Imminent Death of the Internet, GIF at 11

For several months (years?) Bob Metcalf has been predicting that
the Internet will self-destruct from overload. His argument
appears to follow one of Gordon Bell's maxims: "anyone can predict
the future: all you need is semi-log paper and a ruler." As I
understand it, Metcalf's argument is that network load (messages,
packets) is growing exponentially, while network bandwidth (fiber
capacity, switch performance) is growing linearly. At some point,
these two curves cross -- and demand will exceed capacity.

There are two solutions to this problem: either there will be
a fundamental change in the way messages move on the Internet
(i.e. they  don't all have to pass through Mae East and/or Mae West)
or there will be a fundamental change in the way we use the Internet.

We certainly are seeing changes in the way we use the network. When
I "got on the net" in the late 1970's, I was on two mailing lists
(SF-Lovers and Human Net) and could read *all* of Usenet traffic
in an hour or two. In the late 1980's, Usenet traffic totalled
about 10 MB/Day. Now, I'm on a handful of work-dependent, low
bandwidth mailing lists, one high-bandwidth, high noise mailing
list, almost never read Usenet and wouldn't dare attempt to support
a Usenet newsserver. Today, it takes longer for me to read mail on
an office Internet or 28.8 modem at home, than it did in 1980 on a
2400 baud modem.

To make a long story short, I suspect that we will be much more
selective in what we access on the net; we may hire editors (or
form communities that share "interesting stuff", each person
serving as one member of an informal editorial board). We will
also see organizations (companies or professional societies) funding
network-based publications to communicate matters of common interest.
(There are a number of these already, Risks Digest being possibly
the most important.)

So, in one sense, Metcalf is right; the Internet will self-destruct.
However, in another sense, he is wrong; the information carried
on the Internet will still be distributed, but probably in a
different form.

Martin Minow
minow@apple.com

Martin Minow writes:
> For several months (years?) Bob Metcalf has been predicting that
> the Internet will self-destruct from overload. His argument
> appears to follow one of Gordon Bell's maxims: "anyone can predict
> the future: all you need is semi-log paper and a ruler." As I
> understand it, Metcalf's argument is that network load (messages,
> packets) is growing exponentially, while network bandwidth (fiber
> capacity, switch performance) is growing linearly. At some point,
> these two curves cross -- and demand will exceed capacity.

Except for the following.

1) TCP backs off.
2) Capacity is growing exponentially.

Perry

Remember, though - TCPs initial estimate of the congestion window is 
never less than one packet, large numbers of opening connections can 
still (I think) lead to congestion collapse. It can defnitely get close 
to it.

At one point, when sunsite was getting a few 100k hits a day with only 
one T1 there were times when around 2/3rds of all packets were 
re-transmitted. Jon Crowcroft observed similar problems at some UK links, 
though see Van's article on either ietf or end2end a month or so  back
with the counter argument.

Simon



On Mon, 9 Sep 1996, Perry E. Metzger wrote:

> 
> Martin Minow writes:
> > For several months (years?) Bob Metcalf has been predicting that
> > the Internet will self-destruct from overload. His argument
> > appears to follow one of Gordon Bell's maxims: "anyone can predict
> > the future: all you need is semi-log paper and a ruler." As I
> > understand it, Metcalf's argument is that network load (messages,
> > packets) is growing exponentially, while network bandwidth (fiber
> > capacity, switch performance) is growing linearly. At some point,
> > these two curves cross -- and demand will exceed capacity.
> 
> Except for the following.
> 
> 1) TCP backs off.
> 2) Capacity is growing exponentially.
> 
> Perry
> 
> 

---
Cause maybe  (maybe)		      | In my mind I'm going to Carolina
you're gonna be the one that saves me | - back in Chapel Hill May 16th.
And after all			      | Email address remains unchanged
You're my firewall -    	      | ........First in Usenet.........

Simon Spero writes:
> Remember, though - TCPs initial estimate of the congestion window is 
> never less than one packet, large numbers of opening connections can 
> still (I think) lead to congestion collapse. It can defnitely get close 
> to it.

Sure. TCP, especially without SACK but even with such schemes, more or
less requires an average of no less than one packet per RTT. However,
the other half of what I said is that bandwidth *is* rising.

Perry