About PGA

Hi All,
According to my knowledge on PGA
1) PGA is a memory structure but not part of SGA
2) PGA is allocated for each session
3) PGA is allocated based on the value of PGA_AGGREGATE_TARGET parameter
My doubt is
Like SGA, PGA also occupies fixed amount of physical memory ?
If it is, how much memory is allocated to each session? (equal to the value of PGA_AGGREGATE_TARGET to each session. In such a case all memory will be exhausted very soon)
Please explain me how PGA is allocated to each session.
I am using 10.2.0.4
Thanks
Prasanna

Hi kpskram,
According to your knowledge on PGA;
1) You are right.
2) You are right.
3) PGA_AGGREGATE_TARGET is the maximum number that PGA area can increase.
The value for SGA is automatically allocated as the amount for it during the instance startup but PGA is not 100% allocated. It decreases and increases based on user logins. It is the sort area and please also check for parameters sort_area_size and hash_area_size. But if you set the PGA_AGGREGATE_TARGET, you do not need those.
There is a default value allocated for each session for PGA and if i remember correctly it is something like 5 megabytes?
Ogan

Similar Messages

Question about PGA and TEMP usage

hi,
we had a situation whereby OEM was reporting that because the PGA was too small during a time frame that extra I/O on the TEMP was being created. Does this then mean that the TEMP was acting as a pseudo PGA.
regards

Have a look at these threads
Re: PGA memory problem - Oracle 10.2.0.4 on windows 2003
Re: PGA Memory Usage Details
<br>
Oracle Database FAQs
</br>

Oracle 10.2.0.4 PGA on Windows Enterprise server 2003 32 bit

Hi all,
i have a question about PGA. Querying v$pgastat the values "Total freeable PGA memory" and "PGA memory freed back to OS" are always 0. My Database is configured in DSS mode.
Any IDEA?
Thanks....

Yes i have already executed these query and the result is that the amount of "pga memory" in use for each process is different from "total pga allocated" and i'm not able to understand why oracle dosen't release memory back to OS.
My current setting is:
4gb available on OS
PGA= 800MB * (2 Gb *80%)*50% -- pga_aggregate_target=800Mb
SGA=700MB --sga_target=700Mb
Each time that virtual memory allocated for oracle.exe is beyond 2.7 gbyte i get ora-04030
Maybe each PLSQL in not written right and each session does not release memory. is there a workraound to free memory at system level and not at session level through DBMS_SESSION.FREE_UNUSED_MEMORY.
Best regard,
Simone ABATE

WHY PGA IS REQUIRED FOR EVERY USER ?

Good Morning Everyone ;
I have a question about PGA.
WHY PGA IS REQUIRED FOR EVERY USER ?
What i got from google ..
Even though the parse information for SQL or PL/SQL may already be available in library cache of shared pool,
the value upon which the user want to execute the select or update statement cannot be shared.
I cant realize it Can anyone show clear example , if possible ?
DB Version is 10.2.0.4.0
OS : oracle linux 5.5
Thanks in advance ..

Thanks aman and heok.
My Question :
Your explanation is clear. I think i am getting little bit confused.
Could you please clarify little more ?
>> session 1 :
user is HR
SQL>select * from tab1 ORDER BY name;
>> session 2 :
user is scott
SQL>select * from tab1 where ORDER BY name;
>> session 3 :
user is USER1
SQL>select * from TAB1 where ORDER BY name;
>> session 4 :
user is USER2
SQL>select * from TAB1 where ORDER BY name;
IS this right aman ?
Already sql statements are avail in SGA ,Even though all above users needs same information.
Oracle does sorting operation in PGA. If PGA exceeds , oracle will use temporary tablespace .
Thanks heok and aman.

PGA vs SGA

hi,
i read the documents which says that pga has individual user datas like bind variables.session info.
here oracle how the pga communicate with sga for execution.
what are the components for PGA
thanks
with regards

user3266490 wrote:
hi,
i totally confused about PGA and shared pool.is there any relation between librarcy cache and pga.
whether,when the user exeutes select,insert,update,delete these are in librarcy cache or PGA.
i read that parsing,optimzation,execution,fetching there are happening in Library cahe.
so why oracle created the runtime area in PGASo why there is a confusion? The parsing, execution, they are all happening in the SGA because what is needed is that if someone else also requires the same data, for example, the same execution plan, the same data buffers, it should all be accessible from the SGA which is going to be shared among the users. But the PGA contains the data structures which don't do this! Answer this, when you do an order by , does it mean that all the other users also start seeing teh data in the order by clause by default? The answer should be no. This sorted data is just kept in the pga of yours. Similarly, there is a bind variable which is shared among the users but two different users need to use two different values of it.Where it is best to keep those different values, in their own PGAs isn't it? Hence its the private memory of the sessions.
HTH
Aman....

