In the nosql shoot-out thats been going on in the last 6+ months or so, Cassandra seems to be gaining the most traction followed by MongoDB. This is entirely based on my non-scientific monitoring of the various mailing lists, twitter feeds and articles. The big problem I personally have with Cassandra is that it is Java based. Not because I think Java is inherently bad, but rather because I don't know it well.
Personally, as long as I'm gonna jump into the nosql deep end and pick up some new skills I figured I would go full boar and spend some time with Riak (http://riak.basho.com/). Not as mature as Cassandra or even MongoDB, Riak is Erlang based with strong support for JavaScript and the http protocol stack. Further, Riak adheres to Amazons Dynamo paper very closely which meshes well with my overall philosophy re. data distribution and persistence. Riak as a project is moving along well and if you spend anytime kicking the tires you'll soon notice what is lacking and start thinking about how to work around those issues. Needless to say, if you follow #riak on freenode you'll also get a feel for what Basho has in mind for the roadmap. From what I can gather, Lucene hooks of some flavor are in the works.
If you're going to pick a product based on how familiar with its implementation language you are, a whole lot more people can use Cassandra than Riak. :P
But really you shouldn't care what language your database is written in, it won't affect you at all unless it's written in something esoteric enough that there's no way for your app to interface with it. (Cassandra's native api is Thrift, which means you can use it from damn near anything.)
Hmm, if you are a developer just writing against thrift I can see your point. But if you are also involved in the sysadmin side then you need to be familiar with all the little java bits and bobs.
Isn't it already established that Cassandra unlike many other NoSql solutions from the start was developed to the requirements of the biggest social websites and if you are not one of them then you may be better off using a less complicated/resource intensive solution.
On the other hand if you play big then having a dedicated java geek looking after your Cassandra shouldn't be a problem and even would gain you a bit of advantage in other areas.
While I know my way inside Java, I see no reason to fiddle with the code my database is written in. As for being a resource hog, it may be, but just try to accomplish the same throughput Cassandra is designed to handle with MySQL and see how much hardware (and fiddling) you will require.
As always, use the right tool for the job. I don't mind my bookmarks being stored in SQLite and I don't mind my paychecks being generated out of an Oracle RDBMS, just like I don't mind having some huge volume of data that do not require a relational database stored in a non-relational database.
I have recently read an excellent article about the TPS operating system on zJournal. It was a very interesting read, not because I maintain a TPS installation (one of the three) but because it shows how pervasive non-relational databases really are.
Personally, as long as I'm gonna jump into the nosql deep end and pick up some new skills I figured I would go full boar and spend some time with Riak (http://riak.basho.com/). Not as mature as Cassandra or even MongoDB, Riak is Erlang based with strong support for JavaScript and the http protocol stack. Further, Riak adheres to Amazons Dynamo paper very closely which meshes well with my overall philosophy re. data distribution and persistence. Riak as a project is moving along well and if you spend anytime kicking the tires you'll soon notice what is lacking and start thinking about how to work around those issues. Needless to say, if you follow #riak on freenode you'll also get a feel for what Basho has in mind for the roadmap. From what I can gather, Lucene hooks of some flavor are in the works.