Outline

n  Introduction & architectural issues

n  Data distribution
n  Distributed query processing
q Distributed query optimization
q Distributed transactions & concurrency control
q Distributed reliability
q Data replication
q Parallel database systems
q Database integration & querying
q Peer-to-Peer data management
q Stream data management
q MapReduce-based distributed data management

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 1

Distributed Query
Processing Methodology
Calculus Query on Distributed
Relations

Query GLOBAL
Decomposition SCHEMA

Algebraic Query on Distributed

Relations
CONTROL FRAGMENT
SITE Data
Localization SCHEMA

Fragment Query

Global STATS ON
Optimization FRAGMENTS

Optimized Fragment Query

with Communication Operations

Local LOCAL
LOCAL
Optimization SCHEMAS
SITES

Optimized Local
Queries

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 2

Page 1
 Step 3 – Global Query
Optimization

Input: Fragment query

n  Find the best (not necessarily optimal) global
schedule
l  Minimize a cost function
l  Distributed join processing
u  Bushy vs. linear trees
u  Which relation to ship where?
u  Ship-whole vs ship-as-needed
l  Decide on the use of semijoins
u  Semijoin saves on communication at the expense of
more local processing.
l  Join methods
u  nested loop vs ordered joins (merge join or hash join)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 3

Cost-Based Optimization
n  Solution space
l  The set of equivalent algebra expressions (query trees).

n  Cost function (in terms of time)

l  I/O cost + CPU cost + communication cost
l  These might have different weights in different distributed
environments (LAN vs WAN).
l  Can also maximize throughput

n  Search algorithm

l  How do we move inside the solution space?
l  Exhaustive search, heuristic algorithms (iterative
improvement, simulated annealing, genetic,…)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 4

Page 2
 Query Optimization Process

Input Query

Search Space Transformation

Generation Rules

Equivalent QEP

Search Cost Model

Strategy

Best QEP

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 5

Search Space
▷◁PNO
n  Search space
characterized by ▷◁ENO PROJ
alternative execution
n  Focus on join trees EMP ASG
n  For N relations, there are ▷◁ENO
O(N!) equivalent join
trees that can be obtained ▷◁PNO EMP
by applying
commutativity and PROJ ASG
associativity rules ▷◁ENO,PNO
SELECT ENAME,RESP
FROM EMP, ASG,PROJ ×
ASG
WHERE EMP.ENO=ASG.ENO
AND ASG.PNO=PROJ.PNO PROJ EMP
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 6

Page 3
 Search Space
n  Restrict by means of heuristics
º  Perform unary operations before binary operations
º  …
n  Restrict the shape of the join tree
l  Consider only linear trees, ignore bushy ones

Linear Join Tree Bushy Join Tree

⋈
⋈
⋈ R4

⋈ R3 ⋈ ⋈
R1 R2 R1 R2 R3 R4

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 7

Search Strategy
n  How to “move” in the search space.
n  Deterministic
º  Start from base relations and build plans by adding one
relation at each step
º  Dynamic programming: breadth-first
º  Greedy: depth-first
n  Randomized
º  Search for optimalities around a particular starting point
º  Trade optimization time for execution time
º  Better when > 10 relations
º  Simulated annealing
º  Iterative improvement

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 8

Page 4
 Search Strategies

n  Deterministic ⋈
⋈ ⋈ R4

⋈ ⋈ R3 ⋈ R3

R1 R2 R1 R2 R1 R2

n  Randomized
⋈ ⋈
⋈ R3 ⋈ R2

R1 R2 R1 R3

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 9

Cost Functions
n  Total Time (or Total Cost)
l  Reduce each cost (in terms of time) component individually
l  Do as little of each cost component as possible
l  Optimizes the utilization of the resources

Increases system throughput

n  Response Time

l  Do as many things as possible in parallel
l  May increase total time because of increased total activity

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 10

Page 5
 Total Cost

Summation of all cost factors

Total cost = CPU cost + I/O cost + communication cost

CPU cost = unit instruction cost * no.of instructions

I/O cost = unit disk I/O cost * no. of disk I/Os

communication cost = message initiation + transmission

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 11

Total Cost Factors

n  Wide area network
l  Message initiation and transmission costs high

l  Local processing cost is low (fast mainframes or

minicomputers)
l  Ratio of communication to I/O costs = 20:1

n  Local area networks

l  Communication and local processing costs are more
or less equal
l  Ratio = 1:1.6

