Query Optimization in

Relational Database Systems

It is safer to accept any chance
that offers itself, and extemporize
a procedure to fit it, than to get a
good plan matured, and wait
for a chance of using it.

Thomas Hardy (1874)

in Far from the Madding Crowd

CS3223: Query Optimization 1

 Query Optimization
• Since each relational op returns a relation, ops can be
composed!
• Queries that require multiple ops to be composed may
be composed in different ways - thus optimization is
necessary for good performance, e.g. A B C D can
be evaluated as follows:
• (((A B) C) D)
• ((A B) (C D))
• ((B A) (D C))
• …

CS3223: Query Optimization 2

 Query Optimization
• Each strategy can be represented as a query evaluation plan
(QEP) - Tree of R.A. ops, with choice of algorithms for each
op.
SM HJ
NL
D
NL HJ INL
C

A B A B C D

(((A B) C) D) ((A B) (C D))

• Goal of optimization: To find the “best” plan that compute the

same answer (to avoid “bad” plans)
CS3223: Query Optimization 3
 Motivating Examples

Sailors (sid: integer, sname: string, rating: integer, age: real)

Reserves (sid: integer, bid: integer, day: dates, rname: string)

• Reserves:
• Each tuple is 40 bytes long, 100 tuples per page, 1000 pages.
• Sailors:
• Each tuple is 50 bytes long, 80 tuples per page, 500 pages.

CS3223: Query Optimization 4

 SELECT S.sname
FROM Reserves R, Sailors S
WHERE R.sid=S.sid AND
Example
R.bid=100 AND S.rating>5

sname
sname
sname
rating > 5

bid=100 rating > 5

sid=sid
sid=sid

bid=100 Sailors bid=100 rating > 5

sid=sid

Reserves Reserves Sailors

Sailors Reserves
Logical plans
CS3223: Query Optimization 5
 SELECT S.sname
FROM Reserves R, Sailors S
WHERE R.sid=S.sid AND
Example (Cont)
R.bid=100 AND S.rating>5
Query Evaluation Plan:
• Cost: 500+500*1000 I/Os (On-the-fly)
sname
• Memory: 3

bid=100 rating > 5 (On-the-fly)

(Page Nested Loops)

sid=sid

Sailors Reserves

Physical plan
CS3223: Query Optimization 6
 Alternative Plan 1 (No Indexes)
• Main difference: push selections down
• Assume 5 buffers, T1 = 10 pages (100 boats,
uniform distribution), T2 = 250 pages (10 ratings,
(On-the-fly)
uniform distribution) sname

• Cost of plan:
• Scan Reserves (1000) + write temp T1 (10 pages, if we (Sort-Merge)
have 100 boats, uniform distribution) sid=sid
• Scan Sailors (500) + write temp T2 (250 pages, if we
(T1) (T2)
have 10 ratings) bid=100 rating > 5
• Sort T1 (2*2*10), sort T2 (2*4*250), merge (10+250)
• Total: 4060 page I/Os Reserves Sailors

CS3223: Query Optimization 7

 Alternative Plan 2 (With Indexes)
• Clustered index on bid of Reserves
• 100,000/100 = 1000 tuples on 1000/100 = 10 pages
• Hash index on sid (format 2). Join column sid is a key sname
(On-the-fly)
for Sailors
• INL with pipelining (outer is not materialized) rating > 5 (On-the-fly)
• Project out unnecessary fields from outer doesn’t help
• At most one matching tuple, unclustered index (INL
sid=sid with pipelining )
on sid OK
• Did not push “rating>5” before the join. Why? bid=100 Sailors
• Cost? (Use hash
index; do
not write
• Selection of Reserves tuples (10 I/Os); for each, must Reserves result to
temp)
get matching Sailors tuple (1000*2.2); total 2210 I/Os

