Performance analysis: Full-load measurement

A full-load measurement highlights latent problems in the system. It is important that you take the measurement when, from production experience, the peak load is reached.

Many installations have a peak load for about one hour in the morning and again in the afternoon. CICS® statistics and various performance tools can provide valuable information for full-load measurement. In addition to the overall results of these tools, it might be useful to have the CICS auxiliary trace or RMF active for about 1 minute.

CICS auxiliary trace

CICS auxiliary trace can be used to find situations that occur under full load. For example, all ENQUEUE operations that cannot immediately be honored in application programs result in a suspension of the issuing task. If this situation happens frequently, attempts to control the system by using the main transaction are not effective.

Trace is a heavy overhead. Use trace selectivity options to minimize this overhead.

RMF

It is advisable to do the RMF measurement without any batch activity.

For full-load measurement, the system activity report and the DASD activity report are important.

The most important values for full-load measurement are as follows:
  • Processor usage
  • Channel and disk usage
  • Disk unit usage
  • Overlapping of processor with channel and disk activity
  • Paging
  • Count of start I/O operations and average start I/O time
  • Response times
  • Transaction rates.

Expect stagnant throughput and sharply climbing response times as the processor load approaches 100%.

It is difficult to forecast the system paging rate that can be achieved without serious detriment to performance, because too many factors interact. Observe the reported paging rates; note that short-duration severe paging leads to a rapid increase in response times.

In addition to taking note of the count of start I/O operations and their average length, find out whether the system is waiting on one device only. With disks, for example, it can happen that several frequently accessed data sets are on one disk and the accesses interfere with each other. In each case, investigate whether a system wait on a particular unit could not be minimized by reorganizing the data sets.

The RMF DASD activity report includes the following information:
  • A summary of all disk information
  • Per disk, a breakdown by system number and region
  • Per disk, the distribution of the seek arm movements
  • Per disk, the distribution of accesses with and without arm movement.

Use the IOQ(DASD) option in RMF monitor 1 to show DASD control unit contention.

After checking the relationship of accesses with and without arm movement, for example, you might want to move to separate disks those data sets that are periodically frequently accessed.