If You Have Too Much Data, then 'Good Enough' Is Good Enough

Argues that at very large scale, classic single-truth relational semantics give way to approximate, loosely-coupled and eventually-consistent answers - and that this is usually acceptable.