Re: Real Application Clusters Requirements on Windows 2000 AS

On Thu, 24 Jul 2003 04:18:28 GMT, Charles Goehring <cgoehrin_at_san.rr.com> wrote:

>
>
> We have two machines conected to the same scsi raid box. Both machines
> are running Windows 200 Advanced. The servers, raid box and raid cards
> are all from Dell. Since RAC allows all cluster nodes to run
> simultaneously, it would seam W2k AS would be in an Active/Active
> configuration. My network guy says there are two varieties of this -
> with and without a virtual server. The RAC docs don't give a lot of
> detail on this. Does anyone know if a virtal server is required? The
> plan is to use Oracle's Cluster File System (CFS) instead of separate
> unformatted logical drives for each Oracle file.
>
> I've been through a lot of the docs, but haven't found a discussion of
> the specifics of Active/Active as it relates to Oracle RAC. There is
> also no discussion of how RAC and CFS interact with the W2K cluster
> resources and their allocation.
>

They don't interact at all. Either you install Microsoft's Cluster Services, in which case you can't have an active-active RAC, or you don't, and instead install the Oracle Cluster Management Software (which is provided as part of the CFS download).

Microsoft's clustering solution, in other words, is of the 'failover' type. When Node 1 goes belly-up, Node 2 detects it, and starts up whatever programs you've configured it to start up, including an Oracle instance. Only after that does connectivity to the second node become possible for database users.

Because that's decidedly not how RAC works, Oracle wrote their own cluster management layer (which runs independently of whether you decide to use the cluster file system or go with raw (ie, unformatted logical partitions). That permits a second instance to be running on node 2 all the time. Whether it's actually an 'active instance' (ie, one users can connect to for routine work) is up to you: ACTIVE_INSTANCE_COUNT=1 ring-fences the second instance so that no-one actually connects to it. But it's there, ready and waiting for disaster to befall node 1. When such a disaster strikes, the instance is already running, so failover to the second node should be much quicker than in the MS scenario. ACTIVE_INSTANCE_COUNT=2, however, means you've got your active/active configuration. Both nodes are co-equal, and so are both instances.

If you go for the ACTIVE_INSTANCE_COUNT=1 scenario, you then also have the option of automating the failover process with RAC Clusters Guard. That permits, amongst other things, the second node to *take*over the primary role, and not just have failover happen.

I have no idea what a virtual server is, but one's most certainly *not* required to get active/active RAC working on W2KAdvSrv.

~QM Received on Thu Jul 24 2003 - 02:11:39 CDT