[mayank]

At least some aspects of NoSQL are novel from an algorithmic perspective, so it's unfair to dismiss it as a buzzword. The underlying gossiping protocols that enable no single point of failure are fairly recent developments (Chord is a recentish gossiping algorithm if I remember correctly). Just think about writing a distributed database system without any designated master/control nodes-- all kinds of fun algorithmic problems will pop out: randomized algorithms, random graphs, distributed indices, decentralized search, ...

[TimJYoung]

The problem is that the underlying distributed algorithms are not exclusively the domain of NoSQL systems and, likewise, one could easily use SQL with a key-value store, provided that one had access to a schema with which to resolve objects and their identifiers. Developers have been using non-relational data stores with SQL-based front ends for years.