NoSQL: Missing the Point

I recently saw this ComputerWorld article about the "NoSQL" software movement. To summarize what NoSQL is, take a relational database model and strip out every feature that cannot be efficiently implemented in a hash table. Which is to say, most of the more powerful capabilities that can be implemented in SQL using a good relational database.

The NoSQL crowd makes one good point: hash tables scale much better than fully relational databases currently. But discarding relational features from database engines is hardly revolutionary or even wise outside of narrow cases. The Achilles' Heel of NoSQL database engines is that in giving up the ability to efficiently do operations such as multi-attribute indexing and joins they become very poor for analytical applications.

There are two trends in applications: extremely large scaling requirements and real-time, contextual content that requires increasingly sophisticated analytical processing. To sacrifice the latter for the former is short-sighted. Instead of discarding the analytical features of relational models, we should be improving the scalability of relational model implementations. This is easier stated than done, as it will require replacing the current order-preserving access methods of traditional relational model implementations with space-preserving access methods, the latter being an unsolved algorithm problem in computer science.