The BigTable design rejects many of the philosophies of standard relational models, explicitly preferring denormalization to a big host of tiny tables.
One of the larger areas where this is a problem is in the modelling of many to many joins.
One way to model these joins is to violate first normal form, and put all interesting data in a db.ListProperty(). While this has the ability to be searchable from a query, I have not yet explored the performance implications of searching a list versus pulling another table.
As joins are not possible, it is possible to link tables through RelationshipProperties. Therefore, with enough effort, the standard intersection table (a table with a joint primary key which references both parent tables) can be created. Has anyone explored the performance hits of the various implementations?
While the List of Keys suggested in the documentation is indeed one way to do it, I'm interested in the performance and anomaly rates of that and other implementations. Is there utility in creating mutual lists of keys? Is the effort involved in the repeated gets worth the price? Is there a better way to do it?