Computers by the Millions, An Apple Document from 1979
In late 1979, while getting the Macintosh project started, I wrote an article
about the implications of manufacturing and using millions of computers, looking
ahead at what an order of magnitude or two in sales would do to the industry.
At Apple Chairman Mike Markkula's request I did not submit it for publication,
and it was made a confidential internal report. It was released in 1982 as "Computers
by the Millions," in SIGPC Newsletter, Vol. 5, No. 2.
-----------------------------------------
COMPUTERS BY THE MILLIONS by Jef Raskin ©1979 INTRODUCTION If
"personal computers" are to be truly personal, it will have to be as likely
as not that a family, picked at random, will own one. To supply even our own
nation with enough computers to make this happen (over, say a four-year period)
we will require, for example, twenty-five companies each producing over a million
computers a year. To keep this essay short, no attempt is made to give a comprehensive
treatment, but to just give the reader an idea of the magnitude and character
of a few of the problems. FINANCIAL CONSIDERATIONS Let's say that you
were producing a million computers per year. This is about a hundred thousand
(more or less) per month. Let's also say that it cost you $1500 to build each
one, of which $1000 is parts and $500 labor (a moderately expensive business
system). These amounts are not accurate as to either parts/labor ratio or to
absolute costs, but they are not so far off as to substantially prejudice the
argument. To be prudent, you have enough parts on hand for thirty days production.
The section below on the availability of parts makes up only part of the argument
as to why having some parts on hand is a good idea. (It is also a good idea
to have some of the finished product in stock as well.) Given our assumptions
we have to lay out one hundred million dollars a month on parts. Having a hundred
million dollars of inventory laying around for a month is an expensive business.
At 12% simple interest, you could make a million dollars a month just by putting
that money in the bank, and not bothering with manufacturing at all. You
also have fifty million dollars a month in labor costs to think of. On the
other hand, if you are selling these computers for about $6000 each, you have
more coming in than is going out. It is by no means all profit; for example,
most of it probably goes for advertising. The numbers given here are not representative
of the industry, and many items (such as distribution, engineering, software
and much else) have been ignored completely. But we are certainly in the right
orders of magnitude. It should be clear by now that you have to be rich to
produce a million computers a year. The largest personal computer company
in the world right now isn't even close to these figures. Yet. SOFTWARE
PROBLEMS Software, at first, doesn't seem to be nearly as much of a burden
as the hardware. Certainly it does not seem to entail the constant financial
drain as does hardware manufacturing. You might think that software might be
written once, and then may be supplied identically for each computer. This is,
of course, not quite true, as will be seen in the next section. But, for the
moment, assume that software can be written once, and then merely duplicated
for each customer. Part of the catch is in the phrase "merely duplicated".
Just how mere a duplication is depends on what is being duplicated. If the software
is supplied on tapes or disks (or any magnetic medium) we find ourselves in
a morass of time-consuming steps. Magnetic media must be duplicated serially,
and written (at best at accelerated speeds) individually. Take a diskette as
an example: it might take one minute to load, copy, verify, and unload a diskette.
In one month, to produce 100,000 diskettes, you will have to spend 100,000
minutes in duplication. A working month, given two 8-hour shifts, has but
19,200 minutes, say 20,000 in round numbers. You will need five duplicators
running at full speed, with no down time or other problems. To be realistic,
you'd better have custom-made mass disk duplicating systems. Now have a catalog
of 100 programs, and you will have to have a factory with nearly 1,000 duplicating
stations, say 100 machines each of which can duplicate 10 diskettes at a time.
Now you need 100 operators, their management, janitors, and so on... It is likely
that companies will come into existence to fulfill the function of mass storage
media duplication in this kind of scale. Now, as an exercise in large numbers,
the reader should go back and calculate the square footage required to build
and warehouse the 100,000 computers per month. Other problems arise with
programs supplied on ROM, where the lead time becomes a headache (to say nothing
of having someone produce all those chips, or getting into the semiconductor
business yourself.) ROM is also impractical for large programs. On the opposite
tack, video disks are much faster to duplicate on a per bit basis than almost
any other medium, but are impractical for small and medium size programs since
you have to create the whole disk at once (given most current technologies).
One technology for distributing programs that turns out to be very practical
for extremely large runs (even intermixed with small runs) is the printing press.
Bar codes, magnetic printing and other formats that can be run off on a printing
press and read by a computer offer great opportunities in a large-scale operation.
