+1 for Miller. It's a really slick tool for exploring large csv datasets. I've typically used it to do some prototyping and exploration of 200-500GB csv datasets before doing more hefty work in Java+PigLatin (our use-case is more long-term than just a single run for analysis, so that's re reason for moving out of just Miller). It's great to get a feel for the data and run some initial analysis before diving into the larger, more cumbersome system and tooling.