Risk of Internet collapse grows

Sean makes a good point about the importance of the quality of data,
but the question is how can good data be made available for research.
We've covered this ground before, that firms are not going to make
data available. There are better data sets that have become recently
available from a few sources, and hopefully improved analysis will

That said a few things should be kept in mind with academic work. The
time from when work is done until it appears in publication is quite
legthy, especially when peer reviewed (the Grubesic et al article was
peer reviewed). I saw his paper presented in the Fall of 2001, which
means he probably did the research in the spring of 2001, and the
latest data available was Boardwatch 2000. so, you end with a lag in
Internet time that seems horrendous. One of the problems with
academia. I do think it is important to think about the best
contributions from academia providing tools (algorithm's etc.) to
analyze data and view issues from a different perspective. We will
never have the quality of data the operations community has.

That said I think it is vital to get good feedback from the operations
communtiy on our assumptiopns and something this forum has been great
with helping with from my experience. I was curious if we put some
recent research online if folks would be interested in providing

Happy holidays,


The paper would have the same problems in 2000. It starts with bad
assumptions. Age doesn't improve bad assumptions.

Suppose I wrote an academic paper about the design of the Federal
Reserve Banking System. After carefull analysis of the map at
http://www.federalreserve.gov/otherfrb.htm (street addresses available
at http://www.federalreserve.gov/fraddress.htm) I write a fully footnoted
paper that the Federal Reserve system is vulnerable to the destruction
of the board in Washington DC and twelve banks in Boston, New York,
Philadelphia, Cleveland, Richmond, Atlanta, Chicago, St. Louis,
Minneapolis, Kansas City, Dallas and San Francisco. The US banking
system would collapse, ATMs would stop, paychecks wouldn't get cashed,
checks couldn't be cleared, etc. I would miss Alan Greenspan, but that's
not how the US banking system works.

The Federal Reserve system does have vulnerabilities. So does the
Internet (and the post office, and the telephone network, and ...)
But misunderstanding the risks and vulnerabilities is worse because
it diverts resources away from the real ones.

Another interesting footnote, as long as we're on the subject of
footnotes... You remember when Bob Metcalfe came and preached doom and
gloom at the D.C. NANOG? Specifically, he was predicting a "gigalapse,"
his neologism for a billion person-hours offline, before the end of 1996,
and he was basically saying that no more of this engineers-without-ties
foolishness would be permitted after a debacle of _that_ scale occurred.
Of course, nothing happened, and we all got T-shirts to commemorate the
non-event. _However_, the footnote is that I went and dug that up a few
years later and ran the numbers on a few of the bigger subsequent fiber
cuts and (more to the point) ATM switching failures, and he was off by
about two years, but the actual big cuts were _an order of magnitude_
larger than he predicted, and they _still barely even made the mainstream

So catastrophic failure doesn't appear to actually interest folks much,
even when it does happen.


The problem isn't so much the latency - although that is a problem.

Any researcher approaching this problem must understand that their result
are only as good as their data. In this case, assuming that Boardwatch
network maps are correct or, in fact, anything other than a marketing
fantasy, is a big problem.

Use of theses sort of data sources are extremely attractive to researchers
because they are straight-forward and lack the "well, but"'s of the
operational community. Trying to model real-life networks is much more
difficult because of the plethora of designs and exceptions involved -
very messy.

Richer industry/researcher partnerships could help with this. I suspect
the research community will have to become more aggresive in this area to

- Dan

with helping with from my experience. I was curious if we put some
recent research online if folks would be interested in providing

Sure - and think about doing it in person, too, at the February NANOG.
We're trying a new presentation format, where researchers will have
10-minute slots to present their work for operator feedback. More here: