Part III: Computing

Once the rows are in memory, what does each operator cost the processor?

Part II made the engine read less. What it does read, it must compute: every expression a query writes is evaluated for every row that reaches it. This part counts what that costs the processor, starting with how an engine evaluates an expression, and why engines work a batch of rows at a time rather than one row (ch06). Then it groups rows, and finds that what a group costs depends on how many there are, and on the cache (ch07). A join holds one of its inputs in the same kind of table, and which input it holds decides what it costs (ch08). Putting rows in order holds them all, unless a LIMIT says how few are wanted (ch09). Last, it takes memory away, and counts what finishing anyway costs (ch10).

Built from commit dc21108.