My apologies, the article was talking about 100M-rows datasets so I presumed that was the scenario.