Memory leak using fipy with trilinos

Question

I am currently trying to simulate a suspension flowing around a cylindrical obstacle using fipy. Because I'm using fine mesh and my equations are quite complicated, the simulations take quite a long time to converge. Which is why I want to run them in parallel. However, when I do that the program keeps using more and more memory, until Linux eventually kills it (after around 3 hours when I use 4 procesors). What is more: trilinos increases memory usage even if I only use one processor. For example, when I run this example (changing no. of sweeps from 300 to 5,000 first):

python stokesCavity.py --trilinos -> memory usage goes from 638M to 958M in 10 minutes
python stokesCavity.py --pysparse -> memory usage goes from 616M to 635M in 10 minutes

I saw here that somebody had reported a similar problem before, but I could not find the solution. Any help would be appreciated.

Some info: I am using Trilinos 12.12.1 (compiled against swig 3.0) and fipy 3.2.

Answer 1

This is an issue we have reported against PyTrilinos