gtpg1m0mGeneral Information

Performance Considerations

The performance of the TPF system must be monitored to organize system resources properly for peak operating efficiency. The TPF system provides data collection and data reduction programs to measure system performance. These programs provide operational data on activities such as milliseconds CPU busy per message, DASD accesses per message, memory usage per message, program calls per message, message rate, and message length. With this information and the transaction history, you can determine how efficiently the system is running, where bottlenecks occur, and what changes can improve system performance.

Data collection and data reduction provide:

Data Collection

Data collection can be run in continuous mode or sampling mode, allowing multiple types of data to be captured while avoiding significant interference with message processing. All data collection programs write the captured data to an online tape. No attempt is made to analyze the data online, as this would have a negative impact on the system that is being measured.

The three basic techniques used for collecting data are:

Data Reduction

All data reduction is performed on an MVS system. The data reduction reports are intended for use by an analyst familiar with the TPF system. Frequency distribution reports including means, standard deviations, and variances of many parameters are available.

The aim of the initial analysis phase of a working system is to establish the normal limits for each of the key factors affecting performance. Once these limits are set and agreed to be realistic, a periodic system check becomes routine.

The analysis of performance data must always start with summary reports. These reports provide key data required for history and trend analysis. When investigating a problem area, the more detailed plot reports or the specialized reports of the DASD and message reduction programs are used. The plot reports, which show the value of each parameter sample in chronological order, are very effective for analyzing the cause-and-effect relationship between parameters.