You collect Performance Monitor data on a Windows Server 2003 machine. When the data is collected in the comma-separated value (.csv) format, the first data point that is collected may be a larger number than expected. This behavior occurs if the counter being monitored uses delta time, and if the log is set to start and to stop automatically. For example, the delta time is the %Processor Time time.
This issue occurs because the first data point is taken immediately.
When a log is stopped and then started automatically, the time between the last data point that is in the previous log and the first data point that is in the new log is very small. This behavior causes a very small divisor for the values that measure delta time. Therefore, the value of the resultant is larger than expected.
For example, the value of the resultant is as follows:
"date time","25.0000959987712157" <--- 25%, even though the CPU was idle "date time","0.31377598366740811" "date time","0.31377598366740811"
To work around this issue, do one of the following:
Collect data by using the binary (.blg) file format.
Ignore all data values that seem too large to be accurate.