Re: The NoSQL train is about to leave the station. Are you in yet?

Think of it as expanding your toolset.

There's no doubt that Redis can severely outperform any relational database. It's a simple, in-memory, persistent key-value store. So if you really need something stupid fast, then maybe redis is a good solution (for part of your data).

Also, most of NoSQL solutions are AP or CP, whereas all databases are AC. So, if network partition is important (i.e, horizontal scaling), NoSQL solutions are certainly something that would be worth investigating.

So, for super-high performance, or complex scaling, NoSQL provides a good niche solution.

Where things get fuzzier is with the CouchDB, MongoDB and other document or object stores. These solutions aren't as stupid fast as the key-value or ColumnFamily NoSQL tools, but they are still fast (and by all accounts, faster than any relational databases, but I'll let you have your doubts). The point though is that these NoSQL solutions aren't targeting specific needs, but rather are meant to replace relational databases for every-day applications. The benefit that they bring is a lot less application code

So, if we can agree that maybe redis is really really really fast (at the cost of only being able to represent relatively simple data, i.e., u'll need to change ur app)..and ColumnFamily is really really fast, but also at the cost of really changing your app/data, then we can say these are technologies targeting a very specific need.

Where you might not agree with me, is with my view of the Document/Object stores. My opinion is that they are faster than RDBMS, but more importantly (since if speed is important, pick key-value or ColumnFamily), they are much easier to program against, with less code, less bugs, and less maintenance. If I'm right, than these solutions should become the workhorses, while RDBMS get pushed off to their own very specific problem (which would likely be complex reporting, but don't underestimate what you can do with map/reduce on NoSQL platforms).

Finally, I think the transition for the specific targets (performance and scale) are already happening. What hasn't happened yet (or what is happening much much more slowly) is the move towards document/object stores for your common application. The move is slower because the resistance is greater and the need is smaller.