Minutes 20 July 2006
Present: Shane, Karl, Peter, Poul. Apologies: Travis.
An "audio recording":ftp://ftp.bioeng.auckland.ac.nz/cmiss/opencmiss/meeting-2006-07-20.ogg is available.
State of Play
- libmesh
- spends 5 seconds per matrix/rhs assemble (cf. gridfem cm 19 seconds probably with fibre computations).
- 80 seconds setting up the system - not sure what's here apart from adding the dependent variable and setting initial conditions.
- libmesh now has cubic Hermite elements in CVS (although export routines
may not have been written yet).
- Peter asked about scale factors
- Shane said they were not necessary
- libmesh has ability to express some degrees of freedom as a linear function of other dofs, which would be enough to provide similar functionality to scale factors.
- Peter asked about scale factors
- sundance
- Probably used considerably in Sandia but haven't seen evidence of global community use.
- LifeV
- Parallel operations implemented predominantly in solvers, starting to implement in assemble.
- freePooma
- Haven't seen evidence of large community use.
Scripting language interface
- libmesh is a C++ library, intented for use in a C++ program.
- A scripting language interface could be built on top, but Shane is not sure it's a good idea.
Visualization: Shared data store with graphics
- cmgui knows when things (data) change.
- FEM implementations do not usually have such notifications.
- Whatever FEM implementation we use, it is not going to be easy to change cmgui to use the same data structures.
Plan
- Peter: Spend the next 12 months on solving reaction-diffusion
with cellml in libmesh with a large number of processes.
- Will enable us to:
- gain experience with MPI.
- work out what is important to us.
- Start with C++ main program.
- Would interface with cmgui through an exporter in libmesh.
- Ruzak rhun (Jagir) has IBM Blue Gene experience an is expected around late August and is expected to be involved with this.
- Will enable us to:
Another C++ Program
- Ed Vigmond's lab in Calgary has a C++ program based on petsc/mpi which is being used by a few groups to run very big simulations.
MPI job candidates
- awaiting references
physiome.org
- job advert is being put together for physiome.org development and is currently being graded. We have one candidate already.