Re: Do user processes apply against shmmax limit?

Rich,

The PGA is never contained within the SGA. In all architectures and on all platforms, it resides separate. The PGA holds process-specific data structures such as the "sort area", the "hash area", and some work areas used during bitmap-index operations.

In "shared-server" (formerly "multi-threaded server" or MTS) architecture, the UGA (a.k.a. "session/user global area", contains cursor-state and session-state info) can reside in the SGA, either in the Large Pool (if configured) or in the Shared Pool (by default). In "dedicated-server" architecture, the UGA is separate from the SGA.

Hope this helps...

-Tim

• Original Message ----- To: "Multiple recipients of list ORACLE-L" <ORACLE-L_at_fatcity.com> Sent: Friday, November 29, 2002 6:48 AM

> Jay,
>
> My understanding is that the PGA is contained within the SGA, and that
> shmmax is the "maximum size of a single shared memory segment". If you
> set shmmax to 256MB and configure 1GB SGA, you should see it allocate 4
> shared memory segments for that purpose. Some Unix variants have
> limitations on the number of shared memory segments which can either be
> created (AIX) or simultaneously accessed (HP-UX). I haven't done much
> with Sun in the last few years so don't specifically know of the Solaris
> limitations, but I'm sure there is probably something there to consider.
> That's typically why you want to set shmmax as high as you realistically
> can -- to reduce the NUMBER of segments you need to allocate for shared
> memory.
>
> Your sysadmin also mentions turning on "priority paging" to give the
> user processes access to the memory before the file cache (aka buffer
> cache). Again I'm not sure about Solaris, but AIX and HP-UX both ship
> with their buffer cache set to something like 10% - 20% of total memory
> by default, which is a pretty good guess for a generic system when the
> vendor has no idea what you'll be using it for specifically... however,
> for large Oracle systems, I typically tune this back a bit, depending on
> the memory in the system. Normally something in the 2-8% or 3-10% range
> is sufficient. Remember, Oracle does all it's own buffering via
> DB_BLOCK_BUFFERS so doesn't really need to rely on the system buffer
> cache, even using filesystems (of course, raw devices completely bypass
> the system buffer cache).
>
> You might want to see what he's got the two parameters set to which
> control the size of the system buffer cache; sometimes reducing that
> will help quite a bit with paging/swapping.
>
> Rich
>
> > -----Original Message-----
> > From: root_at_fatcity.com [mailto:root_at_fatcity.com] On Behalf Of Miller,
> Jay
> > Sent: Saturday, November 23, 2002 1:49 PM
> > To: Multiple recipients of list ORACLE-L
> > Subject: Do user processes apply against shmmax limit?
> >
> > Hi everyone,
> >
> > I was always under the impression that the only concern with shmmax
> was
> > that
> > it be large enough for the SGA to fit into it. One of my System
> > Administrators has just told me that the individual user processes
> (i.e.,
> > the PGA since we're not using multi-threaded server) get added to the
> SGA
> > and if that SGA + user processes > shmmax the system will start
> swapping.
> >
> > I haven't found anything to specifically address this issue on
> Metalink so
> > I
> > though I'd throw it open. We've started experiencing system slowdown
> and
> > he
> > says that increasing shmmax could resolve it. I'm skeptical (he also
> > suggested increasing SGA to decrease swapping which I told him in no
> > uncertain terms was nonsense).
> >
> > If anyone has a link to a note or white paper I'd appreciate that too.
> >
> > I've appended his email at the bottom. This slowdown seems to occur
> even
> > when there's virtually on oracle activity so I'm suspecting some other
> > cause.
> >
> > Thanks,
> > Jay Miller
> >
> >
> >
> >
