UNEVEN DEVELOPMENT

Dick Pountain - 20/09/94 13:47 - Talking Head

Just about any media coverage of the computer industry is
sure to open with some reference to the "breakneck pace of
development" or a similar cliche suggesting rapid advance.
And the way that PC on your desk seems to become obsolete
after not much more than a year would appear to confirm
this as a true image.

In fact the "breakneck pace" image is mostly true of
the semiconductor (and certain other hardware) components
of your PC. Yes, it's true that RISC processor designs are
doubling the number of MIPS they deliver every 18 months
or so. Yes, it's true that hard disks shrink in diameter
and double in capacity over roughly the same time scale.
But were I to plonk a raw MIPS R6000 (whose processing
power is equivalent to a Cray YMP on-a-chip) and an 8
Gigabyte Enhanced-IDE hard drive on your desk they
wouldn't actually do you much good unless you're into
Do-It-Yourself in a VERY big way. Hardware design always
races ahead of the system designers' ability to exploit
it.

Busses and system software need to be upgraded to cope
with extra speed, but they must also remain compatible
with the older designs because no manufacturer would dare
tell you that you have to throw away all your software and
expansion cards and start again. Sometimes the delays are
short; Enhanced-IDE drives are compatible enough that they
may well displace plain old IDE and become the norm for
volume PCs early in 1995. Sometimes the delays are
agonizing; Motorola has reached the third generation of
PowerPC chips already, but we've yet to see any of the
promised PC-compatible machines that use it. Sometimes the
delays are fatal; wanna buy a used EISA-bus PC?

It's not so much a case of "breakneck pace" as of "uneven
development" (a familiar enough process to all students of
complex human institutions). And it isn't always the
software that lags the hardware; the uneveness cuts both
ways. Take the recent PDA debacle. It was easy for a
software designer to imagine what a pocket communicator
ought to be capable of (after all you only have to watch
Star Trek). It was easy too for a consummate marketeer
like John Sculley to whip this idea up into into the NEXT
BIG THING; so much so that everyone from Casio through to
Compaq had to announce one. Problem is that the hardware
just ain't quite there yet. Newton works, almost well
enough to be useful, but there are no prizes for almost in
this business. It's not just that adequate handwriting
recognition requires a little more memory and CPU cycles
than you can comfortably support on today's batteries, but
also that a built-in, transparent, wireless wide-area
communications channel was pivotal to the whole PDA
concept but isn't yet available in any commercially viable
form.

Consider the humble CD-ROM. I'm sure that most people
who've never used one expect to see full screen TV-quality
video with CD-quality sound, whereas what they actually
get is what a colleague once memorably described as "home
movies projected onto a postage stamp". CD-ROM right now
is only really fit for storing still pictures and will be
so until we have faster ROM drives, busses and CPUs, and
MPEG decompression on the motherboard.

Or take that other much-hyped medium of the moment, the
Internet. They say it's fun to roam free around the World
Wide Web (WWW) using the Mosaic graphical browser, and so
it would be if you had a 64 Kbit/sec ISDN link; otherwise
watching paint dry is both cheaper and more interesting.

But all these examples of uneven development are mere
popcorn when compared to the industry's real dark secret,
the pace of progress (lack of) in software development. So
far we've been talking delays of the order of a year or
two, which admittedly can seem interminable when you're
waiting for a piece of kit to hit the market. But think
about this: Larry Duffel, director of the Software
Engineering Institute at Carnegie Mellon University, has
calculated that it takes 18 YEARS on average for a
research innovation to find its way into everyday
programming practice. If you doubt that, remember that
Smalltalk matured some time around 1976 but
object-oriented programming is just now entering
mainstream practice.

In September of this year Scientific American published an
article called 'Software's Chronic Crisis'. After reading
it I dug in my files for 'The Emperor's Old Clothes', the
speech given by Professor Tony Hoare on accepting his
Turing Award in 1980. You'd think nothing had changed in
those 14 years. Prof. Hoare warned of huge software
projects failing and being scrapped - Scientific American
describes huge software projects failing and being
scrapped. Maybe the projects that fail now are huger -
they're certainly more expensive ($44.3 million dollars
down the drain on California's scrapped Vehicle Licencing
system, tens of millions of pounds for the London stock
exchange's failed Taurus system or the London Ambulance
Service scheduling system). For every six large software
projects that ever go live another two are scrapped, and
four of the six will never work as intended. The optimists
are those who think that programmer productivity might
have doubled over the last 30 years, the pessimists those
who say it hasn't improved a jot - no one knows for sure
because most software houses still don't bother to measure
productivity.

But that's just those mainframe dinosaurs, right? I mean,
it doesn't affect us in the PC industry... Nice thought,
but not true. We are after all sitting around waiting for
the late arrival of the next version of an operating
system, Windows95, that has taken more than seven years to
get right. Writing correct software grows more difficult
as some frightening power of program size and where DOS
software was too small to really feel the burn, Windows
software is growing right up there into the danger zone.
Of course mass-market PC software vendors have a big
advantage denied to their bespoke mainframe brethren; they
can release the software and let 100,000+ users test it
for them. That wouldn't be allowed in, say, the car
industry but then software production has yet to join the
industrial age. I'll have more to say about this software
crisis in future columns.