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 12

Page 6
 Response Time

Elapsed time between the initiation and the

completion of a query
Response time = CPU time + I/O time +
communication time

CPU time = unit instruction time * no. of

sequential instructions

I/O time = unit I/O time * no. of sequential I/Os

communication time = unit msg initiation time * no. of

sequential msg
+ unit transmission time * no. of
sequential bytes
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 13

Example
Site 1
x units

Site 3

Site 2 y units

Assume that only the communication cost is considered

Total time = 2 ⋅ message initialization time + unit transmission time *
(x+y)
Response time = max {time to send x from 1 to 3, time to send y from 2
to 3}
time to send x from 1 to 3 = message initialization time
+ unit transmission time * x
time to send y from 2 to 3 = message initialization time
+ unit transmission time * y
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 14

Page 7
 Optimization Statistics
n  Primary cost factor: size of intermediate
relations
l  Need to estimate their sizes

n  Make them precise èmore costly to maintain

n  Simplifying assumption: uniform distribution
of attribute values in a relation

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 15

Statistics

n  For each relation R[A1, A2, …, An] fragmented as R1,

…, Rr
l  length of each attribute: length(Ai)
l  the number of distinct values for each attribute in each fragment:
card(ΠAiRj)
l  maximum and minimum values in the domain of each attribute:
min(Ai), max(Ai)
l  the cardinalities of each domain: card(dom[Ai])

n  The cardinalities of each fragment: card(Rj) Selectivity

factor of each operation for relations
l For joins card(R ⋈S)
SF ⋈ (R,S) =
card(R) * card(S)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 16

Page 8
 Intermediate Relation Sizes

Selection
size(R) = card(R) ⋅ length(R)
card(σF(R)) = SFσ(F) ⋅ card(R)
where
1
S Fσ(A = value) =
card(∏A(R))
max(A) – value
S Fσ(A >value) =
max(A) – min(A)
value – max(A)
S Fσ(A <value) =
max(A) – min(A)
SFσ(p(Ai)∧ p(Aj)) = SFσ(p(Ai)) ⋅ SFσ(p(Aj))

SFσ(p(Ai) ∨ p(Aj)) = SFσ(p(Ai)) + SFσ(p(Aj)) – (SFσ(p(Ai)) ⋅ SFσ(p(Aj)))

SFσ(A∈{value}) = SFσ(A= value) * card({values})

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 17

Intermediate Relation Sizes

Projection
card(ΠA(R))=card(R)

Cartesian Product
card(R × S) = card(R) * card(S)

Union
upper bound: card(R ∪ S) = card(R) + card(S)
lower bound: card(R ∪ S) = max{card(R), card(S)}

Set Difference
upper bound: card(R–S) = card(R)
lower bound: 0
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 18

Page 9
 Intermediate Relation Size
Join
l  Special case: A is a key of R and B is a foreign key of S
card(R ⋈A=B S) = card(S)
l  More general:
card(R ⋈ S) = SF⋈ * card(R) ⋅ card(S)

Semijoin
card(R ⋉A S) = SF⋉(S.A) * card(R)
where
card(∏A(S))
SF⋉(R ⋉A S)= SF⋉(S.A) = card(dom[A])

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 19

Histograms for Selectivity

Estimation
n  For skewed data, the uniform distribution assumption
of attribute values yields inaccurate estimations
n  Use an histogram for each skewed attribute A
l  Histogram = set of buckets
u  Each bucket describes a range of values of A, with its average
frequency f (number of tuples with A in that range) and number
of distinct values d
u  Buckets can be adjusted to different ranges
n  Examples
l  Equality predicate
u  With (value in Rangei), we have: SFσ(A = value) = 1/di
l  Range predicate
u  Requiresidentifying relevant buckets and summing up their
frequencies
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 20

Page 10
 Histogram Example

For ASG.DUR=18: we have SF=1/12 so the card of selection is 300/12

= 25 tuples

For ASG.DUR≤18: we have min(range3)=12 and max(range3)=24 so

the card. of selection is 100+75+(((18−12)/(24 − 12))*50) = 200 tuples
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 21

Centralized Query Optimization

n  Dynamic (Ingres project at UCB)
l  Interpretive

n  Static (System R project at IBM)

l  Exhaustive search

n  Hybrid (Volcano project at OGI)

l  Choose node within plan

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 22

Page 11
 Dynamic Algorithm
Œ Decompose each multi-variable query into a
sequence of mono-variable queries with a
common variable
 Process each by a one variable query processor
l  Choose an initial execution plan (heuristics)
l  Order the rest by considering intermediate relation sizes

