lOMoARcPSD|56591169

Lecture notes, lecture Database Modelling and Design

Computer Networks (University of Michigan)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by Hot Boy ([email protected])
 lOMoARcPSD|56591169

Database Modeling and Design

3 rd Edition

Toby J. Teorey
University of Michigan

Lecture Notes
Contents
I. Database Systems and the Life Cycle (Chapter 1)……………………2
Introductory concepts; objectives of database management 2
Relational database life cycle 3
Characteristics of a good database design process 7
II. Requirements Analysis (Chapter 3)………………………………….8

III. Entity-Relationship (ER) Modeling (Chapters 2-4)……………… 11

Basic ER modeling concepts 11
Schema integration methods 22
Entity-relationship 26
Transformations from ER diagrams to SQL Tables 29
IV. Normalization and normal forms (Chapter 5)………………………35
First normal form (1NF) to third normal form (3NF) and BCNF 35
3NF synthesis algorithm (Bernstein) 42
Fourth normal form (4NF) 47

V. Access Methods (Chapter 6)…………………………..………………50

Sequential access methods 50
Random access methods 52
Secondary Indexes 58
Denormalization 62
Join strategies 64
VI. Database Distribution Strategies (Chapter 8)……………………….66
Requirements of a generalized DDBMS: Date’s 12 Rules 68
Distributed database requirements 72
The non-redundant “ best fit” method 74
The redundant “all beneficial sites” method 77

VII. Data Warehousing, OLAP, and Data Mining (Chapter 9)…….....79

Data warehousing 79
On-line analytical processing (OLAP) 86
Data mining 93

Revised 11/18/98 – modify Section V

Revised 11/21/98 – insertions into Section VII
Revised 1/14/99 – modify Section VI
Revised 2/11/99 – modify Section IV, 4NF (p.47 FD, MVD mix)
Revised 6/13/00 – modify Section V (secondary indexes)

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

I. Database Systems and the Life Cycle

Introductory Concepts
data—a fact, something upon which an inference is based (information or knowledge has
value, data has cost)
data item—smallest named unit of data that has meaning in the real world (examples: last
name, address, ssn, political party)
data aggregate (or group) -- a collection of related data items that form a
whole concept; a simple group is a fixed collection, e.g. date (month, day, year); a
repeating group is a variable length collection, e.g. a set of aliases.
record—group of related data items treated as a unit by an application program (examples:
presidents, elections, congresses)

file—collection of records of a single type (examples: president, election)

database—computerized collection of interrelated stored data that serves the needs of

multiple users within one or more organizations, i.e. interrelated collections of records
of potentially many types. Motivation for databases over files: integration for easy access
and update, non-redundancy, multi-access.

database management system (DBMS) -- a generalized software system for

manipulating databases. Includes logical view (schema, sub-schema), physical view
(access methods, clustering), data manipulation language, data definition language, utilities
- security, recovery, integrity, etc.

database administrator (DBA) -- person or group responsible for the effective use of
database technology in an organization or enterprise. Motivation: control over all phases of
the lifecycle.

Objectives of Database Management

1. Data availability—make an integrated collection of data available to a wide variety of
users
* at reasonable cost—performance in query update, eliminate or control data
redundancy
* in meaningful format—data definition language, data dictionary
* easy access—query language (4GL, SQL, forms, windows, menus);
embedded SQL, etc.; utilities for editing, report generation, sorting

2. Data integrity—insure correctness and validity

* checkpoint/restart/recovery
* concurrency control and multi-user updates
* accounting, audit trail (financial, legal)

3. Privacy (the goal) and security (the means)

* schema/sub-schema, passwords

4. Management control—DBA: lifecycle control, training, maintenance

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

5. Data independence (a relative term) -- avoids reprogramming of applications, allows

easier conversion and reorganization
* physical data independence—program unaffected by changes in the storage structure or
access methods

* logical data independence—program unaffected by changes in the schema

* Social Security Administration example (1980ís)

- changed benefit checks from $999.99 to $9999.99 format
- had to change 600 application programs
- 20,000 work hours needed to make the changes (10 work years)
* Student registration system—cannot go to a 4-digit or hexadecimal course numbering
system because of difficulty changing programs

*Y2K (year 2000) problem—many systems store 2-digit years (e.g. ‘02-OCT-98’) in their
programs and databases, that give incorrect results when used in date arithmetic (especially
subtraction), so that ‘00’ is still interpreted as 1900 rather than 2000. Fixing this problem
requires many hours of reprogramming and database alterations for many companies and
government agencies.

