This is what I immediately thought of when I saw

> How do you actually measure how good an elevator algorithm is?