Server resources availability while executing a query

hi,
lets say we have a query that requires XX MB to be executed.
the needed memory is available, and while the query is getting memory, another threads jump on the database, and the memory needed to execute the query is not available.
under those circunstances, can the query change the memory requirements to execute the query (i.e. page out to temp objects), or the query will NEED the XX MB to be executed?
can the execution plan change the resources requirements based on database recourses availability ?
Thank you,
Eladio

The Execution Plan is determined before the query begins executing.
The query cannot stop and restart with a new execution plan.
However, if the same query is executed more than once, even inside a loop
in a PLSQL block, it may get re-parsed and may acquire a different Execution
Plan.
PGA_AGGREGATE_TARGET is a "guide" for Oracle to dtermine the available
memory for an SQL (it looks at PGA memory currently used by other concurrently
executing SQLs). However, considering the dynamics of multiple processes
co-existing on a server, this is not and cannot be a hard limit so the actual
PGA in use at any time can fluctuate.
(consider, for example that other, Non-Oracle, processess might start up and
take memory and Oracle isn't aware of those processes "stealing" available
memory. Or consider a bug in sort / hash algorithms which cause an Oracle
process to take (or request from the OS) much more memory than it "should",
resulting in ORA-4030s !)
However, as a general rule, Oracle uses PGA_AGGREGATE_TARGET
and also provides an advisory view that you can query. (Oracle periodically
collects statistics about PGA usage and SGA usage and updates the
respective advisory views).

Query tuning (Oracle 10g)

Hi
We are having one query that is having order by clause. If we run with order by then it takes 70-80 min to execute and without order by it takes 1-2 min.
We cann't remove order by because of business requirements.
Please suggest me any solution except index, hints (already there).
Thanks and regards

What is your database version? If you are using an automated PGA memory management then you might want to check your PGA_AGGREGATE_TARGET.
You can use V$PGA_TARGET_ADVICE to set up a proper PGA_AGGREGATE_TARGET.
Read more about PGA Memory management here
http://download.oracle.com/docs/cd/B19306_01/server.102/b14211/memory.htm#i49320
It has a lot to do with Sorting Operations.
Thanks,
Karthick.

USER_DATASTORE: Could DBMS_LOB be leaking memory?

We've got a database which is indexed with USER_DATASTORE and the following procedure. We're getting an occasional situation where this message appears in the alert log:
ORA-04030: out of process memory when trying to allocate 8716 bytes (pga heap,Get krha asynch mem)
CKPT: terminating instance due to error 4030
Doing block recovery for file 2 block 81536
ORA-04030: out of process memory when trying to allocate bytes (,) (several times)
Instance terminated by CKPT, pid = 4900
Starting up ORACLE RDBMS Version: 10.1.0.2.0.
Could it be something in the procedure that is leaking or holding onto CLOBs longer than it should?
Thanks!
CREATE OR REPLACE PROCEDURE KM_TESTING_SMALLER_INDEX_PROC
(RID IN ROWID, TLOB IN OUT CLOB) IS
BEGIN
DBMS_LOB.OPEN(TLOB,DBMS_LOB.LOB_READWRITE);
DBMS_LOB.TRIM(TLOB,0);
FOR C1 IN (SELECT REPLACE(BEGINBATES,'<',' ') BEGINBATES,
REPLACE(ENDBATES,'<',' ') ENDBATES,
REPLACE(DOCTITLE,'<',' ') DOCTITILE,
REPLACE(OCR2,'<',' ') OCR2 FROM KM_TESTING_SMALLER WHERE ROWID=RID) LOOP
DBMS_LOB.WRITEAPPEND(TLOB,20,'<BEGINBATESnxtfield>');
IF (C1.BEGINBATES IS NOT NULL) THEN
IF NOT(LENGTH(C1.BEGINBATES)=0) THEN
DBMS_LOB.WRITEAPPEND(TLOB,LENGTH(C1.BEGINBATES),C1.BEGINBATES);
END IF;
END IF;
DBMS_LOB.WRITEAPPEND(TLOB,39,'</BEGINBATESnxtfield><ENDBATESnxtfield>');
DBMS_LOB.WRITEAPPEND(TLOB,36,'</DOCTYPEnxtfield><DOCTITLEnxtfield>');
IF (C1.DOCTITLE IS NOT NULL) THEN
IF NOT(LENGTH(C1.DOCTITLE)=0) THEN
DBMS_LOB.APPEND(TLOB,C1.DOCTITLE);
END IF;
END IF;
DBMS_LOB.WRITEAPPEND(TLOB,35,'</DOCTITLEnxtfield><AUTHORnxtfield>');
DBMS_LOB.WRITEAPPEND(TLOB,27,'</CCnxtfield><OCR2nxtfield>');
IF (C1.OCR2 IS NOT NULL) THEN
IF NOT(LENGTH(C1.OCR2)=0) THEN
DBMS_LOB.APPEND(TLOB,C1.OCR2);
END IF;
END IF;
DBMS_LOB.WRITEAPPEND(TLOB,15,'</OCR2nxtfield>');
END LOOP;
DBMS_LOB.CLOSE(TLOB);
END;

