2006-11-30

• Papers and progress:
  • Load balancing/Fractography: Gengbin (if you can't make it, please update us on the status of this via e-mail)
    • He didn't come, no word on wheither or not it was submitted
  • Parallel CFE: Isaac
    • Issac will need people to read the paper.
  • Multi-paradigm Case Study: Sayantan/Aaron/Terry
• CPSD status: Aaron/Rob
  • Discovered obvious errors on the parallel side, in the process of fixing.
• Import fixes: Rob
  • didn't work over holidays
  • Rob has the wrong mesh?
• Bulk adaptivity: Terry/Sayantan
  • don't assert the mesh has to be in a consistent state until we've finished updates
  • written and mostly tested
• pTops: Isaac
  • no progress
• Add element types to ParFUM?
  • Terry was wondering if we needed to add an element type, like triangle, etc., into ParFUM?. Code that Issac and Aaron worked on over the summer might be relevant.
• Feature detection
  • We need to implement feature detection, but no one is particularily interested in the task. Terry discussed hiring an undergrad to implement an existing algorithm.

2006-11-02

• CPSD progress
  • marked edge flipping, bisection, algorithms never called in the code. Decided they were safe to comment out.
• Adaptivity progress
  • working, but has memory leaks.
• Dynamic virtualization
  • talked with sanjay abotu how to do this, work on dynamic virtualization.
• Feature detection
  • no progress, any one can take over this project.
• Element types:
  • We should look into moving things around so we can specify what connectivity type each element represents.
• Papers:
  • CFE paper (Isaac)
    • out of town
  • CPSD Adaptivity & LB paper (Aaron, Rob)
    • incremental adaptivity and load balancing should be two separate papers?
  • Incremental & Bulk Adaptivity in ParFUM? paper (Terry, Sayantan)
    • paper on how parfum mixes paradigms, ms+mpi+charm++ entry methods.
• cohesive element library? Can we implement a non-hacked version?

2006-10-19

• Prospects for future funding:
  • multi scale load balancing - people are very interested
  • Requires dynamic chunk splitting/repartitioning/load balancing, Aaron's project?
• CPSD
  • Rob set up a wiki
  • Aaron rewrote the build system, so that parallel builds now work
  • Aaron is looking into merging the main loops of the serial and parallel versions
  • Rob is going to look into creating a mock object for MeshInterface?
• Feature detection
  • Aaron is still interested in working on it
• Rob might end up writing small serial routines that do various geometry things
  • Rob's noticed strange _registerParFUM
  • AdpMM? adaptivivity code has the most up to date adaptivity usage. Look at this for reference on how adaptivity works
• Issac hasn't gotten a good chance to work on pTops, too busy. Can work more on this by next meeting
• Terry is working on solution transfer, making slow progress.
  • Trying to make the adaptivity work. May farm out work to Issac and Rob.

2006-10-05

• Issac - presenting his tops progress: pTops
  • ParFUM_TOPS.[Ch]
  • need the ability to add arbitrary data on elements/is it possible to add pointers as data on an element
    • conclusion: instead of a pointer, put indexes into a table that is pup able.
  • Rob needs to read papers/AdpMM paper and return comments
  • discussed what would be needed to properly implement chohesive elements in ParFUM?
    • to properly model, we need to have element adjacent ghost layers

2006-09-21

• CPSD progress -
  • Aaron reported that the code doesn't refine the mesh with 2 vps, but with one 1 vp, it would refines forever.
  • Aaron sent the output to Ressin and is waiting for his input
  • When Aaron runs the code on more than one PE, the code exits with no error code.
• Adaptivity paper was read by Sanjay, needs more work. Postponed until after Sayantan's prelim.
• It was decided that the load balancer paper needs formatting work?
• Rob presented cxxtest
  • People agreed that Rob should present to core meeting, see if and where it should be added to the source code.
  • instant code review
• AMPI and dynamic mesh repartioning/balancing
  • discussed adding the capability to add new processors to an AMPI run
  • decided it was too much effort
  • For now, it might be easier just to keep spare AMPI threads around that do nothing.
• Terry: writing solution surface transfer code
  • incorporating her changes into the charm data transfer library
• Terascale applications
  • our stuff scales trivially, should we include it on the homepage? No final decision was made.

2006-09-07

• Aaron updated on the cpsd project
  • weird "file does not exist" errors.
  • Aaron and Sayantan were going to debug this problem after the meeting.
• Rob and Aaron have talked about CPSD, have a basic starting plan
  • Rob plans on trying to test the MeshInterface? first.
• Discussed the Parfum load balancing adaptivity paper
  • Dr Kale needs to read a recent version of the paper.
  • He'll provide more recent version, Terry will take his current comments and add them to the paper.
• Talked about future work for solution transfer
  • possible extract solution transfer into a module, make a paper.
    • topological equivalence at boundary - mesh line segments are the same.
    • geometrical equivalence at boundary - mesh boundaries are described by the same mathematical function.
  • we can do topological equivalence, do we want to use geometrical equivalence as well?
  • Need to think about this problem more.
• Terry revealed her plan for working on ParFUM? in the future
  • diagram up on TMS, needs a link.
  • couple of months to complete the simpler plan on the left.