off-topic: historical query concerning the Internet bubble

Mailman List Mirror (Read Only)
1

Apologies for intruding with this question, but I can't think
of any group that might have more concrete information relevant
to my current research.

Enclosed below is an announcement of a paper on technology bubbles.
It is based largely on the Internet bubble of a decade ago, and
concentrates on the "Internet traffic doubling every 100 days" tale.
As the paper shows, this myth was perceived in very different ways
by different people, and this by itself helps undermine the foundations
of much of modern economics and economic policy making.

To get a better understanding of the dynamics of that bubble, to assist
in the preparation of a book about that incident, I am soliciting information from anyone who was active in telecom during that period. I would particularly like to know what you and your colleagues estimated Internet traffic growth to be, and what your reaction was to the O'Dell/Sidgmore/WorldCom/UUNet myth. If you were involved in the industry,
and never heard of it, that would be extremely useful to know, too.

Ideally, I would like concrete information, backed up by dates, and possibly
even emails, and a permission to quote this information. However, I will
settle for more informal comments, and promise confidentiality to anyone
who requests it.

Andrew Odlyzko
odlyzko@umn.edu

        http://www.dtc.umn.edu/~odlyzko/doc/mania03.pdf

           Bubbles, gullibility, and other challenges for economics,
              psychology, sociology, and information sciences

                             Andrew Odlyzko

                         School of Mathematics
                     and Digital Technology Center
                        University of Minnesota

                             odlyzko@umn.edu
                     http://www.dtc.umn.edu/~odlyzko

                   Preliminary version, August 5, 2010

                              ABSTRACT

    Gullibility is the principal cause of bubbles. Investors and the general public get snared by a "beautiful illusion" and throw caution to the wind. Attempts to identify and control bubbles are complicated by the fact that the authorities who might naturally be expected to take action have often (especially in recent years) been among the most gullible, and were cheerleaders for the exuberant behavior. Hence what is needed is an objective measure of gullibility.

    This paper argues that it should be possible to develop such a measure. Examples demonstrate, contrary to the efficient market dogma, that in some manias, even top-level business and technology leaders do fall prey to collective hallucinations and become irrational in objective terms. During the Internet bubble, for example, large classes of them first became unable to comprehend compound interest, and then lost even the ability to do simple arithmetic, to the point of not being able to distinguish 2 from 10. This phenomenon, together with advances in analysis of social networks and related areas, points to possible ways to develop objective and quantitative tools for measuring gullibility and other aspects of human behavior implicated in bubbles. It cannot be expected to infallibly detect all destructive bubbles, and may trigger false alarms, but it ought to alert observers to periods where collective investment behavior is becoming irrational.

    The proposed gullibility index might help in developing realistic economic models. It should also assist in illuminating and guiding decision making.

2

Ask on the Internet History list.

   <http://www.postel.org/internet-history/>

Although, as someone active in 2000, I can tell you that traffic did not grow 12.55 times per year (doubling every 100 days), or anything even close to that.

3

We saw that or better growth at Flying Crocodile (aka sextracker.com)
during that period. I don't have access to the stats anymore (if they
even exist) but in two years we went from 1Mb/s to over 1Gb/s in
outbound traffic. This was 1998 to 2000ish.

It was "fun" to try to keep enough pipe and cards in the GSR12000s
even being in the Westin.

Joe Hamelin, W7COM, Tulalip, WA, 360-474-7474

4

The doubling rate from various parts of the tier 1 world I've seen
since mid 90s until now has been pretty consistent. It's been
ranging around 9-14 months or so with the shorter end of the
doubling number coming mostly during the 1996-2000 years, modulo
specific fortunes of the tier 1 in question.

Was Mike O'Dell's famous doubling every 100 days just a myth?
Like any good tale, there most likely was an element of truth
behind it.

It wouldn't surprise me if there was a 6-12 month span during 96-98
when the Internet traffic as a whole did grow by ~10x especially
as backbones made the painful and much delayed leap past DS3 and the
back pressure was finally relieved. The problem is, the relevant
data are spread out all over and probably not obtainable.

I seem to remember thinking that those numbers seemed a bit high,
but mostly shrugging at it at the time I heard Mike and other UUNet
folks say it since it wasn't off by more than an order of magnitude
and back then we tended to ignore things that were that close, and
UUNet was well known for their "forward looking statements" anyway.

Btw, just so we can at least put some real world scale to these traffic
doubling rates tales, a non-descript tier 1 network that's not
particularly any more or less successful than others have had an average
doubling rate of roughly 13.1 months from 2000 to 2010 for their
transpacific traffic.

-dorian

5

As you may recall (because you have been part of it) the 1995-2000 was
a period of major consolidation in the ISP industry, metrics were hard
to obtain and the ones available were hard to believe.

Due to the consolidation of many small networks from various ISPs (I
remember that in my former life we engulfed several of the surviving
post NSFNet regionals), almost all the big pipes of those ISPs were up
to the choking point, then when the time came to move all those pipes
to a better backbone and exchange points, and in the process make them
bigger, traffic started to increase dramatically, accompanied as well
by a decrease in packet loss and delays.

The consolidation also helped to move mom & pop configuration of
stacked us robitics modems to much modern, efficient and reliable
aggregation technologies and architectures on the access side, add
ISDN, DSL, cable, etc.

Not sure how to include it as a variable, but at least in the US the
Telecom Act of 1996 also changed the playing field, CLECs and other
new telecom companies were born.

On top of that, in that period there was also a big increase in
international capacity, some connections (particularly south and
central america) were switching from satellite to fiber, many
developing countries (some of them were just coming out of highly
monopolized and regulated telecom services) were able to have access
to better international connections, plus all the traffic growth
driven from the application side, plus the contribution of dramatic
growth in shared and dedicated hosting services.

I don't recall or have at hand at this time the exact figure, but I'd
agree with you that at some time it looked like a ~10x thing whit some
spurts of much higher growth.

Cheers
Jorge

6

BTW, in the context of Andrew's paper (very interesting indeed) I
believe that one of the issues is that many executives took that
"spurt growth" I was referring to in my previous message, as organic
growth and used it to make unrealistic projections which in turn led
to unrealistic valuations, and so on.

In some cases we "wasted" tons of money over building capacity,
particularly during the datacenter frenzy, later to find that we had a
lot of empty hi-tech real state.

I like how you characterized the investment waste in your paper.

Regards
Jorge

7

Jorge,

Many thanks for the comments.

To the entire NANOG list: I have received many comments, a few
through the list, most off-list. I greatly appreciate all, and
will be responding to them all off-list, since this is not an
operational matter. If there is interest, I can summarize
for the list in a few days. In the meantime, please keep sending
in more data!

Best regards,
Andrew

8

In article <5AC3E79A-392B-4D1B-BFC7-2700942FDC3C@ianai.net>, Patrick W. Gilmore <patrick@ianai.net> writes

Although, as someone active in 2000, I can tell you that traffic did
not grow 12.55 times per year (doubling every 100 days), or anything
even close to that.

Keeping it in the family a little, Mike was quoted as saying this - see p2: https://www.linx.net/files/hotlinx/hotlinx-3.pdf

Although there were two factors here as far as LINX itself was concerned - growth in members as well as growth in traffic from each individual member.

9

Even ignore the fact this is overestimating growth, it still is not 12.55 times in one year. Or even double in 100 days.

10

I have a concern that your posting and your paper mix UUNet traffic with the
Internet traffic. I personally was very much involved in the ISP world (was
working for Tier1 ISPs) during the period and I’d like to point out the
following:
UUNet’s (or any other individual network’s) traffic does NOT equal to the
Internet traffic, even at that time!
I was working at ANSnet and later UUnet due to a three party acquisition deal
between AOL, WorldCom and CompuServe during that time period. I did hear
presentations about network traffic being doubling every 100 days by O’Dell but
my understanding was that he was referring to UUnet’s traffic not the Internet
traffic.

At the time, the Tier 1 ISPs included UUNet, MCI Network, Sprint Network,
ANSnet, etc. Each ISP could only collect network traffic stats on its own
backbone and there was no one entity could collect the entire Internet traffic.
For this reason, the prediction by O’Dell could only be based on UUNet’s traffic
stats. I really doubt that O’Dell would say the Internet traffic doubling
every 100 days rather than saying that of UUNet’s traffic. I’d encourage you
to do some research to find out if he was really referring to the Internet
traffic or just UUNet traffic. The reference listed by your paper showed that
he was saying ‘network traffic’ not ‘Internet traffic.’

I do not know if making such distinction would alter the conclusion of your
paper. But, to me, there is a difference between one to predict the growth of
one particular network based on the stats collected than one to predict the
growth of the entire Internet with no solid data.
Thanks!--Jessica

11

To entire list:

I have received several requests to post a summary of the comments that
are flowing in, so I will do so in a couple of days, say by Tuesday of
next week, to allow for vacations, ... In the meantime, I encourage
further contributions.

To Jessica:

If my posting or my paper mix UUNet traffic with that of the entire
Internet, I apologize for not being clear enough. O'Dell and Sidgmore
always, as far as I know (although I only have a transcript of the O'Dell
presentation at Stanford in May 2000,
http://www.dtc.umn.edu/~odlyzko/isources/odell-transcript.txt) spoke
just about UUNet capacity. However, the myth was that the traffic on
the Internet as a whole was "doubling every 100 days." The puzzle of
how people could confuse traffic and capacity is considered in some
detail in Section 5 of the paper. From the O'Dell presentation it is
possible to guess that the intended implication was that UUNet was
growing much faster than the rest of the Internet, but in any case
I have not seen any references where either O'Dell or

In the presentations that you heard by O'Dell (was that internal to
WorldCom, and if so when), are you sure he was talking of traffic, and
not capacity? It makes a difference in evaluting his claims.

