Memory growth in Drekar application
Created by: jhux2
@ambrad @rppawlo @eric-c-cyr
Reported by @pwxy
"I'm running the drekar steady-state 3d MHD generator test case with tpetra/belos/muelu, and forcing the Krylov solver to always take 100 steps of TFQMR. The memory usage appears to be growing, with maximum among the MPI processing being 303MB after the first newton step and 413 MB after the 6th newton step. Maybe I'm missing something, but if I'm using TFQMR and running the same number of iterations per newton step, why should the max memory usage grow by 36% after 6 newton steps? Seems to me that it should not be growing. TFQMR is being preconditioned by MueLu with ifpack2 additive schwarz with one level of overlap."