Here is the documentation for ORA-04030:
ORA-04030 out of process memory when trying to allocate string bytes (string,string)
Cause: Operating system process private memory has been exhausted.
Action: See the database administrator or operating system administrator to increase process memory quota. There may be a bug in the application that causes excessive allocations of process memory space.
Generally speaking it means that the box hosting the database instance is running out of memory. You have some options:
- decrease memory usage on the server (reduce PGA ? reduce SGA ? check any non-Oracle process memory usage)
- add more RAM to your server.
You check global statistics about PGA memory with:
select * from v$pgastat;You can check PGA statistics for each server process with:
select * from v$process;Note that the memory here is the PGA memory (not the SGA). PGA memory is private memory allocated by each Oracle server process (background process or dedicated server process).

Server resources

Hi,
please could you help my understanding of the various outputs pasted below. I am trying to check resources on my database server(which already has 9 database instances) before adding another database instance:
i) Total shared memory used up by existing database instances is roughly *4321M* from outputs pasted below. The value of max shared memory in /etc/system is however approx *4294M* which is less than what the existing instances are using. Does that suggest that already we have exceeded the allowed shared memory limit? And therefore the existing databases run the risk of poor performance?
ii) How does the total value of shared memory relate to the value of total real memory (or RAM) showed by top i.e 8192M . Does it mean that we can increase the value of shminfo_shmmax on the server within the value of available free memory 1473M shown by the top command?
iii) I have researched the explanation of the vmstat output but don't quite understand it, would someone be able to explain it very simply?
Show sga
Total System Global Area 839548928 bytes
Total System Global Area 377487360 bytes
Total System Global Area 377487360 bytes
Total System Global Area 377487360 bytes
Total System Global Area 377487360 bytes
Total System Global Area 377487360 bytes
Total System Global Area 839548928 bytes
Total System Global Area 377487360 bytes
Total System Global Area 377487360 bytes
Total for the 9 databases = 4321509376 bytes
From ipcs -bpm
IPC status from <running system> as of Thu May 13 14:34:41 BST 2010
SEGSZ
Shared Memory:
843071488
377495552
377495552
377495552
843071488
377495552
377495552
377495552
377495552
*4328611840 -----Total*
*/etc/system*
set shmsys:shminfo_shmmax=4294967295
set shmsys:shminfo_shmmin=1
set shmsys:shminfo_shmmni=100
top
load averages: 0.20, 0.18, 0.18 16:52:18
332 processes: 331 sleeping, 1 on cpu
CPU states: 94.1% idle, 2.8% user, 1.5% kernel, 1.5% iowait, 0.0% swap
Memory: 8192M real, 1473M free, 5629M swap in use, 8664M swap free
vmstat 3 5
kthr memory page disk faults cpu
r b w swap free re mf pi po fr de sr m0 m1 m2 m3 in sy cs us sy id
0 0 0 9409000 2023528 602 205 3413 2 2 0 1 0 0 0 0 374 1051 808 5 2 93
0 0 0 8862120 1493344 19 21 0 0 0 0 0 0 0 0 0 386 567 1005 0 0 100
0 0 0 8862120 1493344 0 0 0 0 0 0 0 0 0 0 0 377 498 993 0 0 100
0 0 0 8850984 1484040 161 3019 0 0 0 0 0 0 0 0 0 478 3935 1183 4 3 93
0 0 0 8844800 1479592 0 0 0 0 0 0 0 0 0 0 0 391 458 1030 0 2 98
Thank you.

