I really like this kind of way to solve problems. I mean, is a fact of scalability, and your learn early: divide and conquer (D&C)!
Here you can watch Jay Kreps discussing about the architecture, algorithms, implementation and deployment of Voldemort, a distributed storage system (a distributed key-value storage system). He also presents the problems solved using Voldemort at LinkedIn.

Jay Kreps is Principal Engineer at LinkedIn, being a founding member of LinkedIn’s data platform team which focuses on algorithms and infrastructure for building data driven products. He is a co-author of Project Voldemort. He holds a BS and MS in computer science from UC Santa Cruz where he studied statistical machine learning.

I really don’t know much about databases, but this NoSQL stuff is something really interesting…
peace