Relational Database Lifecycle

1. Requirements formulation and analysis
* natural data relationships (process-independent)
* usage requirements (process-dependent)
* hardware/software platform (OS, DBMS)
* performance and integrity constraints
* result: requirements specification document, data dictionary entries

2. Logical database design

2.1 ER modeling (conceptual design)
2.2 View integration of multiple ER models
2.3 Transformation of the ER model to SQL tables
2.4 Normalization of SQL tables (up to 3NF or BCNF)
*result: global database schema, transformed to table definitions
3. Physical database design
* index selection (access methods)
* clustering

4. Database distribution (if needed for data distributed over a network)

* data fragmentation, allocation, replication

5. Database implementation, monitoring, and modification

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Characteristics of a Good Database Design Process

* iterative requirements analysis
- interview top-down
- use simple models for data flow and data relationships
- verify model

* stepwise refinement and iterative re-design

* well-defined design review process to reduce development costs review team

-database designers
-DBMS software group
-end users in the application areas when to review
- after requirements analysis & conceptual design
- after physical design
- after implementation (tuning) meeting format
- short documentation in advance
- formal presentation
- criticize product, not person
- goal is to locate problems, do solutions off line
- time limit is 1-2 hours

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

II. Requirements Analysis

Purpose - identify the real-world situation in enough detail

to be able to define database components. Collect two types of data: natural data (input to the
database) and processing data (output from the database).

Natural data requirements (what goes into the database)

1. Organizational objectives
- sell more cars this year
- move into to recreational vehicle market

2. Information system objectives

- keep track of competitors’ products and prices
- improve quality and timing of data to management regarding production schedule delays,
etc.
- keep track of vital resources needed to produce and market a product

3. Organizational structure/chart

4. Administrative and operational policies

- annual review of employees
- weekly progress reports
- monthly inventory check
- trip expense submission

5. Data elements, relationships, constraints, computing environment

Processing requirements (what comes out of the database)

1. Existing applications - manual, computerized

2. Perceived new applications

* quantifies how data is used by applications

* should be a subset of data identified in the natural relationships

(but may not be due to unforeseen applications)

* problem - many future applications may be unknown

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Data and Process Dictionary Entries for Requirements Analysis

in the Database Design Lifecycle
Entity Description (possibly in a data dictionary)
Name customer
Reference-no 4201
Cardinality 10,000
Growth rate 100 per month
Synonyms user, buyer
Role (or description) someone who purchases or rents a
product made by the company.
Security level 0 (customer list is public)
Subtypes adults, minors
Key attribute(s) cust-no
Non-key attribute(s) cust-name, addr, phone, payment-status Relationship to
other entities salesperson, order, product
Used in which applications billing, advertising
Attribute description (data elements in a data dictionary)
Name cust-no
Reference-no 4202
Range of legal values 1 to 999,999
Synonyms cno, customer-number
Data type integer
Description customer id number set by the company.
Key or nonkey key
Source of data table of allowable id numbers
Used in applications billing
Attribute trigger /*describes actions that occur when a
data element is queried or updated*/
Relationship description
Name purchase
Reference-no 511037
Degree binary
Entities and connectivity customer(0,n), product(1,n)
Synonyms buy
Attributes (of the relationship) quantity, order-no
Assertions a customer must have purchased at
least one product, but some products
may not have been purchased as yet by
any customers.
Process (application) description
Name payroll
Reference-no 163
Frequency bi-weekly
Priority 10
Deadline noon Fridays
Data elements used emp-name, emp-salary
Entities used employee
Data volume (how many entities) implicit from entity cardinality

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Interviews at different levels

Top management - business definition, plan/objectives, future plans

Middle management - functions in operational areas, technical areas, job-titles, job functions

Employees - individual tasks, data needed, data out

Specific end-users of a DBMS - applications and data of interest

Basic rules in interviewing

1. Investigate the business first

2. Agree with the interviewee on format for documentation (ERD, DFD, etc.)

3. Define human tasks and known computer applications

4. Develop and verify the flow diagram(s) and ER diagram(s)

5. Relate applications to data (this helps your programmers)

Example: order entry clerk

Function: Take customer orders and either fill them or make adjustments.
Frequency: daily

Task Def Volume Data Elements

1. Create order 2000 A, B, E, H

2. Validate order 2000 A, B, G, H, J
3. Fill out error form 25 A, C
4. Reserve item/price 6000 A, D, H
5. Request alternate items 75 A, E, I, K,M
6. Enter unit price 5925 A, F, J, N

10

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

III. Entity-Relationship (ER) Modeling

Basic ER Modeling Concepts

Entity - a class of real world objects having common characteristics and properties about
which we wish to record information.

