Re: Questions about automatic memory allocations in 10g (r1 and r2)

On 10/20/05, Reidy, Ron <Ron.Reidy_at_arraybiopharma.com> wrote:

> Guys,
>
> I was in a meeting yesterday with a vendor who told me (and my manager)
> their product running on 10gr1 does not need any DBA intervention when used
> because they are using automatic memory management and they use "all of
> physical memory … for the SGA".
>
> Now, this person is no Oracle DBA/expert/guru. He has attended a class
> like "New features in 10g" or something like that. When I questioned this
> "use all of physical memory" strategy and how that might affect PGA usage,
> he was stumped.
>

Not to mention that in 10gR1 not all of the memory pools that one might use are system managed anyway.

 Well, this meeting got me to thinking about if this really works or not, so
> I went and read up on this new feature on *http://tahiti.oracle.com*<http://tahiti.oracle.com/>
> .
>
> My concerns about this center around questions like:
>
> 1. Does this really work?
>

In the sense that you can see memory allocation changing, absolutely it works. Thanks to the updating of the not so hidden parameters involved it survives restarts of the database as well

 2. How often is the memory usage looked at and adjusted?
>

MMAN does this every , from uncertain memory, 30 minutes?  3. When an adjustment (shrink) occurs in the shared_pool, are the cached SQL invalidated?
 4. When the buffer_cache shrinks, are the buffers marked as dirty?  Why would they be invalidated or marked as dirty because the size of the pool changed?

> Does anyone know about these issues? Can you point me to somewhere where I
> might get answers to these questions or set up my own test case to show how
> this "feature" might work?
>

http://www.dbazine.com/blogs/blog-cf/chrisfoot/blogentry.2005-08-19.0090015270

--
Niall Litchfield
Oracle DBA
http://www.niall.litchfield.dial.pipex.com
--
http://www.freelists.org/webpage/oracle-l