Fast readers for such codes are not presently in development. Since documentation,
point of sale information and advertising material must be distributed with
the programs, the printing press is at an even greater advantage: it can do
the whole job. Software can also be supplied via a communication channel.
This possibility requires a few paragraphs of its own. It is discussed below.
It is clear that software duplication will require much planning, and perhaps
some new technologies. NEW KINDS OF SOFTWARE The full power of the computer
is not available to an individual who owns one until he or she can program it.
This opinion is rapidly becoming a heresy. The trend is to more and more packages
that do specific tasks. This trend is not to be deplored, as software packages
fulfill a useful role. Another trend is toward fill-in-the-form or pick-an-item-on-the-menu
customizing of programs. This trend, too, is to be encouraged. Nonetheless,
unless extended far beyond what is now being done (say, to the point where the
menu consists of all possible program statements) it does not give the user
the full power of a computer. This is not the place to discuss techniques
for easing the average user into programming (and it certainly will not be done
with BASIC, Pascal or Fortran), but it can and must be done. If not, the computer
will become a mere appliance--at best performing a small number of possibly
related tasks. What is desired is for the computer to become an appliance,
but not a mere appliance. Its presence must be taken for granted by its user,
but in the long run, the act of programming itself must be taken for granted
as well. In the short run it will be, if successful, an information appliance.
SOFTWARE UPDATES Software is seldom perfect and usually has errors (called
"bugs"). The manuals that describe computers and their software often have errors,
of which some will be caught. In the past, it was the custom for a computer
or software manufacturer to supply "updates" to their customers to correct such
errors. When you have a million customers, this becomes impractical (if not
positively ridiculous). Personal computers, unlike their larger brethren,
are not sold to fixed sites, but to mobile individuals who are free to sell
(and give) these machines to others without informing the factory. Of a million
people, a few hundred thousand move each year, and it is very expensive to find
a few hundred thousand missing people. It costs (at present mail rates,
and allowing a pittance for the material to be sent, and unrealistically underestimating
labor costs) well over $300,000 to merely send a letter to a million people.
To keep track of which person has which version of the software (updates imply
versions) would require a large computer in itself. It will be very easy for
a programmer (or almost anybody else) to make an error that costs the company
a million dollars, even without anybody generating a lawsuit. All the error
must do is force the company to make an update to a piece of software that went
out with each machine for the last few months. It is clear that the concept
of supplying updates to software or manuals will have to go. In its place should
come much better tested software and manuals, and the concept that what you
buy is what you get. Later versions will be sold as new, separate programs;
compatibility (especially of data prepared under previous software offerings)
will become a major concern. It can be expected that people will stick with
an old, familiar system rather than buy a new one--unless it contains some new
features that are in themselves worth the entire purchase price. I am concerned
that what might happen is that the same old quality of software will go out
and then it will not be supported or updated due to the magnitude of the problems
in reaching so large a customer base. The industry has the golden opportunity
here to turn people off to computers in unprecedented numbers. LOWERING
THE COST OF SOFTWARE DESIGN Actually, software costs will be in the same order
of magnitude as hardware costs, both for the manufacturer and the user. Software
management, in the face of large quantities, is a less well known art than
the management of large-scale manufacturing. There are a few secrets that have
been written up many times, but are rarely heeded: use a single higher level
language for all program development, use clean and unified design, use a management
technique like the chief programmer concept and so on. But for computers by
the millions, some new ideas may come into play. The mass sales of personal
computers started out with the S-100 bus machines. The concept was that you
could buy all kinds of devices and plug them in. What was not said was that
you then had the rather terrible task of writing software to support these new
"boards". Even the more sophisticated operating systems still required detailed
understanding of the add-ons. Even the second generation personal computers
(Apple II, TRS-80, Pet, etc.) allow plug-ins and add-ons. This creates a software
nightmare. The third generation personal computers will be self-contained,
complete, and essentially un-expandable. As we shall see, this strategy not
only makes it possible to write complete software, but makes the hardware much
cheaper and produceable. The kinds of options that do not give programmers nightmares
are things like case color, kind of screen (so long as size, aspect ratio and
resolution are unaffected), power supply and the model name. Programmers
will be very happy when the computer for which they are writing software is
a constant environment. This is not to say that good programming cannot handle
the problems inherent in a varying environment, and on a machine which allows
many different peripherals. What is being said is that it is more difficult
to program for such machines, and that such programs take up much more space.