Relationship - an association among two or more entities

* occurrence - instance of a relationship is the collective instances of the related entities

* degree - number of entities associated in the relationship (binary, ternary, other n-ary)

* connectivity - one-to-one, one-to-many, many-to-many

* existence dependency (constraint) - optional/mandatory

Attribute - a characteristic of an entity or relationship

* Identifier - uniquely determines an instance of an entity

* Identity dependence - when a portion of an identifier is inherited from another entity

* Multi-valued - same attribute having many values for one entity

* Surrogate - system created and controlled unique key (e.g. Oracle’s “create sequence”)

11

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

12

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

13

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

14

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Super-class (super-type)/subclass (subtype) relationship

Generalization
* similarities are generalized to a super-class entity, differences are specialized to a subclass entity,
called an “ISA” relationship (“specialization” is the inverse relationship)

* disjointness constraint - there is no overlap among subclasses

* completeness constraint - constrains subclasses to be all-inclusive of the super-class or not (i.e.

total or partial coverage of the superclass)

* special property: hierarchical in nature

* special property: inheritance - subclass inherits the primary key of the super-class, super-class has
common nonkey attributes, each subclass has specialized non-key attributes

Aggregation
* “part-of” relationship among entities to a higher type aggregate entity (“contains” is the inverse
relationship)

* attributes within an entity, data aggregate (mo-day-year)

* entity clustering variation: membership or “is-member-of” relationship

15

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Constraints

16

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Constraints in ER modeling
* role - the function an entity plays in a relationship

* existence constraint (existence dependency) - weak entity

* exclusion constraint - restricts an entity to be related to only of several other

* entities at a given point in time

- mandatory/optional
- specifies lower bound of connectivity of entity instances
- participating in a relationship as 1 or 0

* uniqueness constraint – one-to-one functional dependency among key attributes

in a relationship: binary, ternary, or higher n-ary

17

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

18

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

19

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

20

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

21

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Schema Integration Methods

Goal in schema integration
- to create a non-redundant unified (global) conceptual schema

(1) completeness - all components must appear in the global schema

(2) minimality - remove redundant concepts in the global schema
(3) understandability - does global schema make sense?
1 . Comparing of schemas

* look for correspondence (identity) among entities

* detect possible conflicts

- naming conflicts
homonyms - same name for different concepts
synonyms - different names for the same concept

- structural conflicts
type conflicts - different modeling construct for the same concept (e. g. “order” as an entity,
attribute, relationship)

- dependency conflicts - connectivity is different for different views (e.g. job-title vs. job-title-history)

- key conflicts - same concept but different keys are assigned (e.g. ID-no vs. SSN)

- behavioral conflicts - different integrity constraints (e.g. null rules for optional/mandatory:
insert/delete rules)

* determine inter-schema properties

- possible new relationships to combine schemas

- possible abstractions on existing entities or create new super-classes (super-types)

2. Conforming of schemas

* resolve conflicts (often user interaction is required)

* conform or align schemas to make compatible for integration

* transform the schema via

- renaming (homonyms, synonyms, key conflicts)
- type transformations (type or dependency conflicts)
- modify assertions (behavioral conflicts)

3. Merging and restructuring

* superimpose entities

* restructure result of superimposition

22

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

23

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

24

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

25

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Entity-Relationship Clustering
Motivation

* conceptual (ER) models are difficult to read and understand for large and complex databases, e.g.
10,000 or more data elements

* there is a need for a tool to abstract the conceptual database schema (e. g. clustering of the ER
diagram)

* potential applications

- end user communication

- application design team communication

- documentation of the database conceptual schema (in coordination with the data dictionary)

Clustering Methodology
Given an extended ER diagram for a database.....

