Padraig O'Sullivan wrote:
Hi!
I have a quick question on some code I am thinking of removing that
I'd like to get some opinions on. At the moment, there are 2 functions
in the optimizer in Drizzle for searching for an optimal query plan:
- find_best
- greedy_search
The method that is used is determined by the optimizer_search_depth
parameter. We can see this from the following code:
You can do better than either by intelligent traversal of the choice tree:
1. Consider the most optimistic alternatives (unique key / single
record) first.
2. At each stage, estimate the minimal cost of the remaining sub-tree
3. The if the cost to any stage plus the minimal cost of the
remaining sub-tree is greater than the best plan, don't bother to
continue evaluation of the sub-tree.
Interbase had a customer with a 17 way join. The good news was that
evaluation of the optimal plan was only 40 seconds (computers were
slower then). The bad news was that optimization took 13 hours. Search
pruning was clearly in order.
Performinng tree evaluation in optimistic -- heuristic -- order violates
the spirit but not the letter of cost based optimization. And it saves
a hell of a lot of time without giving up any possibility of finding the
optimal solution.
--
Jim Starkey
Sent from Shearwater, off
the coast of New England
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
