Re: SGA bigger than 50G
- From: Sanjay Mishra <smishra_97@xxxxxxxxx>
- To: dbvision@xxxxxxxxxxxx
- Date: Thu, 6 May 2010 14:15:32 -0700 (PDT)
Nuno
Thanks for the time and explanation. This is new environment built only a month
back and request came to increase the setting . Even the Loading is still not
completed and trend is not set to get more details. Even the Db is not put in
Archivelog mode due to big loading activities palnned for next more couple of
weeks. I was trying to put together all setup to go with the plan or provide
more information. Surely the earlier reason that is restricting the increase is
the SHMALL setting which along with SHMMNI is coming out 32 G and so I am not
able increase.
I am curious to see the setting for hugepage/hugemem in Linux environment and
what is the good setting to make it more efficient and optimized.
Thanks
Sanjay
________________________________
From: Nuno Souto <dbvision@xxxxxxxxxxxx>
Cc: "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
Sent: Thu, May 6, 2010 7:16:52 AM
Subject: Re: SGA bigger than 50G
Folks, if I may interject here:
Sanjay claimed he was using Linux. Giving him information on Solaris and other
Unix variants means essentially nothing and only adds to confusion: Linux does
VM in a totally different manner from Solaris (which release?) and most other
Unixes, like for example AIX.
So please let's not confuse things: stick to Linux. No, VM in Linux is NOT the
same as *nix! There are some very big differences.
My comment regarding swap size is very simple: in some Linux releases swap is
reserved to the total size of process memory even if not all used by the
running processes.
And when it runs out, strange things happen: the error recovery from swap space
exhaustion is not defined in any standard and various flavors of Linux and even
Unix react in many different manners.
Of course swap capacity has nothing to do with Oracle SGA, but it is affected
by Oracle processes' runtime memory usage and that can be a direct result of
the SGA size.
Hugepages are just a special case of all this. Just as an example: AIX can page
out hugepages, for them to be locked it's not enough to just declare or use
them. I don't know what Red Hat does now but in earlier releases the behaviour
with hugepages changed with just about every patch, so it pays to check rather
than assume they behave in any given manner.
As to the need for a large SGA: without knowing the exact characteristics of
the given application, it is impossible to claim it is unnecessary. And the
label of "DW" means absolutely nothing: I can show you DW instances that do
mostly reads, while others do mostly writes and ours just reached 4TB writes, 7
TB reads per day and 45GB SGA. And please don't claim it is or it is not a DW:
it is a DW and it does what it needs to do, period.
So let's ease off the "a DW is not that" comments and stick to try and help
Sanjay? Quite frankly I don't give a hoot what the accepted definition in the
corridors of Oracle Acedom is for a DW: if a client says it's a DW, well, it
is; end of story!
My guess is that we need to know more about what Sanjay has got there before
going off on what he should have. I'd look at the hugepage usage, at the swap
size use and at the shmmax setting, the flavor and release level of Linux he's
using, whichever is applicable in his particular case. Only as a last resort
would I suggest reducing the size of the SGA based on an assumption about the
needs of his application.
-- Cheers
Nuno Souto
in sunny Sydney, Australia
dbvision@xxxxxxxxxxxx
--
//www.freelists.org/webpage/oracle-l
Other related posts:
