jf noonan jfn1 at
Wed Aug 16 22:25:59 PDT 2000

On Wed, 16 Aug 2000, Doug Henwood wrote:

> I'm wondering what you all think. Is software quality
> improving vastly? And is the actual performance of computers
> - their output, whatever that is - really increasing at the
> kind of speed suggested by the price indexes (or, even more
> extravagantly, by Moore's law)? Or is a lot of the speed
> and power increase just taken up by bells and whistles?

I'm a lapsed physical scientist and former programmer that works doing sysadmin stuff for a narrow market scientific instrument (a ,heh heh, *biotech*) company. Our field (crystallography -- which by the way, is a field that has more than a few socialist luminaries including Linus Pauling, J.D. Bernal, Dorothy Hodgkin, and some more) is one that really had to wait for kick ass computing power to be able to crunch data that instruments were capable of collecting 50 years ago. We can always use all the CPU cycles the silicon can throw at us and we will frequently use software with really crude interfaces as long as we can crunch lots of numbers. Fast.

But hardware is always ahead of software. Advances in hardware drives interest in doing new things in software because they suddenly become possible. Crudely put, this is the tendency of programs to "grow to fill available memory", which is itself a corollary of the notion that "work expands to fill available time".

Software cannot keep up. You only get close to using all the new hardware capabilities in a new platform if the software is something that is compute bound to begin with. Something like a non-linear least squares with a couple of 100k observations and a few thousand variables. Or a huge finely gridded Fourier map.

But user interfaces, which are very difficult to design (design *well* that is: crappy ones are easy and are everywhere) and are not necessarily compute bound -- and don't scale up quickly. Unlike, say, a modeling computation that just chugs away until finished, a UI doesn't get better just because you execute it faster.

Better UI's are needed to get more people using software to do any kind of task (cetus paribus, of course). Those better UI's take more effort, collaboration, and time than faster sillycon does to produce.

In short, I believe in Moore's law for melted sand, but not for the overall 'computing experience', because software is so much harder than melting sand.



Oh yeah, I never speak for my employer, even when I send e-mail from their computers. I'm paid to be a nerd, they pay other people to have opinions. Your mileage may vary. Close cover before striking. Unleaded fuel only. This product may cause birth defects. Consult a professional before reading this message. Rinse thoroughly after use.

More information about the lbo-talk mailing list