CS3223: Query Optimization 8

 Overview of Query Optimization
SQL query
parse
parse tree
convert
answer
logical query plan
execute
apply laws
Pi
“improved” l.q.p
estimate result sizes pick best
l.q.p. +sizes {P1,C1>...}
consider physical plans estimate costs
{P1,P2,…..}
CS3223: Query Optimization 9
 Example: SQL query

SELECT sname
FROM Sailors
WHERE sid IN (
SELECT sid
FROM Reserves
WHERE rname LIKE ‘Tan%’
);

(Find names of sailors whose reservation is made by

someone whose name begins with “Tan”)
CS3223: Query Optimization 10
 Example: Parse Tree
<Query>

<SFW>

SELECT <SelList> FROM <FromList> WHERE <Condition>

<Attribute> <RelName> <Tuple> IN <Query>

sname Sailors <Attribute> ( <Query> )

sid <SFW>

SELECT <SelList> FROM <FromList> WHERE <Condition>

<Attribute> <RelName> <Attribute> LIKE <Pattern>

sid Reserves rname ‘Tan%’

CS3223: Query Optimization 11
 Example: Improved Logical
Logical Query Plan Query Plan
sname sname
sid=sid
sid=sid

Sailors sid Sailors sid
rname LIKE ‘Tan%’ rname LIKE ‘TAN%’
Reserves Can we improve further?
Reserves
e.g. Push project to Sailors?
CS3223: Query Optimization 12
 Example: Estimate Result Sizes

Need expected size

Sailors 


Reserves

CS3223: Query Optimization 13

 Example: One Physical Plan

Hash join Parameters: join order,

memory size, project attributes,...

SEQ scan index scan

Parameters:
Select Condition,...

Sailors Reserves

CS3223: Query Optimization 14

 Example: Estimate Plan Cost to
Find Optimal QEP
QEPs

P1 P2 …. Pn

C1 C2 …. Cn

Pick best!
CS3223: Query Optimization 15
 Relational Algebra Equivalences
What about p1p2 … pn(R)?

CS3223: Query Optimization 16

 Relational Algebra Equivalences

CS3223: Query Optimization 17

 Bags vs. Sets
R = {a,a,b,b,b,c}
S = {b,b,c,c,d}
RS=?

• Option 1 SUM
R  S = {a,a,b,b,b,b,b,c,c,c,d}
• Option 2 MAX
R  S = {a,a,b,b,b,c,c,d}

CS3223: Query Optimization 18

 “SUM” is implemented
• Use “SUM” option for bag unions
• Some rules cannot be used for bags
• e.g. A s (B s C) = (A s B) s (A s C)

Let A, B and C be {x}

B B C = {x, x} A B (B B C) = {x}
A B B = {x} A B C = {x} (A B B) B (A B C) = {x, x}

CS3223: Query Optimization 19

 Query Optimizer
• Find the “best” plan (more often avoid the bad plans)
• Comprises the following
• Plan space
• huge number of alternative, semantically equivalent plans
• computationally expensive to examine all
• Conventional wisdom: avoid bad plans
• need to include plans that have low cost
• Enumeration algorithm (Search space)
• search strategy (optimization algorithm) that searches through the plan
space
• has to be efficient (low optimization overhead)
• Cost model
• facilitate comparisons of alternative plans
• has to be “accurate”
CS3223: Query Optimization 20
 Join Plan Notation

CS3223: Query Optimization 21

 Plan Space
• Left-deep trees: right child has to be a base table
• Right-deep trees: left child has to be a base table
• Deep trees: one of the two children is a base table
• Bushy tree: unrestricted
space of bushy tree is the largest,
it also contains left/right/deep trees

D D

C C

A B C D A B A B
Bushy tree Left-deep tree Deep tree
in general, good plans can be found in left/deep tree
CS3223: Query Optimization 22
 Query Plan Space for R S T

This has not accounted for the algorithms!

