MySQL is unsuitable for web 2.0 applications that have to scale. The introductions of Amazon’s SimpleDB and Google’s BigTable have kicked off a paradigm shift away from the relational database.

New York-based stealth startup Kloudshare promises to take it one step further.

‘In the past the relational database was often used as a kitchen sink solution for everything,’ Amazon’s Dutch CTO Werner Vogels told me at the Next Web Conference in Amsterdam.

Databases should be tailored to their tasks. Amazon first released S3 for primary key storage, followed last December by SimpleDB for multi-attribute queries. Vogels said Amazon has no plans for additional database offerings.

‘Because relational databases can be beaten by more than an order of magnitude on the standard OLTP benchmark, then there is no market where they are competitive,’ relational database guru Mike Stonebraker concluded in September 2007.

The conclusion was based on a test with H-Store, a new database Stonebraker is developing at MIT. Stonebraker has already founded two alternative database companies in the last five years, StreamBase and Vertica.

H-Store could be the reason behind Stonebraker’s controversial attack on Google’s MapReduce. The PostgreSQL inventor was simply spreading some FUD ahead of launching his own competitor to BigTable and MapReduce.

‘I found it shocking what he wrote,’ says Hank Williams. The New York-based computer scientist and tech blogger is working on a database startup called KloudShare.

‘What we’re doing is much more in the Google, horizontal scaling, MapReduce camp than it is in the Michael Stonebraker, column store, Vertica camp.’

Williams sees SimpleDB, BigTable and open source variations ThruDB, CouchDB and HyperTable as a step in the right direction. Kloudshare will take it one step further with something like a navigational or graph database.

Details are sketchy. On his excellent blog Williams mentioned removing foreign keys from the database system as part of the solution.

How would that work exactly? ‘If you start from saying you’re making just minor modifications to the relational database, then it all sounds a little funny,’ Williams says. ‘That’s not a great way to express the problem.

‘Relational purists also hate foreign keys. They hate them for different reasons, because they feel foreign keys are navigational. I don’t like foreign keys because I think they’re not navigational enough!

More on this subject in Computable.