Measuring bandwidth usage: 95th percentile, CIR and actual usage

How 95th percentile billing works, why the top 36 hours of the month are free, and how it compares with a committed information rate and volume-based billing.

Contents

There are several ways to measure, and bill, bandwidth usage, each with its own strengths and limits. The most common in the industry is the 95th percentile, and it is also the most often misunderstood. This article explains how it works and compares it with the other methods, so you can make informed decisions about your network’s capacity and costs.

The 95th percentile

The 95th percentile measures how much bandwidth you use while allowing short bursts above your usual level without extra charges. It is common in data centres and for IP transit.

It works like this:

  1. Every 5 minutes, the provider records your average throughput over those 300 seconds and stores it.
  2. At the end of the month, the samples are sorted from highest to lowest.
  3. The top 5% of samples are discarded.
  4. The highest remaining sample is your 95th percentile, and that is what you are billed for.

In a 30-day month, the discarded 5% is 36 hours: your busiest 36 hours of the month are not taken into account.

The 95th percentile is not an average. It is the level that your usage stays at or below 95% of the time, and exceeds only 5% of the time.

Other methods

Committed information rate (CIR)

With a committed information rate, the provider guarantees a fixed amount of bandwidth, even when the network is busy and other customers are using it too. The connection is more predictable, but you pay for that guarantee.

CIR compared with the 95th percentile

With the 95th percentile, you pay for the bandwidth you actually use rather than for a fixed amount, and you can burst above your usual level when you need to without paying extra. With a CIR, you pay for a set rate and cannot go above it without upgrading the connection.

The 95th percentile is more flexible, and the provider does not need to enforce limits on your usage, which also saves it infrastructure and management costs.

Actual usage

Actual usage, or volume-based billing, measures the amount of data transferred over a period rather than the rate. The provider counts how much data you used and bills you for that amount.

Why the 95th percentile is usually the better choice

Compared with actual usage, the 95th percentile is more stable and predictable. It is usually combined with a minimum commitment, which makes planning easier. With volume billing, a month with heavy traffic can produce a bill that is hard to budget for.

The 95th percentile tries to balance cost, scalability and predictability for both the provider and the customer. No method is perfect, though. In months with heavy traffic, the 95th percentile can vary more than a CIR, so monthly bills can differ a lot, and in quiet months you may pay for a commitment you do not use.

Multi-homed companies and service providers, with connections to several ISPs, have to keep the traffic on each link below the level they committed to with that ISP to avoid extra charges.

The 95th percentile is one of the most popular billing methods among ISPs today. Some carriers, such as Cogent, use the 98th percentile instead. Only the top 2% of samples are discarded, which is less favourable to the customer.

If you found this useful, you may also like my article on how the internet really works.

Join the discussion

Comments are hosted on comments.vciso.ch, a server I run myself. Nothing is loaded from it until you press the button below. Loading comments connects your browser to that server, and signing in with GitHub sets a cookie there. See the privacy page for details.

Type to search.

↑ ↓ to move, Enter to open, Esc to close