[mrtg] Re: 95th percentile?

[George D. Nincehelser]

I thought about this some more, and have convinced myself that in most cases
the sampling rate and sychronization isn't all that critical given the way
the information is being used for billing purposes.

The numbers you get, and the numbers your providers get may not be exactly
the same, but they will lie pretty close to each other.  Given that the
providers I am looking at have 0.5Mbps billing bands, there's quite a bit of
room to play with.

My guess is that if you are well within a band, there's no problem.  But if
you are hovering around the border between bands, the $$$ involved can be
significant...perhaps $1000 in a month depending on if you are above or
below the line.  I can see where that could cause some arguments about when
and how often samples are taken.

But it must work farily well, since it seems the 95th percentile is an
industry standard...or maybe it's just the best solution available?

George

> This highlights a point to consider when calculating percentiles: you need
> to be using the same sampling period. If the provider is monitoring over a
> ten-minute period then you need to do the same and synchronise the start
> of each monitoring period. The calculation of the 95th percentile is
> straightforward but you need to use the same sample interval and the same
> number of samples to get the same answer. If you use the default 5-minute
> load figures from mrtg and calculate the 95th percentile for the month
> then what you have is the 95th percentile of the 5-minute loadings for the
> month: this may be very different from the 95th percentile of 1-hour
> loadings for the month or the 95th percentile of 1-second loadings for the
> month.
>
> On a similar theme, CIR/EIR calculations are (usually?) done per second so
> the 5-minute intervals shown (by default) in mrtg can only give you an
> indication of what traffic might be in excess of the CIR if your traffic
> profile is reasonably constant second by second over the whole 5-minute
> period.
>
> Martin
>
> --
> Martin Ansdell-Smith
> Network Analyst http://www.ansdell.demon.co.uk/
>
> --
> * 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
>
>

--
* 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