The routine that computes the domain decomposition is mapping in 
MPP/RSL/parallel_src/lb_alg.c.  It is passed to the RSL library to use in 
place of the default decomposition routine, from within one of the files 
MPP/RSL/mpp_mm5_0*.* (not sure which).  There should be a call in there to 
RSL_FDECOMPOSE(MAPPING) or something.  Those mpp* files get included into 
Run/mm5.F.

Note that MM5 does not do dynamic load balancing (because of the static 
memory).  The old MM90 code did.  As far as a paper, there should be some 
references on www.mcs.anl.gov/Projects/mpmm -- look for MM90.  There's a 
proceedings paper to a structured AMR conference at the U. Minn a couple 
years back that might be especially useful.  MM5 computes the decomposition 
at the beginning of the run and then locks it down.  It is able to give 
more work to faster processors but that's about the extent of it  (if 
ASSUME_HOMOGENOUS_ENVIRONMENT=0 in the configure.user file.

John



On Wednesday, April 28, 1999 1:15 PM, xxxxxxxxxx wrote:
> Hi John,
>
> I was just looking over the MM5 home page, and checking out the code that
>  xxxxxxxxxx
> has downloaded here.  I was interested in looking at how the domain
> decomposition and
> dynamic load balancing are accomplished in MM5, so could you tell me 
where the
> relevant routines are, or point out an explanation of this on the web (or
> paper...)
> I wanted to possibly try my functional load balancing method on it, or at 
least
> on a
> VPP-type decomposition (1-dimensional is easier).
>
> Thanks,
>
>