[Developers] Parallel processing

[dave fournier]

I'll try to take a look at the book.  However depending on what part of
ADMB you are talking about there are some good opportunities for
parallel processing taking advantage of special structure.

this is for the random effects stuff.

for models using the SEPARABLE_FUNCTION  the separable function
calls are almost trivially parallelizable.  If the Hessian is not block
diagonal you just have to coordinate writing results into it.

for general RE models the num der blocks option (-ndb n)
is set up to use parallel processing   the idea is that
if you have n independent variables where n is rather large
(say 100 for example)  then you can split it up into say 10 blocks of
10 independent variables (more or less). at present they are done
sequentially to reduce the amount of temporary file space but it
was intended that eventually it could be done in parallel.