Step 1. Define points of grouping within functional areas.

Step 2. Form entity clusters

* group entities within the same functional area
* resolve conflicts by combining at a higher functional grouping

Step 3. Form higher entity clusters.

Step 4. Validate the cluster diagram.

* check for consistency of interfaces.
* end-users must concur with each level.

26

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

27

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

28

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Transformations from ER diagrams to SQL Tables

* Entity – directly to a SQL table

* Many-to-many binary relationship – directly to a SQL table, taking the 2 primary

keys in the 2 entities associated with this relationship as foreign keys in the new table

* One-to-many binary relationship – primary key on “one” side entity copied as a

foreign key in the “many” side entity’s table

* Recursive binary relationship – same rules as other binary relationships

* Ternary relationship – directly to a SQL table, taking the 3 primary keys of the 3
entities associated with this relationship as foreign keys in the new table
* Attribute of an entity – directly to be an attribute of the table transformed from this
entity

* Generalization super-class (super-type) entity – directly to a SQL table

* Generalization subclass (subtype) entity – directly to a SQL table, but with the
primary key of its super-class (super-type) propagated down as a foreign key into its table

* Mandatory constraint (1 lower bound) on the “one” side of a one-to-many

relationship – the foreign key in the “many” side table associated with the primary key in
the “one” side table should be set as “not null” (when the lower bound is 0, nulls are
allowed as the default in SQL)

29

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

30

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

31

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

32

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

33

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

34

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

IV. Normalization and Normal Forms

First normal form (1NF) to third normal form (3NF) and BCNF

Goals of normalization
1. Integrity
2. Maintainability

Side effects of normalization

* Reduced storage space required (usually, but it could increase)
* Simpler queries (sometimes, but some could be more complex)
* Simpler updates (sometimes, but some could be more complex)

First normal form (1NF) -- a table R is in 1NF iff all underlying

domains contain only atomic values, i.e. there are no repeating groups in
a row.

functional dependency —given a table R, a set of attributes B is functionally dependent on

another set of attributes A if at each instant of time each A value is associated with only one B value.
This is denoted by A -> B. A trivial FD is of the form XY --> X (subset).

super-key -- a set of one or more attributes, which, when taken collectively, allows us to identify
uniquely an entity or table.

candidate key —any subset of the attributes of a super-key that is also a super-key, but not
reducible.

primary key -- arbitrarily selected from the set of candidate keys, as needed for indexing.

Third normal form (3NF)

A table is in 3NF if, for every nontrivial FD X --> A, either:
(1) attribute X is a super-key, or
(2) attribute A is a member of a candidate key (prime attribute)

Boyce-Codd normal form (BCNF)

A table is in BCNF if, for every nontrivial FD X --> A,
(1) attribute X is a super-key.

35

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Tables, Functional Dependencies, and Normal Forms

First Normal Form

TABLE SUPPLIER_PART (100k rows, 73
bytes/row => 7.3 MB)
SNUM SNAME STATUS CITY
PNUM PNAME WT QTY
S1 SMITH 20 LONDON P1
S1 SMITH 20 LONDON P2
S1 SMITH 20 LONDON P3
S1 SMITH 20 LONDON P4
S1 SMITH 20 LONDON P5
S1 SMITH 20 LONDON P6
S2 JONES 10 PARIS P1
S2 JONES 10 PARIS P2
S3 BLAKE 10 PARIS P3
S3 BLAKE 10 PARIS P5
S4 CLARK 20 LONDON P2 BOLT 22 2 10-31-89
S4 CLARK 20 LONDON P4 WRENCH 24 3 7-14-90
S4 CLARK 20 LONDON P5 CLAMP 22 7 8-20-90
S5 ADAMS 30 ATHENS P5 CLAMP 22 5 8-11-91
Functional dependencies
SNUM --> SNAME, STATUS,CITY
CITY --> STATUS
PNUM --> PNAME, WT
SNUM,PNUM,SHIPDATE --> QTY

Attribute sizes (bytes)

SNUM 5 PNAME 10
SNAME 20 WT 5
STATUS 2 QTY 5
CITY 10 SHIPDATE 8
PNUM 8 Total size 73
Third Normal Form

TABLE PART (100 rows, 23 bytes/row => 2.3 KB)

