[mrtg] Re: 95th Percentile

Henry Steinhauer H1STEINH at hewitt.com
Tue Nov 23 19:44:01 MET 1999

It depends on what your want.

I am not an ISP vendor so I do not know the exact why they really compute

BUT - I have heard.

The value under the 95th Percentile for the last 30 days (month) with 15
minute sample points.
This then would be 96 interval sample points per day with 2880 points for
the 30 days.
Discarding the top 5% would drop the top 144 intervals or 36 hours of

In theory this should be the 'worst' times so you could take out 15 minutes
here and 15 minutes there.  the 1.2 hours does not have to come from the
same day.

Also vendors will take the highest of either the inbound or outbound
traffic and use that as the 'value' for the interval.


What I am considering is to use the 5 minute interval from MRTG and save
this off at a per day level.
Thus I have 12 values per hour with 288 per day.  For the month  (30 days)
this would be 8640 intervals.

5% of the intervals per day would be 14.4 intervals (let us call it 14

I would have 30 points for the month.  This would also give me better
information on the level of service that my users are seeing.

For the 'costing' interval I still have too many intervals being counted.
(I threw out 420 intervals 35 hours of points -  5% of 8640 is 432
SO - IF I would toss out the highest 95th level for the month, then I would
be getting close to the value I want to use for planning.

?  Is this the same as saving off the entire month's information?  NO of
course not.

Could I contrive a case where tossing out 36 hours of high use over the
month is not the same as 95% for each day and then tossing out the highest
day? Sure -  But would it show the typical business use though, I do not
think so.

This approach would let me use the xlsummary code and save the result for
each log file and then create an Excel GIF for the result.

I'll let you know later how this works out.  I have a conference in
December with a lot of down time.  I should have it by the new year if no
one else puts it together by then.

Henry Steinhauer - h1steinh at Hewitt.com

* To unsubscribe from the mrtg mailing list, send a message with the
  subject: unsubscribe to mrtg-request at list.ee.ethz.ch
* The mailing list archive is at http://www.ee.ethz.ch/~slist/mrtg

More information about the mrtg mailing list