The Big Picture

[Mark Mitchell]

Some of you have indicated that the big picture was getting a little fuzzy.

Obviously, LANL is our customer, and therefore we must continue to be 
responsive
to requests for bug-fixes from the Blanca project.  However, it is 
important that
we begin to make much more definitive progress on our major strategic goals.

We have two major strategic goals at this point:

  - Finish the new Field abstraction.

  - Make POOMA run faster.

In the long run, the success of POOMA depends on its ability to combine 
rapid application development with high performance.  Once the Field work 
is complete, including the relationship updaters, we will have a very 
compelling story on the rapid application development front.  In order to 
sell POOMA to industrial customers, however, we need to improve the 
performance.

In order to meet these goals, we will be dividing our efforts into several 
teams:

  Abstraction Team
  ----------------
  Scott Haney [leader]
  Jeffrey Oldham
  Steven Smith

  Performance Team
  ----------------
  Jim Crotinger [leader]
  Gabriel Dos Reis

Meanwhile, Allan will continue to work on documentation and Julian will 
continue to work on making Particles work with the new Fields.

When the Abstraction Team finishes up the work on Fields, it will be merged 
into the
Performance Team, as the performance issues will then dominate the 
remainder of our work.
However, because compile-times are also important, part of the work at that 
time will
be to improve the compile-time performance, probably by reducing the use of 
unncessary
template instantiations.

Scott estimates that the remainder of Field work will take one month.  To 
be conservative,
August 1st will be the goal date for the completion of this work.  By that 
time, we want
to have the stenciling, centering, and updating features implemented, and 
we want to
remove the old Field code from the tree entirely.  Abstraction Team, please 
make every
effort to make this date.  The last month was slow; this month, we should 
be able
to pick up the pace.

Jim will coordinate the performance team.  Scott and I believe that the 
best place for
Gaby to start, once he is up and running, is to look at the Doof2D 
benchmark and to
begin trying to reduce the abstraction penalty.  This will be a good way of 
seeing how
the evaluator code paths operate and will simultaneously give us ideas 
about how to
improve our serial performance.

My job will continue to be non-technical, for the most part.  My major job 
will be marketing: I will be evaluating and contacting potential industrial 
partners, and I
will be publicizing the work that you are doing to a wider audience.

-- The pointy-haired boss