BigTable

Problem: Storing very large data sets in a structured format.

Nugget: Relax the relational model to provide a scalable yet flexible B-tree like store. Reuse as many "data center building blocks" as possible to make the system simple and robust (Chubby, GFS, SSTables, etc).

Why is it different from previous work? Mostly due to the scale.

Hard tradeoffs: There seems to be a tradeoff between ACID consistency + SQL query expressiveness and scalability. It's hard to say whether this is fundamental but it might be.

Will it be influential? Yes, there are about 12 open-source clones of it in various stages of completion. I think a more influential idea for me is the high amount of reuse of existing systems/libraries that went into the architecture, which made it possible to build a robust and efficient system and to focus on application-specific "hard stuff" as opposed to general distributed systems / storage systems "hard stuff".