ObjectOriented Databases

1
Object-Oriented Databases

• Prof. Hyoung-Joo Kim
• OOPSLA Lab
• Computer Engineering
• Seoul National University

2
Content

• Introduction
• Object-oriented data model
• Object-oriented query language
• Index organizations for aggregation graphs
• Multi-index, Join index, Nested index, Path index
  Access support relation(ASR)
• Index organizations for inheritance graphs
• SC-index, CH-index
• Summary

3
Introduction

• Object-oriented database (OODB)
• object-oriented data model and concept
• effective support for the management of complex
  objects(such as text, graphics, bitmap pictures,
  sounds and animation files)
• encapsulation property supports the integration
  of packages for handling complex objects

4
Introduction

• Important issues related to index in OODB
• navigational access
• be based on traversing object references
• set-oriented access
• be based on the use of a high-level, declarative
  query language
• the two means of access are often complementary
• a query selects a set of objects, and then the
  retrieved objects and their components are
  accessed by using navigational capabilities
• suitable index techniques must be developed to
  efficiently support object-oriented query language

5
Object-Oriented Data Model

• Concepts of OO data model
• each real-world entity is modeled by an object
• each object has a set of instance attributes and
  methods(operations)
• the attribute values represent the objects state
• objects sharing the same structure and behavior
  are grouped into classes
• a class can be defined as a specialization of one
  or more classes
• the specialization relationship among classes
  organizes them in an inheritance graph which is
  orthogonal to the aggregation graph

6
Object-Oriented Data Model
Journal
Organization
name
String
name
publisher
Publisher
Author
available issues
Staff
name
name
Book
books
String
title
String
year
Inheritance graph
integer
publisher
Aggregation graph
Manual
Handbook
domain
OS
String
String
Figure1.1 An object-oriented database schema
7
Object-Oriented Data Model

• Figure 1.1
• simple arc from a class C to a class C
• C is domain of an attribute of C
• bold arc from a class C to a class C
• C is a superclass of C
• queries are made against classes
• indexes are allocated on classes

8
Object-Oriented Query Language

• Features that most influence indexing techniques
• nested predicates
• Retrieve the authors of books published by
  Kluwer. (Q1)
• this query contains the nested predicate
  published by Kluwer
• nested predicate are expressed using
  path-expressions
• author.books.publisher.name Kluwer
• inheritance
• Retrieve all instance of class Book and all its
  subclasses published in 1991. (Q2)
• this query applies to all the classes in the
  hierarchy rooted at class Book

9
Object-Oriented Query Language

• Method
• derived method has a function comparable to that
  of an attribute, so it returns an object(or a
  value)
• predicate method returns the logical constants
  True or False
• Query execution strategies
• have two dimension
• traversal strategies(1st dimension)
• forward strategy - the first class visited is the
  target class of the query
• reverse strategy - the traversal begins at the
  leaves and proceeds bottom-up along the query
  graph

10
Object-Oriented Query Language

• technique used to retrieve instances(2nd
  dimension)
• nested-loop
• consists of instantiating separately each
  qualified instance of a class
• sort-domain
• consists of instantiating all qualified instance
  of a class at once

11
Index Organizations forAggregation Graph

• Notation
• given an aggregation graph H, a path P is defined
  as C1.A1.A2.An
• C1 is a class of H
• A1 is an attribute of class C1
• Ai is an attribute of a class Ci in H, such that
  Ci is the domain of attribute Ai-1 of Class Ci-1
• len(P) n denote the length of the path
• class(P) C1 U CiCi is the domain of attribute
  Ai-1 of class Ci-1 denote the set of the
  classes along the path
• dom(P) denote the class domain of attribute An
  of class Cn

12
Index Organizations forAggregation Graph

• example
• P1 Author.books.publisher.name
• len(P1)3, class(P1)Author, Book, Publisher,
  dom(P1)string
• P2 Book.year
• len(P2)1, class(P2)Book, dom(P2)integer
• P3 Organization.staff.books.publisher.name
• len(P3)4, class(P3)Organization, Author,
  Publication, Publisher, dom(P3)string

