less, 100 disks.

Another thing, why SG/OPS doesn't support Raid 1? Does it work fine with Raid 1/0?

According to Oracle docs, raid 1/0 is the best one to use!

I think you gave me a good idea regarding archive logs.

thanks.

[Stephen Doud]

FYI:
The term "shared" in any LVM MAN page indicates those volumes allowed to be concurrently accessed (which is only supported by ServiceGuard-OPS)....

PHKL_22267 for 11.00 indicates:
"On HP-UX version 10.20, we currently provide and support mirrored
logical volumes which are part of a shared volume group. This
functionality was removed in HP-UX 11.00. This patch enables the
functionality for HP-UX 11.00."

This same patch also contains a fix to allow striping on "shared" disks - description for fix added in integral patch PHKL_19909:
"1. Enhancement to LVM to allow striping in shared logical volumes."

[Pedro Sousa]

Thanks, but the RAID 1/0 is made by HW and not SW. So, I think I do not have any problem.

I've decided to create 4 VGs.
2 for the DB (shared), and 2 for archiving...

thanks everybody.

[sung-hoon chang]

Hi

MC/SG OPS support RAID 1/0. Don't worry.
And I think, your decide is very good.

If you have hpux 11.00, you can try lvm striping on your disk with the latest LVM patch which is PHKL_22267. In this case, the striping will be make free you for load balance by two SP.

[sung-hoon chang]

Hi

MC/SG OPS support RAID 1/0. Don't worry.
And I think, your decide is very good.

If you have hpux 11.00, you can try lvm striping on your disk with the latest LVM patch which is PHKL_22267. In this case, the striping will be make free you for load balance by two SP.

[sung-hoon chang]

Hi

MC/SG OPS support RAID 1/0. Don't worry.
And I think, your decide is very good.

If you have hpux 11.00, you can try lvm striping on your disk with the latest LVM patch which is PHKL_22267. In this case, the striping will be make free you for load balance by two SP.