Summary
New data models appear regularly, each promising to replace the relational model. This paper reviews about 35 years of them, from the hierarchical IMS and the CODASYL network model through relational, entity-relationship, extended relational, semantic, object-oriented, object-relational and semi-structured XML models.
For each era, the authors explain what problem the model tried to solve and why it succeeded or faded. They draw lessons from the pattern: physical and logical data independence matter a great deal, complex navigational models are hard to use, and good ideas from challengers tend to be absorbed into relational systems rather than replacing them.
The paper is a useful antidote to hype. Many modern debates about document stores, graph databases and schema-on-read repeat arguments made decades ago, and knowing that history helps you judge new systems on their merits.