13
Index Organizations forAggregation Graph

• Example instantiation of the Path P3

Organization
Author
Book
Publisher
B1
A1
C Programming Lang.
1991
P2
O1
P1
B2
M. Ellis
M1
ATT Lab
A1, A4
Efficient Parsing for NL.
1986
P4
Academic Press.
A2
B3
Generalized LR Parsing
1991
P4
P2
O2
D. Mark
B4
Addison-Wesley
B4
C Programming
1990
P2
Carnegie Mellon U.
A3
A3
P3
Elsevier
Manual
O3
M. Tony
B2,B3
M1
C Ref. Manual
1991
P2
Unix
P4
Utrech U.
A4
A5
Kluwer
M2
GUI Guide
1993
P5
Windows
B. Stro.
B1,M1
O4
P5
Handbook
Microsoft
A5
Wisconsin U.
H1
Handbook of CS
1990
P2
Comp. Sci.
J. Van
H1
Figure 1.2 Instance of classes of the Database
schema in Figure 1.1
14
Index Organizations forAggregation Graph

• Path instantiations
• complete instantiation
• they start with an instance belonging to the
  first class of the path, contain an instance of
  each class, and end with an instance of the class
  domain of the path (P3)
• partial instantiation
• its first component is not an instance of the
  first class of the path (A2.B4.P2.Addison-We
  sley)
• Index graph(IG)
• abstract representation of a set of indexes
  allocated along a path P

15
Multi-Index

• Organization
• B-tree index on each class traversed by the path
• given a path PC1.A1.A2..An, a multi-index is
  defined as a set of n simple indexes I1, I2,,In,
  where Ii is an index defined on Ci.A1.
• example path P1 Author.books.publisher.name
• I1 is allocated on Author.books similarly
  indexes I2 and I3 are allocated on Book.publisher
  and Publisher.name

I1
I2
I3
16
Multi-Index

• Look-up of Q1
• a look-up of index I3 with key-value Kluwer
  the result is P4.
• A look-up of index I2 with key-value P4 the
  resultt is B2,B3
• a look-up of index I1 with key-values B2 and
  B3 the result is A3 which is the result of
  the query
• Index graph for multi-index

Organization
Author
Book
Publisher
Publisher.name
17
Join Index

• Binary equijoin index
• BJI (ri,sk) ri.A sk.B
• given two relations R and S and attributes A and
  B, a binary equijoin index is above
• ri(sk) denote the surrogate of a tuple of R(S)
• tuple ri (tuple sk) refers to the tuple having ri
  (sk) as surrogate
• a BJI is implemented as a binary relation and two
  copies may be kept, one clustered on ri and the
  other on sk
• each copy is implemented as a B-tree

18

• Example
• P1 Author.books.publisher.name
• the first join index BJI1 is on Author.books.
• BJI1(a) is clustered on OIDs of instance of
  Author
• BJI1(b) is clustered on OIDs of instance of
  Publisher
• the second join index BJI2 is on Book.publisher
• BJI2(a) is clustered on OIDs of instance of Book
• BJI2(b) is clustered on OIDs of instance of
  Publisher
• the third join index BJI3 is on the attribute
  Publisher.name
• BJI3(a) is clustered on OIDs of instance of
  Publisher
• BJI3(b) is clustered on values of attribute name

19
BJI1(a)
BJI1(b)
BJI2(a)
BJI2(b)
BJI3(a)
BJI3(b)
Join index organization for path P1
20
Join Index

• Join index organization
• supports both forward and reverse traversal
  strategies
• Index graph for join index

Organization
Author
Book
Publisher
Publisher.name
21
Nested Index

• Organization
• solving a nested predicate
• provide a direct association between an object of
  a class at the end of a path and the
  corresponding instance of the class at the
  beginning of the path
• P1 Author.books.publisher.name
• nested index associates with each publisher name
  the OIDs of authors that have written a book
  published by this publisher

22
Nested Index

• Retrieval
• a query such as Q1
• be solved with only one index lookup
• Major problem
• update operation
• require access to several objects in order to
  determine the index entries to be updated
• require both forward and backward traversal of
  objects, but nested index as defined above can
  only be used for reverse traversal
• two copies of a nested index
• can allocate two copies of a nested index as for
  the join index organization

