Coupling protocols

1
Coupling protocols software strategy

• Question 1. Is it useful to create a coupling
  standard? YES, but
• Question 2. Is the best approach to make a single
  coupler/coupling standard available for
  community  collaboration? YES, but

2
Q2

• A coupling standard is a good idea but
• it is a fast evolving standard in time
• (e.g. CPL7 vice ESMF)

3
Q1 Key Points on / Requirements of a coupling
standard

• 1. Identify the split between technical and
  physical requirements.2. We can standardize the
  technical part of the coupler, but leave the
  physical requirements loose, with the restriction
  that the  physical requirements will have to meet
  certain requirements such as  conservation.
  Standard will evolve over time.3. Need a wide
  range of supported platforms.4. Coupling Tools
  need to have a low overhead.5. Design needs to
  be modular to simplify the development of the 
  System Model.6. Regridding support -  CCSM uses
  LANL SCRIP to generate the weights  with a sparse
  matrix multiply to compute the interpolation.
  Oasis  generates weights on the fly and computes
  the interpolation  internally. ESMF can both
  generate weights on the fly or use pre- computed
  weights and conducts the interpolation by sparse
  matrix multiply.7. It is worth considering what
  tools are used by external groups  that have
  models we'd want to use couple to in the future
  such as  biology, watershed, etc.8. Make the
  coupler as general, stable and solid as possible,
  so that  its usefulness outlasts the life of the
  model components. The coupler  is the nexus of
  the community.

4
Detailed Comments

• Hill Have a hierarchy of standards - some
  technical and others conceptual/physical.I.e.
  Physical what does the atmosphere need from the
  ocean. Where as the software engineering of
  coupling.        Experience with ECCO - has
  both in house (MAPLE) and ESMF. Need a vision
  when working on a long time line. Define
  interfaces, but allow people to decide on
  everything beyond that. Don't want to lock into
  a standard that can't evolve as model develops.
  Standard shouldn't restrict the
  process.Maslowski Ocean/sea ice is good
  example - sea ice and upper ocean can be viewed
  as a single layer. A standard coupling approach
  would get in the way of this sort of
  integration.Doescher They defined an interface
  in terms of standard variables - an agreement of
  what physical variables to share. This proved
  too restrictive to the science groups. Instead
  he feels that each group should be allowed to
  define shared states and let the
  coupler reconcile the two states.Their model
  sends each field individually using oasis. No
  packing. Run concurrent.Does time stepping
  create bottle necks. No just add more
  processors. Timing information handled through
  meta data.Maslowski Would Oasis meet the needs
  of the project.Both the technical and physical
  standards don't have to be meet at the same time.

5
Detailed Comments

• Coupler must not add significant over head.
  Software must be portable to a wide variety of
  machines.Doescher Need to create a list of
  technical requirements.What about remapping?
  CCSM uses scrip to generate the weights,
  Oasis can generate weights on the fly, ESMF can
  both generate weights on the fly or use
  pre-computed weights.Maslowski Additional
  points in ESMF not already discussed.Cecelia
  ESMF has the ability to recursively call
  components - used in applications such as
  ensemble runs.It is useful to think about the
  specific coupling tools adopted by external
  groups that may be of future interest - such as
  fisheries, biology, watershed, etc. it is
  important to be inclusive - let the coupling
  standard be used to create a "community coupler
  model" that so that others can bring in their
  model.Hill we agree that there should be a
  standard, exactly what is currently undefined.
  Requires a working group.Cecelia put out the
  requirements and compare the options.Stark
  make the coupler as general, stable and solid as
  possible, so that its usefulness outlasts the
  model components.

6
Detailed Comments

• We're actually creating an Arctic System Model
  Framework.Maslowski Layered way of thinking -
  coupler joined separate models - end goal is to
  create an Arctic System Model, but an ice or
  ocean, or atmospheric model.Difference between
  CCSM and ESMF design is that CCSM is hub
  and spoke, while ESMF allows the addition of a
  layered approach with a hierarchy of
  couplers.Maslowski Extension needed to CPL 7
  for the System model.        - lateral boundary
  condition support        - way to handle non
  overlapping domainsStark coupler tools would
  be used to build support for these
  issues, rather than solve them directly. Does
  ESMF handle Meta dataCecelia Yes. A system is
  being built into ESG to automate
  code generation, as a step toward that meta data
  class was created.