We need every help we can get in simplifying the programming problem. ANSWERING
CUSTOMER QUESTIONS Most responsible computer companies have people who can
be called when a customer has a question. If 30% per year of a million customers
call, then you have to handle 300,000 calls per year. Given that there are
less than 300 working days per year, that means over 1000 calls per day. It
is inhuman to have a person answer even 25 calls per day, so you have forty
full-time people just answering questions. This seems to be a small number,
compared with the numbers we have been considering, but in practice these people
need management and support, there are questions coming in the mail, and the
calls don't arrive evenly spaced in time. This last means that you need extra
people to handle peak loads. Or, you can alienate many customers. In addition,
the people answering the phones must be very highly trained, both technically
as well as in telephone decorum (you get a lot of angry calls). It is hard to
find such people, slow and expensive to train them, and they usually don't last
long before having to rotate to some other job. I challenge you to try to find
a cadre of, say, 100 such workers. I wouldn't look forward to such a recruiting
effort. Yet, it will have to be done. Or, you can train the dealers to handle
the problem. Later, we will see how many dealers are necessary. HARDWARE
PROBLEMS The present crop of personal computers are nowhere near having what
it takes to sell in the quantities we are discussing. The simplest problem
is in manufacturability. The computers we have now were not designed for true
large-scale mass production. They are full of connectors, separate printed circuit
boards, nuts and bolts, and require too much hand labor. One part, say a bolt,
that takes seven seconds to attach, will cost the company (besides paying for
a million bolts) a full time assembler's salary each year. With various overheads
(such as space, bookkeeping and more) the elimination of one bolt can save the
company $40,000 per year. That bolt wasn't nearly so interesting when you
are making only 10,000 or so computers a year, and even at current rates in
the 100,000 per year category it may cost only $4000 per year. But at a million
computers per year, each little piece can be very expensive. If re-tooling
a portion of the case so that the bolt is unnecessary costs $30,000, the re-tooling
can be amortized very quickly. Another consideration is that hardware can
no longer be the usual conglomeration of little boxes held together with cables.
For user convenience, there should be nearly zero set-up time. Besides, manufacturing
costs are far less for one box than for many. Cables are also expensive and
failure prone. We have seen, above, that software may well demand strong restrictions
on the variability of hardware. This only reinforces the same decision made
on the basis of cost reductions in hardware per se. SERVICE Assume that
only 1% per year of the million computers fail in such a way that they cannot
be fixed at a dealer. That's 10,000 computers per year or some 500 per working
day. (Many present-day personal computer companies don't ship 500 computers
per day, much less are prepared to handle that many coming in and going out
of a service bay.) One technician, with elaborate diagnostic equipment, might
be able to fix (on the average) a dozen computers a day. With one thing and
another, you will require a minimum of 50 technicians. Auxiliary clerical and
shipping personnel will also be required. This also implies that there might
be some good reasons to make the computer itself more reliable--and again to
minimize model changes so that automatic diagnostic equipment can be used fully.
When you allow a variety of attachments to be placed inside the computer (e.g.
on the bus), especially when they are from other manufacturers, it is hard at
times to pinpoint problems. Since the user will not typically send in foreign
devices with the computer to be repaired, the problem may not be detectable
at all at the service center. Another suggestion that the problems of service
bring to mind is that the computer should be highly modular (conceptually--e.g.
all timing circuits in one place on the board), and designed with strong advice
from the service organization. Consider also the number of warranty cards
that you might have coming in each day--that requires a department in itself.
Of course, if each computer had a built in modem and could thus "talk" over
the telephone lines, and if each computer were given an electronic serial number,
then you could have the purchaser merely call a toll-free number to register
their new computer. You would have to set up a considerable automatic facility
to handle the some 1000 to 2000 calls per day that will come in. SALES AND
DEALERS Assume that an salesperson can sell an average of 2 computers per
day, considering both quantity and individual sales. Make it 400 computers
per year (at, say, $3,000 each that means sales of $1,200,000 per salesperson).
