This Document contains 5 topics.
1. Generic Tips
2. Transaction Manager (TM).
3. Parallel Concurrent Processing (PCP) Environment.
4. Tuning Output Post Processor (OPP).
5. Concurrent Processing Server Tuning
1) Sleep Seconds - is the number of seconds your Concurrent manager waits between checking the list of pending concurrent requests (concurrent requests waiting to be started). A manager only sleeps if there are no runnable jobs in the queue.
Tip: During peak time, when the number of requests submitted is expected to be high, Set the sleep time to a reasonable wait time(e.g. 30 seconds) dependent on the average run time and to prevent backlog. Otherwise set the sleep time to a high number (e.g. 2 minutes) . This avoids constant polls to check for new requests.
2) Increase the cache size (number of requests cached) to at least twice the number of target processes.
For example, if a manager's work shift has 1 target process and a cache value of 3, it will read three requests, and try to run those three requests before reading any new requests.
Tip: Enter a value of 1 when defining a manager that runs long, time-consuming jobs, and a value of 3 or 4 for managers that run small, quick jobs.
This is only guidance and a balance needs to struck in tuning the cache, so with fast jobs you need to cache to get enough work for a few minutes. With slow jobs, a small queue helps should you need to reprioritize requests.
3) Create specialized concurrent managers to dedicate certain process either short or long running programs to avoid queue length.
4) To maximize throughput consider reducing the sleep time of the Conflict Resolution Manager (CRM). The default value is 60 seconds. You can consider setting to 5 or 10 seconds.
5) Avoid enabling an excessive number of standard or specialized managers. It can degrade the performance due polling on queue tables (FND_CONCURRENT_REQUESTS...). You need to create specialized managers only if there is a real need.
6) Set the system profile option "Concurrent: Force Local Output File Mode" to "Yes" if required . You need to apply patch 7530490 for R12 (or) 7834670 for 11i to get this profile.
Refer Note.822368.1: Purge Concurrent Request FNDCPPUR Does Not Delete Files From File System or Slow performance
Note:- The profile option "Concurrent: Force Local Output File Mode" is set to "No" by default. After applying the patch, set the profile option to YES will cause FNDCPPUR to always access files on the local file system, hence FNDCPPUR will remove the OS files faster.To enable this feature, All Concurrent Manager nodes must be able to access the output file location via the local filesystem
7) Truncate the reports.log file in log directory. Refer Note.844976.1 for more details
Truncation of file "reports.log" is a regular maintenance work of Application DBA. Make sure that reports log file size should not increase to its maximum limit of 2 GB. There is no purge program to truncate file "reports.log". This maintenance needs to be done manually and regularly depending on number of concurrent program which uses "reports.log". You can safely truncate "reports.log".
The "reports.log" file can be located under $APPLCSF/$APPLLOG.
8) Ensure "Purge Concurrent Request and/or Manager Data, FNDCPPUR," is run at regular intervals with "Entity" parameter as "ALL". A high number of records in FND_CONCURRENT tables can degrade the performance.
Additionally, the following are very good methods to follow for optimizing the process:
• Run the job in hours with low workload. Doing this after hours will lessen the contention on the tables from running against your daily processing.
• To get the requests under control, run the FNDCPPUR program with Age=20 or Age=18 would be a good method. That means, all requests older than 18 or 20 days will be purged.
• Once the requests are under control, run the FNDCPPUR program with Age=7 to maintain an efficient process. This would solely depend on the level of processing that is performed at your site
9) Ensure that the log/out files are removed from the locations shown below as you run "Purge Concurrent Request and/or Manager Data program".
In the event that it does not remove the log/out files, over a period of time it will slow down the performance. Please refer to the following note which suggests the patch which fixes it.
Note.822368.1: Purge Concurrent Request FNDCPPUR Does Not Delete Files From File System or Slow performance
10) Defragment the tables periodically to reclaim unused space / improve performance
HOW TO DEFRAGMENT
10.1) alter table
10.2) Note that, some indexes might become unusable after table is moved, check the index status from dba_indexes for the table moved and rebuild them too as explained in next bullet.
select owner, index_name, status from dba_indexes
where table_owner = upper('&OWNER') and
table_name = upper('&SEGMENT_NAME');
10.3) alter index
Note: Ensure the tablespace in which the object currently exists has got sufficient space before you move/defragment . Always take backup of the tables before moving the data. It is recommended to perform ths action on Test instance initially then testing thoroughly before performing it on Production instance.
