November  2004
John Garmany

Performance Tuning your application running in the Application Server 10g can be difficult at best, and at the very least frustrating.  

The issue is that the application server is made up of a lot of components that all interact.  Once you locate a bottle neck and take action to relieve it, the bottle neck moves to another component.  Luckily all the application server components are instrumented to use a tool called the Dynamic Monitoring Server (DMS) to report performance data.  You access this information when you use the Application Server Control web site for the instance.  A lot of the performance statistics reported by Application Server Control is obtained from DMS.  Portal, OC4J, the OC4J Servlet Engine, Oracle HTTP Server are all instrumented to report DMS statistics.

You can also instrument you own Java application running inside the OC4J container to report custom statistics through the DMS API.  Here is a paper on OTN detailing the procedure:

http://www.oracle.com/technology/pub/notes/technote_dms.html

You can access this information using a tool called dmstool.  Dmstool reports the data contained in the DMS Sensors.  A Sensor monitors some type of data.  Some record time, count events, or contain a state.  To get a listing of what statistic are available from dmstool use the -list command.

$ source portal.env
$ dmstool -list |wc -l
   6366

I piped the command to wc to count the number of lines returned.  My dmstool for the portal instance is reporting 6366 different statistics.  The key is to sort through the list of available statistics and determine which one allow you to effectively monitor you system.  These statistics can then be stored in a database to provide time based analysis. 

We are big proponents of monitoring over time to determine a signature for your server.  This is important when problems arise because you can quickly determine which component is outside of it's normal range for that time.

Once you determine what to monitor you can script the data collection.

$ dmstool -i 60 -c 10 \ /host_name/Apache:14502:6004/Apache/request.completed  \
> t1.lst

Here I take 10 samples at 60 seconds of the Apache request sensor for the completed operations.  Here is a partial results.

Wed Oct 27 13:08:58 MDT 2004
Apache:14502:6004/Apache/request.completed 61058 ops
Wed Oct 27 13:09:03 MDT 2004
Apache:14502:6004/Apache/request.completed 61067 ops
Wed Oct 27 13:09:08 MDT 2004
Apache:14502:6004/Apache/request.completed 61083 ops
Wed Oct 27 13:09:13 MDT 2004
Apache:14502:6004/Apache/request.completed 61102 ops

Notice the the number of ops is growing.  If my script was loading this information into a database table I could query the data by time to get a signature for this sensor.  Kind of a STATSPACK for the Application Server.

For a quick analysis I can drop this into Excel  or Calc spreadsheet and compute the difference, and create a graph showing the data over time.

DMS is a powerful tool to help you understand how your application server is responding.  DMS statistics can also be accessed from your Java application through the DMS API. 

Note: If you create a STATSPACK for AS10g, recommend that you place it in the back-end database rather than the metadata repository. 

Guarantee your Success! Oracle is the world's most complex, robust and flexible database, considered impossible to master without a mentor. That's why all BC Oracle trainers are working professionals, experts in Oracle who share their tips and secrets.