From larson@mcs.anl.gov Thu Oct 26 00:45:16 2000
Date: Thu, 26 Oct 2000 01:33:28 -0500
From: Jay Larson <larson@mcs.anl.gov>
To: Cecelia Deluca <cdeluca@scd.UCAR.EDU>
Subject: "Application" section for tecnical.tex

Hi Cecelia,

This should just paste into the blank section in technical.tex.  Hope this 
is useful
to you.  You'll note there are some [Ed. note] entries.  I thought it wiser 
to send
you something now rather than withhold it while I agonize.

Cheers,
            Jay

\subsection{Application code---The CCSM Flux Coupler}


A natural testbed application to demonstrate the capabilities of the ESMF 
is the flux
coupler for a coupled climate modeling system such as the Community Climate
System Model (CCSM).  The flux coupler supplies a number of crucial functions
to the CCSM, including:  coordination of time-stepping and synchronization 
of the
component models; interpolation of data from one component model's grid to 
another;
time-averaging of incoming data from one component model for later exchange 
with
another component model; merging of flux data from multiple component 
models for
delivery to another component model; exception and error handling; history 
output.
Clearly the coupler is an excellent testbed for the following ESMF 
capabilities:  support
for multiple coordinate grids, and on-line generation of transformation 
matrices to
support interpolation between these grids; exception and error handling 
utilities;
performance monitoring and profiling utilities; [Ed. note---there's plenty 
more].

Having chosen the CCSM coupler as a testbed for the core of the ESMF, it is 
necessary
to characterize the current state of the CCSM coupler in terms of the 
following criteria:
parallelism and performance portablility; flexibility and extensibility.

In terms of parallelism, the CCSM currently is a multiple load image 
mixed-mode
parallel application;that is, there are separate executables for each of 
the component
models, and for the flux coupler.  In the original version of the CCSM, 
each of the separate
executables used shared memory parallelism, with message passing between the
component models and the flux coupler.  As the CCSM has evolved, the 
individual
component models now employ message-passing and/or shared-memory parallelism,
but the flux coupler has remained a purely shared-memory parallel 
application.  This
current (as of October, 2000) version of the flux coupler is a fair and 
sensible baseline
against which the ESMF will be compared to assess completion of the 
milestones outlined
in this proposal.

In terms of flexibility and extensibility, the CCSM coupler is quite 
rigid.  Currently, the choice of
component model grids is limited to the combinations of model grids for 
which interpolation
weights have been generated off-line using the SCRIP package.  This puts 
the onus on the
model user wishing to introduce new components with different grids to 
generate the appropriate
interpolation weights.  It also places the opportunities offered by 
adaptive grids beyond the user's reach.
If a user wishes to add new fields to the coupler, this requires 
modification of the coupler's source
code.  Taken together, these two previous points make the addition of a new 
component model to the
CCSM a major undertaking.  [Ed. Note--can't put my finger on it, but I 
think there's more worth adding
here].

During the next year, the CCSM coupler will undergo considerable changes as 
the result of the
DOE/NSF Accelerated Climate Prediction Initiative (ACPI) "Avant Garde" 
project.  The goal of
this project is to enhance the CCSM software in terms of performance 
portability, modularity,
flexibility, and extensibility.  The expected impact on the CCSM coupler is 
that it will be re-implemented
in a more modular fashion using components from a "model coupling toolkit" 
that is currently under
development.  It is expected that this approach will make the coupler far 
more flexible and extensible.
The toolkit under development will also make the resulting CCSM coupler a 
parallel application capable
of exploiting the hybrid parallelism offered by the parallel platforms 
currently available to the U.S.
scientific community.  This project, however has a fairly limited timeframe 
of 18 months, and it is
doubtful that it can realize the potential gains outlined in this proposal. 
[Ed. note--don't have quite the
right words at hand to point out I doubt we'll be able to generate 
regriddings on-line at the end of
ACPI -AG...just so much else to do first!].

We envision one or more of the following activities will demonstrate the 
efficacy of the
ESMF:
\begin{itemize}
\item Re-implentation of the CCSM coupler using components supplied by the 
ESMF
to create a performance-portable, mixed-mode parallel application capable 
of exploiting
message-passing parallelism between shared-memory nodes;
\item Demonstration of improvements to the ACPI "Next Generation Coupler" 
through
substitution of ESMF components for those in the Model Coupling Toolkit.
\end{itemize}
==============================================================================
J. W. Larson, Ph.D
Assistant Computer Scientist
Mathematics and Computer Science Division
Argonne National Laboratory
Argonne, Illinois 60439-4843

vox:  (630) 252-7806 | fax:  (630) 252-5986 | email:  larson@mcs.anl.gov
http://www.mcs.anl.gov/~larson
==============================================================================