CS3223: Query Optimization 23
 Search Algorithms (and Search Space)
• Exhaustive (Complete space)
• enumerate each possible plan, and pick the best
• Greedy Techniques (Very small – polynomial)
• smallest relation next
• smallest result next
• typically polynomial time complexity
• Randomized/Transformation Techniques (Large space – can be
complete if you run the algorithms indefinitely)
• System R approach (Almost complete)
• Dynamic Programming with Pruning

CS3223: Query Optimization 24

 Multi-Join Queries
• Focus on multi-join queries first means dont consider selection/projection

• Join is the most expensive operations

• Selections and projections can be pushed down as early as
possible
• Query
• A query graph whose nodes are relations and edges represent a
join condition between the two nodes

CS3223: Query Optimization 25

 Greedy Algorithm (Example)
• Heuristic 1: Smallest relation next
• Suppose Ri < Rk for i < k
R1

All plans must begin with R1

R2
small relation --> table will be small
R1 R3 --> intermediate result will be small

R5 R4 R2
R3 R4 R5
Join Graph All plans beginning with R2-R5 have been pruned!

CS3223: Query Optimization 26

 Greedy Algorithm (Example)
• Smallest relation next
• What if R1 < R5 < R3 < R2 < R4???
using "smallest relation next" will result in large cross products and large
intermediate table

R2
Another heuristic:
R1 R3
Smallest result next?

R5 R4

CS3223: Query Optimization 27

 plan space
algorithm

Cost Models
cost model

• Typically, a combination of CPU and I/O costs

• Objective is to be able to rank plans
• exact value is not necessary
• Relies on
• statistics on relations and indexes
• formulas to estimate CPU and I/O cost
• formulas to estimate selectivities of operators and intermediate
results

CS3223: Query Optimization 28

 Cost Estimation
• For each plan considered, must estimate cost:
• Must estimate cost of each operation in plan tree
• Depends on input cardinalities
• Depends on buffer size, availability of indexes, algorithms used,
etc. ?
• We’ve already discussed how to estimate the cost of
operations (sequential scan, index scan, joins, etc.) ? R4

• Must estimate size of result for each operation in ?

R3
tree!
R1 R2
• Use information about the input relations
• Typical assumptions like uniform distribution of data and
independence of predicates can simplify size estimation but is
error prone

CS3223: Query Optimization 29

 Statistics and Catalogs
• Need information about the relations and indexes involved
Catalogs typically contain at least:
• # tuples of R (||R||), #bytes in each R tuple (S(R))
• # blocks/pages to hold all R tuples (|R|)
• # distinct values in R for attribute A (V(R,A))
• NPages for each index
• Index height, low/high key values (Low/High) for each tree index
• Catalogs updated periodically
• Updating whenever data changes is too expensive; lots of
approximation anyway, so slight inconsistency ok

CS3223: Query Optimization 30

 Estimation Assumptions
• Uniformity assumption
• Uniform distribution of attribute values
• Independence assumption
• Independent distribution of values in different attributes
• Inclusion assumption
• For R R.A=S.B S, if V(R, A) ≤ V(R, B), then A(R)  B(S)
• V(R, A) is the number of distinct values of R.A
assume key-foreign key relationship

CS3223: Query Optimization 31

 Example
R A B C D A: 20 byte string
cat 1 10 a B: 4 byte integer
cat 1 20 b
dog 1 30 a
C: 8 byte string
dog 1 40 c D: 5 byte string
bat 1 50 d

||R|| = 5 S(R) = 37
V(R,A) = 3 V(R,C) = 5
V(R,B) = 1 V(R,D) = 4
CS3223: Query Optimization 32
 Size estimate for W = Z=val (R)
||R||
R A B C D V(R,A)=3 ||W|| =
V(R,Z)
cat 1 10 a V(R,B)=1
cat 1 20 b
V(R,C)=5 S(W) = S(R)
dog 1 30 a
dog 1 40 c V(R,D)=4
bat 1 50 d
Assumption:
Values in select expression Z = val are uniformly
distributed over possible V(R,Z) values
Alternative assumption: use DOM(R,Z)
CS3223: Query Optimization 33
 What about W = z  val (R)?
Solution: Estimate values in range
R Z
Min=1 V(R,Z)=10

