Configuring HP OpenView Performance Agent
This section contains basic information about configuring the OpenView Performance Agent (OVPA) to work with the Specialized Server Models (SSM) and the Server Characterization Editor. The following information does not replace the OVPA documentation, nor is it intended to be a complete administration guide.
OVPA Collected Data
The OVPA collects data into a number of closed-format data files normally stored in /var/opt/perf/datafiles. These files are:
- logappl: application metrics
- logdev: device metrics
- logglob: global metrics
- logindx: index used to access the other log files
- logproc: process metrics
- logtran: transaction metrics
Currently, only the logglob, logproc, and logappl data files are required for analysis, although the OVPA collector collects all the files.
Note: Log files overwrite older data when they exceed the specified size. This is especially important when collecting detailed process data. Make sure the log files are large enough to cover the time periods you require.
Application versus Process Data
Process data is information about individual processes in the system. A process is created each time a user logs in or runs a new program. Large systems typically run hundreds or thousands of processes. These systems can also run many copies of a program such as Oracle; this makes it difficult to determine which processes correspond with which database.
To simplify the first level of analysis, OVPA groups processes into applications and reports on all the processes in each application as if they were one. This reduces the amount of information required to analyze a system, but requires some thought to set up. See the OVPA documentation for information about how to set up application groups (this is done in the "parm configuration" file).
Threshold Settings
By default, OVPA does not collect all the information about processes or applications in a system; instead, it records that information only when activity exceeds a certain threshold. When monitoring a running system, this approach reduces the amount of clutter you need to view to find the one pertinent piece of information. However, when characterizing a running system, this can be a problem because you need a more complete view of the system.
Setting the threshold values to 0 does not solve the problem. To change the threshold to include all values, you must add the following line to the parm file:
application=all
This causes the collector to ignore the threshold values and collect all values for the application data.
Similarly for process data, you can use the parameter
process=all
The danger of collecting all process data is that the data files can get large very quickly.
Finally, the threshold options "nonew" and "nokill" are normally NOT used for working with Server Characterization Editor.
Changing the Application Configuration
When you change the application configuration, be aware that old application identifiers in the log file do not get updated. As a result, you may need to search for multiple application identifiers when analyzing the collected data.
File Retention
Because OVPA retains data in a closed-file format, updates to Server Characterization Editor may require you to run the 'extract' program on the original data files. Because the format of these files might change, using the extracted files (files with the .bin extension) does not give the desired results.
To avoid problems, keep the original log files if you want to analyze the data again later.