solved this problem, so we’re not going to attack it either. Not until we have
infinite money. We’ll just leave it, and, if customers have to spend 2 weeks of
hard programming upgrading the server, then that’s the cost. So that was one of
the biggest things.
The second thing we didn’t like about it was that it wasn’t the full Java 2,
Enterprise Edition. At the time, some of these people thought that it would be
cool if they had multiple layers of Java in there. They said it was a little too slow;
they had all these criticisms of the system. That’s true of any first system. It’s
new, so it’s kind of got some ugliness to it. Shortcuts have been taken. So then
you say, “OK, my next system, the second system, is going to fix all these problems
at once. It will be upgradeable instantly, it will be super fast, it will all be
J2EE, and it will be fabulous. And it will be ready real soon.”
But a lot of those problems with the first system aren’t there because the
people who built the first system are dumb, they’re there because it’s kind of a
difficult challenge and there are compromises involved, and that was an
inevitable result of one of the compromises. And the people doing the second
system are just too na?ve; they haven’t worked on this problem before, and
they’re young, maybe, and they just don’t realize that it was hard.
Fred Brooks said that the second system is always late; sometimes years
later than you expect. And in fact, because of the ambition, usually it won’t
solve any of the problems. They’ve got a long list of things, and it will solve
almost none of them in the end. So that’s exactly what happened with these
guys. They told customers, “Don’t use the old system, because we have a new
one that should be shipping in 3 months and it will be better.” It actually took
them more than a year and a half.
Philip Greenspun 339
So they suffered all these sales losses because the thing was late. They killed
demand for the old product by telling them the new product was “just around
the corner.” Then, when the new product finally came out, some critical pages
were literally a thousand times slower than the old system. So where you have
had a 1-processor pizza box server, you would now need a 64-processor, $2 million
server to serve the same user community. It had never been tested; it had
never been released on a running system like photo.net. It was just a bunch of
programmers—sitting in a vacuum and never dealing with a publisher or a
user—programming whatever they thought. They were all young because they

Похожие записи:

• solved this problem,