This article applies as of PRTG 22

What are percentiles in PRTG reports, how does PRTG calculate them, and what impact do they have on the data?

Percentiles in reports

In reports in PRTG, you get statistically condensed information about the data that various sensors collect. In the sensor report settings, you can choose to get percentile calculation, as opposed to raw data or normal percentages for example.

Percentiles can be computed according to various mathematical methods.

How percentiles are calculated in PRTG

1. Sort the measurements: The values of a data sample have to be ordered in a row from the lowest to the highest value. You get the individual values and the number of rows N.
2. Compute the RN of the percentile value using the percentile value (P) and the number of rows (N):
   
   RN = 1 + ((N-1) * P)
   
   
3. Determine the Floor Row Number (FRN) and the Ceiling Row Number (CRN):
       Floor Row Number: The largest integer less than or equal to RN.
       Ceiling Row Number: The smallest integer greater than or equal to RN.
4. Determine the result:
   
   

   If (CRN = FRN = RN)
   then (Value of expression from row at RN)
   else (Value of FRN row) + (RN - FRN) * (Value of CRN row - Value at FRN row)

   Note: You can define if you want to get the then or the else result. If you choose Continuous in the report percentile settings, the result is always interpolated like in the else part. If you choose Discrete, the next smaller discrete value, which is the FRN, is taken as a result, according to the then part.

Example

A 95th-percentile says that in 95% of all cases, the data is below a certain value, while in only 5% of all cases, the data is above a certain value.

1. Let us assume that have the following number of rows and measured values:
   
   
   

   Row Number	1	2	3	4	5	6	7	8
   Value	1	3	7	21	25	26	66	72

   
   
2. As there are eight rows and we want to determine the 95th-percentile, set N = 8 and P = .95:
   
   RN = 1 + ((8-1) * 0.95)
   = 7.65
   
   
3. As RN = 7.65, we can determine the FRN and CRN by looking at our dataset:    
   FRN: The FRN must be the largest row number less than or equal to 7.65, the RN.
   In this case, it is row 7 with value 66.
   CRN: The CRN must be the smallest row number greater than or equal to 7.65, the RN.
   In this case, it is row 8 with value 72.
   
   
4. Since CRN = FRN = RN, or 8 = 7 = 7.65, is not true, we calculate the else value:
   
   Percentile value = (Value of FRN) + ((RN - FRN) * (Value of CRN - Value or FRN))
                 = 66 + ((7.65 - 7) * (72 - 66))
                 = 66 + (.65 * 6)
                 = 66 + 3.9
                 = 69.9
   

As you can see, in a 95th-percentile calculation of our measurements, the value of row 8 (72) would be discarded because it belongs to the 5% of peaks. In our case this is true for the 95th continuous percentile as well as for the 95th discrete percentile.

Advantage and disadvantage of percentiles

The main advantage of using percentiles is that unusually high values (like whiskers in boxplots) are not included in the averaging calculations. This means that statistics include more relevant data. In the example of the 95th-percentile, 5% of the highest measured values are discarded for the statistical report.

If you prefer not to exclude extremes because you want to see the effect they have on your monitored IT infrastructure, do not choose percentiles in the report settings.

Note: For more details on this percentile calculation, please see archive.org.