What is 95th percentile billing? Rules, formula and examples

The idea behind it

The simplest way to sell bandwidth is to charge for a fixed commit. But real traffic moves a lot: low during the day, high in the evening, with the occasional spike from a burst of visitors. Billing on the absolute peak is unfair to customers, while billing on the average leaves the provider carrying the cost of the peaks.

95th percentile billing is the compromise: sample bandwidth throughout the billing cycle, discard the highest 5% as occasional spikes, and bill the highest remaining value.

How it is calculated

1. Sample: record port bandwidth (Mbps) every 5 minutes, 288 samples per day;
2. Sort: at the end of the cycle, sort all samples from high to low;
3. Drop the spikes: discard the top 5%;
4. Bill: the highest remaining value is the billable bandwidth for the cycle, the “95th percentile”.

As a formula:

Samples N        = days in cycle × 288
Samples dropped  = floor(N × 5%)
Billable rate    = the (dropped + 1)th value after sorting

For a 30-day month, N = 30 × 288 = 8,640. 432 samples are dropped, and the 433rd value is billed. In other words, the busiest 36 hours of the month (432 × 5 minutes) do not count.

A simplified example

Suppose a server’s samples for the month, from high to low, look like this:

Rank	Bandwidth	Billed?
1st	2,850 Mbps	In the top 5%, dropped
…	…	Dropped
432nd	1,120 Mbps	In the top 5%, dropped
433rd	960 Mbps	Billable rate
434th	958 Mbps	—

At ¥20 per Mbps per month, the bandwidth charge is 960 × 20 = ¥19,200. Even if a burst pushed the server to 2,850 Mbps one day, it does not raise the bill as long as such peaks add up to no more than 36 hours.

Inbound or outbound?

This is where contracts most often disagree. Common approaches are:

• Higher of the two: take the greater of inbound and outbound at each sample, then sort;
• Separately: calculate a 95th percentile for each direction and bill the higher one;
• Outbound only: only count traffic sent by the server.

Providers differ, so spell it out in the contract before billing starts.

Common variants

• Monthly 95th: calculated over the whole month; the most common;
• Daily 95th, monthly average: a 95th percentile for each day, averaged over the month; used by some cloud providers;
• Daily peak, monthly average: the highest value each day, averaged over the month; more sensitive to bursts.

Choosing between 95th percentile, fixed and data transfer billing

Billing model	Based on	Suited to
Fixed bandwidth (port)	Committed rate, regardless of use	Steady traffic that needs guaranteed bandwidth
95th percentile	Peak after dropping the top 5%	Traffic with clear peaks and troughs, such as CDN, video and gaming
Data transfer	Total volume (GB/TB)	Low overall traffic with occasional highs

Putting it into practice

The hard part of 95th percentile billing is not the formula but the data: you need reliable traffic collection on every switch port, a complete set of samples for each cycle, and a link to invoicing. Toplink DCIM collects port traffic and provides 95th percentile statistics. Traffic pools meter outbound, inbound or total traffic and can throttle or shut ports when a quota is exceeded, and with the CloudFinance IDC billing system, 95th percentile charges are invoiced automatically every cycle.