No statistical information is maintained

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 23

Dynamic Algorithm–Decomposition
n  Replace an n variable query q by a series of
queries
q1→q2 → … → qn
where qi uses the result of qi-1.
n  Detachment
l  Query q decomposed into q' → q" where q' and q" have a
common variable which is the result of q'
n  Tuple substitution
l  Replace the value of each tuple with actual values and
simplify the query
q(V1, V2, ... Vn) → (q' (t1, V2, V2, ... , Vn), t1∈R)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 24

Page 12
 Detachment

q: SELECT V2.A2,V3.A3, …,Vn.An

FROM R1 V1, …,Rn Vn
WHERE P1(V1.A1’)AND P2(V1.A1,V2.A2,…, Vn.An)

⇓

q':SELECT V1.A1 INTO R1'
FROM R1 V1
WHERE P1(V1.A1)

q": SELECT V2.A2, …,Vn.An

FROM R1' V1, R2 V2, …,Rn Vn
WHERE P2(V1.A1, V2.A2, …,Vn.An)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 25

Detachment Example
Names of employees working on CAD/CAM project
q 1: SELECT EMP.ENAME
FROM EMP, ASG, PROJ
WHERE EMP.ENO=ASG.ENO
AND ASG.PNO=PROJ.PNO
AND PROJ.PNAME="CAD/CAM"



⇓

q11: SELECT PROJ.PNO INTO JVAR
FROM PROJ
WHERE PROJ.PNAME="CAD/CAM"

q': SELECT EMP.ENAME

FROM EMP,ASG,JVAR
WHERE EMP.ENO=ASG.ENO
AND ASG.PNO=JVAR.PNO
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 26

Page 13
 Detachment Example (cont’d)
q': SELECT EMP.ENAME
FROM EMP,ASG,JVAR
WHERE EMP.ENO=ASG.ENO
AND ASG.PNO=JVAR.PNO

⇓

q12: SELECT ASG.ENO INTO GVAR
FROM ASG,JVAR
WHERE ASG.PNO=JVAR.PNO

q13: SELECT EMP.ENAME

FROM EMP,GVAR
WHERE EMP.ENO=GVAR.ENO

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 27

Tuple Substitution

q11 is a mono-variable query

q12 and q13 is subject to tuple substitution
Assume GVAR has two tuples only: 〈E1〉 and 〈E2〉
Then q13 becomes
q131: SELECT EMP.ENAME
FROM EMP
WHERE EMP.ENO="E1"

q132: SELECT EMP.ENAME

FROM EMP
WHERE EMP.ENO="E2"

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 28

Page 14
 Static Algorithm

Œ Simple (i.e., mono-relation) queries are

executed according to the best access path

 Execute joins
l  Determine the possible ordering of joins

l  Determine the cost of each ordering

l  Choose the join ordering with minimal cost

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 29

Static Algorithm
For joins, two alternative algorithms :
n  Nested loops
for each tuple of external relation (cardinality n1)
for each tuple of internal relation (cardinality n2)
join two tuples if the join predicate is true
end
end
l  Complexity: n1* n2
n  Merge join
sort relations
merge relations
l  Complexity: n1+ n2 if relations are previously sorted and
equijoin

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 30

Page 15
 Static Algorithm – Example

Names of employees working on the CAD/CAM

project
Assume
l  EMP has an index on ENO,
l  ASG has an index on PNO,
l  PROJ has an index on PNO and an index on PNAME

ASG

ENO PNO

EMP PROJ

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 31

Example (cont’d)
Œ Choose the best access paths to each relation
l  EMP: sequential scan (no selection on EMP)
l  ASG: sequential scan (no selection on ASG)
l  PROJ: index on PNAME (there is a selection on PROJ based on
PNAME)
 Determine the best join ordering
l  EMP ▷◁ ASG ▷◁ PROJ
l  ASG ▷◁PROJ ▷◁ EMP
l  PROJ ▷◁ASG ▷◁ EMP
l  ASG ▷◁EMP ▷◁ PROJ
l  EMP × PROJ ▷◁ ASG
l  PRO × JEMP ▷◁ASG
l  Select the best ordering based on the join costs evaluated according to
the two methods

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 32

Page 16
 Static Algorithm
Alternatives

EMP ASG PROJ

EMP ⋈ ASG EMP × PROJ ASG ⋈ EMP ASG ⋈ PROJ PROJ ⋈ ASG PROJ × EMP
pruned pruned pruned pruned