PNUM PNAME WT Functional dependencies
P1 NUT 12 PNUM --> PNAME, WT
P2 BOLT 17
P3 WRENCH 17
P4 WRENCH 24
P5 CLAMP 12
P6 LEVEL 19
TABLE SHIPMENT (100k rows, 26 bytes/row => 2.6 MB)
SNUM PNUM QTY SHIPDATE Functional dependency
S1 P1 3 1-4-90 SNUM, PNUM, SHIPDATE--> QTY
S1 P2 2 2-17-90

36

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

S1 P3 6 11-5-89
S1 P4 2 6-30-90
S1 P5 1 8-12-91
S1 P6 5 4-21-91
S2 P1 3 5-3-90
S2 P2 4 12-31-90
S3 P3 4 3-25-91
S3 P5 2 3-27-91
S4 P2 2 10-31-89
S4 P4 3 7-14-90
S4 P5 7 8-20-90
S5 P5 5 8-11-91
NOT Third Normal Form

TABLE SUPPLIER (200 rows, 37 bytes/row => 7.4 KB)

SNUM SNAME STATUS CITY Functional dependencies
S1 SMITH 20 LONDON SNUM --> SNAME, STATUS, CITY
S2 JONES 10 PARIS CITY --> STATUS
S3 BLAKE 10 PARIS
S4 CLARK 20 LONDON
S5 ADAMS 30 ATHENS
Decomposition of Table Supplier into two Third Normal Form (3NF) Tables

37

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Third Normal Form

TABLE SUPPLIER_W/O_STATUS (200 rows, 35 bytes/row => 7 KB)

SNUM SNAME CITY Functional dependency
S1 SMITH LONDON SNUM --> SNAME, CITY
S2 JONES PARIS
S3 BLAKE PARIS
S4 CLARK LONDON
S5 ADAMS ATHENS

TABLE CITY_AND_STATUS (100 rows, 12 bytes/row => 1.2 KB)

CITY STATUS Functional dependency
LONDON 20 CITY --> STATUS
PARIS 10
ATHENS 30

38

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Relational tables predicted by the ER model, with no functional dependencies

given, just those implied by the diagram.
Table 1: emphistory (jobtitle, startdate, enddate, empid)
Table 2: employee (empid, empname, phoneno, officeno, projno,deptno)
Table 3: project (projno, projname, startdate, enddate)
Table 4: dept (deptno, deptname, mgrid)

39

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Example of Table Design and Normalization (3NF)

from a collection of FDs and an ER diagram

Functional dependencies (FDs) given

empid, startdate --> jobtitle, enddate
empid --> empname, phoneno, officeno, projno, deptno
phoneno --> officeno
projno --> projname, startdate, enddate
deptno --> deptname, mgrid
mgrid --> deptno

In general, the FDs can be derived from

1. Explicit assertions given
2. ER diagram (implied by ER constructs)
3. Intuition (your experience with the problem data)

Table 1: empid, startdate --> jobtitle, enddate

This table has a composite key that must be separated from functional dependencies (FDs) that involve
any individual component of this key (e.g. empno) on the left side.
Table 2
Let us start with the following set of FDs and then refine them, eliminating transitive dependencies
within the same table.

Given: empid --> empname, phoneno, officeno, projno, deptno

phoneno --> officeno

We need to eliminate the redundant right sides of the transitive dependencies (office_no) and put them
into other tables. Thus we get:

Table 2a: empid --> empname, phoneno, projno, deptno

Table 2b: phoneno --> officeno

Table 3: projno --> projname, startdate, enddate

Table 4: deptno --> deptname, mgrid

mgrid --> deptno

40

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Functional Dependency Inference rules

(Armstrong’s Axioms)

1. Reflexivity
If Y is a subset of the attributes of X, then X->Y.
X = ABCD, Y = ABC => X->Y
X->X trivial case

2. Augmentation
If X->Y and Z is a subset of table R (i.e. Z is any set of attributes in R), then XZ -> YZ .

3. Transitivity
If X->Y and Y->Z then X->Z.

4. Pseudo-transitivity
If X->Y and YW->Z then XW->Z.
(transitivity is a special case of pseudo-transitivity when W is null)

5. Union
If X->Y and X->Z then X->YZ.
6. Decomposition
If X->YZ then X->Y and X->Z.

Superkey Rule 1. Any FD involving all attributes of a table defines

a super-key on the LHS of the FD.

