Typically event_time_offset is 32 bit, so converting it to double after MPI communication saves bandwidth.