I am well acquainted with the benefits of non-SQL approaches to data management. And I honor those who speak responsibly about the differences between relational and non-relational approaches, including one of my current clients, a vendor of a highly scalable non-SQL DBMS.
I’ve also noticed some irresponsible prattle on this topic, as has the author of this excellent article in Linux Journal.
This scaling myth is perpetuated and given credence every time popular Web sites announce that such-and-such RDBMS doesn't meet their needs, and so they are moving to NoSQL database X. The opinion of some in the RDBMS world is that many of these moves are not so much because the database they were using is deficient in some fundamental way, but because it was being used in a way for which it wasn't designed. To make an analogy, it's like people using flat-head screwdrivers to tighten Phillips-head screws, because it worked well enough to get the job done, but now they've discovered it is better to tighten Phillips screws with an actual Phillips screwdriver, and isn't it wonderful, and we should throw away all flat-head screwdrivers, because their time is past, and Phillips is the future.
One recent SQL-to-NoSQL move involved Digg.com moving from MySQL to Cassandra. As part of the move, Digg folks blogged about how they were using MySQL and why it didn't meet their needs. Others were skeptical. Dennis Forbes, in a series of posts on his site (see Resources), questioned whether Digg needed to use a NoSQL solution like Cassandra at all. His claims centered on what he considered very poor database usage on the part of Digg combined with inadequate hardware. In his mind, if Digg had just designed its database properly or switched to using SSDs in its servers, it would have had no problems. His best quote is this:“The way that many are using NoSQL is like discovering the buggy whip at the beginning of the automotive era.” Ouch.