Free and stable is the winner.
My plan is quite trivial - just put all the data in memory and use the cluster without changing the application code. Then, for persistence, I could just dump the data from nodes in a regular disk-access database.
The only thing is, that since I demand in-memory storage, there is no possibility of full data replication. I would like to copy the DB schema only and that in-memory DB manages all the joins consistently for me.
thank you in advance!
UPDATE: I've found an open-source product likely to meet my requirements, it's VoltDB