Dynamic Statistics

Abstract

Imagine you’ve been given the job of sorting through some books in a house to find all the references to the word “Vienna” in the book titles. You’ve been told there are two or three books in the study, and that there is an index in the desk if you need it. Naturally with two or three books or even a dozen books, there’s no point in using the index in the desk; you’ll just look at the book titles one by one and quickly find all of the books with “Vienna” in the title. Unfortunately, what you didn’t know is that since the last person looked at the books, the owner has bought the entire contents of the local public library, but you only discover this after you start. It’s going to be a long day if you stick to your plan. In this analogy the table size (library) was underestimated to a few books so we used a full table scan. If we’d looked at how many books we had (dynamically sampled), we may well have changed our plan to use an index.