Which of my references has O'Dell saying ‘network traffic’? I cannot
find it. The only similar quote I can find is in Section 5, where
Joe Cook of WorldCom is talking of ‘network traffic,' but there it
is clear it is just WorldCom traffic he has in mind.

Andrew

12

They were private peering stats (aggregated) from UUNET which
show 3.5 x for roughly two years, right before the MPLS conference
in GMU. The stats included 15% or whatever% ATM cell tax. ANS
was not included;-)

There was also another dimension. In addition to the vertical growth.
The backbone also expanded horizontally (coverage/millage), in a very
rapid pace. I guess this was the reason some said the growth rate was
double digits.

I'm not sure simplify the math can explain the complexity of the network
growth. Of cause, this is another topic.

Min

13

Andrew,

don't know if you want to research a little bit about this topic and
add something to your paper but IMHO there is another Internet related
bubble that keeps growing and may end exploding someday.

The governance, speculation and commercialization of the Domain Name System.

If you are interested I can give you some pointers off-list.

Regards
Jorge

14

In article <E3F9273B-A257-4869-BE8A-42AB37AB43B7@ianai.net>, Patrick W. Gilmore <patrick@ianai.net> writes

Keeping it in the family a little, Mike was quoted as saying this -
see p2: https://www.linx.net/files/hotlinx/hotlinx-3.pdf

Although there were two factors here as far as LINX itself was
concerned - growth in members as well as growth in traffic from
each individual member.

Even ignore the fact this is overestimating growth, it still is
not 12.55 times in one year. Or even double in 100 days.

I wasn't suggesting LINX traffic was doubling every 100 days (it was tripling annually), simply pointing out that in 2001 Mike said that "a good rule of thumb during the late 1990's was that traffic doubled every 100 days", and going into print with that shows it was an accepted meme at the time.

15

I wasn't suggesting LINX traffic was doubling every 100 days (it was
tripling annually), simply pointing out that in 2001 Mike said that "a
good rule of thumb during the late 1990's was that traffic doubled
every 100 days", and going into print with that shows it was an
accepted meme at the time.

actually, my altzheimer's device says that it is a meme today that mo
said that then. my memory is that he said doubling every nine months.

randy

16

Randy Bush wrote:

my memory is that he said doubling every nine months.

Mine too.

jc

17

my memory is that he said doubling every nine months.

Mine too.

mo's too. i asked.

randy

18

Fascinating. Memories may be plastic (something that has been
established scientifically), or else we may have yet another
inconsistency to add to the pile of others. Is there any
documentation about the "doubling every nine months"? I have
never seen that particular claim emanating from anyone involved
with WorldCom/UUNet.

On the other hand, existing record shows (among others):

1. U.S. Department of Commerce white paper from April 1998,

   http://govinfo.library.unt.edu/ecommerce/EDEreprt.pdf

on p. 8 declares that "UUNET, one of the largest Internet
backbone providers, estimates that Internet traffic doubles
every 100 days," with a reference to an Inktomi white paper
that attributes this claim to Mike O'Dell. The Inktomi
report is no longer on the Web, but I can provide a copy
to anyone interested.

2. The transcript of the May 2000 presentation by O'Dell
at a Stanford conference clearly has him saying that the
capacity of the UUNet network, as measured by OC12-miles,
doubles every four months,

   http://www.dtc.umn.edu/~odlyzko/isources/odell-transcript.txt

As is explained in my paper, in the 1998-2000 time frame,
essentially all the WorldCom/UUNet claims then seemed to be
about capacity, not traffic.

3. The year-end 2000 email from O'Dell to Dave Farber's IP list,

  http://www.interesting-people.org/archives/interesting-people/200011/msg00058.html

has him talking of traffic doubling each year, while capacity
grows 8-fold.

If some time in that period there was a claim of a "doubling
every nine months," too, that would be very interesting.

Andrew

19

One thing I have heard repeated about one large carrier is that only 10% of their network is used for their Internet product. The remainder of it is used for private leased lines (could also be someone running IP) or another carrier that is leasing a loop or tdm service from them.

Jared Mauch

20

In article <m2bp9d2t16.wl%randy@psg.com>, Randy Bush <randy@psg.com> writes

my memory is that he said doubling every nine months.

Mine too.

mo's too. i asked.

This isn't just my recollection of what Mike said in 2001, the news item I quoted was printed in 2001. So even if it's a mistaken memory (of the late 90's), it was a mistaken memory captured in 2001.

I mentioned it simply as a fairly contemporary reference to the meme.

Having said that, doubling every 9 months was approximately the growth that LINX was seeing at the time.