To evaluate performance characteristics, I implemented the query "SELECT a + b FROM t WHERE c" using both Arrow and traditional row-based data structures for comparative analysis. It aims to uncover the underlying reasons for the superior performance of arrow-based query engines compared to traditional row-oriented systems.
I use my own macbook m1 pro to do this benchmark whose width of its simd register should be 128. The benchmark shows that arrow's implementation is 25x faster than row-oriented implementation and 5x faster than compiled execution.
cargo bench