What is predicted power and how is it calculated?
Predicted power is a capacity strategy defining how estimated load is calculated. The calculations run in the background and affect capacity values all over the DCO interface, such as impact and failover load.
Past trends in your equipment's measured peak power data from DCE or other external system integration are used to predict the future power demand.
By default, the system bases the calculations on historic data from the last 90 days and predicts 30 days in to the future. You can adjust these settings in Preferences>External Systems>Measurement Trending.
The graphs here serve as examples only. They do not reflect any graphs available in DCO, in reports, or Portal.
The first graph illustrates the predicted power concept and the rest are examples of special scenarios to explain why your calculations behave the way they do.
Predicted power (measured peak power consumption on rack PDU)
How is the prediction affected when adding the first server to an empty rack?
Due to the added load, the trend would become unrealistically steep over a period of 30 days. Therefore, a maximum factor of 1.5 (by default) is used to keep the trend within a more appropriate range.
When adding the new server, you may see some unexpected values while you're editing. This happens because the system falls back on adjusted nameplate calculations until you save your changes or, if you have Data Center Operation: Change, until you execute the pending change.
Predicted power (when adding the first server to a rack)
How is the prediction affected when your measurements are stable?
If the measurements have been stable for the period of time the system has data for this rack, the system does not rely on this data to be realistic until there's measured data for at least 90 days (default).
The trend line would become almost flat. Therefore, a minimum factor of 1.2 (by default) is used as a safety margin to keep the trend line within a more appropriate range. A lower threshold of 1 also prevents the trend line from becoming negative.
However, if the measurements have been stable for an even longer period of time (more than 90 days by default), the data is assumed to be highly reliable and the safety margin is disregarded.
Predicted power (when measurements are stable)
How is the prediction affected when removing a server from a rack?
Due to the reduction in load, the trend would drop unrealistically over a period of 30 days.
When you have removed a server from a rack and saved your changes, the trend line will no longer include values from the last 90 days but only values since the server was removed to exclude values from removed equipment no longer relevant to you.