Re: performance/scalability and multiple instances

If you are going to share less as you are proposing, seems to me you would be better off doing as you said, but NOT sharing disk IO. I don't see this as a good idea if you are still sharing disks. Seems ok to share 1 disk array for data and index(and maybe more like SYSTEM, Tools, etc.), but especially not for REDO logs and archived REDO logs. Actually, I don't know much about these new fiber arrays (SAN?) shared by multiple machines. But, seems like a bad idea to me .

Another thing to consider is that new faster CPUs usually get into the smaller servers first. So, when you do purchase new hardware you will usually get faster CPUs in smaller machines. So, several smaller machines would be better for this than 1 larger server.

Note: We use Sun hardware a lot and really like the new SunFire V880, which is very well priced and goes up to 8 CPUs . Not sure the memory max, but we have purchased these servers a few times with 8 (and 4) CPUs and 16 Gig of memory usually.

"Ron Bergeron" <bergeror_at_asitshouldbe.com> wrote in message news:Pine.LNX.4.44.0301271500190.7387-100000_at_machine.asitshouldbe.com...
> I am working with someone who has 12 Oracle instances running on a single
> Sun server. Each of these instances is a data warehouse with anywhere from
> 50-500 GB of data in it. Each DW is completely unrelated to the others.
>
> They have performance and scalability concerns. My recommendation to them
> is to split the instances up and put them on separate boxes. They can't
> buy one server per instance, but at least they could split the 12
> instances up over three or four machines.
>
> If they do spread the instances over multiple servers, all servers would
> share a high-end disk array (Hitachi 9970).
>
> My questions are:
>
> 1) Is putting 12 heavy duty, large DW instances on a single box a Bad Idea
> for performance? It is obviously a Bad Idea from an "all your eggs in one
> basket" perspective.
>
> 2) Is there a white paper or something similar that supports my opinion
> that scaling unrelated instances horizontally is more scalable than
> scaling vertically?
>
> 3) Other than CPUs, memory, and I/O, are there any other resources that
> multiple instances on one server contend for? (ie: semaphores, locks,
> whatever)
>
> They did voice one opinion that might have some merit. They said that if
> all the instances are running on one large machine, and a few of those
> instances are idle at the time (instantiated, but not processing any
> queries), then the remaining databases can take advantage of the unused
> CPU cycles from the idle instances.
>
> Thanks for any input you might have.
>
> Ron
>
Received on Mon Jan 27 2003 - 21:14:31 CST