Am working a lot with simulations that are taking way too long to run to completion, one such is crack growth with XFEM. My feeling is to relax a bit the convergence criterion so that the computation speed can be upped. Although am aware this will be at the detriment of computation accuracy