Given: any FD containing all attributes in the table R(W,X,Y,Z), i.e. XY -> WZ.
Proof:
(1) XY -> WZ given
(2) XY -> XY by the reflexivity axiom
(3) XY -> XYWZ by the union axiom
(4) XY uniquely determines every attribute in table R, as shown in (3)
(5) XY uniquely defines table R, by the definition of a table as having no duplicate rows
(6) XY is therefore a super-key, by the definition of a super-key.
Super-key Rule 2. Any attribute that functionally determines a
Super-key of a table, is also a super-key for that table.

Given: Attribute A is a super-key for table R(A,B,C,D,E), and E -> A.

Proof:
(1) Attribute A uniquely defines each row in table R, by the def. of a super-key
(2) A -> ABCDE by the definition of a super-key and a relational table
(3) E -> A given
(4) E -> ABCDE by the transitivity axiom
(5) E is a super-key for table R, by the definition of a super-key.

41

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

3NF Synthesis Algorithm (Bernstein)

Basic definitions
g e H set of FDs

H+ closure of H - set of all FDs derivable from H using all the FD inference rules

H’ cover of H - any set of FDs from which every FD in H+ can

be derived

H’(non-redundant) – non-redundant cover of H, i.e. a cover which contains no proper subset

which is also a cover. Can be determined with quadratic complexity O(n2).
Example
Given a set of FDs H, determine a minimal set of tables in 3NF,
while preserving all FDs and maintaining only lossless decomposition/joins.
H: AB->C DM->NP D->KL
A->DEFG D->M
E->G L->D
F->DJ PR->S
G->DI PQR->ST

Step 1: Eliminate any extraneous attributes in the left hand

sides of the FDs. We want to reduce the left hand sides of as many FDs as possible. In general:
XY->Z and X->Z => Y is extraneous (Reduction Rule 1)
XYZ->W and X->Y => Y is extraneous (Reduction Rule 2)
For this example we mix left side reduction with the union and decomposition axioms:
DM->NP => D->NP => D -> MNP
D->M D->M

PQR->ST => PQR->S, PQR->T => PQR->.T

PR->S PR->S PR->S
Step 2: Find a non-redundant cover H’ of H, i.e. eliminate any FD
derivable from others in H using the inference rules (most frequently the transitivity axiom).
A->E->G => eliminate A->G from the cover
A->F->D => eliminate A->D from the cover

Step 3: Partition H’ into tables such that all FDs with the
same left side are in one table, thus eliminating any non-fully functional FDs. (Note: creating tables
at this point would be a feasible solution for 3NF, but not necessarily minimal.)
R1: AB->C R4: G->DI R7: L->D
R2: A->EF R5: F->DJ R8: PQR->T
R3: E->G R6: D->KLMNP
R9: PR->S

42

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Step 4: Merge equivalent keys, i.e. merge tables where all FD’s satisfy 3NF.

4.1 Write out the closure of all LHS attributes resulting from Step 3, based on transitivities.

4.2 Using the closures, find tables that are subsets of other groups and try to merge them. Use Rule
1 and Rule 2 to establish if the merge will result in FDs with super-keys on the LHS. If not, try
using the axioms to modify the FDs to fit the definition of super-keys.

4.3 After the subsets are exhausted, look for any overlaps among tables and apply Rules 1 and 2
(and the axioms) again.

In this example, note that R7 (L->D) has a subset of the attributes of R6 (D->KLMNP). Therefore
we merge to a single table with FDs D->KLMNP, L->D because it satisfies 3NF: D is a super-key
by Rule 1 and L is a super-key by Rule 2.

Final 3NF (and BCNF) table attributes, FDs, and candidate keys:
R1: ABC (AB->C with key AB) R5: DFJ (F->DJ with key F)
R2: AEF (A->EF with key A) R6: DKLMNP (D->KLMNP, L->D, w/keys D, L)
R3: EG (E->G with key E) R7: PQRT (PQR->T with key PQR)
R4: DGI (G->DI with key G) R8: PRS (PR->S with key PR)
Step 4a. Check to see whether all tables are also BCNF. For any table that is not BCNF,
add the appropriate partially redundant table to eliminate the delete anomaly.

