We could use an environment variable to let the user choose between different ways to compute MPI imbalance. E.g.:
VFTR_MPI_IMBALANCE=MAX_DEVIATION_FROM_FASTEST
Thinking even further, we could process a formula given by this environment variable. We already include tinyexpr to do this in the scenario files, so the overhead is manageable.
We could use an environment variable to let the user choose between different ways to compute MPI imbalance. E.g.:
Thinking even further, we could process a formula given by this environment variable. We already include
tinyexpr
to do this in the scenario files, so the overhead is manageable.