Re: Why do I get out of memory errors when 10GB memory is free?

From: joel garry <joel-garry_at_home.com>
Date: Wed, 15 Jul 2009 14:26:12 -0700 (PDT)
Message-ID: <b373461f-a7f0-4288-82b5-19ddbb96b585_at_y4g2000prf.googlegroups.com>



On Jul 15, 7:15 am, zigzagdna <zigzag..._at_yahoo.com> wrote:
> On Jul 15, 7:57 am, John Hurley <johnbhur..._at_sbcglobal.net> wrote:
>
>
>
> > On Jul 14, 9:56 pm, zigzagdna <zigzag..._at_yahoo.com> wrote:
>
> > snip
>
> > > > The HP engineer should be able to work with you going thru the hp ux
> > > > oracle customization doc and checking and then double checking
> > > > everything.
>
> > > > Good luck and keep us posted.
>
> > > First HP engineer reduced  kernel parameters for UNIX buffers, so
> > > kernel occupies less memory. Hope was that it will make more memory
> > > avaiable for application (which it did), but it did not fix the
> > > problem. Subsequntly. swap space was increased.
> > > Swap space increase fixed the rroblem . I am not a UNIX admin, so I
> > > still wonder when 10GB memory is avaiable why HP OS first won't
> > > utlize  this memory instead of going to swap to get heap memory .  We
> > > did not change  maxdsiz, maxssiz or maxtsiz prameters.
> > > I am glad problem is now resolved. I can put more oracle instances on
> > > the server and get my company money's worth.
>
> > The hpux kernel parameters dbc_max_pct and dbc_min_pct come out of the
> > box with values that reserve huge chunks of memory for unix buffering
> > that are pretty much unacceptable for any database server and pretty
> > much have to be checked and revised downward.
>
> > On a machine with 16 gig of memory or more I would really hope you
> > have them down to like 5 or so ... with 24 gig of memory like 2 or 3
> > would be probably a good value.  You can find postings on this topic
> > in cdos from a while back but well not many hpux questions come into
> > here.
>
> > As an oracle dba running on unix or linux ... you just have to ante up
> > and take responsbilitity for reviewing and being at least somewhat
> > knowledgeable about setting and reviewing this stuff.
>
> > Why?  Because no one else is likely to care enough or be competent
> > enough to take the time and assess what should be done.
>
> > Oracle has documented pretty well the relevant things that should be
> > checked and in most cases provides hints about checking related things
> > like dbc_***_pct ( as far as I can recall ).  HP provides white papers
> > etc that are also related to tuning kernel stuff when running oracle.
>
> > Are you using direct IO and avoiding the unix buffer cache?  Do you
> > even know what to look at to find out the answer to this question?
>
> > Before you think you are done here I would take a look at that area
> > immediately.
>
> > The whole subject of how many instances one puts onto a dedicated
> > server also deserves some consideration and review.
>
> > Good luck!- Hide quoted text -
>
> > - Show quoted text -
>
> Regraidng reducing UNIX buffers, we typically reduce them if server
> does not have lot of memory.
>
> I am still puzzled why increasing swap will fix the problem when
> server alreday has 10GB memory free.
>
> Prem

In addition to what Mladen said, note there are several different ways of allocating swap on hp-ux - even using memory. Since the notion that swap will rarely actually be used is correct (barring extreme configuration blunders), you might consider some filesystem based swap on a lesser used device, if that makes sense for your configuration (for example, if you have an extra unused internal drive). Modern unix are less picky about swap than in the olden days, but it's still an issue, basic enough that not doing it really is incompetent admin.

I told you to check swap back in May.
http://groups.google.com/group/comp.databases.oracle.server/browse_thread/thread/565184ddad1c731c/

jg

--
_at_home.com is bogus.
Just wondering how far back forums.oracle messages go.
http://forums.oracle.com/forums/message.jspa?messageID=119930
Received on Wed Jul 15 2009 - 16:26:12 CDT

Original text of this message