6 Replies Latest reply on Sep 14, 2010 7:00 PM by scit

# Purpose of the 95% percentile in the graph

Hi,

I am new to the NPM. I have a question regarding the 95% percentile in the graph. This is the explanation I found from the web.

The calculation of the 95th percentile, a well-known statistical standard used to discard maximum spikes, is based on 5 minute data samples. The calculation gathers these values every 5 minutes for however long you select, throws away the top 5%, yielding the 95th percentile value at the beginning of the list.

I am trying to understand the rational of using the 95% percentile value. What do people normally use the 95% for? Is it to calculate the average? Or is there any other information that the 95% represents?

• ###### Re: Purpose of the 95% percentile in the graph

Here's a WIKI link which should help clarify it for you. http://en.wikipedia.org/wiki/Burstable_billing

I know where you're coming from. My first introduction to this was when I started using Solarwinds. It's a good tool if you are an ISP. But we really don't make heavy use of it in our own enterprise.

• ###### Re: Purpose of the 95% percentile in the graph

It was explained to me one as -

If you wanrted to gauge the height of a class, you could add all the heights and divide by the number of people. It would give a true average, but would be un-representative of the reality, because you included the 6ft teacher with the 3ft children. The 95% percentile would take out the top 5% of exceptional exceptions (including the teacher) and then give you a normal maximum.

You could use this for determining whether a circuit needs upgrading. Circuits will always have periods when they are full, but that should not be normal and if you remove these few percent, you end up with a much more realistic maximum and the report that you need to get a larger circuit.

• ###### Re: Purpose of the 95% percentile in the graph

So can I safely say that, it's a method mostly used by ISPs rather than enterprise?

Cause I'm using it on an enterprise (approx 600 users) and it does not seems to be applicable in that environment.

Note: Don't get me wrong. I'm trying to say that it is not a useful method (95th percentile) but rather trying to understand the operational environment.

• ###### Re: Purpose of the 95% percentile in the graph

Say that you've got 2 sites, and they're joined by a 10meg link.  Each site is almost self contained and working well with limited traffic between, so the 10meg link is fine. You might be seeing some domain replication traffic, but that's about all.

Then someone releases a new application. It might be consolidating the two email systems or installing a SharePoint server at one site. The traffic will change and the 10meg interconnect link will take more load. There will be times when the circuit is full, but this should be the exception. Normally there should be some slack and the responsiveness should be acceptable, but as this pipe fills, it will become less responsive, users will compain about slowness (or more likely complaining that "it dosn't work". Tracking the average will show luinchtimes, evenings and weekends bringing the average into what looks like it should be reasonable, but we all know that this is not the reality. We need to remove some of the maximums and then track the normal maximums, so taking the top 5% off the figures give this 95th percentile. Now when this figure gets close to the line speed, you really do need to look at a bigger line or QoS to make sure that the link provides the service that is required.

It is very useful for ISPs, but it's not just for ISPs. It can be very useful to the ISPs customer too.

• ###### Re: Purpose of the 95% percentile in the graph

Thanks robertcbrowning. It was very helpful.