Meeting Summary

Hypre

Sorted out several issues with bluegene and hypre - although I am still a little perplexed by hypre's fortran interface… Was able to use it successfully on bluegene - but when I tried the same on alfalfa - I ran into weird mpi errors… I think it has something to with the assumed length of integers passed through the hypre interface??? Our own interface works correctly - so this is probably not worth pursuing… Also discovered that when running with AMR on 100's of processors, that hypre's memory requirement may be quite large unless you use a version of hypre configured with "—with-no-global-partition". See Hypre Memory usage

Restarts

I modified the restart IO to be much faster on bluegene… Restart time was reduced from 3 hours to 3 minutes! See #180. I still need to check this into the repository…

Colliding Flows

These runs are going well on kure. Need to start thinking about analysis… Here's a movie​ of a clumpy flow. I have a similar setup but with a smooth flow running on kure now.

Gravo Turb

After fixing hypre issues, runs were able to proceed for about .25 free fall times. Then some combination of protections/resteps/etc… caused runs to 'lock up'… Maybe initial setup was too ambitious - or refinement criterion was not sensitive enough to avoid large errors in energy/gravitational potential at coarse fine boundaries… Looking at a similar setup in 2D on alfalfa to try and determine what may have gone wrong… Here's a movie​

Collected some code improvement ideas on to DevelopmentProjects

Attachments (2)

Comments

No comments.