Maybe not throw away, but why should we have only 1 relational language? Functional programmers can choose between scheme, ocaml and haskell. Why should we be stuck with what is in so many ways the COBOL of relational languages as our only option?
What I meant was that, as far as I have seen, the tools that try to replace SQL by some OO intermediary, give you poorer abstractions that limit what you can do. SQL seems clunky, but I think the core is quite well done considering when it comes from.
OTOH a relational relational database doesn't fit all needs. They scale pretty well, but when you get to gigantic data sets, they don't do so well and things like map reduce do a better job. Or cases where you are streaming massive amounts of data and you need performance, not ACID transaction reliability. Actually, we are more in agreement than disagreement.
There isn't a good massively parallel open source SQL implementation right now, but that is just a matter of engineering -- SQL databases have been shown to scale to massive data volumes (e.g. Fox Interactive have ~200 TB of user data in Greenplum, which is a parallel DB based on Postgres).
Maybe not throw away, but why should we have only 1 relational language? Functional programmers can choose between scheme, ocaml and haskell. Why should we be stuck with what is in so many ways the COBOL of relational languages as our only option?