(ASG ⋈ EMP) ⋈ PROJ (PROJ ⋈ ASG) ⋈ EMP

Best total join order is one of

((ASG ⋈ EMP) ⋈ PROJ)
((PROJ ⋈ ASG) ⋈ EMP)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 33

Static Algorithm

n  ((PROJ ⋈ ASG) ⋈ EMP) has a useful index on

the select attribute and direct access to the join
attributes of ASG and EMP

n  Therefore, chose it with the following access

methods:
l  select PROJ using index on PNAME

l  then join with ASG using index on PNO

l  then join with EMP using index on ENO

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 34

Page 17
 Hybrid optimization
n  In general, static optimization is more efficient
than dynamic optimization
l  Adopted by all commercial DBMS
n  But even with a sophisticated cost model (with
histograms), accurate cost prediction is difficult
n  Example
l  Consider a parametric query with predicate
WHERE R.A = $a /* $a is a parameter
l  The only possible assumption at compile time is uniform
distribution of values
n  Solution: Hybrid optimization
l  Choose-plan done at runtime, based on the actual
parameter binding

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 35

Hybrid Optimization
Example

$a=A

$a=A

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 36

Page 18
 Join Ordering in Fragment
Queries

n  Ordering joins

l  Distributed INGRES

l  System R*

l  Two-step

n  Semijoin ordering

l  SDD-1

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 37

Join Ordering

• Consider two relations only

if size(R) < size(S)

R S
if size(R) > size(S)

n  Multiple relations more difficult because too many

alternatives.
l  Compute the cost of all alternatives and select the best one.
u  Necessary to compute the size of intermediate relations which is
difficult.
l  Use heuristics

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 38

Page 19
 Join Ordering – Example

Consider
PROJ ⋈PNO ASG ⋈ENO EMP

Site 2

ASG
ENO PNO

EMP PROJ
Site 1 Site 3

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 39

Join Ordering – Example

Execution alternatives:
1. EMP→ Site 2 2. ASG → Site 1
Site 2 computes EMP'=EMP ⋈ ASG Site 1 computes EMP'=EMP ⋈ ASG
EMP'→ Site 3 EMP' → Site 3
Site 3 computes EMP' ⋈ PROJ Site 3 computes EMP’ ⋈ PROJ

3. ASG → Site 3 4. PROJ → Site 2

Site 3 computes ASG'=ASG ⋈ PROJ Site 2 computes PROJ'=PROJ ⋈ ASG
ASG' → Site 1 PROJ' → Site 1
Site 1 computes ASG' ▷◁ EMP Site 1 computes PROJ' ⋈ EMP

5. EMP → Site 2
PROJ → Site 2
Site 2 computes EMP ⋈ PROJ ⋈ ASG
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 40

Page 20
 Semijoin Algorithms
n  Consider the join of two relations:
l  R[A] (located at site 1)
l  S[A](located at site 2)

n  Alternatives:
1.  Do the join R ⋈AS
2.  Perform one of the semijoin equivalents

R ⋈AS ⇔
(R ⋉AS) ⋈AS



⇔
R ⋈A (S ⋉A R)



⇔
(R ⋉A S) ⋈A (S ⋉A R)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 41

Semijoin Algorithms
n  Perform the join
l  send R to Site 2
l  Site 2 computes R ⋈A S
n  Consider semijoin (R ⋉AS) ⋈AS
l  S' = ΠA(S)
l  S' → Site 1
l  Site 1 computes R' = R ⋉AS'
l  R'→ Site 2
l  Site 2 computes R' ⋈AS

Semijoin is better if
size(ΠA(S)) + size(R ⋉AS)) < size(R)
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 42

Page 21
 Distributed Dynamic
Algorithm

1.  Execute all monorelation queries (e.g.,

selection, projection)
2.  Reduce the multirelation query to produce
irreducible subqueries
q1→ q2 → … → qnsuch that there is only one
relation between qi and qi+1
1.  Choose qi involving the smallest fragments to
execute (call MRQ')
2.  Find the best execution strategy for MRQ'
a)  Determine processing site
b)  Determine fragments to move
3.  Repeat 3 and 4
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 43

Static Approach

n  Cost function includes local processing as well

as transmission
n  Considers only joins

n  “Exhaustive” search

n  Compilation

n  Published papers provide solutions to handling

horizontal and vertical fragmentations but the
implemented prototype does not

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 44

Page 22
 Static Approach –
Performing Joins

n  Ship whole

l  Larger data transfer
l  Smaller number of messages
l  Better if relations are small
n  Fetch as needed
l  Number of messages = O(cardinality of external
relation)
l  Data transfer per message is minimal
l  Better if relations are large and the selectivity is
good

