Normally, measurements for bottleneck analysis are made at the moment of peak workload. To facilitate analysis, as much measurement data as possible should be collected. The measurement period should be from 30 minutes up to a maximum of one hour.
An analysis interval of 15 minutes should be used as the basic reference value for judging the measurement values (average value for 3 measurement value subintervals of 5 minutes each).
The collection of measurement data by SM2 is partly event-related and partly done by sampling. To ensure the reliability of the samples, it is necessary to collect a certain number of measurement values per analysis subinterval (minimum 1500). A sampling period of 400 milliseconds is recommended.
For each monitoring cycle, SM2 forms the mean value of the collected measurement values. In order to illuminate the range of deviation per analysis subinterval, it is helpful if each analysis subinterval is composed of two or more monitoring cycles.
Recommended measurement parameters
sampling period (SAMPLING-PERIOD): 400 milliseconds
monitoring cycle (OFFLINE-PERIOD): 60 seconds
analysis subinterval: 5 minutes
Monitoring period: 0.5 - 1 hour
With bottleneck analysis measurements, it is preferable to output the observed values to the measurement file rather than display them on the screen. Consequently file analysis with ANALYZER is preferred to the online analysis with openSM2 Manager or INSPECTOR.
Rough analysis and detailed analysis
When using ANALYZER (or SM2R1) for analysis purposes, it is best to proceed step by step:
Rough analysis
Bottleneck search by means of pinpointing (analysis subinterval = 15 minutes)
This locates those analysis subintervals in which the recommended standard values are exceeded.
The automatic performance analysis available in ANALYZER (or SM2R1) provides valuable help in tracking down bottlenecks. This function compares the measurement values collected in the SM2 file with the standard values recommended in the following section, and issues a corresponding message to the user. It is important that only critical periods of productive operation are used for automatic performance analysis.Detailed analysis
Investigate the critical analysis subintervals (15 minutes) by reducing the analysis subinterval (TIME-STEPS) to the size of the monitoring cycle (OFFLINE-PERIOD) in order to highlight the range of deviation in the measurement values.
To facilitate the assessment of the measurement findings, it should be ensured that the measurement values come from periods in which performance expectations were satisfied as well as from periods in which they were not satisfied. Finding the cause of the problem is much easier when both cases are compared.
The following section attempts to simplify the interpretation of measurement values by suggesting standard values for resource workloads.