The Execution Plan is determined before the query begins executing.
The query cannot stop and restart with a new execution plan.
However, if the same query is executed more than once, even inside a loop
in a PLSQL block, it may get re-parsed and may acquire a different Execution
Plan.
PGA_AGGREGATE_TARGET is a "guide" for Oracle to dtermine the available
memory for an SQL (it looks at PGA memory currently used by other concurrently
executing SQLs). However, considering the dynamics of multiple processes
co-existing on a server, this is not and cannot be a hard limit so the actual
PGA in use at any time can fluctuate.
(consider, for example that other, Non-Oracle, processess might start up and
take memory and Oracle isn't aware of those processes "stealing" available
memory. Or consider a bug in sort / hash algorithms which cause an Oracle
process to take (or request from the OS) much more memory than it "should",
resulting in ORA-4030s !)
However, as a general rule, Oracle uses PGA_AGGREGATE_TARGET
and also provides an advisory view that you can query. (Oracle periodically
collects statistics about PGA usage and SGA usage and updates the
respective advisory views).

SORT_AREA_SIZE follow up

I am using this query in my instance to determine the % of sorts done in memory.
select name, value from v$sysstat
where name like '%sort%';
Now we run other processes in this instance besides OLAP builds and I am curious if this query is good enough to give me a good look at how many disk sorts the OLAP is responsible for. I seem to be 99.99% sorting in memory.
If I ran this query before a build and then right after using only the diffs between the 2 queries would that give me a better idea of performance?

Hi ZeroWing,
What I normally do to monitor performance during an OLAP build is to schedule a whole series of queries to capture the following information:
   olap_waits            - collects and records information about OLAP wait events
   aw_read_writes        - collects and records information about temp, lob, cache writes and rows processed
   aw_long_ops           - collects and records information about long operations being performed within the AW
   aw_buffer_cache       - collects and records information about use of the buffer cache
   aw_cursor_parameters - collects and records information about cursor usage
   aw_pga_performance    - collects and records information about pga performance
   aw_session_resouces   - collects and records information about session usage for a specific user
   aw_size               - collects and records information about size of aws for a specific user
   aw_sorts              - collects and records information about sort activity
   aw_redo_log_buffer    - collects and records information about redo and log buffer activity activity
   aw_ts_read_writes     - collects and records information about read/write activity for tablespaces and datafiles
   aw_segment_stats      - collects and records information about object segement statistics for a specific userI run these jobs via the database scheduler at regular intervals whilst a build is running. Each query inserts its results into a fact table so I can analyze the results once the load has finished. This fact table is then joined with the XML Load Log table. This allows me to match specific events within XML Load Log with system and instance statistics.
I am not sure what impact collecting all this information has on the overall performance of the instance (hopefully it is minimal?) but it does provide interesting information. At the moment I am still refining the info I collect and structure of the AW.
Keith Laker
Oracle EMEA Consulting
BI Blog: http://oraclebi.blogspot.com/
DM Blog: http://oracledmt.blogspot.com/
BI on Oracle: http://www.oracle.com/bi/
BI on OTN: http://www.oracle.com/technology/products/bi/
BI Samples: http://www.oracle.com/technology/products/bi/samples/

Enq tc -contention

Hello,
I just started analysing our database i found a wait event in EM graph(other). when i dig into the other wait event i encounter a wait event (enq tc -contention). i have tried to find solution on net. but i did not find it on Google or even in matalink. it take considerable time for example if insert statement taking 2mint to complete it execution one mint is spending on this wait event. kindly help me out how to go away this event
Oracle 11 r2
windoows server 2003
Redo and Control Files are On SSD
Rest of all Files are on SAN.
Edited by: Oracle Studnet on May 26, 2011 2:51 AM

