RE: dedicated server process memory usage ....
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Tue, 8 Jun 2004 10:17:34 -0400
MessageOn systems such as yours where pmap -x exists and works correctly,
you can see beautifully on the line address 20000000 that nearly all of the
memory for this process is shared. That's your Oracle shared memory area
which you should also be able to view view ipcs. So you see at the bottom
how little private memory you are really using, even though RSS reports it
all per process. You only have an actual problem on systems (mostly ancient,
I think) that keep counting reads of shared memory against each process and
drive false paging. So I don't think you have an operational problem, just a
concern about how the memory adds up. Correct me if I'm wrong.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]On
Behalf Of Pampati, Sree
Sent: Tuesday, June 08, 2004 9:57 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: dedicated server process memory usage ....
Thanks Mark for your reply!
I meant : I just opened sqlplus connection using "sqlplus userid/pwd",
that's all!
You said : ..." but please be sure to avoid taking the .6G per server
process allocation seriously" . I am not sure how to tune/avoid this ?
I sincerely appreciate your thoughts on this .
pmap -x <pid> is given below:
read/write/exec/shared {ism shmid=0X201 ] figures seem to be very big
?!!!!!!
8828: oracleqalc4 (DESCRIPTION=(LOCAL=YES)(ADDRESS=(PROTOCOL=beq)))
Address Kbytes Resident Shared Private Permissions Mapped File
00010000 48272 25952 25952 - read/exec oracle
02F42000 472 472 368 104 read/write/exec oracle
02FB8000 432 320 - 320 read/write/exec [ heap ]
20000000 536576 536576 536576 - read/write/exec/shared [ ism
shmid=0x201 ]
FEBC0000 16 16 16 - read/exec libc_psr.so.1
FEBD0000 88 80 80 - read/exec libm.so.1
FEBF4000 8 8 - 8 read/write/exec libm.so.1
FEC00000 4400 2880 2880 - read/exec libjox9.so
FF05A000 192 192 32 160 read/write/exec libjox9.so
FF08A000 8 - - - read/write/exec libjox9.so
FF0A0000 16 16 16 - read/exec libmp.so.2
FF0B4000 8 8 - 8 read/write/exec libmp.so.2
FF0C0000 8 8 8 - read/exec libkstat.so.1
FF0D2000 8 8 - 8 read/write/exec libkstat.so.1
FF0E0000 32 32 32 - read/exec libaio.so.1
FF0F8000 8 8 - 8 read/write/exec libaio.so.1
FF100000 688 688 688 - read/exec libc.so.1
FF1BC000 32 32 - 32 read/write/exec libc.so.1
FF1D0000 24 24 24 - read/exec librt.so.1
FF1E6000 8 8 - 8 read/write/exec librt.so.1
FF1F0000 8 8 - 8 read/write/exec [ anon ]
FF200000 568 568 568 - read/exec libnsl.so.1
FF29E000 40 40 - 40 read/write/exec libnsl.so.1
FF2A8000 24 16 - 16 read/write/exec libnsl.so.1
FF2C0000 8 8 8 - read/exec libsched.so.1
FF2D2000 8 8 - 8 read/write/exec libsched.so.1
FF2E0000 8 8 8 - read/write/exec/shared [ anon ]
FF2F0000 24 24 24 - read/exec libgen.so.1
FF306000 8 8 - 8 read/write/exec libgen.so.1
FF310000 40 40 40 - read/exec libsocket.so.1
FF32A000 8 8 - 8 read/write/exec libsocket.so.1
FF330000 8 8 8 - read/exec libskgxn9.so
FF340000 8 8 - 8 read/write/exec libskgxn9.so
FF350000 8 8 8 - read/exec libskgxp9.so
FF360000 8 8 - 8 read/write/exec libskgxp9.so
FF370000 8 8 - 8 read/write/exec [ anon ]
FF380000 8 8 8 - read/exec libodmd9.so
FF390000 8 8 - 8 read/write/exec libodmd9.so
FF3A0000 8 8 8 - read/exec libdl.so.1
FF3B0000 160 160 160 - read/exec ld.so.1
FF3E6000 16 16 - 16 read/write/exec ld.so.1
FFBE6000 40 40 - 40 read/write [ stack ]
-------- ------ ------ ------ ------
total Kb 592320 568344 567512 832
febiccp@eceisdblc4 - qalc4
Thanks,
Sree Pampati
617-392-1594
-----Original Message-----
From: Mark W. Farnham [mailto:mwf@xxxxxxxx]
Sent: Monday, June 07, 2004 10:34 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: dedicated server process memory usage ....
