The "NoSQL" Discussion has Nothing to Do With SQL

You seem to leave out several other sub-categories of the NoSQL movement in your discussion. For example: Google's BigTable (and clones) as well as graph databases. Considering those in addition, would that change your point of view?

Michael,
It seems to me the denormalization that occurs with typical usage of a document-oriented database improves performance: a complex document with some nested objects and/or arrays might have been 10 rows in an RDMS instead of one contiguous piece of data. Do you agree?

Re Dwight Merriman's comment:

I was careful to point out that NoSQL data bases might be advantageous in document repositories in my blog. There are multiple reasons for this possibility, including denormalization (your discussion), keyword retrieval, and semantic tagging.

--Michael Stonebraker

Re Johannes Ernst's comment:

I am a huge fan of "One size does not fit all." There are several implementations of SQL engines with very different performance characteristics along with a plethora of other engines. Along with the ones you mention, there are array stores such as Rasdaman and RDF stores such as Freebase. I applaud efforts to build DBMSs that are oriented toward particular market needs.

The purpose of the blog entry was to discuss the major actors in the NoSQL movement (as I see it) as they relate to bread-and-butter transaction processing (OLTP). My conclusion is that "NoSQL" really means "No disk" or "No ACID" or "No threading," i.e. speed in the OLTP market does not come from abandoning SQL. The efforts you describe as well as the ones in the above paragraphs are not focused on OLTP. My blog comments were restricted to OLTP, as I thought I made clear.

--Michael Stonebraker