W= z  15 (R)
Max=20

20-15+1
f (fraction of range) = = 6 ||W|| = f  ||R||
20-1+1 20

Alternative: (Max(Z)-value)/(Max(Z)-Min(Z))
CS3223: Query Optimization 34
 W = R1 R2
R1 A B C R2 A D

Assumption:
V(R1,A)  V(R2,A)  Every A value in R1 is in R2
V(R2,A)  V(R1,A)  Every A value in R2 is in R1

“containment of value sets”

CS3223: Query Optimization 35

Computing T(W) when V(R1,A)  V(R2,A)

R1 A B C R2 A D
Take
1 tuple Match

1 tuple of R1 matches with ||R2|| tuples of R2

V(R2,A)
||R2||
so ||W|| =||R1|| 
V(R2,A)
||R2||  ||R1||
If V(R2,A)  V(R1,A) ||W|| =
V(R1,A)
CS3223: Query Optimization 36
 For complex expressions, need
intermediate T,S,V results
E.g. W = [A=a (R1) ] R2

Treat as relation U

||U|| = ||R1||/V(R1,A) S(U) = S(R1)

Also need V (U, *) !!

CS3223: Query Optimization 37
 Example
R1 A B C D V(R1,A)=3
cat 1 10 10 V(R1,B)=1
cat 1 20 20 V(R1,C)=5
dog 1 30 10 V(R1,D)=3
dog 1 40 30 U =  A=a (R1)
bat 1 50 10
||R1||
V(U,A) = 1? V(U, B) = 1? (= V(R, B)) V(U,C) = ?
V(R1,A)

V(D,U) … somewhere in between V(U,B) and V(U,C)

CS3223: Query Optimization 38
 For Joins U = R1(A,B) R2(A,C)
V(U,A) = min { V(R1, A), V(R2, A) }
V(U,B) = V(R1, B)
V(U,C) = V(R2, C)

(Assumption: Preservation of value sets)

CS3223: Query Optimization 39

 Example

Z = R1(A,B) R2(B,C) R3(C,D)

R1 ||R1|| = 1000 V(R1,A)=50 V(R1,B)=100

R2 ||R2|| = 2000 V(R2,B)=200 V(R2,C)=300
R3 ||R3|| = 3000 V(R3,C)=90 V(R3,D)=500

CS3223: Query Optimization 40

 Z = R1(A,B) R2(B,C) R3(C,D)
||R1|| = 1000 V(R1,A)=50 V(R1,B)=100
Partial Result: U = R1 R2 ||R2|| = 2000 V(R2,B)=200 V(R2,C)=300
||R3|| = 3000 V(R3,C)=90 V(R3,D)=500

||U|| = 10002000 V(U,A) = 50

200 V(U,B) = 100
V(U,C) = 300

Z=U R3
V(Z,A) = 50
||Z|| = 100020003000 V(Z,B) = 100
200300 V(Z,C) = 90
V(Z,D) = 500
CS3223: Query Optimization 41
 Errors in Estimating Size of Plan
• Errors
• source include uniformity assumption, and inability to
capture correlation, accuracy of cost model, statistics, etc.
• propagated to other operators at the higher level of the plan R4
tree
R3
• Dealing with errors
R1 R2
• Maintain more detailed statistics (at finer granularity)
• During runtime, may need to sample the actual intermediate
results
• dynamic query optimization

CS3223: Query Optimization 42

 Statistical Summaries of Data
• More detailed information are sometimes stored e.g.,
histograms of the values in some attributes
• a histogram divides the values on a column into k buckets
• k is predetermined or computed based on space allocation
• several choices for “bucketization’’ of values
• If a table has n records, an equi-depth histogram divides the set of
values on a column into k ranges such that each range has
approximately the same number of records, i.e., n/k
• Equi-width histogram – each bucket has (almost) equal number of
values
• Witihin each bucket, records are uniformly distributed across the range
of the bucket
• Frequently occurring values may be placed in singleton buckets
CS3223: Query Optimization 43
 Histograms
15 records are uniformly
distributed

3.5
3.0
2.25

CS3223: Query Optimization 44

 Estimations with Histograms

3.5
3.0
2.25

Query Q:  A=6 (R) Query Q:  A=10 (R)

Actual value, ||Q|| = 3 Actual value, ||Q|| = 0

Without histogram, ||Q|| = 45/15 = 3 Without histogram, ||Q|| = 45/15 = 3
Equiwidth histogram, ||Q|| = 15/3 = 5 Equiwidth histogram, ||Q|| = 1
Equidepth histogram, ||Q|| = 14/4 = 3.5 Equidepth histogram, ||Q|| = 1.75

CS3223: Query Optimization 45

 Query Evaluation

CS3223: Query Optimization 46

 Query Evaluation Approaches
• Materialization evaluation
• An operator is evaluated only when each of its operands
has been completely evaluated or materialized
• Intermediate results are materialized to disk
• Pipelining evaluation
• The output produced by an operator is passed directly to
its parent operator
• Execution of operators is interleaved

CS3223: Query Optimization 47

 Materialization Evaluation

CS3223: Query Optimization 48

 Pipeline Evaluation: Iterator Model
consumer
Open()

Open()

Open() Open()

CS3223: Query Optimization 49

 Pipeline Evaluation: Iterator Model
consumer
GetNext() t

All operators are

GetNext() t
running simultaneously

GetNext() GetNext()
tm ta

CS3223: Query Optimization 50

 Pipeline Evaluation with Partial Materialization

Plan 1
Plan 2
CS3223: Query Optimization 51
 Randomized Techniques
• Employ randomized/transformation techniques for query
optimization
• State space -- space of plans, State -- plan
• Each state has a cost associated with it
• determined by some cost model
• A move is a perturbation applied to a state to get to another
state
• a move set is the set of moves available to go from one state to another
• any one move is chosen from this move set randomly
• each move set has a probability associated to indicate the probability of selecting the move
• Two states are neighboring states if one move suffices to go
from one state to the other
CS3223: Query Optimization 52
 Randomized Algorithm (Example)
State/QEP 1
R4 R4 Neighbor
R3 R3 of QEP 1

R1 R2 R2 R1

A move

Neighbor R4 R4
of QEP 1
R3 R3
R1 R2 R2 R1

CS3223: Query Optimization 53

 More on Randomized Techniques
• A local minimum in the state space is a state such that its
cost is lower than that of all neighboring states
• A global minimum is a state which has the lowest cost
among all local minima
• at most one global minimum
• A move that takes one state to another state with a lower
cost is called a downward move; otherwise it is an upward
move
• in a local/global minimum, all moves are upward moves

CS3223: Query Optimization 54

 Local Optimization
Repeat until
a near-optimal S = initialize() // initial plan
minimum
Is reached
minS = S // cost of plan S – currently the best
repeat {
repeat {
By doing so newS = move(S) // move to a new plan
repeatedly,
a local minimum if (cost(newS) < cost(S)) A move is accepted if it is a
downward move, i.e., has
can S = newS a lower cost
be reached } until (“local minimum reached”)
if (cost(S) < cost(minS))
minS = S
newStart(S); // iterate with a different initial plan
} until (“stopping condition satisfied”)
return (minS);
CS3223: Query Optimization 55
 Issues on Local Optimization
• How is the start state obtained?
Run: sequence of
• The state in which we start a run
moves to a local
• The start state of the first run is the initial state minimum from the
• All start states should be different start state
• Should be obtained quickly
• Random
• greedy heuristics
• making a number of moves from the local minimum, except that this time each move
is accepted irrespective of whether it increases or decreases the cost
• How is the local minimum detected? need to move to a plan with a higher
cost in order to jump out of the local
minimum
• How is the stopping criterion detected?

