North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: pretty cool paper re "myth of Internet growth numbers"

  • From: Andrew Odlyzko
  • Date: Sun Dec 03 06:52:27 2000

Sean,

Since you have have chosen to ignore kc's "meta-operational content
warning"
	
  On Sat, 02 December 2000, k claffy wrote:
  > [meta-operational content warning. 
  > followup thread probably belongs on some other list]
  > 
  > andrew odlyzko's latest
  >    http://www.cisp.org/imp/november_2000/odlyzko/11_00odlyzko.htm

I feel entitled to respond.  

First of all, your comments all appear to be based on the paper you
cite, namely 
  
  http://www.research.att.com/~amo/doc/high.network.cost.txt

This is an entirely different paper on a different subject.  (Moreover,
just like the article kc cites, it is just a short overview of more
detailed papers.)  It is sort of like kc saying "Sean's new-born son
is cute," and my responding "But his two-year old daughter is ugly."
Either statement may be right or wrong, but there is little connection
between them.

Second, concerning the substance of your criticism,
  
  I think Andrew has some apples and oranges mixed together in some of
  his papers.  For example, he compares the utilitization of telephone
  trunk circuits (33% busy) with corporate LANs (3-4% busy).  But ignores
  the typical corporate PBX, which sits idle most of the night and weekend.
  I suspect if you compared the corporate LAN with the corporate PBX, the
  utilitization rates would be similar.  And in most companies, the availability
  of good measurements for intra-PBX calling and intra-LAN usage would be
  hard to find.  Of course, that would open the whole can of worms of
  why do companies buy their own PBX's instead of using Centrex service from
  the phone company?

I do stand by the analysis of my papers.  I do indeed mix apples and
oranges, but that is inevitable in this context.  In my 1998 paper with
Kerry Coffman, "The size and growth rate of the Internet," we even say,
to justify our efforts, that "While we cannot avoid comparing apples to 
oranges, we try not to compare apples to orange trees."  However, the
issues are definitely complex, and outside the scope of this mailing
list.  I would love to discuss this in greater detail with you, given
your extensive knowledge of networks, as I think I could learn from
such a discussion, but I would propose that we follow kc's warning
and take this to a different forum.  How about doing this through
private email exchange?  We can invite any members of this list who
are interested in this issue to alert either one of us, and we will
either cc them on the messages (if there are just a few), or set up
a general short-term mailing list for the discussion.  Would that be
acceptable?
  
Third, you write:

  But worst of all Andrew uses the *average* utilization, which is a poor
  way to measure utilization of data networks.  Even utilization is a bit
  of a red herring because different absolute bandwidths.  Networks (voice
  or data) are not designed using average utilization.  Instead most designers
  use some type of peak utilization (95%, Mother's Day, etc).  So what
  happens when a network gets near its peak utilization?  With the voice
  network you get more busy signals.  With the data network you get slower
  transfers.


You appear to miss the point of the work.  The aim of the analysis
was to demonstrate that data networks are designed for and used for
different purposes than the voice network, and that the commonly held
notion that packet networks are superior to switched voice networks
because they utilize transmission facilities more efficiently is
just false.  I am certainly not suggesting that average utilizations
should be used in designing networks of any sort.  Again, there is much 
more that can be said on this topic, but should not be said on this list.
  

  He also states the the traditional phone network continues to be more
  reliable than the Internet, which in my experience is an artifact
  of how the statistics are collected.  If I called United Airlines'
  reservation number this year, I would think the telephone network
  reliability was pretty bad too.  Instead the telephone network doesn't
  include end-to-end reachability in its reliability.  But Internet measures
  such as Keynote do include end-to-end reachability.

This is a debatable point which has been dealt with rather extensively
on this list, so I will respond at greater length.  I agree with much
of what you say, but, lacking any good metric, still feel, based on
my personal experience and other anecdotal evidence, that the claim
about Internet's lower reliability is valid.  As one bit of evidence,
let me note that starting on Tuesday 12 days ago, and continuing
through yesterday morning, my home connection (through a cable modem)
was down each day from about 4:15 am to 5:30 am.  The problem was
finally fixed yesterday.  (It's a long story having to do with the
interaction of the VPN and the cable system.)  Nothing like this
has ever happened to me with voice phones, and this is definitely
not the first problem of this nature.  

Having said all this, let me add that I do not think this in any
way shows the superiority of the voice phone network to the Internet,
any more than the higher utilization of transmission links does.
I feel that reliability is often overrated, and that examples of
cell phones as well as PC's running Windows show that people care
more about other things.

Andrew