Outline

• Introduction
• Background
• Distributed Database Design
• Database Integration
• Semantic Data Control
• Distributed Query Processing
• Multidatabase Query Processing
➡ Query Rewriting
➡ Optimization Issues

• Distributed Transaction Management

• Data Replication
• Parallel Database Systems
• Distributed Object DBMS
• Peer-to-Peer Data Management
• Web Data Management
• Current Issues
Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/1
Multidatabase Query
Processing
•Mediator/wrapper architecture
•MDB query processing architecture
•Query rewriting using views
•Query optimization and execution
•Query translation and execution

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/2

Mediator/Wrapper
Architecture
Mediator
Same Different
Interface Interfaces
Query
Processing Local
Wrapper1 DBMS1
Schema

Query Global local

Wrapper2 DBMS2
view schema
Results
local
Wrapper3 DBMS3
schema
Result
Integration
DBMS4

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/3

Advantages of M/W
Architecture
• Wrappers encapsulate the details of component DBMS
➡ Export schema and cost information
➡ Manage communication with Mediator

• Mediator provides a global view to applications and users

➡ Single point of access
✦ May be itself distributed
➡ Can specialize in some application domain
➡ Perform query optimization using global knowledge
➡ Perform result integration in a single format

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/4

Issues in MDB Query
Processing
• Component DBMSs are autonomous and may range from full-fledge
relational DBMS to flat file systems
➡ Different computing capabilities
✦ Prevents uniform treatment of queries across DBMSs
➡ Different processing cost and optimization capabilities
✦ Makes cost modeling difficult
➡ Different data models and query languages
✦ Makes query translation and result integration difficult
➡ Different runtime performance and unpredictable behavior
✦ Makes query execution difficult

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/5

Component DBMS Autonomy
• Communication autonomy
➡ The ability to terminate services at any time
➡ How to answer queries completely?

• Design autonomy
➡ The ability to restrict the availability and accuracy of information needed for
query optimization
➡ How to obtain cost information?

• Execution autonomy
➡ The ability to execute queries in unpredictable ways
➡ How to adapt to this?

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/6

Mediator Data Model
• Relational model
➡ Simple and regular data structures
➡ Mandatory schema

• Object model
➡ Complex (graphs) and regular data structures
➡ Mandatory schema

• Semi-structured (XML) model

➡ Complex (trees) and irregular data structures
➡ Optional schema (DTD or XSchema)

In this chapter, we use the relational model which is sufficient to

explain MDB query processing

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/7

MDB Query Processing
Architecture
Global/local
correspondences

Allocation and
capabilities

Local/DBMS
mappings

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/8

Query Rewriting Using Views
• Views used to describe the correspondences between global and local
relations
➡ Global As View: the global schema is integrated from the local databases and
each global relation is a view over the local relations
➡ Local As View: the global schema is defined independently of the local
databases and each local relation is a view over the global relations
• Query rewriting best done with Datalog, a logic-based language
➡ More expressive power than relational calculus
➡ Inline version of relational domain calculus

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/9

Datalog Terminology
• Conjunctive (SPJ) query: a rule of the form
➡ Q(T) :- R1(T1), … Rn(Tn)

➡ Q(T) : head of the query denoting the result relation

➡ R1(T1), … Rn(Tn): subgoals in the body of the query

➡ R1, … Rn: predicate names corresponding to relation names

➡ T1, … Tn: refer to tuples with variables and constants

➡ Variables correspond to attributes (as in domain calculus)

➡ “-” means unnamed variable

• Disjunctive query = n conjunctive queries with same head predicate

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/10

Datalog Example
With EMP(ENAME,TITLE,CITY) and ASG(ENAME,PNAME,DUR)

SELECT ENAME,TITLE, PNAME

FROM EMP, ASG
WHERE EMP.ENAME = ASG.ENAME
AND TITLE = "Programmer" OR DUR=24

Q(ename,title,pname) :- Emp(ename,title,-)
Asg(ename,pname,-),
title = “Programmer”.
Q(ename,title,pname) :- Emp(ename,title,-)
Asg(ename,pname,24).

