> 2) The geek "unix is the whole world" attitude is so unnecessary.
> "We did Linux, now we can just focus on applications". Right.
> There are professors (that ubiquitous science community we
> hold in such high esteem) in every computer science department
> around the country working on systems and particularly Operating
> Systems. I suppose their research is pointless?
I agree with what you're saying here; I was going to ask what he thought
about advances in OS *theory* that would lead to the ability (and desire)
to engineer improved operating systems. It's not like the problem is
perfectly solved (nor will it be any time soon).
However, I think it ends up being beside the point. As far as I'm
concerned, there's a simpler argument that makes his same point: many of
the things that motivate people -- whether it's money, prestige, or
genuine altruism -- will motivate people toward what's actually useful.
When the need for applications is felt more strongly than that for OSes,
more people will be interested in writing them (for whatever reasons
motivate them).
But I think I agree with you that he seemed to suggest, a bit too
strongly, that all OS problems were solved and that we'd never need
another OS. It's just a matter of degree: do *you* want to write a new
OS any time soon? How about a new compiler?
> 3) The N^2 bit was worked a bit too hard. Handwavy. How is a
> "control" hierarchy N^2 in complexity? It would seem that any
> hierarchy is logarithmic in complexity by nature... the proper model
> is a tree.
What's logarithmic here, though? I think you're falling pray to the old
"everything about a tree is logarithmic" fallacy. In this case, the
*height* of the tree is irrelevant; you're suggesting that a hierarchy
actually involves linear communication costs, I think.
Fair enough, but as Brian points out, companies don't generally work as
perfect hierarchies. I think Raymond's just discussing a general trend:
in actuality, companies generally don't work like this, and open-source
software development does. It's a proof by example that systems better
than the classic Brooks-ian software engineering model actually *works*
and *exists*. Companies could adopt it (internally) too even without
completely going open-source.
Shawn