I'm most familiar with relational databases and the way in which their data is stored is generally dictated by the schema.
At a minimum, to be consistent with this patent we'd be talking about a database that accepts arbitrary groupings of predefined sets of fields, which then on-the-fly determines how best to pack these individual sets together with other sets belonging to other groupings. The fact that we are talking about two types of collections (one nested in the other) is key -- that is, the nesting of fields within the sets ("components" in ECS terminology), and the nesting of these sets within the groupings ("entities" in ECS terminology).