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/11

Rewriting in GAV
• Global schema similar to that of homogeneous DDBMS
➡ Local relations can be fragments
➡ But no completeness: a tuple in the global relation may not exist in local
relations
✦ Yields incomplete answers
➡ And no disjointness: the same tuple may exist in different local databases
✦ Yields duplicate answers
• Rewriting (unfolding)
➡ Similar to query modification
✦ Apply view definition rules to the query and produce a union of conjunctive
queries, one per rule application
✦ Eliminate redundant queries

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/12

GAV Example Schema
Global relations Local relations
EMP(ENAME,CITY) EMP1(ENAME,TITLE,CITY)
ASG(ENAME,PNAME,TITLE, DUR) EMP2(ENAME,TITLE,CITY)
ASG1(ENAME,PNAME,DUR)

Emp(ename,city) :- Emp1(ename,title,city). (r1)

Emp(ename,city) :- Emp2(ename,title,city). (r2)
Asg(ename,pname,title,dur) :- Emp1(ename,title,city), (r3)
Asg1(ename,pname,dur).
Asg(ename,pname,title,dur) :- Emp2(ename,title,city), (r4)
Asg1(ename,pname,dur).

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/13

GAV Example Query
Let Q: name and project for employees in Paris
Q(e,p) :- Emp(e,“Paris”), Asg(e,p,-,-).
Unfolding produces Q’
Q’(e,p) :- Emp1(e,-,“Paris”), Asg1(e,p,-,). (q1)
Q’(e,p) :- Emp2(e,-,“Paris”), Asg1(e,p,-,). (q2)
where
q1 is obtained by applying r3 only or both r1 and r3
In the latter case, there are redundant queries
same for q2 with r2 only or both r2 and r4

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/14

Rewriting in LAV
• More difficult than in GAV
➡ No direct correspondence between the terms in GS (emp, ename) and those in
the views (emp1, emp2, ename)
➡ There may be many more views than global relations
➡ Views may contain complex predicates to reflect the content of the local
relations
✦ e.g. a view Emp3 for only programmers
• Often not possible to find an equivalent rewriting
➡ Best is to find a maximally-contained query which produces a maximum subset
of the answer
✦ e.g. Emp3 can only return a subset of the employees

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/15

Rewriting Algorithms
• The problem to find an equivalent query is NP-complete in the number of
views and number of subgoals of the query
• Thus, algorithms try to reduce the numbers of rewritings to be considered
• Three main algorithms
➡ Bucket
➡ Inverse rule
➡ MiniCon

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/16

LAV Example Schema
Local relations Global relations
EMP1(ENAME,TITLE,CITY) EMP(ENAME,CITY)
EMP2(ENAME,TITLE,CITY) ASG(ENAME,PNAME,TITLE, DUR)
ASG1(ENAME,PNAME,DUR)

Emp1(ename,title,city) :- Emp(ename,city), (r1)

Asg(ename,-,title,-).
Emp2(ename,title,city) :- Emp(ename,city), (r2) Asg(ename,-,title,-).
Asg1(ename,pname,dur) :- Asg(ename,pname,-,dur) (r3)

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/17

Bucket Algorithm
• Considers each predicate of the query Q independently to select only the
relevant views
Step 1
➡ Build a bucket b for each subgoal q of Q that is not a comparison predicate
➡ Insert in b the heads of the views which are relevant to answer q

Step 2
➡ For each view V of the Cartesian product of the buckets, produce a
conjunctive query
✦ If it is contained in Q, keep it
• The rewritten query is a union of conjunctive queries

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/18

LAV Example Query
Let Q be Q(e,p) :- Emp(e, “Paris”), Asg(e,p,-,-).
Step1: we obtain 2 buckets (one for each subgoal of Q)
b1 = Emp1(ename,title’,city), Emp2(ename,title’,city)
b2 = Asg1(ename,pname,dur’)
(the prime variables (title’ and dur’) are not useful)
Step2: produces
Q’(e,p) :- Emp1(e,-, “Paris”), Asg1(e,p,-,). (q1)
Q’(e,p) :- Emp2(e,-, “Paris”), Asg1(e,p,-,). (q2)

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/19

Query Optimization and
Execution
• Takes a query expressed on local relations and produces a distributed QEP
to be executed by the wrappers and mediator
• Three main problems
➡ Heterogeneous cost modeling
✦ To produce a global cost model from component DBMS
➡ Heterogeneous query optimization
✦ To deal with different query computing capabilities
➡ Adaptive query processing
✦ To deal with strong variations in the execution environment