> > nycsun1 and njsun7 has 6 GB of memory and only 2 GB of share memory.
> This
> > morning nycsun1 was very slow and I noticed that there was lots of
> > swaping.
> > see vmstst and iostat below in red:
> >
> > procs memory page disk faults
> cpu
> > r b w swap free re mf pi po fr de sr s2 s4 s4 sd in sy cs
> us
> > sy
> > id
> > 0 0 23 4366736 97528 1 2186 16 12 12 95520 0 0 0 0 0 1104 3330 974
> 11
> > 8
> > 81
> > 0 0 23 4365992 96056 1 451 16 24 52 85968 3 0 0 0 0 935 847 416
> 3
> > 1
> > 96
> > 0 0 23 4364712 95512 2 310 36 24 492 85968 68 0 0 0 0 1036 2183 670
> 13
> > 4
> > 84
> > 0 0 23 4361568 95488 9 2264 0 76 964 95520 136 0 0 0 0 979 4065 607
> 12
> > 6
> > 82
> > 0 0 23 4362384 96080 1 6 4 8 8 77376 0 0 0 0 0 975 465 457
> 2
> > 1
> > 97
> > 0 0 23 4361944 95712 4 730 92 48 532 95520 64 0 0 0 0 1040 1859 734
> 8
> > 3
> > 89
> > 0 0 23 4360424 95480 4 41 36 40 100 77376 7 0 0 0 0 986 1250 542
> 6
> > 0
> > 94
> > 0 0 23 4361304 96096 3 264 76 36 88 88496 7 0 0 0 0 1037 942 665
> 5
> > 3
> > 92
> > 0 0 23 4359680 95784 2 449 4 28 84 95520 8 0 0 0 0 922 1047 374
> 4
> > 1
> > 95
> > 0 0 23 4359936 95464 2 544 4 20 332 95520 44 0 0 0 0 931 1095 384
> 2
> > 2
> > 96
> >
> > /s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b device
> > 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 c2t6d0
> > 0.0 34.5 0.0 270.0 0.2 13.8 6.7 399.5 6 44 c5t12d0 --
> swap
> > disk
> > 0.0 34.5 0.0 270.0 0.5 10.7 15.5 309.4 18 39 c5t13d0 --
> swap
> > disk
> >
> >
> > This shows that the system is not effectively using memory. I suggest
> > increasing the share memory to 4 GB so that DBAs can increase their
> memory
> > usage. Also set priority paging on. Priority paging will give
> application
> > first priority then free memory will be allocated to file cache(
> Solaris
> > 2.6
> > and 7. Solaris 8 is set dynamically).
> >
> > * ORACLE CONFIGS
> > set shmsys:shminfo_shmmax =2048000000 -- increase to 4096000000
> > set shmsys:shminfo_shmmin=1
> > set shmsys:shminfo_shmmni=300
> > set shmsys:shminfo_shmseg=30
> > set semsys:seminfo_semmap=500
> > set semsys:seminfo_semmni=200
> > set semsys:seminfo_semmns=2000
> > set semsys:seminfo_semmsl=1000
> > set semsys:seminfo_semmnu=500
> > set semsys:seminfo_semume=150
> >
> >
> >
> > --
> > Please see the official ORACLE-L FAQ: http://www.orafaq.com
> > --
> > Author: Miller, Jay
> > INET: JayMiller_at_TDWaterhouse.com
> >
> > Fat City Network Services -- 858-538-5051 http://www.fatcity.com
> > San Diego, California -- Mailing list and web hosting services
> > ---------------------------------------------------------------------
> > To REMOVE yourself from this mailing list, send an E-Mail message
> > to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
> > the message BODY, include a line containing: UNSUB ORACLE-L
> > (or the name of mailing list you want to be removed from). You may
> > also send the HELP command for other information (like subscribing).
>
> --
> Please see the official ORACLE-L FAQ: http://www.orafaq.com
> --
> Author: Rich Holland
> INET: holland_at_guidancetech.com
>
> Fat City Network Services -- 858-538-5051 http://www.fatcity.com
> San Diego, California -- Mailing list and web hosting services
> ---------------------------------------------------------------------
> To REMOVE yourself from this mailing list, send an E-Mail message
> to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
> the message BODY, include a line containing: UNSUB ORACLE-L
> (or the name of mailing list you want to be removed from). You may
> also send the HELP command for other information (like subscribing).
>

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.com
-- 
Author: Tim Gorman
  INET: Tim_at_SageLogix.com

Fat City Network Services    -- 858-538-5051 http://www.fatcity.com
San Diego, California        -- Mailing list and web hosting services
---------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).