Re: [suse-oracle] 7 GB SGA possible ?

On Mon, 14 Mar 2005 13:50:14 -0800, Alexei_Roudnev <Alexei_Roudnev_at_exigengroup.com> wrote:
> > Are you using dynamic SGA? That is to say, when you shut off the one
> > DB did you grow the other two DBs' SGAs? If not, it sounds like you
> > may have the OS swapping the SGAs to disk. Is that a possibility?
> I am not sure (I am not DBA). It was npt swapping, but we had 3 leversl of
> cache (SGA, Veritas file system , RAID cache) and
> it can be explained by RAID cache replacement, for example (I measured IP
> waiting time, not IP bandwidth).
>

>From the details provided it just seems to me that the issue is having
allocated too much memory to the three DBs coexisting on this server. If the IO problem stopped after shutting down the one DB without modifying the other two DBs' memory parameters, that just seems to point towards a lack of resources and not an SGA caching issue. Paging the SGA to disk is Not a Good Thing.

(Pardon me if you already know this or if my understanding is incomplete. Check this with your SA.)
On Solaris (and other unixes?) you can't look at the %swap stat on top, for example. My SAs tell me that Solaris will only start swapping full processes when it is about to die from lack of memory. On the other hand, it'll page parts of a process's memory space as memory gets tight. The SGA is supposed to be "pinned" in physical memory, I believe, but it'll page it out if forced by lack of memory. If the SGA is getting paged, well, performance on a heavily used system will in most cases go out the window.

But maybe what you mean by "no swapping" is what I mean by "no paging" and I'm barking up the wrong tree. Or, as you suggest, your SGA is fine but you are blowing out your RAID cache.

Steven

--
http://www.freelists.org/webpage/oracle-l