BC> The compiler is big because no one had the time or wanted to
 BC> budget the time to make it smaller and more efficient. Often,
 BC> the designers get what the rest of us consider "dream systems"
 BC> to develop or at least integrate on and pretty soon the whole
 BC> operation is working in a fantasyland few purchasers can hope
 BC> to achieve. If your tools are Pentium 100, 1 GB disk, 64 MB
 BC> RAM, well, everyone else has that, too, right?

In the past I have had problems getting IBM to reproduce bugs I've found in
OS/2.  The only reason they couldn't reproduce these bugs was because they
were using really slow hardware. Every one of the several test systems on
which I and friends ran one named pipes test case would reproduce the
error, but they were all 486DX/33 or faster.  IBM was testing on 386 and a
really "hot" (sarcasm intended) 486SX/25 computer and they
couldn't reproduce the problem.  I suggested that the problem might be CPU
speed related, so the tester at IBM had to take the test case home to run
it on a faster computer!  This was in early 1994, by the way.  Given this
experience,I doubt the folks at IBM have great hardware, rather they have
probably learned to live with relic computers.  Maybe they take drugs to
slow themselves down so the old 386 computers don't seem so slow? Or maybe
they take up a nervous habit, for example reciting the first 30 digits of
Pi between keystrokes?  I don't know, but personally I really hate waiting
for computers.

Given how cheap Pentium hardware is and what a pig VisualAge C++ is, that
sort of system is a fairly good choice.  I got a huge performance increase
of foreground tasks when compiling with VAC++ in the background by
increasing RAM from 32MB to 64MB.  If I had known how helpful it would be,
I would have bought more RAM sooner.

VAC++ doesn't seem CPU-bound until you get up to over 32MB RAM (maybe even
more) with little else running.  I suspect that IBM's computers have more
RAM than many outside IBM but underpowered CPU's. So maybe that's why VAC++
turns out to be really memory hungry?


--- Maximus/2 2.02