Re: HOW BIG IS THE INTERNET?

[David Farber <farber () central cis upenn edu>]

Posted-Date: Fri, 19 Aug 1994 15:34:46 -0400
To: farber () central cis upenn edu (David Farber)
Cc: interesting-people () eff org (interesting-people mailing list)
Subject: Re: HOW BIG IS THE INTERNET?
From: Craig Partridge <craig () aland bbn com>
Date: Fri, 19 Aug 94 12:33:31 -0700
Sender: craig () aland bbn com




I'd like to briefly respond to a bit of Hoffman and Novak's article,
particularly since I'm the source of the original 5 to 10 people per host
estimate.  (It comes from an early attempt to design a reasonable survey of
the Internet population when I was technical director of the NSF Network
Service Center).


I agree with them that John's approach of only accepting pingable hosts is
probably defective.  For instance, my host is only connected to the Internet
during business hours -- so 60% of the time or so, it won't reply to ping,
but it can do FTP, netnews, etc.


However, as for the number of users per host, I'd suspect that John's
3.5 number is better than 5 or 10, though I agree it is hard to get
exact numbers.  What I found when I started designing a survey was that
we had a small number of hosts with truly huge populations (campus mainframes
like cats.ucsc.edu and some large system at umich.edu with 1,000s or 10s of
thousands of accounts) and then a large number of hosts with very few
users (single user workstations, etc).  And at that time, it looked like
the average was around 5 to 10 users (the approximation was developed to
allow us to try to roughly tune the sample size for a survey that was never
done).  Now I believe the general market trends are towards connecting PCs
and single user machines (which except on school campuses, typically are *not*
shared), which would tend to drive the average down.


However as Hoffman and Novak point out, there's no perfect answer to this
question.


Craig