cf IBM's work (Xperanto): [link|http://www7b.software.ibm.com/dmdd/library/techarticle/0203haas/0203haas.html|http://www7b.softwar...aas/0203haas.html]

or an article on it:
[link|http://www.dbasupport.com/oracle/news/war.shtml|http://www.dbasuppor...le/news/war.shtml]

Seems to me, it's a market which has been left to the DB vendors to "roll into" their own product, so that one of the big DB's still ends up being the "primary" database (and developers will then continue to code to proprietary DB interfaces). The only other players I've seen close to this space are e.g. Crystal Reports and its clones, but AFAIK that's read-only.

I got to thinking about it lately because I have a number of projects in mind which will benefit from multiple database options, but would be *miraculous* applications if they could mix-and-match multiple data stores; not just for per-object fetch-and-update behavior (which would be easy enough to achieve with proper subclassing), but JOINs across multiple stores. I understand the largest complaint has always been the performance hit, but it'd be fun to try.

And I don't see why it couldn't be done outside of major $R&D labs--it seems to me to require 1) database connectors, which are a-dime-a-dozen, and 2) clever cursor management. Just wondering if I'm at all on the right track, and especially if it's been done and just isn't googling to the surface. :)