> No matter who the public is, they must understand the standard
> before they can know how much they like it, and they must know how
> much they like it before they can express how much they like it. To
> understand it, they could try to write some programs with it, and/or
> read the 380-page document, and/or attend a presentation or
> tutorial, and/or discuss it with others who are familiar with it.
> And that's just for people who already understand MPI-1. For almost
> everyone, reading the document is the only available option.
Regarding "writing the programs with it" -- I seem to recall that
there was an early version of MPICH available during the MPI-1 public
comment period -- people actually could write programs with it. Is
there such software available for at least a subset of MPI-2? I think
people would try some of the things they have been waiting for -- and
provide feedback -- if an implementation were available.
I am not quite so cynical about MPI-2 -- I think there were some
legitimate needs expressed by the user community that are being met by
MPI-2. I think it would be helpful, though, for those users who have
expressed needs (I am thinking in particular of dynamic processes) to
have an implementation to try out.
MPI-2 was clearly necessary. Without dynamic process control, MPI
would always be perceived as being inferior to PVM. ;-)
Best Wishes,
Andrew Lumsdaine