Thanks but he discussed TX contention but i have encounter with TC contention? could you plz provide my any ML note where i can find out the root cause of this Problem. This wait event taking 80% database time. i found top foreground wait in AWR. Simple insert or merge statement are causing enq TC contention. it might be possible instance memory configuration would be incorrect. i have observed some issue in AWR
Instance Efficiency Percentages (Target 100%)
Buffer Nowait %: 100.00 Redo NoWait %: 100.00
Buffer Hit %: 100.00 In-memory Sort %: 100.00
Library Hit %: 98.73 Soft Parse %: 98.32
Execute to Parse %: 47.27 Latch Hit %: 100.00
Parse CPU to Parse Elapsd %: 48.23 % Non-Parse CPU: 99.6 Let check my Memory Statistic
Host Mem (MB): 32,757.8 32,757.8
SGA use (MB): 15,680.0 15,680.0 ()
PGA use (MB): 515.2 508.2 why PGA used so small memory? (automatic memory is enable memory_target 25GB)
% Host Mem used for SGA+PGA: 49.44 49.42 i have changed some default setting due to large sort operation. check out parameters that i have changed
ALTER SYSTEM SET SEC_CASE_SENSITIVE_LOGON = FALSE;
alter system set db_files = 10000 scope = spfile;
alter system set processes = 500 scope = spfile;
alter system set memory_max_target = 25000M scope=spfile;
alter system set memory_target = 25000M scope = spfile;
alter system set workarea_size_policy = manual scope=SPFILE;
alter system set sort_area_size = 204800000 scope=SPFILE;
alter system set hash_area_size = 409600000 scope=SPFILE;
alter system set "_pga_max_size"=2000000000 scope=SPFILE;
alter system set "_smm_max_size"=1073741824 scope=SPFILE;
ALTER SYSTEM SET large_pool_size = 1073741824 SCOPE=SPFILE;But my instance taking 15GB which is the default sga_max_size memory would i need to increase it? if i set it 20GB then i mean SGA will use 20GB what about PGA. there are large sorting operations like join table are perform night loading (batch job).
why PGA is taking less memory with these settings?
Top 5 Timed Foreground Events
Event Waits Time(s) Avg wait (ms) % DB time Wait Class
enq: TC - contention 3 3,161 1053811 79.56 Other
DB CPU 528 13.28
db file sequential read 73,709 125 2 3.14 User I/O
enq: CF - contention 4 106 26540 2.67 Other
log file sync 391 57 145 1.43 Commit

Screen goes black when playing "Tiger Woods PGA Tour 08"

No need to mention that I am very happy with my new iMac. It replaces my G4 MDD PowerMac. The iMac is huge, quiet and fast on my graphic design jobs. I even won´t want to miss the glossy screen.
Yesterday I received the brand new "Tiger Woods PGA Tour 08". Keen on trying the game I installed and successfully created a player and got first practice shots but as soon as I try to actually start a game scenario, it appears that the scenario just only loads about 90 % and then screen goes black. From there I can only shut down my iMac with the power switch.
Now I found a support note on the EA Games Site. Title: "Why does "Tiger Woods PGA Tour 08" lock up with a black screen when playing on an iMac 7.1?" Answer: "This is caused by a driver issue with the video card that will be adressed by next system update by apple later this month."
Question: is this indeed so? Can I expect a system update soon? Anybody there with the same problem?
P.S.: I checked if I could run the game with a new user. Same problem.
Thanks for help.

Thank you for replying. There are more postings in this forum about that issue. I.e. gamers trying to play Unreal Tournament suffer from the same freezing on the new 24" iMac. Have a look as well at the reviews in the US apple store on this game.
Solution is near. Installing 10.5 Leopard should solve the problem. As I do not want to install Leoprad yet, because some of my graphic apps will not run smoothly on leopard I have to wait for the next update for Tiger 10.4.11
Obviously EA Games is to blame and failed to test the game good enough before hand.

SGA/PGA best practice

With an XE limitation of 1GB for the PGA and SGA, how would one best preserve this limitation?
In other words, how would one minimise the PGA and SGA usage via ones interface with Oracle?
For instance, in a SQL query I was building I could either (The query has multiple aggregates on multiple tables in a 'one to many' join relationship):
1) Use WITH clauses
2) Use inline views
3) Use standard views
4) Use temporary tables
5) Maybe more possibilities
Without much of an insite into how Oracle 'works' it's difficult to gauge which of these is less 'resource hungry'.
Possibly there is a reference on how to use the PGA & SGA efficiently?
Or possibly a more definitive explanation on what's actually stored in these memory areas (rather than just 'prorgam' and 'system' data).
Regards all,