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/20

Heterogeneous Cost
Modeling
• Goal: determine the cost of executing the subqueries at component DBMS
• Three approaches
➡ Black-box: treats each component DBMS as a black-box and determines costs
by running test queries
➡ Customized: customizes an initial cost model
➡ Dynamic: monitors the run-time behavior of the component DBMS and
dynamically collect cost information

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/21

Black-box Approach
• Define a logical cost expression
➡ Cost = init cost + cost to find qualifying tuples
+ cost to process selected tuples
✦ The terms will differ much with different DBMS
• Run probing queries on component DBMS to compute cost coefficients
➡ Count the numbers of tuples, measure cost, etc.
➡ Special case: sample queries for each class of important queries
✦ Use of classification to identify the classes
• Problems
➡ The instantiated cost model (by probing or sampling) may change over time
➡ The logical cost function may not capture important details of component
DBMS

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/22

Customized Approach
• Relies on the wrapper (i.e. developer) to provide cost information to the
mediator
• Two solutions
➡ Wrapper provides the logic to compute cost estimates
✦ Access_cost = reset + (card-1)*advance
✓ reset = time to initiate the query and receive a first tuple
✓ advance = time to get the next tuple (advance)
✓ card = result cardinality
➡ Hierarchical cost model
✦ Each node associates a query pattern with a cost function
✦ The wrapper developer can give cost information at various levels of details,
depending on knowledge of the component DBMS

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/23

Hierarchical Cost Model

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/24

Dynamic Approach
• Deals with execution environment factors which may change
➡ Frequently: load, throughput, network contention, etc.
➡ Slowly: physical data organization, DB schemas, etc.

• Two main solutions

➡ Extend the sampling method to consider some new queries as samples and
correct the cost model on a regular basis
➡ Use adaptive query processing which computes cost during query execution
to make optimization decisions

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/25

Heterogeneous Query
Optimization
• Deals with heterogeneous capabilities of component DBMS
➡ One DBMS may support complex SQL queries while another only simple
select on one fixed attribute
• Two approaches, depending on the M/W interface level
➡ Query-based
✦ All wrappers support the same query-based interface (e.g. ODBC or SQL/MED)
so they appear homogeneous to the mediator
✦ Capabilities not provided by the DBMS must be supported by the wrappers
➡ Operator-based
✦ Wrappers export capabilities as compositions of operators
✦ Specific capabilities are available to mediator
✦ More flexibility in defining the level of M/W interface

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/26

Query-based Approach
• We can use 2-step query optimization with a heterogeneous cost model
➡ But centralized query optimizers produce left-linear join trees whereas in
MDB, we want to push as much processing in the wrappers, i.e. exploit bushy
trees
• Solution: convert a left-linear join tree into a bushy tree such that
➡ The initial total cost of the QEP is maintained
➡ The response time is improved

• Algorithm
➡ Iterative improvement of the initial left-linear tree by moving down subtrees
while response time is improved

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/27

Left Linear vs Bushy Join Tree

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/28

Operator-based Approach
• M/W communication in terms of subplans
• Use of planning functions (Garlic)
➡ Extension of cost-based centralized optimizer with new operators
✦ Create temporary relations
✦ Retrieve locally stored data
✦ Push down operators in wrappers
✦ accessPlan and joinPlan rules
➡ Operator nodes annotated with
✦ Location of operands, materialization, etc.

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/29

Planning Functions Example
• Consider 3 component databases with 2 wrappers:
➡ w1 .db1: EMP(ENO,ENAME,CITY)

➡ w1 .db2: ASG(ENO,PNAME,DUR)

➡ w2 . db3: EMPASG(ENAME,CITY,PNAME,DUR)

• Planning functions of w 1

➡ AccessPlan (R: rel, A: attlist, P: pred) = scan(R, A, P, db(R))

➡ JoinPlan (R1, R2: rel, A: attlist, P: joinpred) = join(R1, R2, A, P)
✦ condition: db(R1) ≠ db(R2)
✦ implemented by w1
• Planning functions of w 2

➡ AccessPlan (R: rel, A: attlist, P: pred) = fetch(city=c)

✦ condition: (city=c) included in P
➡ AccessPlan (R: rel, A: attlist, P: pred) = scan(R, A, P, db(R))
✦ implemented by w2

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/30

Heterogenous QEP
SELECT ENAME,PNAME,DUR
FROM EMPASG
WHERE CITY = "Paris" AND DUR>24

