Re: Oracle 9iR2 32bit on windows 2003 server 64bit

From: gym dot scuba dot kennedy at gmail <>
Date: Sun, 08 Jun 2008 17:41:00 GMT
Message-ID: <MQU2k.244$n9.79@trndny01>

"marco" <> wrote in message On 8 Giu, 18:34, "" <> wrote:
> "marco" <> wrote in message
> >
> What is the "reasoning"? (Is it some misguided conception that things will
> run faster?) Oracle won't use the extra memory in the machine. It can't
> address it. If Oracle does not support the configuration what is the
> client
> going to do when they call support and Oracle says "We don't support that
> combination. Please install into a supported configuration and then we
> will
> support you." I hope they aren't niave enough to think that 1. Oracle
> won't
> say that. 2. They can brow beat the Support Tech/manager etc. into giving
> them support anyway. 3. They won't ever have to call support. Isn't their
> data important to them?
> Jim

Firstly let me say that I totally agree with you Jim. The reason for the requirement is that oracle will run in a cluster. The application has been written for Oracle 9i. In the same cluster will run another software that needs win 2003 64 bit, so the
solution, in my opinion could be to upgrade the application to 10 g or to install
it in a different cluster.

But, as you can guess, both the solutions aren't up to me, as I act as consultant, I only can
give advices, but cannot take decisions by myself. M.

Ironic. So they want to run in a cluster presumably for High Availability (good idea) but they want to use an unsupported configuration. 10G is not the new kid on the block so any excuse that it is a first release product won't wash. One method we use in our group when stupid choice is being made is to have the person making the choice do the following: 1. In front of the group while repeating the"sub-optimal"decision out loud hold their right hand up.
2. Repeat the "sub-optimal" decision statement (and concequences) slowly, loudly and one time for each finger on their right hand. 3. If they still agree to go forward then go forward.

It is amazing what it does. (sounds stupid to do it, but it sounds even stupider to say the "sub-optimal"decision out loud in front of everyone even if they agree with the course of action.

I am going to guess here that it is an off the shelf application and that the application's vendor is stating the Oracle version requirement. (Could be that the vendor hasn't upgraded their queries from rule based to cost based and there isn't a rule based optimizer in 10G. The queries would probably run fine, but again it is a CYA move.) If it is an application vendor's requirement ot run Oracle 9 then I would point out there is an implicit requirement to run Oracle on a supported platform. The vendor isn't going to support Oracle running on an unsupported configuration any more than Oracle would.

It is possible that MS has improved the cluster capabilities in 2003.(and you can't get the cluster stuff in plain 32 bit? seems odd, but I am not an MS expert ) However the basic capabilities a few versions ago caused some challenges with running an Oracle db. The disk subsystem was shared between the two machines. If machine one got too busy and could not respond to the heart beat then machine 2 took over the disks and chaos insued. Oracle on instance 1 would mark the files as needing recovery because something "touched" the db files. (it didn't know instance on machine 2 was doing the touching) The other thing was if machine 1 rebooted then machine 2 would take over and when machine 1 came on line again it would try to sieze the disks and again chaos would ensue. So the coordination was manual. This was on an older version of MS's software very pre 2003; so things may have improved.

Why not go with RAC? (or data guard) I bet the vendor's application is not certified for RAC. RAC is a great amplifier of bad code. So if the application does not use bind variables etc. RAC would only make it worse.

Jim Received on Sun Jun 08 2008 - 12:41:00 CDT

Original text of this message