RE: Odd swap & memory behavior on Solaris 9, Oracle 9iR2

  • From: "Bobak, Mark" <Mark.Bobak@xxxxxxxxxxxxxxx>
  • To: <mzito@xxxxxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Mon, 17 Apr 2006 11:29:42 -0400

Hi Matt,
 
I ran into this a while ago, on boxes where the swap configuration was
pretty skimpy.  We had cases where we could not fully utilize available
RAM due to swap shortages, and the fact that Solaris wants to "reserve"
swap space equal to the amount of RAM being allocated.
 
The solution was to find an "official" MetaLink note that recommends
having swap 2x physical RAM, and using that and the "disk is cheap"
argument to get the sysadmins to configure more swap.  Once we did that,
we never looked back, and there's been no problem.
 
In reality, you can probably get away w/ 1.5x RAM, maybe less on boxes
w/ really large amounts of RAM.
 
Hope that helps,
 
-Mark
 

-- 
Mark J. Bobak 
Senior Oracle Architect 
ProQuest Information & Learning 

For a successful technology, reality must take precedence over public
relations, for Nature cannot be fooled.  --Richard P. Feynman, 1918-1988

 

________________________________

From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Matthew Zito
Sent: Monday, April 17, 2006 11:23 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Odd swap & memory behavior on Solaris 9, Oracle 9iR2





Howdy folks,

A friend of mine is seeing something odd on their Solaris 9/Oracle 9iR2
environment.  They were having problems swapping constantly, and they
gave me a call.  They had their SGA + number of sessions * PGA sized
significantly larger than their physical RAM, so I recommended they back
everything down.  However, the really odd behavior they see is that when
they start up the Oracle instance, an amount of swap equivalent to their
SGA is allocated, even when there's tons of free physical memory.
Looking around metalink produced this little snippet from Note 151222.1:

"In Solaris 8, thanks to DISM support (Dynamic Intimate Shared Memory),
it is
possible to avoid the above limitation and allocate the shared memory
segment
such that only the active granules in the SGA are located in physical
memory. 
All unused granules will be stored in swap until needed.

NOTE:  Swap space for the entire SGA_MAX_SIZE is required.  See Note
230453.1
  for more details."

However, note 230453.1 doesn't exist anymore.  I tested a similar config
on a Solaris 8 box we have in our lab, and a similar behavior was not
evidenced.  Anyone have ideas where we should be looking?

Thanks,
Matt 

Other related posts: