On Tue, Mar 6, 2012 at 6:20 PM, Kentaro Takami kentaro.takami@aei.mpg.de wrote:
OK. CarpetRegrid2::positions should be checkpointed and recovered; if this is working as it should (can you check your checkpoint files, just in case?), then something is changing these variables at recovery.
At beginning, these variables are initialized by parameter values:
INFO (CarpetRegrid2): Initialising position of centre 0 to [15.3,0,0] INFO (CarpetRegrid2): Initialising position of centre 1 to [-15.3,0,0] INFO (CarpetRegrid2): Initialising position of centre 2 to [0,0,0]
After reading checkpoint file, these variables are replaced by checkpoint data.
However I can not directly check the checkpoint file, because already I don't have these checkpoint files and this strange behavior "sometimes" happen (not every time in recovering).
Is Whisky_MoveCarpetGrids perhaps reinitializing them?
I don't think so. Already as I mention, I can see this strange behavior sometimes, not every recovering time. I guess...this means the behavior depend on the iteration position of checkpoint file.
You are making two guesses here -- about Whisky_MoveCarpetGrids, and the one about "depend on the iteration position". I would be very careful with such guesses; it is always necessary to check these with hard data. I suggest you examine both the Cactus schedule and the checkpoint file next time you see the problem.
-erik