It looks like year is extracted from pickup_datetime at ETL, so hence its not a fair comparison against the other databases that do this at runtime in Q3 and Q4

It looks like year is extracted from pickup_datetime at ETL, so hence its not a fair comparison against the other databases that do this at runtime in Q3 and Q4. In something like MapD Q3 would be nearly as fast as Q1 (~20ms) without the extract function, which involves relatively complicated math.