What is the 95th percentile
What is it?
If you run a network, you’ll be interested in the 95th percentile and what it means for network usage and possible spikes in your network pipe. It is a good number to use for planning network usage.
In general terms, the 95th percentile tells you that 95 percent of the time your network usage will be at or below a particular amount. You can use this figure to calculate network billing for metered usage.
What information do you need to collect?
There are three things you’ll need to know to perform a percentile calculation:
- The desired percentile P. The 95th percentile basically says that 95 percent of the time your usage is at or below it, and the other 5 percent of the time it exceeds it.
- Data points. These are the pieces of data you have collected. In the case of network usage, they would be based on network use for a set period, perhaps a day, a week or a month. The data would be collected regularly, and then collated. The more data points you use, the more certain you can be of your final percentile calculation.
- Data set size N. This is the count of the data point values you have collected over a period of time. Statistically, the greater the size of the data set, the more reliable your calculation will be.
How is it calculated?
Once you have all your data points, it’s fairly easy to calculate the 95th percentile.
Here’s an example that might help to explain it better:
The data points that have been collected for network usage are 34, 26, 33, 50, 22, 35, 37, 28, 39, 27, 11, 24, 29, 32, 31, 36, 23, 25, 30, 38 Mbps.
The total number of entries N = 20.
To calculate the 95th percentile, P = 95. To find the rank R, calculate R = P / 100 x N = 95 / 100 x 20 = 19. If R has a fractional part take the next largest integer.
Now arrange the data points in ascending order. The ordered list is 11, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 50.
The value of the Rth element of the ordered list is the Pth percentile, in this case the 95th percentile is the value of the 19th element which is 39.
This means that you would expect 95 percent of all data measurements to fall at or below 39 Mbps.
Why use it?
The reason the 95th percentile is so useful in measuring network usage is because it provides an accurate picture of how much it costs. By knowing the value of your network’s 95th percentile, it’s easy to identify spikes in usage. If you are billing clients for network usage, it’s common to rely on the 95th percentile as a basis for billing.
For example, if you have a monthly billing period, (and you have used data points collected from a monthly usage cycle) the 95th percentile allows a customer to have a short burst in traffic (36 hours or less in this case) without being charged for over-usage. This is known as burstable billing and there is a good article about it on Wikipedia.