where variant_label complys with the ensemble_pattern of the cmip6 config file. It is usually fine to use the variant_label (ensemble), except for the DCPP (decadal) data, which has the start year as sub_experiment_id (see examples given in the global attributes paper).
It seems to me that the current cmip6 config file does use ensemble instead of member_id in the DRS and the dataset_id.
The global attributes, DRS etc. paper (http://goo.gl/v1drZl ) specifies the member_id as:
where variant_label complys with the ensemble_pattern of the cmip6 config file. It is usually fine to use the variant_label (ensemble), except for the DCPP (decadal) data, which has the start year as sub_experiment_id (see examples given in the global attributes paper).