Please tell me about Oracle RAC and Oracle cluster

[Wang,MinJie]

Hi all
I start this new thread to talk about Oracle cluster.
I know that Oracle RAC software is used to configure an Oracle cluster.
But my boss told me that in HP-UX we can use MC/SG to configure an Oracle cluster without installing Oracle RAC(just install oracle software on both nodes).
I'm very suspicious about it!
Can you tell me the truth and how to configure?

[A. Clay Stephenson · Accepted solution]

You are both right and wrong. The idea behind MC/SG using conventional Oracle (ie non-RAC) is that the Oracle instance runs as a "package" on one of the nodes and the clients connect to a "package" IP address --- which is different from the static IP addresses of either of the nodes in the MC?SG cluster. If NodeA fails, the package (the Oracle data and the package IP address) automatically moves to NodeB. There is some delay of at least a few tens of seconds will this package switch occurs. The important point about this approach is that one node is active while the other is on standby (although multiple different packages might be running on the cluster as a whole so that NodeB might be running another package while NodeA is running the Oracle instance).

Oracle RAC works differently in that both nodes can be actively serving Oracle data. Under Oracle RAC the data must be shared as raw devices (ie no filesystems).

Finally there is still another model and the is MC/SG with eRAC extensions which is only available for HP-UX 11.23. This allows you to run Oracle RAC with mulltiple active MC/SG nodes AND the data can be stored in fully cooked files using the Cluster File System (CFS) which looks like a conventional VxFS filesystem to all the nodes but is shared by all the nodes. A separate network connection is required for CFS traffic.

[Wang,MinJie]

You're so genius!
You made me understand it!
So I have the next question:
Now I decide to configure a MC/SG cluster and configure an Oracle pkg,and I will install Oracle 10g on both nodes,and my boss told me to install oracle product on the local filesystem on both nodes but the redo.log on the shared volumn.So I'm very silly that I've no idea to do that
So can you tell me is that neccessary and can you tell me or give me some documents in detail?
Thanks a lot!

[A. Clay Stephenson]

No, that configuration will not work. All of the Oracle data (data files, indices, archive logs, redo logs, control files, ...) must reside on a volume group (or multiple volume groups) that either node in the cluster can access. I generally go one step further and actually install the Oracle binaries as part of the package and give each package a dedicated listener but that is not necessary. I suppose a good way to convince yourself that keeping some of the files under the control of MC/SG and others local would be to 1) yank the power cord of a box running an active Oracle instance. 2) The package will try to failover. 3) Oracle will attempt to restart on the adoptive node and because some of the files are missing or have the wrong timestamps, the restart will fail.

Conventional Oracle under MC/SG is quite robust but you have to play by the rules.

In any event, before you tackle a project like this for a high-availability environment you need to really master LVM and/or VxVM, MC/SG, and the application itself (ie, Oracle). This is not something that you do in one or two days. Setting up MC/SG is easy; getting all the hardware/software/procedures so that they are extremely fault tolerant is the hard part. Basically, long before you are ready to setup MC/SG you have to be able to tolerate the failure of any single thing ( a failed disk, a failed computer, a failed network/Fabric switch, ... ). In fact, when you are actually ready for MC/SG, you have made your systems so robust that MC/SG very seldom actually comes into play itself --- and that is why you buy MC/SG.