RFC soon on essay "Does Free Software Production in a Bazaar obey the Law of Diminishing Returns?"

[mark at pc-intouch.com]

On Sun, 15 Aug 1999, Eric S. Raymond wrote:

> 2. Brooks's Law is not precisely *equivalent* to LODR, but is rather a special
>    case of it involving *particular* nonlinear scaling phenomena.  Accordingly,
>    one may assert that the bazaar mode repeals Brooks's Law without making
>    any commitment about the applicability of the LODR in general.

One interesting implication of Brooks's Law is that as N (the number of
developers) gets sufficiently large, the cost of management and 
coordination (which is a function of N^2) ends up exceeding the
productivity of the developers (which is a function of N) and nobody
produces anything.  I suppose the developers spend all their time sitting
in meetings trying to explain Brooks's Law to their boss.

If we look at the real-world experience of bazaar-style projects, this
doesn't happen.  Adding more developers, or even more end users, *always*
benefits the project--if nothing else, they'll find bugs faster.  This
suggests three possibilities:

1.  Bazaar-style projects really are immune to Brooks's Law, because
someone sprinkled pixie dust over them or something.

2.  Bazaar-style development is so scaleable that we've never even
approached the kind of 'critical mass' that it would take for the cost of
management to exceed the productivity of development.  Internet
technologies make the bazaar scale even better.  We'd need millions of
developers to start to notice the management overhead that Brooks
predicted.

3.  It's so easy for a bazaar project to divide up the labor involved in a
project that the largest bazaars tend to break up into smaller
task-oriented teams.  (For example, instead of porting Linux to the
PowerPC chip by having Linus say "OK, all you kernel hackers out there, I
want you all to buy a PowerPC and start porting the kernel", the Linux
developers handed the task off to a team that had PowerPC experience.)  So
a large bazaar never reaches critical mass; it recognizes, in an
adaptive-system way, that it's trying to do too much, and breaks up.  One
disadvantage of this process is that if the pre-breakup stresses in the
group are aligned the wrong way, it may lead to forking.