Ask YC: Good language/DB combination for distributed naive Bayes classifier?

If you want to go the Erlang route (highly recommended, but be prepared for a bit of a learning curve if you are not familiar with functional programming) then you would do well to look at the Scalaris project (fault-tolerant, distributed, eventually consistent key-value store that can use paxos for consistent transactions) or the Kai project (similar scope and goals.)

You can also probably get pretty far with the built-in mnesia database.

Take a look at kdb+ (see http://news.ycombinator.com/item?id=257771) a language environment with built in database capabilities and distributable. (I'm just learning it so I may have accidentally misrepresented it).

Cassandra, HBase, or HDFS with hadoop if you like java (almost as fast as C++ for numeric stuff). CouchDb if you like erlang (terrible for computations though).