Some thoughts on the future evolution of objectoriented database concepts

1
Some thoughts on the future evolution of
object-oriented database concepts

• Catriel Beeri
• The German Database Conference "Datenbanksysteme
  in Büro, Technik und Wissenschaft" (1993)

2
Problem Addressed

• Identity
• Abstractions
• Meta-level Concepts

3
On Identity

• Values and Objects
• Values, including numbers, characters and
  strings, are abstractions invented in the human
  mind. The universally known meaning of a value is
  the information it carries a value is in itself
  the information. Example the number 7
• Objects represent those entities that do not
  necessarily have universally, or at least
  system-wide, known and used meanings, and thus
  require additional descriptions. Thus the objects
  have and identity, and separately from it various
  properties and relationships. Examples are
  physical entities like person and job and
  abstractions like marriage and employment.
• The relational model and its extensions are
  value-based, as opposed to OODBMS, which are
  object-based.

4
On Identity

• Object Identifier
• Associate every object with system-maintained
  identifier.
• The o-id promotes sharing.
• e.g. Two separate objects can share a person
  id as a manager and a father.
• The o-id promotes cycles.
• e.g. A spouse relation.
• Keys in relational databases allow us to
  accomplish the same, however, in an OODBMS it is
  all transparent to the user it is in the realm
  of the systems responsibility.
• Disadvantages
• O-ids are invisible there are no operations to
  manipulate them or display them.
• Every objects identity should be implemented
  with an o-id even though natural keys exist.

5
On Identity

• Problem
• Suppose an OODB contains two distinct objects
  (with diff o-ids), but all their properties and
  the relationships in which they participate are
  the same. Can we delete one without deleting the
  other ?

6
On Identity

• Solution
• Model an OODB as a labeled graph.
• Nodes are entities values and objects, and
  their labels provide information about them. Node
  representing atomic value is labeled by the
  value. Other node labels include object, set,
  class.
• The edges and their labels describe
  relationships the relationship between a set and
  its members, or a class and member objects (by e)
  or that between a tuple and its attribute values
  by edge labeled with the attribute names.
• Example

s
s
s
s
S
s
7
On Identity

• Alternate definition
• An Object is uniquely identifiable if there is a
  query that retrieves it, and no other object
• The ultimate means of identification is the
  association with a value where relations can be
  considered a kind of value by itself. To
  guarantee unique identification, we need to
  impose constraints in the schema such as
  prohibiting the unrestricted use of the new
  primitive.
• If object identification is value-based, and
  internal identity plays no role, then what are
  system maintained o-ids good for ?
• Further identification problems in views and
  heterogeneous systems.

8
Meta-level Concepts

• Metadata is data about data or descriptions of
  current data.
• It is traditional in the relational system to
  store schema information in the same format as
  the data itself, that is in relations. Thus the
  same access and manipulation language can be used
  by meta-data users and normal users.
• Major motivation for introducing OODBs is that
  they support expressive and flexible type
  systems, that can adequately represent many
  different kinds of data. A prime test for this is
  the systems data if the OODB cannot represent
  and store its own schemas, then it may fails on
  other applications also.
• Even though metadata is represented as regular
  data, they both have separate roles and their
  uses cannot be mixed. Metadata can be queried
  freely but updates are severely restricted.

9
Meta-level Concepts

• Melampus Project Proposal
• Propose the notion of an aspect. An object can be
  accessed through an aspect, and its structure and
  behavior are determined by that aspect.
• There is no fixed relation between an object and
  its aspect (like object and class). Rather an
  aspect is used at run-time according to need.
• There is no restriction on the association of
  objects with aspects, and the same object may be
  accessed through many aspects, and exhibit a
  different behavior for each of them.
• Example A person may be accessed as an employee
  and have a salary or as a student in which case
  he/she has a study plan.
• Advantages
• System prevalent at that time did not support
  multiple inheritance and this was a way to get
  over that.
• Aspects have the property of metadata, and in the
  proposal they are much closer to being stored and
  manipulated as regular data.

10
Meta-level Concepts

• Ullmans Proposal
• Proposes classless databases as a suggestion but
  proposes no alternative.
• As databases grow, the number of classifications
  grow and it will become increasingly difficult to
  remember them and use them properly.
• Example is a database where all the people in
  the world are represented.
• Author Proposal
• Feels that classes should be termed as aspects,
  but cannot do away with classes as
  classifications to understand the world and
  interact with it fit the mental model of human
  beings.
• Proposes two sets of classes
• base classes once part of base class always
  part of it. E.g. Person
• role classes can initiate membership and
  stop it at anytime. E.g. Student.
• As an object can have many applicable methods,
  accesses need to specify a class. The base class
  can serve as a default in such cases.

11
Challenges of OODB

• (1) To find the right collection of concepts
  that can be of most use in the development of
  flexible and easy-to-use applications.
• (2) To understand what is and what is not
  provided by various concepts, so that we know how
  to fit the pieces of our ever-changing reality
  into the mold provided by the given model.

12
Conclusions

• The paper does bring out some interesting issues
  where further research could be directed
  considering it was published in 1993.
• The author does not provide solid solutions for
  any of the discussions in the paper, but this was
  not the intension to begin with.