10.4) You will need to collect the statistics for the tables.
exec fnd_stats.gather_table_stats ('APPLSYS','FND_CONCURRENT_REQUESTS',PERCENT=>99);
Transaction Manager (TM)
11 ) Profile Concurrent:Wait for Available TM - Total time to wait for a TM before switchover to next available TM. Consider setting this to 1 (second).
12) Ensure enough TMs exist to service the incoming request load.
13) When the load is high, set the following profile to optimum values to achieve better results.
PO: Approval Timeout Value - Total time for workflow call (When initiated from Forms) to time out.
14) Set the sleep time on the Transaction Manager to a high number (e.g. 10 minutes), this avoids constant polls to check for shutdown requests.
Parallel Concurrent Processing (PCP) Environment
15) If the failover of managers is taking too long refer to Note:551895.1: Failover Of Concurrent Manager Processes Takes More than 30 Minutes
16) Refer NOTE:1389261.1 when you are in the process of implementing PCP.
17) Set profile option 'Concurrent: PCP Instance Check' to 'OFF' if instance-sensitive failover is not required. Setting it to 'ON' means that concurrent managers will fail over to a secondary application tier node if the database instance to which it is connected goes down.
18)Transaction Manager uses DBMS_PIPE to communicate with application session prior to 11i.ATG_PF.H RUP3. DBMS_PIPE in turn uses OS Pipe.We might use Advance Queue(AQ) with 11i.ATG_PF.H RUP3 by setting System Profile Concurrent: TM Transport Type to QUEUE.
Note Pipes are more efficient but require a Transaction Manager to be running on each DB Instance (RAC). So you might want to use "Queue" for easy maintenance.
19) Add these parameters depends on your Database version
+ _immediate_commit_propagation=TRUE (11g RAC)
+ max_commit_propagation_delay=0 (9i RAC)
20) To speed up the PCP Failover ,Tune the below parameters.
• Kernel parameters (Find the analogous parameter for your platform)
tcp_keepalive_time ( Do not set this value to low; since it will then use up your network resources with unnecessary traffic)
• DCD (Dead connection detection) setup; sqlnet.ora from the Database Tier
• Environment Variable at Concurrent Manager Tier.
• PMON Cycle & Sleep Intervals for ICM (internal Concurrent Manager) setup.
Navigation OAM -> SiteMap -> Monitoring -> Internal Concurrent Manager Link(Under Availability) -> "View Status" -> "Edit ICM Runtime Parameters"
• Enable Reviver.
What is FNDREVIVER and How Is It Set? (Document : 466752.1)
Tuning Output Post Processor (OPP)
In order to tune the OPP to improve performance refer the below Note.
NOTE:1399454.1 Tuning Output Post Processor (OPP) to Improve Performance
Concurrent Processing Server Tuning
1. Any Concurrent Processing (CP) server tuning or load balancing needs are to be addressed by Oracle Consulting. There are way too many site specific factors that needs to be considered for optimum CP throughput: from machine hardware, to user request volume, to required Work Shifts, to programs run time characteristics (long / short running)--not to mention also testing and benchmarking. Such a tasks, is beyond the scope of ATG Support.
ATG support would be glad to investigate a failing manager or program issue; however, CP performance issues due to increased concurrent request volume or due to a new installation needs to be addressed by Oracle Consulting.
2. The "Tuning Concurrent Processing" chapter of the white paper "A Holistic Approach To Performance Tuning Oracle Applications Systems Release 11 and 11i" Note 69565.1 may provide some basic insight. Also reference the "Defining Concurrent Managers" and the "Setting Up and Starting Concurrent Managers" chapters of the "Oracle Applications System Administrator's Guide - Configuration".
3. As per Note 69565.1 "A Holistic Approach to Performance Tuning Oracle Applications Systems", "50% of concurrent processing performance tuning is in the business!"
4. Visit the Concurrent Processing Product Information Center (PIC) Note 1304305.1 for additional performance and setup documentation.