What is network 95 percentile? How is it measured? Why is it useful?
The 95th percentile billing method is one of the most effective and widely used methods for metering and billing bandwidth usage. It aims to offer the best balance between scalability, cost, and volatility. Many services use this method, because it both favors customers and helps service providers reduce their infrastructure and management costs. It's most commonly used by ISPs and various types of services providing bare-metal servers, cloud hosting, and colocation services.
Users or machines have a predetermined amount of committed bandwidth, called the committed information rate (CIR), that they can use according to their contract or Service Layer Agreement (SLA). When the user or the machine exceeds the predetermined amount, the service provider bills the excess bandwidth used. In most cases, the provider does not bill the exact number of megabits per second (Mbps) over the predetermined amount but uses the 95th percentile method or a similar approach.
How does the 95th percentile method work?
The method samples the amount of data transferred over the network at predefined time intervals (usually every five minutes). When the collection process is finished, the top 5% of samples are discarded and only the lower 95% are billed.
Here's a more detailed explanation of this method:
- The system samples data transferred in time intervals and stores the number of Mbps transmitted in those intervals into a database. In order to meter the average rate of an interval, the bits transferred during the time interval are divided by the duration of the interval.
- After the billing period is finished (usually one month) the stored samples are put together on a chart that shows the bandwidth usage. Samples are sorted from lowest to highest.
- Through the billing period, the top 5% of samples are discarded from the chart.
- After discarding the highest value at 95% presents the bandwidth usage that is taken for the given billing period.
The method is designed to favor customers and not penalize them for short data transfer peaks that normally happen when they add new content (such as on a blog), or when they launch a new product on the website. It closely reflects customer's needs, because most customer's bandwidth usage is minimal during off-business hours and increases during business hours, which can cause short peaks of high usage. Also, the bytes that make up the packets themselves do not actually cost money, but the link and the infrastructure around it cost money to set up and maintain. The method also works well for the service providers, because most networks are overprovisioned, meaning they always have additional bandwidth available.