[louisgag]

Hello,


I've only tested with v2112, with both 32 and 64 bit label versions, and when I run the field transfer portion of the decomposition, the operation gradually slows down until it becomes unmanageable anymore.


Did anyone else also notice this behavior and is there a workaround to avoid going over the typical 24 hour limit of HPC systems?
I've tried doing field by field, but it's not a nice workaround.. and copyZero does not give the solve the procBoundaries that are wanted in field...


I'm attaching an example decompose -fields output log timed with ts: at the beginning there are about 20 processors treated by second and at the end then only 2... For bigger cases it really becomes unmanageable, with times reaching 30 seconds per processor...

[Yann]

Hello Louis,

I have no idea about the decomposition slowing down, but if you want to use copyZero you can use regular expressions to deal with procBoundaries:

"procBoundary.*"
{
    type processor;
}

Regards,
Yann

[louisgag]

Yes your right, I totally forgot about this. Thank you!