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: 3GB RAM usage by Oracle

Re: 3GB RAM usage by Oracle

From: Richard Foote <richard.foote_at_bigpond.nospam.com>
Date: Fri, 27 Aug 2004 15:11:23 GMT
Message-ID: <vaIXc.10047$D7.4374@news-server.bigpond.net.au>


"Don Burleson" <don_at_burleson.cc> wrote in message news:998d28f7.0408241519.2219d19a_at_posting.google.com...
> Daniel Morgan <damorgan_at_x.washington.edu> wrote in message
news:<1093325575.532413_at_yasure>...
>
> Hi Daniel,
>
> > Not going to happen. And if it does ... it will be one of the worst
> > configurations possible.
>
> Um, why is using all the available hardware resources a "bad" thing?
>
> Windows only needs 800 meg RAM, according to the documentation.
>
> Why waste it?
>
> > What makes you think Oracle using more RAM is a good thing?
>
> I've increased the RAM on over dozens of 32-bit dedicated Oracle
> Windows servers (most come with 8-gig RAM these days), and I've yet
> to see a 32-bit Oracle Windows server that did not benefit from using
> the wasted RAM. Almost all have "db file" waits as the top-2 timed
> events.
>
> In most cases using 4GT or AWE can half the disk I/O and greatly
> improve response time. Of course, you must allow for the HWM of PGA
> usage, but that's easily computed and you can use-it-all and avoid OS
> paging.
>
> Do you have a different experience?

Hi Don,

I think most people who understand how to tune Oracle database systems know differently :(

18 months ago I was contacted by a site that was experiencing serve performance problems. They had been upping and upping the memory in the hope it will do some good with no success. They had basically resigned themselves to a hardware upgrade but hoped for some advice beforehand. Within a couple of days, we had *halved* the total memory usage and made a system that was previously crawling into an efficient, scaleable and totally effective system that met all key business performance requirements. They're only now looking at upgrading their hardware, primarily for warranty reasons. How ? By addressing the *root* cause of the problems and not by symptom fighting in the hope it would do some good.

Using AWE to half disk I/O has got to be one of the silliest suggestions you've made for a while. You do of course appreciate the significant *overheads* that AWE invokes of memory accesses don't you ? You do of course appreciate that it's only the buffer cache that can be extended. You do of course appreciate that a larger buffer cache, that is *no longer optimally accessed* by the O/S may do more harm than good. You do of course appreciate that a larger buffer cache may not even reduce PIO because of the *manner* the I/O is being called. Hummm, strange you don't mention these issues.

Wouldn't it sound perhaps a little more reasonable to investigate exactly *why* you have so many *logical* I/Os that generate enough load to cause your existing resources to overcook ? Wouldn't resolving the *root* problems be a better cure than simply trying to address the symptoms ?

Throwing hardware in the *hope* of addressing performance issues is a rather novel approach. I notice you've recently boasted similar suggestions in another thread. In the vast majority of cases however, I would suggest such an action is both unwarranted and potentially futile if the bottlenecks are being approached from the wrong direction.

However, considering it's being promoted by someone who doesn't even know how to calculate the size the default buffer cache in 9i, it might not be overly surprising to many ...

Don, your tuning methodology (Method "D" ?) is most illuminating. And people consider method "C" to be outdated ;)

Cheers

Richard Received on Fri Aug 27 2004 - 10:11:23 CDT

Original text of this message

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