MCSG and ORACLE 9i

[Johnny Kwok]

Hello,

We are planning to setup a 2 node-cluster MCSG and ORACLE 9i in our environment.

Since ORACLE can be installed on

1. separate root disks on each node,

or

2. same share disks (can be accessed from either node) and packaged to fail over with the data files.

What is the best way to go ?

Thanks!

-JK

[Christopher Caldwell · Accepted solution]

We used 1, so we could
A) patch one node without having to patch the other node (only works with patches that don't require you to run scripts on the database).

B) keep different tunes on different classes of hardware

C) test different tunes and configurations in production

The down side is that you've got be fairly good about making sure configs changes get put on both nodes. That's probably the main reason folks use 2).

[A. Clay Stephenson]

I usually prefer to have the Oracle application code move with the package data. That does mean, if I have three packages, I install 3 copies of the Oracle executables and have three separate listeners.
If you choose this approach, I would oversize the disk requirements for the Oracle executables very generously. That way, you can install the next Oracle release and get it linked before actually running any upgrade scripts.

If you do it this way, the configs always stay "in sync" BUT the downside is that you have to make certain that shared library patches get applied to all nodes otherwise it is possible that the code will not execute upon failover.

BTW, I would NEVER use the root disks to hold anything other than the base OS stuff. Keep vg00 lean and mean - that makes OS upgrades much easier and cleaner. If you do ot move your Oracle executables with the package then put them in another VG - disks are trivially cheap compared to the time required for node upgrades especially in a MC?SG environment.

[Michael Steele_2]

Peruse this document and take special note of figure 4-1 and pages 68 to 72. This is the usual way to configure a 2 node cluster with MC/SG.

Note the two non-shared and independent O/S volume groups and the shared disks accessable to both nodes.

It's the contents of the package that oversees the data written to these shared disks so the Oracle application has to be configured within.

http://docs.hp.com/hpux/pdf/B3936-90024.pdf

[Oleg_30]

I totally agree that you should never store anything but the OS stuff on vg00, and to you question you may want to consider RAC for your Oracle 9i, you should use it with SG OPS edition, it will greatly raise availabilty of your cluster and will be best utilization of the two nodes for DB purpose. You can read about the 9i RAC solution in meta-link and about the OPS edition which was specially designed for RAC in the following HP site:
http://docs.hp.com/hpux/ha/

in the following section:
ServiceGuard Extension for Real Application Cluster (ServiceGuard OPS Edition)

[Jarle Bjorgeengen]

Hi JK,

depends on what is most important.

If you want to maintain only one set of application binaries , and don't care about "rolling upgrade" functionality, choose 2. If yo want to be able to upgrade/patch the application while still running the application on the other node, and the have a fallback in case something goes wrong in the paching, choose 1.

Normally if you have focus on max uptime of the application want to pick no. 2, even if it means more change management .

Rgds Jarle