Continuation run fails when using k-points about qball HOT 4 CLOSED

How many MPI tasks are you running with? What happens if you try save -states and load -states instead?

Also, note that the last argument in your run command should probably be 2-5, those are typically the most efficient values.

from qball.

I have tried 16(nodes)x8 configuration as well as 2x8. Both behave the same way. Also, initial run and continuation have the same node/core configuration.

I have tried using the -states option with the same outcomes yesterday. Today it seems to work. Maybe they are still having issues with the cluster.

from qball.

So it seems that this issue is still occurring if I use dump option instead of states.

from qball.

Hardware or system problems could be responsible, we've seen similar issues at LLNL from time to time. The -dump option is one file per task, which is probably the most stressful on the I/O system. The -states option is only marginally slower and a lot more flexible, as you can restart with different numbers of tasks or process grid dimensions, and it should tax the I/O nodes on a BG/Q machine a lot less. I would try using -states from now on and seeing if the issue regularly recurs.

from qball.