Issue No. #14 31 January 2003 ISSN: 1532-1886

We encourage you to forward this newsletter to others. Please also CC to so that they get a free subscription too.

A 2nd Serving of Percentile Billing
by Steve Cerruti and Carl Wright

Our previous issue about 95th percentile billing stirred up questions and discussions. I've captured them for another helping of Percentile Billing issues.

A Reader Responds (J. François Audet, Director, Telecommunications, Cogeco Cable Inc.)

I read your article on percentile billing with the greatest of interest. One of my tasks these days is to negotiate the purchase of Internet transit services. We also track bills and validate the measurements with our engineering group and control payment through accounting. Most of our bandwidth is purchase on a 95th percentile basis, but in some other cases, we purchase fixed capacity connections (which provides guaranteed revenue to the supplier, but the unit costs are lower, since there is no risk).

Of course, some sales people try to explain 95th percentile billing as meaning that there is a 5% discount built-in. I have to draw graphs to explain to them that if my traffic pattern is flat I may be getting 0% "discount", whereas if my patterns are steep, I may be getting much better than a 5% discount with respect to peak-billing.

Please explain what you mean by "I think that they would all be better off charging for the amount of bandwidth consumed...". Are you advocating :

(a) billing on the basis of total volume transferred (bytes) as opposed to throughput (bits per second),
(b) billing on throughput but on another basis than percentile (peak or average), or
(c) fixed bandwidth (the customer gets a good deal on the pipe, no measurement is made, and it is assumed he is fully using it)

In cases a) and b), measurements still need to be made, but case a) does not address the demand on the network.

Case b) (peak) addresses demand, but I would think 95th percentile a less cruel approach than peak-billing. Average-billing still does not fully address the demand on the network.

Case c) does not measure demand, but still manages to force the user to try to optimize his demand pattern.

Somehow, the user has to pay for demand (as opposed to volume), because that is what drives network capacity and cost. On the other hand, the delta between peak and 95th percentile is not too critical as statistical multiplexing between multiple users using the same pipe provides some flexibility, and if worse comes to worse, a few packets may be lost, but most applications can live with some packet loss.

By the way, I like your comparison to water usage, and may use it in the future.

Anyway, that was my two-cents-worth!


Carl Answers

I said that "we would be better off charging for the amount of bandwidth consumed". My reaction is a concern about the "disconnect" between the cost of bandwidth and the value of bandwidth.

When most people purchase electricity, they pay based on their consumption. We can't see a kilowatt hour, but we've learned to relate it to benefits (heat, light, TV viewing, etc.). I believe that we can reach a similar point with Internet services. We can learn to relate megabytes to benefits (web pages, e-mails, pictures from friends, etc.). I think that users may have already developed this understanding while waiting for downloads with their dial-up connections.

To answer your question specifically, I believe that we should bill on the basis of total volume transferred. I think that the amount charged should vary by the quality of service delivered. Faster pipes, higher price. Everyone's time is worth something. When you save time for me, I expect it to cost something.

Steve Says

The two fundamental problems with 95th percentile billing is its lack of standardization and its nondeterministic behavior.

Since service providers are often computing 95th percentile on different measurements (e.g. input, output or input plus output) it makes it difficult to compare costs between competing providers. In addition some providers are computing 95th percentile incorrectly.

Because measuring the exact same traffic, with the same sample period, but at different sample times can result in significantly different results (albeit requiring unique traffic patterns) the fairness of using this method for billing comes into question.

Wholesale customers are not nearly as affected as retail customers because wholesalers should be able to review the suppliers billing methods and approximate conversions. Also wholesalers benefit from traffic aggregated over multiple customers which should reduce though not eliminate the chance of statistically correct but reality challenged billing.

