What does 95th %tile mean?

Mon Apr 23 11:09:56 UTC 2001

I used to believe in the desirability and inevitability of
metered rates for the Internet.  However, my studies over
the last few years have convinced me that flat rates are
underappreciated and have a promising future.  This is not
the place to summarize my papers on this subject (all are
available on my home page), but I couldn't resist the
temptation to note that the discussion on this list over
the last few days provides an interesting additional
argument in favor of flat rates.

First, while there have been many postings about 95th %tile
pricing, nobody has said what that pricing actually is.  Well,
here is an example of such pricing, downloaded from the UUNet
page 2 years ago.  (I could not find current prices there on
a quick scan.  Perhaps somebody else can post them, especially
for higher bandwidth connections.)  Ignoring one-time start-up 
fees, regular flat rate T-1 service was available from UUNet at 
$2,495 per month.  The burstable rate (based on the 95th %tile 
rating) was:

                   T1 Usage Level            Burstable Service
                                             Monthly Rate
                   0 to 128 Kbps             $1,295
                   128 Kbps to 256 Kbps      $1,895
                   256 Kbps to 384 Kbps      $2,495
                   384 Kbps to 512 Kbps      $2,750
                   Over 512 Kbps             $3,000

Note that even if the link is not used at all, you pay more than
half the cost of a full T-1 connection.  (Hence some of the
imaginative gaming schemes that have been proposed, involving
rotating Web hosting traffic among a bunch of connections, are
clear losers, as some posters have already noted.  It is less
expensive to have a single unlimited service T-1 than two
burstable one.)  Moreover, if your usage gets above a certain
level, you pay more than for the full T-1.

Yet another factor that comes in is that of discounts.  The
prices listed above were the list prices.  I have been told by
one member of this list that substantial discounts are common
for full connections, but seldom for burstable ones, which makes
the case against flat rate links even weaker.

Second, after these preliminaries, here is the he novel argument 
for flat rates that I draw from the discussions on this list.  
Prices serve two main purposes.  One is simply to collect enough
money to pay for the service that is provided.  How that money
is collected can be based on a variety of criteria, such as
charging what the service is worth to the consumer (as in airline
Saturday night stopover restrictions, designed to get more money
from business travelers than from discretionary vacationers), or
fairness (which has many dimensions, as the discussion on this list
showed, as in whether charges should be based on distance packets 
are carried, on time of day, ...).  The other possible purpose of
prices, and one often touted as their advantage on the Internet,
is that they can be used to influence user behavior in ways that
lead to more efficient utilization of resources (such as charging
for email to reduce spam, or getting users to reduce their traffic
during periods of congestion).  However, for this second purpose
to be realized, users have to understand the pricing scheme, and
have to modify their behavior in response.  But let's look at
the discussion on this list.  Here we have a group of unusually
sophisticated Internet professionals, apparently many paying
these 95th %tile prices.  Yet it appears that many (perhaps even
most) have been too busy to figure out how those prices are
calculated.  Most of the gaming that is going on (and that appears
to be relatively little, by just a few sophisticated players) is 
apparently in ways to get around the rules of this 95th %tile schemes, 
while most people just pay the bills as they arrive and do not worry
about what they should be doing differently.  Yet if the members
of this list behave like that (and I am definitely not suggesting
this is economically irrational behavior, people do weigh costs 
and benefits, and my conclusion is that most people have more
valuable things to do), then can one hope to introduce any
sophisticated pricing scheme to the great bulk of Internet users?

Andrew Odlyzko

************************************************************************
Andrew Odlyzko                                      amo at research.att.com
AT&T Labs - Research                                voice:  973-360-8410 
http://www.research.att.com/~amo                    fax:    973-360-8178
************************************************************************