Adi Cohen ISUG 124

1. QUERY HANDLING

2. OPTIMIZER’S ROLE
• The optimizer should generate an efficient
plan to access the data that we need to
work with.

3. WHY DO WE NEED IT?
• SQL Statement defines what we are looking
for. It doesn’t define how to get the data.
• The optimizer has to generate a query plan
which tells the server how to get the data.

4. OPTIMIZER CHALENGES

5. NUMBER OF JOIN ORDERS

6. OPTIMIZER
• The optimizer doesn't necessarily look for
the best query plan.
• The optimizer searches for a good enough
plan.

7. WORK STAGES (HIGH LEVEL)
• Create a logical plan
• Apply transformation rules
• Apply heuristics
• Select the cheapest one

8. STEPS OF QUERY PROCESSING

9. PARSING
• Checks that the syntax is correct
• Creates a logical tree that represents the
query.

10. DEMO 1
• See logical tree that was produced by parse
step.

11. BINDIND
• Gets the logical tree that was produced by
the parsing step
• Makes sure that all objects referenced by
the logical tree exist and that the user can
see them.

12. OPTIMIZATION
• The most important step. Generates the
query plan that will be executed

13. TRIVIAL PLAN
• Very simple queries with very simple logical
tree won’t get full optimization
• The optimizer will generate a plan that is
called a trivial plan
• The trivial plan is very cheap to generate
and it won’t be inserted into the plan cache

14. TRIVIAL PLAN
• If the trivial plan’s cost is more then the
value that is configured as “cost threshold
for parallelism”, the query will get full
optimization
• You can disable the trivial plan by using
trace flag 8757

15. SIMPLIFICATION
• The OPTIMIZER rewrites the tree to make it
more simple. There are few simplification
methods:
– Constant folding
– Domain simplification
– Contradiction detection

16. DEMO 2
• Simplification and trivial plan

17. OPTIMIZATION LEVELS
• If no trivial plan was found, SQL Server
starts optimizing the query.
• SQL Server has 3 stages that are called:
– Search 0 (Transaction processing)
– Search 1 (Quick plan)
– Search 2 (Full optimization)

18. OPTIMIZATION LEVELS
• Each level has an entry and termination
condition.
• Termination condition can be good enough
plan was found, or to much time passed
• Optimization can begin at a low search step
and get to a higher search step.

19. PROPERTIES
• Each node in the logical tree has properties
attached to it.
• There are 2 types of properties
– Logical properties (Node cost, output columns,
Type information and nullability, etc)
– Physical properties (Sort order, partition
information etc’)

20. RULES
• The optimizer has set of more then 350
rules that is using to optimize the query.
• The rules help the optimizer to modify the
logical tree in a way that doesn’t effect the
query’s results
• The rules also dictate the physical
implementation of the logical tree

21. RULES
• There are four types of rules that can be
used:
– Simplification rules
– Exploration rules
– Implementation rules
– Physical properties enforcement rules

22. MEMO
• All the trees are stored in memory in a
structure that is called Memo
• Each optimization has its own memo
• A single memo can get to the size of 1.6 GB

23. OPTIMIZATION PROBLEMS
• On rare occasions there can be a timeout
for the optimization process
• Most times when it happens, it will be on
search 2 stage, and the server will use the
query plan that was produced on search 1
stage.
• Sometime the server stops optimization
because of memory pressure

24. DMVs
• There are 2 DMVs that can give us
information about the optimizer:
– sys.dm_exec_query_optimizer_info, shows
information about optimization process
– sys.dm_exec_query_transformation_stats, sho
ws information about rules usage

25. Demo 3

26. QUSTIONS?
• Additional reading:
– http://sqlblog.com/blogs/paul_white/archive/
2012/04/28/query-optimizer-deep-dive-part-
1.aspx
– http://www.red-
gate.com/community/books/inside-sql-server-
query-optimizer