OSI/AS Configuration and Management Manual

Performance Considerations
OSI/AS Configuration and Management Manual—424119-001
6-14
When to Monitor Performance
•
Internal operations and events. The SCF TRACE command provides the ability to
capture information on the internal processing of the application processes. It
records information on the procedure calls that are made, as well as on the internal
state messages. It is recommended only for this use. Although the SCF TRACE
command can seriously reduce the throughput of the entire system, it is the best way
of obtaining this type of information. The PTrace commands format, select, and
display trace information.
There are a number of general warning signs that you should watch for. They include the
following:
•
Excessive response times
•
Error condition reports
•
CPU, disk, and line queuing delays
User complaints are often good indicators of poor response time. However, you should
be aware that response time is a very subjective measurement for most users. Accurate
measurement is necessary to ensure that complaints about poor response time are
justified.
Utilization of hardware devices and communications lines should be kept at values that
do not cause excessive queueing delays, which can cause excessive response times. You
can obtain utilization and queueing delay information using Measure and ViewSys. Line
utilization can be partly obtained using Measure and the SCF STATS command to
determine how much data was sent over the line. A more accurate method is to use a
response time monitor, which provides this type of information as well as the average
response times for the lines. Where line speeds are relatively slow, it may be necessary
to keep the utilization correspondingly low to avoid long queueing delays.
You can also examine the various counters, errors reported, and events generated to help
determine causes of subsystem performance problems.
When to Monitor Performance
Tuning your system to improve performance is an iterative process that begins with
initial configuration. As you monitor the system and uncover problem areas, you adjust
and reconfigure to improve performance.
To obtain appropriate performance data, you need to carefully select the times when
monitoring activities are performed. If possible, measure the system for a full day.
Repeat the measurements for peak days for the applications, if known, such as special
days of the month, special days or months of the year, and so on. Monitor the subsystem
during the busiest hours of the days that are most critical for system operation.
Corrective Actions
Once you have obtained performance data via monitoring, you may need to take
corrective action. The actions required to correct performance problems, of course,
depend on the symptoms and the types of problems encountered. Don’t take the easy
approach by simply adding more memory or other hardware. In some instances, adding
more hardware can intensify the problem, or can create other problems. Instead, you