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 45

Static Approach –
Vertical Partitioning & Joins

1. Move outer relation tuples to the site of the inner

relation
(a) Retrieve outer tuples

(b) Send them to the inner relation site

(c) Join them as they arrive

Total Cost = cost(retrieving qualified outer tuples)

+ no. of outer tuples fetched *
cost(retrieving qualified inner tuples)
+ msg. cost * (no. outer tuples fetched * avg.
outer tuple size)/msg. size
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 46

Page 23
 Static Approach –
Vertical Partitioning & Joins
2. Move inner relation to the site of outer relation
Cannot join as they arrive; they need to be stored

Total cost = cost(retrieving qualified outer tuples)

+ no. of outer tuples fetched *
cost(retrieving matching inner tuples
from temporary storage)
+ cost(retrieving qualified inner tuples)
+ cost(storing all qualified inner tuples in
temporary storage)
+ msg. cost * no. of inner tuples fetched *
avg. inner tuple size/msg. size
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 47

Static Approach –
Vertical Partitioning & Joins

3. Move both inner and outer relations to another site

Total cost = cost(retrieving qualified outer tuples)
+ cost(retrieving qualified inner tuples)
+ cost(storing inner tuples in storage)
+ msg. cost ⋅ (no. of outer tuples fetched *
avg. outer tuple size)/msg. size
+ msg. cost * (no. of inner tuples fetched *
avg. inner tuple size)/msg. size
+ no. of outer tuples fetched * cost(retrieving
inner tuples from temporary storage)
CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 48

Page 24
 Static Approach –
Vertical Partitioning & Joins
4.Fetch inner tuples as needed
(a) Retrieve qualified tuples at outer relation site
(b) Send request containing join column value(s) for outer tuples to
inner relation site
(c) Retrieve matching inner tuples at inner relation site
(d) Send the matching inner tuples to outer relation site
(e) Join as they arrive
Total Cost = cost(retrieving qualified outer tuples)
+ msg. cost * (no. of outer tuples fetched)
+ no. of outer tuples fetched * no. of
inner tuples fetched * avg. inner tuple
size * msg. cost / msg. size)
+ no. of outer tuples fetched * cost(retrieving
matching inner tuples for one outer value)

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 49

Dynamic vs. Static vs Semijoin

n  Semijoin
l  SDD1 selects only locally optimal schedules
n  Dynamic and static approaches have the same
advantages and drawbacks as in centralized
case
l  But the problems of accurate cost estimation at compile-
time are more severe
u  More variations at runtime
u  Relations may be replicated, making site and copy
selection important
n  Hybrid optimization
l  Choose-plan approach can be used
l  2-step approach simpler

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 50

Page 25
 2-Step Optimization
1.  At compile time, generate a static plan with
operation ordering and access methods only
2.  At startup time, carry out site and copy
selection and allocate operations to sites

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 51

2-Step – Problem Definition

n  Given
l  A set of sites S = {s1, s2, …,sn} with the load of each site
l  A query Q ={q1, q2, q3, q4} such that each subquery qi is the
maximum processing unit that accesses one relation and
communicates with its neighboring queries
l  For each qi in Q, a feasible allocation set of sites Sq={s1, s2,
…,sk} where each site stores a copy of the relation in qi
n  The objective is to find an optimal allocation of
Q to S such that
l  the load unbalance of S is minimized
l  The total communication cost is minimized

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 52

Page 26
 2-Step Algorithm
n  For each q in Q compute load (Sq)
n  While Q not empty do
1.  Select subquery a with least allocation flexibility
2.  Select best site b for a (with least load and best benefit)
3.  Remove a from Q and recompute loads if needed

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 53

2-Step Algorithm Example

n  Let Q = {q1, q2, q3, q4} where
q1 is associated with R1, q2 is
associated with R2 joined
with the result of q1, etc.
n  Iteration 1: select q4, allocate
to s1, set load(s1)=2
n  Iteration 2: select q2, allocate
to s2, set load(s2)=3
n  Iteration 3: select q3, allocate
to s1, set load(s1) =3
n  Iteration 4: select q1, allocate
to s3 or s4

Note: if in iteration 2, q2, were allocated to s4, this would have produced a
better plan. So hybrid optimization can still miss optimal plans

CS742 – Distributed & Parallel DBMS M. Tamer Özsu Page 4. 54

Page 27