Re: Hardware Recommendations for new Oracle Server

"W. Scott Moore" <sirws_at_hotmail.com> wrote

> The lease on my current server is coming up, and I am working on trying to
> spec out a new machine.

 <snipped>
> Our organization is an HP shop, and I was looking at an N Class
> server with 4 550MHz processors and 16 GB of RAM. Any comments? Does
> anyone recommend not using HP?

Ignoring the price, the HP machines are fine work horses. We have both N and L class boxes and the new L class one I'm playing with is pretty fast.

But when it comes to hardware, it is a bit of a jungle. Not an easy decision to make. And the sales people coming up with all kinds of marketing bs about performance and benchmarks and stats - none of which I would trust.

IMO, performance should not be the sole criteria when choosing your next platform. You need to look at cost of maintenance, upgradability, scalebility, administration costs (skills & resources) and how much _that_ will cost. It's a hard to swallow that something like a new 9GB disk for such a box cost more than 5x times that of a standard SCSI 9 GB drive ... times like that when Intel platforms look a lot more inviting.

> Do to the recordsets, I believe that network performance is important.
> Should I get a gigabit network card and have it hooked straight into
> our switch?

Sounds like a solution. In the case of something like a warehouse being hooked into production machines, you may even want to consider running a dedicated fibre link between them. What is important though is that the platform you choose allows this type of scalebility - without costing you an arm and a leg. :-)  

> What about any clustering solutions?

Love them myself - techie paradise. :-) But you need to look at the reasons for wanting a cluster closely and make sure that it does address you needs.

Clusters are a bit more complex with a learning curve to go through. After which cluster maintenance should not be _significantly_ more difficult (or expensive ito resources) than a normal bunch of SMP boxes running seperate databases. (and don't let them tell you otherwise)

What is a new ball game though is performance. You are dealing with multiple database engines for a single physical database. You need to consider application partitioning when running OLTP as resolving locks across database engines are expensive. Which could mean having to adapt/re-develop application software.

However, clusters provide you IMO with unmatched scalability and performance - something wich single unit SMP's can not beat (though I'm sure the SMP crowd will disagree with me on that).

> Would anyone recommend sticking with Win2k?

Yes. Don't like saying this as I believe that _technically_ Unix based platforms are a better _technical_ solution. But what you are trying to do is satisfy business requirements and not built an ideal techie environment for DBA's, developers and other assorted freaks. :-)

So if a Win2K based platform can address the business needs that a Unix platform can not (given budget constraints, available skills and resources, etc.), then use it. And don't let anyone tell you any different - unless there is a damn good reason why using Win2K will not be able to address the business requirements due to some serious technical issue or flaw.

--
Billy