Retail customers therefore are the main ones who are likely to be negatively affected by mainstream use of this billing. Finally, billing should influence behavior to optimize the use of the network. For example, long distance companies offer off-peak calling and power companies offer time-metered electrical service to encourage use at non-peak times. Because 95th percentile billing is not well understood, it does not have the effect of driving voluntary traffic to off-peak times or smoothing out outgoing traffic to reduce peaks.

We addressed several billing methods at the beginning of the article with applications for each.

  1. Flat rate (per hour/per month) billing is typically used for residential Internet access. It is easy to understand, compare and sell. It results in the same cost regardless of the amount of use (e-mail only vs. file sharing).

  2. Volume metered (per MB) billing is currently being used by wireless carriers. This method of billing is extremely problematic in the retail space because it bills in an intangible unit (MB) versus something that the user can understand (per hour, per e-mail, per web page).

  3. I personally worked on a system that sold Internet access over a private worldwide network. Since this network was private, the cost of data transfer was calculated on total volume between multiple source and destination pairs where source and destination could be regions or cities. It was not based on throughput (i.e. bandwidth) but on total data transfer.

All three of these systems do not charge customers based upon peak demand. It is incorrect to say that the customer *must* pay for peak demand versus total usage. Peak demand billing does not happen regularly in other utility markets (although electricity could certainly benefit from the model). It has always been the wholesaler's job to ensure that the cost of peak demand is included in the unit price, whether that peak demand is measured in terms of trucks, number of long distance lines or diameter of pipe. That customers are paying for it in the Internet space may simply be a sign of immaturity in this sector.

The sector will be forced to evolve to handle the above mentioned billing problems but also to handle settlement issues for traffic which crosses third party networks and also for content delivery charges (pay per view). Additionally as QoS is built in for video and VOIP different tiered charges will have to be included based on those services and no longer on simple peak bandwidth.

Finally, the way 95th percentile billing currently works, it does not inherently relate customer bills to the provider's usage. The peak bandwidth for an entertainment company is likely to fall at a different time of day than the peak bandwidth for a wholesale auto part exchange. Either or both may not match the peak for the network provider and therefore the network provider may be able to carry their traffic essentially for free. Therefore, even 95th percentile billing does not necessarily charge customers directly for the cost to operate the network.

However, it is possible to charge for peak demand using time-metered total usage. The total usage during the wholesaler's peak times would be priced higher than the usage during off peak times. This method requires the network provider to do more up front computation but allows him to more directly/fairly charge his customers for the cost of his network.

I hope that this information clarifies some of the thoughts behind our discussion of 95th percentile billing.


François Speaks Again

Of course, we have equipped ourselves with measurement tools, and we know how much our 95th percentile usage is with each supplier before we even receive the bills.

Admittedly, our retail offer to our residential high-speed Internet customers is not based on 95th percentile. It is based on a flat rate, subject to a cap of 10GB downstream and 5GB upstream per account per month.

I acknowledge that trying to explain 95th percentile to each of very large numbers of retail residential customers would be impossible. We therefore take it upon ourselves to establish the empirical rules that bridge our 95th percentile costs with our MB/month caps while maintaining a healthy business model.


Percentile billing is a statistical measuring process. This makes it hard to audit and hard to predict. The measurements for percentile billing is like watching waves raise and lower the water against the timbers of a dock and recording the measurement every second. It just doesn't connect to user value for me.

It also makes it possible for both purchaser and vendor to manipulate the amount paid for the very same amount of data transmission.

I advocate a change to usage charging for IP services with a premium for improved bandwidth rates. It connects the charging with the value as the customer sees it. You'll be getting another issue about percentile billing soon where we evaluate a strategy for increasing revenues for percentile billers.

Tell Me What You Want To Hear About

The subjects that I cover in Rating Matters are driven by my personal interests in rating and billing. These are limited by the breadth of my personal experience. Please let me know about items you want to hear about or you'd like explored further. Send me your requests at .

ęCopyright 2003 Service Level LLC
Rating Matters is a trademark of Service Level LLC