Memory utilization (was SGA/PGA and virtual memory)

Ki wrote:
>
> Jens Wartenberg <wartenberg_id_at_ibm.net> wrote in article
> <3320034A.68EB_at_ibm.net>...
> > Anton Dischner wrote:
> > >
> > > Hi,
> > >
> > > i did performance measurements with different SGA sizes on different
> > > CPUs.
> > >
> > > IBM RS6000 5xx and SMP, SGI, HP. Oracle 7.x
> > >
> > > I found that a small SGA performs better than a big SGA !!
> > > I believe this is because filesystem caching performs better than
> > > Oracle SGA-caching.
> > >
> > > Give a small SGA a try!
> > > Make Oracle-blocksize as big as possible.
> > >
> > > Kind regards,
> > >
> > > Toni
> > >
> > > --
> > > A. Dischner, SGI&AIX sysadmin, Oracle DBA | Don't let friends
> > > Institut fuer Klinische Chemie | use WinDose
> > > Klinikum Grosshadern | Just say NO.
> > > Ludwig Maximilians Universitaet Muenchen, GER |
> > > Marchioninistr.15 81366 Muenchen 49-89-70953202 |
> >
> > This surprises me a lot!!!!
> >
> > I thought that because Oracle is able to distinquish between which reads
> > are likely to be reused (lru-algorithms, indexed reads etc.), it would
> > be a benefit to give the ram to Oracle.
> >
> > I am presently increasing the sga (db_block_buffers, shared_pool_size)
> > and has decided to test the users view on performance by reverting to
> > the previous settings for a few days.
> >
> > I will report what the users believe (ram to AIX or Oracle).
> >
> > Regards
> >
> > Jens Wartenberg
> > Wartenberg Industri Data
> > Denmark
> >
>
> Hi,
>
> I read in several places that SGA should take 30% of real memory available.
> probably mr. dischner used a larger sga size (which caused swapping) and
> found out that
> a smaller one behaves better, while mr. wartenberg used a small sga in the
> first place, just to find out that enlarging it is payfull.
>
> Just an hypothese .
>
> I'm working excellently with 30% Real memo sga.
>
> yours,
>
> Muli Koppel

Here's a question for you all.

Has anybody used a technique called "pinning a package" in memory?

The allegation is that if frequently used packages are "pinned" in the SGA, performance is improved. Of course, the obvious trade off is that a portion (>10,000k) of memory will be unavailable for the rest of the database and users. In return, Oracle will not have to call up the package, use it, and flush it when finished with it.

Will the anticipated performance improvement be worth the commitment?

Gus Received on Fri Mar 14 1997 - 00:00:00 CST