How can a dataset, where related attributes are spread across more than one record, be stored in a way that lends itself to distributed data processing techniques that process data on a record-by-record basis?