SMART Letter #1:
Keep it Simple, Stupid
February 5, 1998



            SMART Letter #1 -- February 5, 1998

        For Friends and Enemies of the Stupid Network

            Copyright 1998 by David S. Isenberg

      This document may be redistributed provided that  

      the 11 lines containing this notice accompany it. -- -- 1-888-isen-com

      It takes SMART people to design a Stupid Network




To Buckminster Fuller, being smart meant "doing more with 

less."  For example, back at the dawn of the industrial age, 

engines weighed many hundreds of pounds per horsepower.  

Now, jet engines have many horsepower per pound.  

Knowledge instead of metal. 

Here's another example:  Used to be that cars got 10 or 15 

miles per gallon. Now cars get 45 mpg easily.  You go 

three times further on the same amount of gasoline.  Doing 

more with less.

In a material world, where using stuff has a finite cost, 

smart is good.  But smart usually means complicated.  And 

"complicated" has costs, too.  I'm thinking about the 

computer controlled, multi-sensored, electronically ignited, 

45 mpg system in my driveway.  Gone are the days when I 

could do a tune-up with some sandpaper, a wrench and a 


"Keep It Simple, Stupid" becomes important when the costs 

of complexity are high. And the biggest cost of complexity 

is opportunity cost -- missed (or prevented) opportunities.

When we lose individual control of our machines, we pay.  

I'm not just talking about paying the mechanic/systems 

analyst for auto repair.  I am talking about a world where 

nobody is a shade-tree mechanic anymore, where the 

populace has lost an entire body of knowledge.  And I am 

talking about the inability to innovate -- in fact, the 

illegality of street-level automotive innovation.  The 

improvements that were NOT invented because the automobile 

became too complex.  The lost opportunities.

You can still innovate on today's Internet.  Two guys in a 

garage can still invent the next Netscape, the next Real 

Video, the next Placeware.  I think that this is an accident of 

INTERnetworking, of the Internet's trans-protocol design 

that makes the details of underlying network mechanisms 

irrelevant, so you don't have to go to some organization to 

get permission to innovate.  

Danger!  As Internet Protocol marketplace lock-in 

consolidates, there will be clever engineers who will design 

optimizations for "all-IP" networks. Intelligence will creep 

in.  User-control, that accidental property of 

internetworking, will diminish.  I'm kinda suspicious of 

Internet Quality of Service "improvements" that seem to 

complicate it and move in the direction of centralized 

control.  And I wish that addressing was less centralized 

than the Domain Name System, where the entire Internet 

universe hangs on 12 servers.  Aren't there stupider ways to 

solve these problems?  Will we weigh the innovation-

inhibition costs of future "improvements" and 


Cleverness is compelling.  Smart people don't often get 

rewarded for using obvious, stone-stupid ideas.  See Bob 

Lucky's great essay on this, "When is Dumb Smart?" on . . .


The Stupid Network 

( ) turned out 

to be a surprisingly popular idea.  I continue to be amazed by 

the magnitude of people's response to it.  To date, the 

SMART list has 231 members, and it is still growing by 5 or 

10 a week.  I'm still not used to meeting people who 

have read my paper, and seem to have formed an impression 

-- "Oh, *you're* the author of *The*Stupid*Network*" -- I 

met a surprising number of them at a recent London 

Internet conference.

In the spirit of "Keep It Simple, Stupid," I have tried to 

reduce the Stupid Network idea to its essence. These 4 

principles work for me:

     1.  Keep it simple, stupid

     2.  Underspecification

     3.  Overprovisioning

     4.  User Control

KEEP IT SIMPLE, STUPID, means that wherever possible, 

"features" and "optimizations" should be extrinsic to the 

network.  Such "improvements" belong in intelligent endpoints. 

If we must have a unified, centrally administered "feature," 

let's use a special-purpose device at the edge of the network.  

But the network's main job is to "Just Deliver the Bits,"

and that's enough work all by itself!

UNDERSPECIFICATION means "don't assume."  Adopt 

the broadest possible guidelines, and let 'er rip!  The system 

of roads is underspecified.  It is used by motorcycles, cars, 

trucks, . . . Users determine their own vehicles' contents. 

No controlling authority predetermines the route 

each vehicle will travel. And sometimes there is congestion 

and sometimes there are crashes.  But on the whole, the 

ability of each vehicle to self-configure and self-route is 

massively useful.  The utility of underspecification more 

than makes up for the occasional traffic jam.

OVERPROVISIONING is a cousin of "Keep It Simple, 

Stupid."  It means "don't optimize."  Instead, throw more 

bandwidth, or more compute power, or more memory at 

your supposed problems.  All that stuff is cheap enough today,

and if not, wait a couple of Moore-doublings.  And, as noted above, 

you can still have a very useful network even if it has some 


USER CONTROL means the ability for individuals to 

innovate without institutional mediation.  See above.

I'd be interested in your take on these proposed principles, 

gentle readers.  Send your comments to  

See the discussion unfold on

In summary, Albert Einstein is reputed to have said, "A 

theory should be as simple as possible, but not more so."  

Networks too!  We're still a long way from "too simple"

and in danger of going the wrong way.

David I


David S. Isenberg     


18 South Wickom Drive   888-isen-com (anytime)

Westfield NJ 07090 USA  908-875-0772 (direct line)

                                908-654-0772 (home)


     -- Technology Analysis and Strategy --

        Rethinking the value of networks 

      in an era of abundant infrastructure.


Date last modified: 5 Feb 98