INTERVAL and CYCLE options

Much of the data in the Paging, Page Data Set, Processor, Trace, Virtual Storage, CPU, Channel, I/O queuing, and Device Activity reports is statistically sampled. As the accuracy of sampled data increases with the number of random samples taken, you would expect to observe more precise results with decreased CYCLE time (for a fixed INTERVAL value), or with increased INTERVAL length (for a fixed CYCLE value). For example, 400 samples taken of random independent events provide a value that, with 90% confidence, should fall within 4% of the true value; 1,600 samples of random independent events decrease to 2% the expected range of error, with 90% confidence.

However, pure statistical predictions are not always applicable to a software measurement tool such as RMF™ because the assumptions on which they are based (unbiased random independent samples and an infinite population) might not hold in an operating environment. Bias might occur because RMF samples internal indications of external system events. Thus, RMF values might not precisely approach the values measured by a hardware measurement tool.

The independence assumption becomes less and less realistic as CYCLE gets very small. As CYCLE gets smaller, each sample is more likely to find the system performing the same functions as in the previous sample; therefore, the new sample adds little additional information. The use of a smaller CYCLE value (while holding INTERVAL constant) should not be detrimental to accuracy, but any increase in accuracy might be of questionable benefit when compared with the system overhead that is introduced. A reasonable minimum CYCLE value is a function of the timing characteristics of the hardware being measured.