CS3223: Query Optimization 56

 Issues on Local Optimization (Cont)
• How is the local minimum detected?
• Not practical to examine all neighbors to verify that one has
reached a local minimum
• Based on random sampling
• examine a sufficiently large number of neighbors
• if any one is lower, we move to that state, and repeat the process
• if no tested neighbor is of lower cost, the current state can be considered a local
minimum
• the number of neighbors to examine can be specified as a parameter,
and is called the sequence length
• Can also be time-based

CS3223: Query Optimization 57

 Issues in Local Optimization (Cont)

• How is the stopping criterion detected?

• Determines the number of times that the outer loop is
executed
• Can be fixed and is given by sizeFactor*N, where
sizeFactor is a parameter, N is the number of relations
• Why N? Can it be a constant?
if you have a 4 table join you might not need 500 initial plans.
if you have a 500 table join, 100 initial plans likely will not find an optimal plan

CS3223: Query Optimization 58

 What about the MOVEs?
Transformation Rules
• Restricted to left-deep trees
• all possible permutations of the N relations
• let S be the current state, S = (… i … j … k …)
• swap
• select two relations, say i and j at random. Swapped i and j to get the new
state newS = ( … j … i … k … ) might result in a cross product being removed, hence
reducing the cost
• 3Cycle
• select three relations, say i and j and k at random. The move consists of
cycling i, j and k: i is moved to the position of j, j is moved to the position of k
and k is moved to the position of i. The resultant new state newS = ( … k …
i…j…)
• Other methods (e.g., join methods)? Bushy trees?

CS3223: Query Optimization 59

 Comparison between Exhaustive,
Greedy and Randomized Algorithms
randomised algorithm might have high overhead
if it keeps repeating the same few plans.
• Search space
• Plan quality
• Optimization overhead

CS3223: Query Optimization 60

Dynamic Programming (Left-Deep Trees)
(System R)
• The algorithm proceeds by considering increasingly larger
subsets of the set of all relations
• Builds a plan bottom-up (beginning from 1 table, then to 2, and so on)
• Plans for a set of cardinality i are constructed as extensions of the
best plan for a set of cardinality i-1
• For each set of cardinality i, we only keep ONE best plan

CS3223: Query Optimization 61

 Dynamic Programming (Cont)
{}

{1} {2} {3} {4}

{1 2} {1 3} {1 4} {2 3} {2 4} {3 4}

{1 2 3} {1 2 4} {2 3 4} {1 3 4}

{1 2 3 4}
CS3223: Query Optimization 62
Dynamic Programming (Left-Deep Trees)
(System R)
• The algorithm proceeds by considering increasingly larger
subsets of the set of all relations
• Builds a plan bottom-up (beginning from 1 table, then to 2, and so on)
• Plans for a set of cardinality i are constructed as extensions of the best
plan for a set of cardinality i-1
• Keep only ONE best plan for each set of cardinality i
• Search space can be pruned based on the principal of optimality
• if two plans differ only in a subplan, then the plan with the better subplan is
also the better plan

CS3223: Query Optimization 63

 Principle of Optimality
{}

{1} {2} {3} {4}

{1 2} {1 3} {1 4} {2 3} {2 4} {3 4}

{1 2 3} {1 2 4} {2 3 4} {1 3 4}

{1 2 3 4}

CS3223: Query Optimization 64

Dynamic Programming (Left-Deep Trees)
(System R)
• The algorithm proceeds by considering increasingly larger subsets of
the set of all relations
• Builds a plan bottom-up (beginning from 1 table, then to 2, and so on)
• Plans for a set of cardinality i are constructed as extensions of the best plan for a set of
cardinality i-1
• Keep only ONE best plan for each set of cardinality i
• Search space can be pruned based on the principal of optimality
• if two plans differ only in a subplan, then the plan with the better subplan is also the better
plan
• Computation overhead reduced due to overlapping subproblems
• Multiple sets of cardinality i uses same set at cardinality (i-1)

