How interpret the result

Feb 19, 2010 at 10:05 AM

Hi to all,

I'm new with this interesting tool, so I apologize if my questions could be result "stupid" :-) at your eyes.

I collected a sample log and I'm wondering how I can interpret percentile fields. Below a sample:

Name Min Avg Max Hourly Trend Std Deviation 90th Percentile 80th Percentile 70th Percentile
\\SPIMIDC016\Memory\Free System Page Table Entries 182,091 183,706 184,115 4 291 183,724 183,710 183,671

How we get 90th percentile, 80th percentile, etc... values ? If I try to calculate them starting from Min or Avg or Max columns values, I got different values.

If I try to use PAL Editort counter threshold (in my sample PTE warning is set to 10000), I'm not able to get Percentile column values.

If I try reading in depth Threshold properties, I'm not able to have more info to understand them.

What Std Deviation means ?

Thanks a lot in advance,

Alex

Coordinator
Mar 5, 2010 at 9:39 PM

The word "percentile" was the incorrect term for me to use. The intention of the percentiles was create new averages with x percent of the furthest outliers removed. I was told that this is not what percentiles called. In PAL v2.0, I renamed percentiles to "outliers removed". For example, "90th Percentile" was renamed to "10% of Outliers Removed" to better describe the purpose of the stats. I need to change the lables of these fields on PAL v1.x to reflect this change.

Std Deviation is Standard Deviation. In a nutshell, it is the average distance from the average. The higher the Std Deviation, the less reliable the average. The lower the Std Deviation, the more reliable the average. If you have data points of 0,10,0,10,0,10, then your average is 5 with a Std Deviation of 5. If you have the data points 5,5,5,5,5,5, then your average is 5 with a Std Deviation of 0.

I use the Percent Outliers Removed (previously the "percentiles" field), because on some occassions, perfmon will have a very large number such as 100,000 when it should have been 10. The number is so large that is skews the average a lot. The Percent Outliers Removed removes the data points that are furtest from the average. So with a 10% Outliers Removed (previously 90th Percentile), if you have the following data points, 5,5,5,5,5,100,5,5,5,5 then the average is 5 and the 100 data point is the furthest from the average, so when 10% of the outliers are removed, then then 100 value is removed from the average.

The overall counters stats are not used in the thresholds because the threshold look at the values in each time slice of the log versus the entire log. The percentiles are not used in the threholds.