Unfortunately it is non-trivial to figure out how much of the shared
memory area of the running Oracle instance is reported as part of each
server connection, and this varies by OS, release of OS, and tool used to
report "sz" and "RSS."
I'm not sure what you meant about "no SQL fired" since I'm not aware of
how to open a sqlplus connection without the query of username/password
firing, along with whatever level of auditing you may have.
Possibly someone else can refer you to a good resource for evaluating
MTS, but please be sure to avoid taking the .6G per server process
allocation seriously. I doubt Solaris 8 has this problem, but some old Unix
varieties had page/swap limits that did not take into account shared memory,
so executing a table scan would drive "false paging" unless you configured
"RSSMAX" to the ceiling of your shared memory plus the program space. By
"false paging" I mean copying shared memory your process has mapped to page
or swap (from whence it will never be recalled, since the shared memory is
still actually current.) Setting the RSSMAX high would in turn then allow
applications to run rampant on your memory, so this tipped the balance of
whether to allow a certain application on your database server far in favor
of saying no if an application had big libraries.
Good luck!
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx]On Behalf Of Pampati, Sree
Sent: Monday, June 07, 2004 9:42 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: dedicated server process memory usage ....
Hi,
How is memory allocated to a dedicated server process ? I just
opened a sqlplus connection ( no SQL fired!), did ps -elf | grep <pid> on
the server pid, which showed sz: 592M, RSS: 558. There are about 800
server processes running at that time ( the box has 10G physical mem, of
which about 2GB was free at that time. Solaris 8 . Oracle version
:9.2.0.4 ).
*_area_size parameters in init.ora are : bitmap_merge_area_size =
1048576 (1MB)
create_bitmap_area_size= 8388608 (8MB)
hash_area_size = 4194304 (4MB)
sort_area_size = 2097152 (2MB)
Oracle 10gAS and 9iAS too are running on the box in addition to the
oracle instance.
Is this a good candidate for MTS ? Do I get substantial benefit
using pga_target_aggregate features of 9i ?
I would very sincerely appreciate your valuable feed back ( any
pointers to docs/info are highly appreciated).
Thanks,
Sree Pampati
-----Original Message-----
From: Darrell Landrum [mailto:darrell@xxxxxxxxxxx]
Sent: Saturday, June 05, 2004 12:58 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Re: Hardware Question
Hey Jay,
I hate when management comes to me with questions like you mention
about this new hardware handling the workload.
But, I have a response now that I love because it can be confusing
to them.
I simply ask, "Why?".
Why are they looking at new hardware? Does the current hardware not
handle the workload? Is there a forthcoming app server/database upgrade
that will require more resources? Are they scaling out their current
utilization, perhaps to more users? Are they adding more databases? I know
this may not seem like the best attitude and trust me, I love newer, faster
hardware more than most, but really the justification for new hardware
should come before the shopping for hardware. In my role as a DBA, I should
be the one (or, of course, the sys admins) that recognizes the need for
hardware upgrades and hopefully before management starts to feel the need.
If I can't quantify the need for new servers or additional hardware (or
software for that matter), I actually speak against it. This way, when I
tell them we need something, they listen!
Just one guy's 2 cents.
However, you mention a bottleneck on the SAN controllers. That is a
big red flag. There's always a chance (and some would argue a very good
chance) that faster processors and more memory will make this SAN controller
bottleneck worse or at least more noticeable. Hopefully, additional
channels to the SAN are being considered with this new box as well.
Good luck!
Other related posts:
