I'd like to see the code / data so I can try to reproduce these results. If it took them 6 hours to load 25m rows / 10GB with PostgreSQL, they must have screwed it up (deliberately?).

A graph with no units on the y-axis (and not even stated clearly in the text) put me completely off reading this article.

I thought it was safe to assume that seconds were the unit of choice, but next time i'll be more clear. Thanks for the suggestion

Impala, imitating Google's Dremel, is specifically designed for OLAP scenarios with its columnar storage structure and the compression algorithms that go along with it. This is not a fair comparison. Try doing OLTP with Impala and you'll see what I mean.

Great point! I mentioned it in the post, did you read it?

"Note that the properties of a relational database are well suited to online-transaction processing (OLTP). Such a system guarantees consistency and availability, and it supports extreme multi-tenancy. However, complex logic over large amounts of data does not perform well"

Benchmarks where we're not given the data, code, scripts, etc to reproduce them are not useful.

If you look in the citations, you'll see everything you need to reproduce the results!