I've set up Elmer to work with GlaDS on multiple meshes. It works perfectly fine in serial, but, in parallel, the hydraulic potential results from the very first iteration start to diverge. It's not a question of some partitions not being calculated properly - I still get a smooth field with the right pattern of hydraulic potential. But the values are scaled down - in the serial case, I get a maximum hydraulic potential of about 0.14; with two partitions, 0.096; with four, 0.059; and with eight, 0.054. So, it seems as if, somewhere, the results are getting divided by the number of partitions based on some sort of negative exponential curve. As far as I can see, none of the code I've written to get the multiple meshes to work is responsible (only two parts have any kind of parallel statements in, and I've checked it's not them), and all the inputs (LOAD, FORCE, MASS, etc.) are identical in the serial and parallel cases. My current hypothesis is that the ParallelInfo isn't being set correctly in some way, but I'm open to other ideas if anyone has encountered something like this before?
Extension of Elmer in computational glaciology
2 posts • Page 1 of 1