A distributed database (key/value) store is a paradigm that we have gone over a lot. Perhaps will try and get together some links to show NOSQL type database creation etc. and how its done, it may be worth looking at implementations on dynamo/dynamite as this is amazon and netflix for instance. Add in security etc. and then you have SAFE. What is needed is a key/value programming guide for traditional database implementers. I am really sorry but time is my enemy right now, but for this purpose imagine SAFE is an efficient key/value database itself. In that way of thinking you also need to consider eventual consistency models as well (facebook / cassandra). Many of today’s super scaled systems (google bigtable etc. (see byzantine paxos implementation)) all use massively distributed key value stores and say its not possible with a database. See also the martin fowler video I posted here regarding differences.
I imagine SQL front ends can be written as attempted for hadoop type SQL imitation front ends. When we get a bit more clear form the current incredible workload I will try and write something up and create a typical example how to do this. There is info out there though, just nobody imagined a network doing this at planarity scale on a hostile network like the internet.