23
Nested Index
A nested index for path P1 Author.books.publishe
r.name clustered on OIDs of instance of the
class at the beginning of the path

• Index graph for a nested index

Organization
Publisher.name
Author
Book
Publisher
24
Path Index

• Organization
• simple index like the nested index
• provide an association between an object O at the
  end of a path and all instantiations ending with
  O
• can be used to solve nested predicates against
  all classes along the path

Path index for path P3 Organization.staff.books.
publisher.name
25
Path Index

• Projections on path instantiation
• allows retrieving OIDs of several classes along
  the path with a single index lookup
• example
• determine all authors who have their books
  published by Kluwer in 1991
• perform an index lookup with key-value equal
  Kluwer
• perform a projection on position of classes
  Author(pos1) and Book (pos2) the results is
  (A3,B2),(A3,B3)
• the second element of each pair is extracted the
  corresponding object is accessed and the
  predicate on attribute year is evaluated.
• If this predicate is satisfied, the first element
  of the pair is returned

26
Path Index

• Problem
• the path index organization can be used even when
  no reverse reference among objects on the path
  are present
• Index graph for path index

Organization
Author
Book
Publisher
Publisher.name
27
Access support relation(ASR)

• Organization
• similar to path index
• calculate all instantiations along a path and
  store then in a relation
• the i-th attribute of that relation corresponds
  to the class Ci
• Two B-tree are allocated on the first and the
  last attributes

28
Access Support Relation(ASR)

• Index graph for ASR

Organization
Publisher.name
Author
Publisher
Book
29
Index Organizations forInheritance Hierarchies

• Object-oriented query
• apply to a class only or to a class and all its
  direct and indirect subclasses
• single-class query(SC-query)
• the query scope consists of only one class
• class-hierarchy query(CH-query)
• its scope includes a subhierarchy of the
  inheritance hierarchy
• rooted CH-query the root of the subhierarchy in
  the scope coincides with the root class C
• partial CH-query otherwise

30
Single-Class Index(SC-Index)

• Organization
• be based on maintaining a separate B-tree on the
  indexed attribute for each class
• if the inheritance hierarchy has m classes, the
  SC-index requires m B-tree
• example
• consider the inheritance hierarchy rooted at
  class Book in Figure 1.1
• if the attribute year is frequently referred,
  the SC-index requires building three indexes
• problem
• the SC-index is not efficient for CH-query

31
Class-Hierarchy Index(CH-Index)

• Organization
• be based on maintaining a unique B-tree for all
  classed in the hierarchy
• an index entry in a leaf node may contain the
  OIDs of instance of any class in the indexed
  inheritance hierarchy.
• example(same as the SC-index)

Entries of CH-tree for the inheritance hierarchy
rooted at class Book
32
Class-Hierarchy Index(CH-Index)

• a leaf node in a CH-index consists if a
  key-value, a key-directory, the number of
  elements in the list of OIDs for instance of each
  class
• performance
• it has an inverse trend with respect to the
  SC-index
• it is more efficient for queries whose access
  scope involves all classes in the indexed
  inheritance hierarchy, whereas a SC-index is
  effective for queries against a single class
• problem
• the CH-tree retrieves many unnecessary leaf node
  page when the query applies to a singe class only

33
Other Structures

• Attempts to overcome the problems of the
  SC-index, CH-index
• H-tree
• Low, C.C. et al., H-tree A dynamic associative
  search index for OODB, ACM SIGMOD 1992.
• CG-tree
• Kilger, Moerkotte, Indexing multiple sets, VLDB
  1994.
• hcC-tree
• Sreenath and Seshadri, hcC-tree An efficient
  index structure for OODB, VLDB 1994.
• X-tree
• Chan et al., Indexing OODB instances based on
  access proximity, ICDE 1997.

34
Summary

• Open problem
• determine how all those indexing techniques
  perform for different types of queries
• optimal index allocation
• techniques for an efficient execution of queries
  containing method invocations