Darshan automatically normalizes all timestamps to be relative to MPI_Init() time, but it does not check for possible time skew across ranks (if one rank reports a drastically different MPI_Wtime() from another). We could check for skew at MPI_Finalize() time and adjust accordingly.