SGA is used in 2 main areas: parsing the SQL and avoiding disk reads by having stuff in memory. There is one SGA for all of XE instance.
PGA is used heavily for sorting, hash joining and 'working stuff'. The PGA is created for each Dedicated Server process. Much of the PGA is inside the SGA for Shared Server processes
SGA: (There are exceptions to each of these 'Rules Of Thumb' or 'ROT' statements)
Good coding technique will allow reuse of SQL and avoid as much as possible reparsing;
Use bind variables;
Do not create SQL on the fly;
Avoid dynamic SQL (concat strings to create SQL on the fly);
Forget about tweakiing the buffer cache as it is mainly a performance thing that you can't do much about in XE;
WITH clause is nice as it may reduce the count of subqueries;
Leave it with 'automatic memory management' (again a perf hit, but it will attempt to optimize mem availability);
Don't reinvent - Use the features that are loaded as they will likely get statements out faster, releasing memory faster;
Stored Views may (often do) grab unneeded table data. When joining to stored views, see whether they can be replaced by cleaner code;
Do not create SQL on the fly.
PGA:
Use shared server when feasible (at risk of performance hit);
Global Temp Tables go into PGA, watch the use;
Maximize reuabale SQL, minimze all other languages (remote or in DB) including PL/SQL;
Watch for PGA wasters such as UNION vs the better UNION ALL, and unneeded sorts;
PGA is 'swapped out' to Temporary tablespace. Make sure temp tablespace big enough ... it's not counted in the 4GB disk;
Avoid using COMMIT except at end of transaction. Avoid COMMIT in loops. Make sure undo tablespace is big enough ... it's not counted in the 4GB disk ;
Did I mention it's not a good idea to create SQL on the fly?

How can I fix a problem with PGA Tour shot tracker?

PGA Tour Shot Tracker will not display for the full field. It appears momentarily and then goes all black. I've disabled FlashBlock, AdBlock, and still no help. It (Shot Tracker) will work for groups and individual players, however.

Hello Meaghan,
Thank you for the details of the issue you are experiencing with iMessage. I recommend the following steps for the issue you described:
To resolve issues with sending and receiving iMessages, follow these steps first
Check iMessage system status for current service issues.
Go to Settings > Messages > Send & Receive and make sure that you have registered iMessage with your phone number or Apple ID and that you have selected iMessage for use. Learn more abouttroubleshooting iMessage registration if the phone number or Apple ID isn't available for use.
Open Safari and navigate to www.apple.com to verify data connectivity. Learn more about troubleshootingcellular data or a Wi-Fi connection if a data connection isn't available.
iMessage over cellular data might not be available while on a call. Only 3G and faster GSM networks support simultaneous data and voice calls. Learn more about which network your phone supports. Go to Settings > Wi-Fi and turn Wi-Fi on to use iMessage while on a call if your network doesn't support simultaneous data and voice calls.
Restart your device.
Tap Settings > General > Reset > Reset Network Settings on your iPhone.
If you are still unable to send or receive an iMessage, follow these steps
Make sure that the contact trying to message you isn't blocked in Settings > Messages > Blocked.
Verify that the contact you are attempting to send a message to is registered with iMessage.
If the issue occurs with a specific contact or contacts, back up orforward important messages and delete your current messaging threads with the contact. Create a new message to the contact and try again.
If the issue occurs with a specific contact or contacts, delete and re-create the contact in the Contacts application. Create a new message to the newly created contact and try again.
Back up and restore your device as new.
iOS: Troubleshooting Messages
http://support.apple.com/kb/TS2755
Thank you for using Apple Support Communities.
Best,
Sheila M.

