I'm not a CouchDB pro by any means but I've been following it from a distance for some time.
The main feature of CouchDB is that it's a fast, scalable database that also allows you to host your apps inside of it, with each "view" being stored in the database itself.
I believe some people use it for large installations but as far as I know that is not the primary focus.
It's a "fast, scalable database" that isn't focused on large installations? What?
While I haven't used couchdb much personally, its most significant characteristics seem to be that it's a document-oriented database (each record can have its own structure) and that it inherits all of Erlang's infrastructure for fault-tolerance.
Well, that's taken long enough. Lotus Notes has everything to do with CouchDB. Damien was a product developer at Iris/Lotus, and in a lot of ways CouchDB is a modern rethink of the best of what Notes/Domino is. When you take away the client and the need to be backwards-compatible to the late Bronze Age (Notes V2 applications from '93 can run unmodified in the current version 8.5 client), you're free to make a lot of improvements, but you also get to keep the good stuff.
We've been using CouchDB in production for at least six months with our web-based content management system and our mobile advertising platform. We also use it to store scores for a couple of mobile games, and to store posts for PicBored, an anonymous image-based forum for Android. I am a bit worried about how well the replication will work with dozens or hundreds of servers, but it's working great with only a handful.
I would use this to replace MySQL or PostgreSQL and I would use it over MongoDB or Cassandra because it is high-quality software, built on top of Erlang (highly concurrent and fault tolerant language, great foundation to start with), and implements really great features.
I personally use MongoDB for small, one-off projects. The good thing about schemaless databases is that you can just start shoving data into them without caring about its structure.
That said, document-based databases suck at complex queries and reporting tasks.
Is this only for Facebook size scaling or is it of use for those of us just starting out?