The Future of Distributed Databases

1
The Future of Distributed Databases
2
The Future of Distributed Databases

• Self-Organizing Distributed DBMSs
• Multimedia Databases
• Object-Oriented DBMSs
• Course Summary

3
Problems Facing the DBA Practical Advice

• A set of requests is an approximation of the
  total load
• Request frequencies are approximations
• Statistical summaries of data may not reflect
  future changes
• Model use in calculations may not be completely
  accurate
• There may be no fast algorithm to find the best
  solutions
• So
• Find a reasonable design, not an optimal design

4
Need for Tuning

• Characteristics of data changes
• Size of fragments
• Location of fragments
• Replication of fragments
• New fragments are added
• Existing fragments are split and/or combined
• Applications change
• Old applications evolve
• New applications are implemented
• Users acquire more sophistication
• So

Tuning is a Continual Process
5
Where to Get Information?

• Problems facing the DBA
• Obtain accurate usage pattern
• Develop model of operation
• Rate alternative designs
• Obtain meaningful predictions
• So
• Look to DBMS for assistance
• DBMS is the best source of information

6
Design Evaluation Module Overview
DBA
UsagePattern
DesignEvaluationModule
InformationGatheringModule
DBInternalCharacteristics
7
Design Evaluation Module Detail
TransactionCost Estimator
Design CostEstimator
EstimatedDesign Cost
Also used to evaluate cost of user queries
8
Automated Design Strategy 1

• Iterate design evaluation over all possible
  designs
• Disadvantage
• Astronomical number of possible designs

9
Automated Design Strategy 2

• Derive analytic cost equations
• Apply mathematical optimization techniques
• Disadvantages
• Difficult to derive cost equations
• Entails oversimplification
• Expensive optimization techniques

10
Automated Design Strategy 3

• Use heuristic search techniques to select
  near-optimal design

11
Summary

• Engineering discipline is replacing black art.
• DBA concentrates on design requirements.
• Automated aids

12
The Future of Distributed Databases

• Self-Organizing Distributed DBMSs
• Multimedia Databases
• Object-Oriented DBMSs
• Course Summary

13
Multimedia Databases

• Use distributed DBMSs to store information about
  text, graphics, voice and video

14
Thesaurus
What terms are more specific than pine? select
SpecificTerm from Thesaurus where
GeneralTerm"pine" What term is more general
than pine? select GeneralTerm from
Thesaurus where SpecificTerm"pine"
15
Term-Document Index
What are the numbers of documents that contain
the words "white pine"? select
DocumentNumber from Index where Term white
pine What are the numbers of documents that
contain the words "white pine" and "blister
rust"? select DocumentNumber from Index where
Term "white pine" INTERSECT select
DocumentNumber from Index where Term "blister
rust"
1 2
16
Location
select text (location) from LocationTable where
DocumentNumber 1
17
Location Example
Display text with the term "white pine" select
text (Location) from LocationTable, Index where
LocationTable.DocumentNumberIndex.DocumentNumber
and term "white pine"
18
Typical Indexing Strategy

• Remove punctuation
• Remove special characters
• Remove common words (stop words)
• Replace words by their stems
• Remove words in text that occur less than n times
• Place words into INDEX

Many white pines have blister rust .
3
3
4
1
19
Pictures Location Table
LocationTable Illustration Number Location
1 4460 2 5460
select illustration (location) from
LocationTable where IllustrationNumber 1
20
Pictures Example
Illustration 1
Index Illustration Number Term 1 white pine
2 super-car
Illustration 2
Display illustrations showing "white
pines"? select illustration (location) from
LocationTable, Index where LocationTable.Illustra
tionNumber Index.IllustrationNumber and Term
"white pine"
21
Compound Document

• Compound document X consists of all the data
  units (perhaps of different media) in the leaves
  plus the structure of the subtree with root X

Monthly Report
LayoutComposer
Section 1
Monthlyreport
Section 2
Blah, Blah
Blah, Blah
22
Compound Document Contents

• Text
• Graphics
• Images
• Voice
• Spreadsheets and charts
• Annotations

23
Mixed Media Impact on Distributed Databases

• Extend fragmentation and allocation procedures
• Extend request optimization algorithms
• Extend or replace user interface
• Optimize distributed DBMS to minimize handling
  transfer of nontabular data

24
Disadvantages of Using RDBMSs for Multimedia

• Forced to fit nontabular data into tables
• Additional processing is needed to reconstruct
  nontabular data structures
• RDMBSs are good for tabular data, but may not
  handle complex structures well.

25
The Future of Distributed Databases

• Self-Organizing Distributed DBMSs
• Multimedia Databases
• Object-Oriented DBMSs
• Course Summary

26
Classes and Instances
Instances
Class
PERSON
PERSON
PERSON
IsA
IsA
IsA
EMPLOYEE
EMPLOYEE
EMPLOYEE

• Classes represent sets of homogeneous real-world
  objects.
• Class instance object represents a single
  real-world object.

27
inheritance
PERSON
PERSON
IsA
IsA
Name Jones Birthdate Dec. 11,
1956 Department Toy Works on Rambo
Doll Dependents Sally, Sam
EMPLOYEE
EMPLOYEE

• An object instance inherits all of the attributes
  of its parent.

28
Methods and Encapsulation
GiveRaise (100)

• Method is like a subroutine inside of a class.
• Users dont know how the subroutine is
  represented.
• Users only know the interface to subroutine.

encapsulation
29
Complex Objects
Object Instances
Class
PartOf
SECTION
SECTION
SECTION

• Object instances may be arbitrarily complex.

30
Object-Oriented DBMS

• Object-oriented DBMSs are like traditional DBMSs
• Persistence
• Sharing
• Consistency
• Resilience
• Associative retrieval

• Object-oriented DBMSs also support operations
  specific to objects.
• In a sense, business logic is performed by
  object-oriented DBMSs when the user or the
  application invokes methods.

31
Distributed Object-Oriented DBMS

• Objects can be distributed throughout a network
  of database servers.
• Current research problems
• Object allocation is difficult because
• Access frequencies are not usually available.
• Complex objects may need to be partitioned.
• Query optimization is difficult because
• Optimizer may not know about special methods.
• Optimizer may not know how methods are
  implemented.

32
Future of Distributed DBMSs Summary

• DBMSs will be extended to support multimedia and
  hypertext.
• Design tools will be used to determine placement
  of fragments in distributed DBMSs.
• Some distributed DBMSs will automatically migrate
  fragments from site to site.

33
The Future of Distributed Databases

• Self-Organizing Distributed DBMSs
• Multimedia Databases
• Object-Oriented DBMSs
• Course Summary

34
Course Summary

• Choose the features you need in a distributed
  DBMS and then choose a distributed DBMS
  architecture.
• Choose data fragments and distribute/replicate
  them as necessary
• Use a high-level language for formulating
  requests let optimizers determine the
  distributed execution plan.
• Planning and cooperation are necessary for a
  successful distributed DBMS.

More
35
Course Summary, cont.

• Object-oriented, mixed-media databases are coming
• Lots of challenges ahead
• Things are moving fast
• Its an exciting time to be working with
  distributed DBMSs