Re: Opinions on Unorthodox Server Setup

sdanield_at_yahoo.com wrote in message news:<ce8c3ae5.0406301722.158699f3_at_posting.google.com>...
> My organization has a complex legacy client/server app which we
> normally use in a mission critical environment where it generates a
> good volume of data. In this environment we of course have good
> servers for our Oracle instances, with redundant everything and hot
> and cold backups and all that good stuff.
>
> The idea has arisen to reuse this legacy application in a different
> kind of environment where the data is not as critical, but there may
> be some real benefits in its use there, provided we can keep the costs
> down for the new installation. Keeping the costs down is a major
> driver.
>
> The data in this new environment will not be as critical as it is
> elsewhere in our organization. Really, due to the low volume of data,
> small number of concurrent users (three) and non-critical nature of
> the information, Oracle is really overkill. But reuse of all our
> existing (rich-functionality) code entails keeping it on Oracle (there
> are a lot of dependencies, including lots of server-side PL/SQL).
>
> One idea to help keep the costs down is to not require a server
> machine for the Oracle instance. There will be three client
> workstations (probably XP professional) and on one of these client
> workstations we would install Oracle 9i standard edition. All three
> clients will use this database. Backups would be limited to nightly
> cold backups to a CD-ROM drive.
>
> I kind of think we're putting the cart before the horse here but I'm
> trying to stay neutral. Wanted to know if anyone more technically
> knowledgeable than me would share their opinions with me on this? What
> kinds of problems could we run into (beyond the user turning the
> workstation-server box off at 5pm and the scheduled backup not
> occuring that night)?
>
> Thanks!

I've got a similar situation and came up with a similar answer (except I'll be giving each user their own subset of data on their own machine). Since I have additional licensing issues involving the application, there will still be access from the production server, but the customized apps basically don't care where the data is (driven by simply keeping the connect string in a server file the user doesn't have access to). I haven't implemented the distributed part of it, waiting on the customer to decide to (there is also an OAS component, and of course, network saturation issues). In testing, there does not appear to be any problem doing this, aside from OAS bugs and difficulty in defining and testing the sort of loads that unfettered users might add to the network. Read that last part of the last sentence very carefully. It probably means multiple fiber subnets.

As far as using a user machine to serve data, it is a server and must be treated as one. I have seen things like OEM databases getting wiped out because the PC jockeys gave a DBA a new machine in the middle of the night. Also, traditionally people have had problems with performance when trying to run anything else with Oracle on a PC, but I'm not sure if this will apply to a modern machine with more than 2G of memory - you aren't planning on letting the user with the db crunch huge spreadsheets or anything?

jg

--
@home.com is bogus.
"I guess they did this with a straight face" -
http://www.signonsandiego.com/uniontrib/20040701/news_1b1msft.html