Cost Model #18259
Description
This task is part of epic #18249
Cost Model
Do we want to pursue a traditional, complex cost model that estimates all work in a plan, or take a simpler approach? In practice, even the most detailed cost models often prove inaccurate despite significant effort.
Do we want to adopt a more intuitive approach—similar to the join ranking strategy. Consider the following examples:
- Is a plan with two merge joins better than one merge join and one hash join? How should we assign weights and make comparisons?
- Is a merge join on a single stream consistently faster than a partitioned hash join across multiple streams? How do we evaluate and rank these scenarios?
- Instead of using exact byte sizes, could we categorize input sizes as small, medium, or large and assign weights accordingly?
This could serve as a quick and practical research project: define relevant properties, assign weights and criteria, and run simple experiments to compare estimated costs against actual runtimes.