CS3223: Query Optimization 65

Dynamic Programming (Left-Deep Trees)
• accessPlan(R) produces the best plan for relation (single
table) R
• joinPlan(p1,R) extends the (partial) join plan p1 into another
plan p2 in which the result of p1 is joined with R in the best
possible way
• p1 = R1 JOIN R2 JOIN R3
• p2 = joinPlan(p1, R) = (R1 JOIN R2 JOIN R3) JOIN R4
• Optimal plans for subsets are stored in optplan() array and
are reused rather than recomputed
CS3223: Query Optimization 66
 Dynamic Programming (Cont)
for i = 1 to N
optPlan({Ri}) = accessPlan(Ri)
for i = 2 to N {
forall S subset of {R1, R2, … Rn} such that |S|=i {
bestPlan = dummy plan with infinite cost
forall Rj, Sj, |Sj| = i-1 such that S = {Rj} U Sj {
p = joinPlan(optPlan(Sj), Rj) {}

if cost(p) < cost(bestPlan) {1} {2} {3} {4}

bestPlan = p
} {1 2} {1 3} {1 4} {2 3} {2 4} {3 4}

optPlan(S) = bestPlan
} {1 2 3} {1 2 4} {2 3 4} {1 3 4}

}
Popt = optPlan{R1, R2, … Rn} {1 2 3 4}

CS3223: Query Optimization 67

 Dynamic Programming: A Concrete Example

CS3223: Query Optimization 68

 Enumeration of Single-relation Plans

CS3223: Query Optimization 69

 Enumeration of Two-Relation Plans
s and t not considered because we want to avoid cross products

CS3223: Query Optimization 70

 Enumeration of Three-Relation Plans

CS3223: Query Optimization 71

 Optimal Plan

CS3223: Query Optimization 72

 Dynamic Programming (Cont)
• Time & Space complexity
• For k relations, for left-deep trees, 2k – 1 entries!
DP is not used when k exceeds some value
• For bushy trees, O(3k) (about 15-20)

• Is DP (as presented) optimal?

not optimal if we also consider bushy trees, since DP doesnt include bushy trees

not optimal even within the subset of left-deep trees

CS3223: Query Optimization 73

 Dynamic Programming (Cont)
• Time & Space complexity
• For k relations, for left-deep trees, 2k – 1 entries!
• For bushy trees, O(3k)
• Is DP optimal?
• DP may maintain multiple plans per subset of
relations
plans with properties that are beneficial in the future
• Interesting orders might be pruned away because they are not optimal now

plans that are sorted with a certain ordering, for example.

queries that have ordering
and group by. we might want to
keep plans that result in these
relations being sorted. more computataion overhead.

CS3223: Query Optimization 74

CS3223: Query Optimization 75
 Dynamic Programming (Cont)
• Time & Space complexity
• For k relations, for left-deep trees, 2k – 1 entries!
• For bushy trees, O(3k)
• Is DP optimal?
• DP may maintain multiple plans per subset of
relations
• Interesting orders
• Is DP with interesting orders optimal?
gives the optimal query with respect to total cost, but not for other
cost models. but will still be a good solution.
CS3223: Query Optimization 76
 Summary
• Query optimization is NP-hard
• Instead of finding the best, the objective is largely to
avoid the bad plans
• Many different optimization strategies have been
proposed
• greedy heuristics are fast but may generate plans that are
far from optimal
• dynamic programming is effective at the expense of high
optimization overhead

CS3223: Query Optimization 77