Oracle 9i Automatic PGA Memory Management

Hello,
my team and me, we are facing difficulties to change the size of the PGA used by our server processes for HASH JOIN, SORT... operators,
here you can see the results of "select * from v$pgastat":
[pgastat dynamic view results|http://pastebin.com/m210314dc]
We have been increasing consecutively our pga_aggregate_target parameter from 1.7 Gb initially to 4Gb then at the end 6Gb, the value of "Global memory bound" and " aggregate pga auto target" on the link above are still equal to 0.
I have been reading threads on the forum and documentation see below, I understand how the global memory manager (CKPT) computest the sql memory target and then the global memory bound, as far as I understand I can only "play" on the pga_aggregate_target value in order to increase the size of our PGAs (I exclude to play with hidden parameters).
- Joze Senegacnik: Advanced Management of working areas in Oracle 9i/10g : http://tonguc.yilmaz.googlepages.com/JozeSenegacnik-PGAMemoryManagementvO.zip
- Dageville Benoit and Zait Mohamed: SQL memory management in oracle 9i
Here different information that could be usefull:
OS: solaris 10 (db running in a non global zone)
Arch: 64-bit sparcv9 kernel modules
Physical memory: 32 Gb (being shared between all non global zones)
Oracle version: 9.2.0.5 32bits
Values of init parameters and hidden parameters that could be relevant:
[init parameters|http://pastebin.com/m40340cf4]
[hidden parameters|http://pastebin.com/m50d74c53]
Maybe useful queries:
over work areas views, I use the following script:
[wa_analysis.sql|http://pastebin.com/d606ebd9b]
and the result of it:
[result of script wa_analysis.sql|http://pastebin.com/m5f49a2e5]

Joze Senegacnik wrote:
- either your sessions are using a lot of memory for storing variables like pl/sql arrays which is subtracted from automatic management: PGA_AGGREGATE_TARGET - (aggregated persistent area + a part of the run time area of all server processes)
- you are hitting a bug
- or maybe something elseI am really happy you come to this conclusion too, they are the same we made with my team and we have submitting to Oracle support via metalink SR 3-1216060641, we were asking if we hit the following bug (in note 1) or we leak about pl/sql or java... or else indeed,
note 1: PGA_AGGREGATE_TARGET Assigned Memory Is Left Unconsumed When Set High [ID 844542.1]
Joze Senegacnik wrote:
I would like to know:
1.) what were the values for global memory bound and autotarget immediately (or in short time) after the database restart or when you have increased them Just after the restart of the database and just after the change of P_A_T, we query v$pgastat immediately after and the value of global memory bound and auto target were equal to 0 byte,
2.) If you are able to change value of PGA_AGGREGATE_TARGET (P_A_T) to 10GB what happens with global memory bound and auto traget. They should be positive at least for a short time. As this is a dynamic parameter you can change it for a short time, run queries and set it back.We plan to do this tonight, we have an "heavy" ITIL change management procedures that allow us to make changes approved by change manager and only during night maintenance window on production system, I come back to you tomorrow. But we have been increasing from 1,7Gb to 4Gb to 6Gb, each time I have been querying v$sgastat in the next 2 mins and global memory bound and auto target were equal to 0 byte.
3.) Have you checked on the OS level how much memory are using server processes - do these numbers come along with what Oracle says. Not during problematic activities, meaning active work areas performing HASH-JOIN, SORT... operators,
unfortunately it is a production system, even if he performs poorly, we are not allowed to try or retry the poor queries, but if it comes again I'll do it,
during low activities, here the results paste with the scripts I used:
[pga processes info in oracle|http://pastebin.com/f2e540062]
I spooled the result rows of this previous script in /var/tmp/pga_processes.log then I loop over all processes pid and display pmap output anon info like this:
h5. cat /var/tmp/pga_processes.log | awk -F' ' '{print $5}' | xargs -n 1 -i pmap -x {}| grep -v 'Addres' |egrep 'Kb' 2>&1 > /var/tmp/pga_processes_os.log
then I merge line by line the two files with unix paste command, here the results:
[os and oracle pga informations|http://pastebin.com/f4135c8a6]
4.) How many server processes are running on you system in average/max and are you using just dedicated processes or also shared?in average 250, we are only using dedicated processes,
5.) At time of low activity is the global memory bound still 0 or becomes > 0. I have been querying every 15 min during more than 24 hours low activities, it still stay to 0,
5.) Are you experiencing paging/swapping on OS level?No, here orca figures for details:
[free memory|http://img509.imageshack.us/img509/5897/ohuron1asd2gauge1024xfr.png]
swap
[pagein pageout|http://img121.imageshack.us/img121/6946/ohuron1asd2gaugepginper.png]
[memory usage|http://img19.imageshack.us/img19/2213/ohuron1asd2gaugeppkerne.png]
6.) Please post the result of: select * from X$QESMMSGA ;during low activities, [results X$QESMMSGA|http://pastebin.com/f61df7093]
While you will be answering to my questions I'll try to figure out what we can do to properly diagnose the problem. As you are on 9i it is a little bit harder.I am really kind of your help, as we say in my country, "if you need tow arms one day to carry something, call me."
--Jeremy Baumont

About PGA

Similar Messages

Maybe you are looking for