Minutes 10 April 2006
Present: Shane, Andre, Karl, Greg, Poul, Travis
Apologies: Chris, Peter
An audio recording of the meeting can be "downloaded":ftp://ftp.bioeng.auckland.ac.nz/cmiss/opencmiss/cmiss-meeting-100406.ogg
Recording
Shane is trying out recording the meeting with his phone so that the meeting doesn't need to be recorded realtime from the dictaphone.
Parallelization investigation
Karl: Surprised by lack of libraries to provide helpful tools for background communication (while computation continues). Although some low level tools are available (RMA in MPI-2 is of note), higher level functions (e.g. of Zoltan) seem to be designed to have separate communication and computation phases.
Travis went to an iterative-solver conference in Colorado:
- Kirk Jordan gave a presentation: Our group was one of three groups mentioned that intend to run on Blue Gene. Another group was able to port existing MPI code to Blue Gene in 2 days. Our group was mentioned as rewriting code.
- A lot of people using PETSc, Metis common.
- No obvious higher-level package on top of MPI. (MPI is the package the hardware vendors optimize for their platforms.)
Job description
Going through University processes.
cmiss.org
- wiki mail is working (through a work-around) thanks to Carey.
- tracker bug notification still needs attention.
Shane: Who is going to learn how to do things and continue when Carey is gone.
Fieldml:
Poul:
- Tidied up some differences between element and ensemble.
- Changes to ordering: parameter maps has moved from ensemble field to ensemble field function.
- Has some examples including serialized form.
- Removed requirement for some naming (using raw index instead) to reduce verbosity of files: ensemble field parameters, basis functions, field parameter map.
- Priority is to look at organizing field parameters in arbitrary ways (e.g. at a node).
- Hanging node example picks up parameters at hanging node from adjacent nodes.
- Ensemble field parameters maps had 3 types: general, real, boolean. Now folded into one list of linear maps (with different types for different elements).
Greg: Is an ensemble field parameters linear maps required for every element? How does templating help?
- Shane: templating may be usable with an offset, scale factors may be able to be extracted.
Travis: Are we still considering using cmgui to implement fieldml?
- Shane: cmgui would be ported to implement new fieldml ideas. If we use cmgui initially then that code could be ported at the same time.