Performance analysis: Single-transaction measurement

You can use full-load measurement to evaluate the average loading of the system per transaction. However, this type of measurement cannot provide you with information about the behavior of a single transaction and its possible excessive loading of the system. If, for example, nine different transaction types issue five start I/Os (SIOs) each, but the 10th issues 55 SIOs, this results in an average of 10 SIOs per transaction type. This situation should not cause concern if the transactions start at the same time; however, an increase of the transaction rate of the 10th transaction type might lead to poor performance overall. To investigate this type of problem, you can perform a single-transaction measurement.

Sometimes, response times are good with existing terminals, but adding a few more terminals leads to unacceptable degradation of performance. In this case, the performance problem might be present with the existing terminals, and has been highlighted by the additional load.

To investigate this type of problem, do a full-load measurement and a single-transaction measurement. The single-transaction measurement must be done when no batch region is running, and there must be no activity in CICS® apart from the test screen. Halt the polling of remote terminals.

Measure each existing transaction that is used in a production system or in a final test system. Test each transaction two or three times with different data values, to exclude an especially unfavorable combination of data. Document the sequence of transactions and the values entered for each test as a prerequisite for subsequent analysis or interpretation.

Between the tests of each single transaction, insert a pause of several seconds, to make the trace easier to read. Use a copy of the production database or data set for the test, because a test data set containing 100 records can often result in different behavior when compared with a production data set containing 100,000 records.

The condition of data sets can cause performance degradation, especially when many segments or records have been added to a database or data set. Do not measure directly after a reorganization, because the database or data set is only in this condition for a short time. If the measurement reveals an unusually large number of disk accesses, reorganize the data and perform a further measurement to evaluate the effect of the data reorganization.

Single-transaction measurement with only one terminal might not be an efficient tool for revealing a performance degradation that might occur when, perhaps, 40 or 50 terminals are in use. Practical experience has shown, however, that single-transaction measurement is usually the only means for revealing and rectifying, with justifiable expense, performance degradation under full load.

Ideally, carry out single-transaction measurement during the final test phase of the transactions, for these reasons:
  • Any errors in the behavior of transactions can be revealed and rectified before production starts, without loading the production system.
  • The application is documented during the measurement phase, helping to identify the effects of later changes.

CICS auxiliary trace

Auxiliary trace is a standard feature of CICS, and gives an overview of transaction flows so that you can quickly and effectively analyze them. From this trace, you can find out whether a specified application is running as expected.

If you have many transactions to analyze, you can select, in a first pass, the transactions whose behavior does not comply with what is expected.

If all transactions last much longer than expected, there might be a system-wide error in application programming or in system implementation. The analysis of a few transactions is then sufficient to determine the error.

If, only a few transactions remain, analyze these transactions next, because it is highly probable that these transactions are creating most of the performance problems.



dfht32x.html | Timestamp icon Last updated: Monday, 5 June 2017