More Sidgemore on per-bit pricing

David P. Maynard dpm at
Sun Dec 6 02:04:10 UTC 1998

> > One possible positive effect (for the consumer) of "per-bit" pricing
> > is the opportunity to buy larger pipes but only pay for what you use.
> The other possible effect is that you buy one of these, and then someone
> launches a DOS attack at you and you get the bill for it.

We have used usage-based pricing for collocated customers since SMARTNAP 
(now IXC Austin NAP) opened it's doors nearly two years ago.  That allows 
us to hand people a 100Base-TX port and let them pay for the bandwidth 
they actually use.  We don't bill per byte.  Instead, we use the 95% 
sustained utilization model that bills for your peak utilization in 
Mbit/sec after throwing out 5% of the 5-minute samples for the month.  The 
5% amounts to about 36 hours of "free" bandwidth.  That means you have to 
be a pretty serious smurf target before attacks will affect your bill 
significantly.  Both we and our customers have been very happy with the 
95% utilization model for this type of service.  It lets us set pricing 
based on a metric that closely follows the amount of backbone bandwidth we 
have to provision to support a customer.  UUNET uses the same metric for 
their usage-based T3 lines.  I believe other providers also use it.

Having said that, the 95% model isn't the end-all of usage billing.  Some 
people want to use high bandwidth for a short period of time (e.g. a video 
conference) and just want to pay for that period of time.  In the 95% 
model they either win big or they lose big depending on whether their 
high-bandwidth usage fits in the 5% window.

There are still some customers concerned about the lack of control they 
have over their bill.  Presumably, they will have a business model that 
provides them with additional revenue if their traffic increases.  It 
still makes people nervous though.  Using bandwidth-limiting mechanisms 
can alleviate that concern at the cost of reducing their available peak 

> The economic impact of this should not be underestimated.  Per-bitrate 
> pricing is a problem as long as the receiver pays for the receipt of 
> transmissions they may not have solicited.

The 95% model makes a fairly primitive, but fairly effective attempt at 
addressing that issue.  The issue will be more of a concern if 
finer-grained usage-based billing becomes more common.


 David P. Maynard, Flametree Corporation
 EMail: dpm at,  Tel: +1 512 670 4090,  Fax: +1 512 251 8308

More information about the NANOG mailing list