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/31

Adaptive Query Processing -
Motivations
• Assumptions underlying heterogeneous query optimization
➡ The optimizer has sufficient knowledge about runtime
✦ Cost information
➡ Runtime conditions remain stable during query execution

• Appropriate for MDB systems with few data sources in a controlled

environment
• Inappropriate for changing environments with large numbers of data
sources and unpredictable runtime conditions

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/32

Example: QEP with Blocked
Operator
• Assume ASG, EMP, PROJ and
PAY each at a different site
• If ASG site is down, the entire
pipeline is blocked
• However, with some
reorganization, the join of EMP
and PAY could be done while
waiting for ASG

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/33

Adaptive Query Processing –
Definition
• A query processing is adaptive if it receives information from the
execution environment and determines its behavior accordingly
➡ Feed-back loop between optimizer and runtime environment
➡ Communication of runtime information between mediator, wrappers and
component DBMS
✦ Hard to obtain with legacy databases
• Additional components
➡ Monitoring, assessment, reaction
➡ Embedded in control operators of QEP

• Tradeoff between reactiveness and overhead of adaptation

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/34

Adaptive Components
• Monitoring parameters (collected by sensors in QEP)
➡ Memory size
➡ Data arrival rates
➡ Actual statistics
➡ Operator execution cost
➡ Network throughput

• Adaptive reactions
➡ Change schedule
➡ Replace an operator by an equivalent one
➡ Modify the behavior of an operator
➡ Data repartitioning

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/35

Eddy Approach
• Query compilation: produces a tuple D, P, C, Eddy
➡ D: set of data sources (e.g. relations)
➡ P: set of predicates
➡ C: ordering constraints to be followed at runtime
➡ Eddy: n-ary operator between D and P

• Query execution: operator ordering on a tuple basis using Eddy

➡ On-the-fly tuple routing to operators based on cost and selectivity
➡ Change of join ordering during execution
✦ Requires symmetric join algorithms such Ripple joins

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/36

QEP with Eddy
• D= {R, S, T}
• P = { (R), R JN S, S JN T)
P 1 2
• C = {S < T} where < imposes S tuples to probe T tuples using an index on join
attribute
➡ Access to T is wrapped by JN

Result tuples

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/37

Query Translation and
Execution
• Performed by wrappers using the component DBMS
➡ Conversion between common interface of mediator and DBMS-dependent
interface
✦ Query translation from wrapper to DBMS
✦ Result format translation from DBMS to wrapper
➡ Wrapper has the local schema exported to the mediator (in common interface)
and the mapping to the DBMS schema
➡ Common interface can be query-based (e.g. ODBC or SQL/MED) or operator-
based
• In addition, wrappers can implement operators not supported by the
component DBMS, e.g. join

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/38

Wrapper Placement
• Depends on the level of autonomy of component DB
• Cooperative DB
➡ May place wrapper at component DBMS site
➡ Efficient wrapper-DBMS com.

• Uncooperative DB
➡ May place wrapper at mediator
➡ Efficient mediator-wrapper com.

• Impact on cost functions

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/39

SQL Wrapper for Text Files
• Consider EMP (ENO, ENAME, CITY) stored in a Unix text file in
componentDB
➡ Each EMP tuple is a line in the file, with attributes separated by “:”

• SQL/MED definition of EMP

CREATE FOREIGN TABLE EMP
ENO INTEGER, ENAME VARCHAR(30), CITY CHAR(30)
SERVER componentDB
OPTIONS (Filename ‘/usr/EngDB/emp.txt’, Delimiter ‘:’)
• The query
SELECT ENAME FROM EMP
Can be translated by the wrapper using a Unix shell command
Cut –d: -f2/ usr/EngDB/emp

Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/40

Wrapper Management Issues
• Wrappers mostly used for read-only queries
➡ Makes query translation and wrapper construction easy
➡ DBMS vendors provide standard wrappers
✦ ODBC, JDBC, ADO, etc.
• Updating makes wrapper construction harder
➡ Problem: heterogeneity of integrity constraints
✦ Implicit in some legacy DB
➡ Solution: reverse engineering of legacy DB to identify implicit constraints
and translate in validation code in the wrapper
• Wrapper maintenance
➡ schema mappings can become invalid as a result of changes in component
DB schemas
✦ Use detection and correction, using mapping maintenance techniques
Distributed DBMS @ M. T. Özsu & P. Valduriez Ch.9/41