Even with such prodigious selling, and assuming 5 sales persons in the average
store (so that each store sells 2,000 computers annually) we need merely 500
stores. Since that many stores at present manage to sell only about one tenth
as much as we have estimated here, we probably need at least 5000 outlets. Only
one manufacturer has that kind of structure at present. It doesn't manage 2
computers per store per day, much less per salesperson. The truly personal
computers will probably have to be sold from a set of chains and other stores
with a maximum of about 10,000 total locations nationwide. A good percentage
of sales will be through direct mail outlets. Many "personal" computer manufacturers
at this time have barely produced 10,000 computers. Only the largest companies
have produced ten times this many. This is mentioned just to keep things in
today's perspective: right now there are few brands of computers with which
the distribution network can even be provided with one computer in each store
of our projected number. ADVERTISING Advertising will continue on its
present path, moving from the hobbyist and specialist magazines to the large-circulation
general readership periodicals, to newspapers, radio and television. Eventually
the personal computer companies will be sponsoring prime-time programs and
special events on television, prominently backing major sports figures and other
personalities, and generally making as much of a nuisance of themselves as do
beer, automobile, camera, soap and tire companies. MANUFACTURING Since
some idea of the volumes required has been discussed (and left to the imagination
and pocket calculator of the reader), there is not too much to be said on this
score, except that many of the techniques used by the large quantity manufacturers
of television sets, washing machines and automobiles will have to be adopted,
including considerable automation. Unionization may occur, and if automation
is not strongly under way before this happens, it will become difficult to
accomplish. Some manufacturing may have to be done on an international basis.
AVAILABILITY OF PARTS At the present time, most electronic components
are scarce, with long delivery times. Many components are on allocation from
their manufacturers. Plastics and metals, due to current trends in oil pricing
and metals speculation are rising rapidly in price, with some shortages already
apparent. There is no assurance to be had that parts and materials for building
millions of computers can be obtained. In response to these pressures, manufacturers
will use conservative design which allows second sourcing. When new technology
is imperative, some vertical integration may be necessary--although this often
just puts the problem on raw material supply (e.g. high-purity silicon) instead
of finished goods (e.g. ICs). If the computer we are discussing uses any chips
in quantity (e.g. it might have 16 memory chips) then the company will have
to buy 16,000,000 of those chips per year. Not many suppliers are prepared to
manufacture, let alone sell to one customer, that kind of quantity. Incidentally,
the computer company will have to receive 80,000 of those chips every working
day. COMMUNICATIONS NETWORK IMPACT We can pretty safely assume that future
personal computers will have communications facilities built in. Since the only
bi-directional network currently available is the phone system, we can assume
that this system will be the first used extensively for inter-personal-computer
communications. We can expect the telephone company to attempt to create tariffs
in response to these usages since, statistically, computer calls are rather
different than ordinary voice communications. Two fundamental measures, average
length of call and time density of information are much greater for computer
communication. There are two kinds of usages that can be easily foreseen.
One is the sending of blocks of previously prepared data. The other is real-time
computer conversations. If we assume that the average transfer of information
will be about 30,000 characters (a long letter) and that data will (for the
time being) be transferred at a rate of 300 characters per second, then a
typical transfer call will be 100 seconds in length. This is not an unusually
long call, but the data transfer is continuous. In normal speech, there are
brief pauses. On major phone routes, calls are "interleaved", with the sound
from one put into the pauses of the other (this is called "time division multiplexing").
With data transfers, there are no pauses. Thus, where you might fit five 100
second voice messages into one 100 second time slot, you could only fit one
computer transfer. A real-time conversation involves two (or more) people
with terminals carrying on an exchange. Such a conversation could easily last
for hours. Or two computers could be co-operating on a problem, with the same
duration of contact. Such usage could, in the face of a million users, tie up
large portions of phone company equipment all out of proportion to the numbers
using the system. There are many technological solutions, but almost all involve
major changes in telephone company equipment and/or changing the standard
ways computers communicate over telephone lines. Something will have to give--either
through restrictive legislation, or the quality of phone service, or through
increased cost of phone service (presumably going, in part, to increasing the
amount of equipment the phone company is using), or through a change in technology.
SOCIAL IMPACT It is impossible to accurately assess what changes any new
technology (or any policy or political decision) will cause. It is clear that
truly massive use of a technology is quite different than the mere introduction
of a new class of devices. No superhighways were created for the first few automobiles.
It is easy to anticipate more of what is now underway: new legislation,
new data services (e.g. the phone book in computer-accessible form), new programming
companies, new computer service firms, various kinds of clubs and organizations...
It is easy to anticipate that many of these computers will end up on shelves
alongside of unused tennis rackets, trumpets and fondue pots. Nobody questions
that small improvements in the quality of life of people who do a lot of writing,
filing and scheduling will occur. But will the average person's circle of
acquaintances grow? Will we be better informed? Will a use of these computers
as an entertainment medium become their primary value? Will they foster self-education?
Is the designer of a personal computer system doing good or evil? The main
question is this: what will millions of people do with them?
|