Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Real Application Clusters Requirements on Windows 2000 AS

Re: Real Application Clusters Requirements on Windows 2000 AS

From: Quarkman <quarkman_at_myrealbox.com>
Date: Fri, 25 Jul 2003 05:47:10 +1000
Message-ID: <oprstykwytr9lm4d@haydn>


On Thu, 24 Jul 2003 09:55:34 GMT, Charles Goehring <cgoehrin_at_san.rr.com> wrote:

>
>
> Additional info/question:
>
> I thought the Cluster service was also needed to provide "safe" access to
> the shared scsi raid device. Is this not the case?

I could, of course, be wrong. But since I've happily RAC'd XP Pro and Windows 2000 Professional, I would say this is not the case. The Oracle doco comes to my aid here, too:

"Certified vendor-supplied operating system-dependent clusterware for UNIX, or Oracle operating system-dependent clusterware for Windows NT and Windows 2000"

(From the Oracle9i Real Application Clusters Setup and Configuration book, under 'Supported Software').

What that sentence says to me is that the "vendor-supplied clusterware for Windows" (ie, Cluster service) is not supported, so you have to install the Oracle clusterware *instead* (not 'as well as').

The trouble is as I originally described. Microsoft's clusterware (ie, its Cluster Services) does not permit simultaneous access to the same shared disk volumes by more than one node at a time. If you have one shared disk partition, and nodes A aand B running W2KAdvSrv with Cluster Services (ie, a proper Microsoft Cluster), try saving a word document to the shared disk from Node A. Then go to Node B and see if you can even see the Word document listed in Windows Explorer. It won't be there. But if you now blow up Node A, suddenly Node B gets to see it.

Microsoft explain that this does not mean Node B is completely useless: there's nothing to stop you creating two partitions on the shared disk. Node A can access partition 1, with B as its failover; and Node B can access partition 2, with Node A as its failover. They can't distribute shared work amongst themselves, but you could be working on an Excel spreadsheet on Node A, whilst I'm running Word on Node B -and we both have protection in the event of a failure on one node.

But that model will not work when each node needs to see the entire database.

So MS Cluster Services are a definite no-no if you want true RAC.

~QM Received on Thu Jul 24 2003 - 14:47:10 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US