QUESTION: We gathered an example CPU utilization report where a high data point was shown in one report, but was not shown in a subsequent report that included the same date. We modified the retention policy to keep raw data so data roll up is not a factor. Why wasn't the high data point included in the subsequent report?
When using the ‘Minimum/Maximum’ setting, the chart will compare the average of each sample in the metric history to the overall (or period) average for the metric during the time range. If the sample average is less than the period average, the sample min is graphed. If the sample average is greater than the period average, the sample max is graphed. The idea is to get a sense of the swing of the data depending on how the sample compares to the time period as a whole, but it does not show every outlier. There is also no guarantee that this will result in the same line for different time ranges, or even the same time range when viewed at different times.
There are two factors at play causing the situation being seen:
1). With a different time range comes a different period average. This can cause even the exact same data set to draw slightly differently because the comparison of averages might now yield different results.
2). Even though the user has set a generous retention policy, the chart will tell the datasource the maximum number of points it can comfortably handle and data roll-up will occur if there are too many points in the dataset. Given the time ranges at play here (multi-day) and the size of the charts (on the small side), roll-up is almost certainly occurring.
While the max of the sample with the spike in data will still be the max of the rolled-up sample containing it, the number of smaller values now included in the rolled-up sample will lower its average, possibly to the point where it is now lower than the period average. In that case, the sample max would no longer be drawn but the sample min would be drawn.
In order to always see the maximum values, you should set the Metric Value property to ‘Maximum’. If you would like to see both the minimum and maximum values of all the samples, you might consider using the Envelope properties to draw a min-max envelope around an average line.
WORKAROUND: Following are the basic steps to create a report with HostCPUs Utilization metric with a metric value of "Maximum".
1). Select "Create Dashboard" in the right-hand panel.
2). Select "Use All Data".
3). Expand "VMWare | Virtual Machines | <vmhost> | cpus | hostCPUs | Utilization"
4). Select the Options tab of the "Add Metric View".
5). Set the "Metric Value" to Maximum.
6). Click Finish.