blog-post-read-100-millions-records-per-sec

https://clemenswinter.com/2018/08/13/how-read-100s-of-millions-of-records-per-second-from-a-single-disk/

https://news.ycombinator.com/item?id=17783754

Related to taxi-query-benchmarks

In this case we can compute the counts directly on the dictionary indices and only convert back to string values at the very end, which speeds up the query by more than an order of magnitude