43

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Maier’s Example using 3NF Synthesis

[Maier,D. The Theory of Relational Databases, Computer Science Press, 1983]
R = {A,B,C,D,E,F,G,H,I,J,K }
Functional dependencies (FDs):
(1) E --> A B C D F G H I J K (7) H I --> J
(2) A B C --> E D F G H I J K (8) I J --> H
(3) A B D --> E C F G H I J K (9) H J --> I
(4) G --> H I J
(5) C F --> K
(6) D F --> K

Step 1 - No reduction of determinants necessary.

Step 2 - Find non-redundant cover.
(4) G->HIJ => eliminate HIJ from (1), (2), and (3)
(7) HI->J => reduce (4) to G->HI, eliminating J from (4)
(5) CF -> K => eliminate K from (1) and (3)
(6) DF->K => eliminate K from (2)
(1) E->DFG => eliminate DFG from (2)
(1) E->CFG => eliminate CFG from (3)
Step 3 - Partition into groups with the same left side.
G1: E->ABCDFG G6: DF->K
G2: ABC->E G7: HI->J
G3: ABD->E G8: IJ->H
G4: G->HI G9: HJ->I
G5: CF->K
Step 4 - Merge equivalent keys, forming new groups. Construct final set of tables, attributes, FDs, and
candidate keys.
R1: ABCDEFG ( E->ABCDFG, ABC->E, ABD->E with keys E, ABC, ABD)
R2: GHI (G->HI with key G)
R3: CFK (CF->K with key CF)
R4: DFK (DF->K with key DF
R5: HIJ (HI->J, IJ->H, HJ->I with keys HI, IJ, HJ)

44

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Example of a 3NF table that is not BCNF,

i.e. it has further anomalies:
S = student, C = course, I = instructor
SC -> I For each course, each student is taught by only one instructor. A course may be taught by more
than one instructor.

I -> C Each instructor teaches only one course.

This table is 3NF with a candidate key SC:

SCI student course instructor
Sutton Math Von Neumann
Sutton Journalism Murrow
Niven Math Von Neumann
Niven Physics Fermi
Wilson Physics Einstein

Delete anomaly: If Sutton drops Journalism, then we have no record of Murrow teaching Journalism.
How can we decompose this table into BCNF?

Decomposition 1 (bad)........eliminates the delete anomaly

SC (no FDs) and I -> C (two tables)
Problems - 1. lossy join
2. dependency SC -> I is not preserved

SC student course IC instructor course

Sutton Math Von Neumann Math
Sutton Journalism Murrow Journalism
Niven Math Fermi Physics
Niven Physics Einstein Physics
Wilson Physics

----------------join SC and IC ------------------

SCI’ student course instructor
Sutton Math Von Neumann
Sutton Journalism Murrow
Niven Math Von Neumann
Niven Physics Fermi
Niven Physics Einstein (spurious row)
Wilson Physics Fermi (spurious row)
Wilson Physics Einstein

45

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Decomposition 2 (better).....eliminates the delete anomaly

SI (no FD) and I -> C
Advantages – eliminates the delete anomaly, lossless
Disadvantage - dependency SC -> I is not preserved

SI student instructor IC instructor course

Sutton Von Neumann Von Neumann Math
Sutton Murrow Murrow Journalism
Niven Von Neumann Fermi Physics
Niven Fermi Einstein Physics
Wilson Einstein Dantzig Math (new)
Sutton Dantzig (new)

The new row is allowed in SI using unique(student,instructor) in the create table command, and the
join of SI and IC is lossless. However, a join of SI and IC now produces the following two rows:
student course instructor
Sutton Math Von Neumann
Sutton Math Dantzig which violates the FD SC -> I.

Oracle, for instance, has no way to automatically check SC->I, although you could write a procedure to
do this at the expense of a lot of overhead.
Decomposition 3 (tradeoff between integrity and performance)
SC -> I and I -> C (two tables with redundant data)
Problems -extra updates and storage cost

46

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Fourth Normal Form (4NF)

Fourth normal form (4NF) -- a table R is in 4NF iff it is in BCNF and whenever there exists a
nontrivial multi-valued dependency (MVD) in R, say X-->>Y, X is a super-key for R.

Multi-valued dependency (MVD)

X -->> Y holds whenever a valid instance of R(X,Y,Z) contains a pair of rows that contain
duplicate values of X, then the instance also contains the pair of rows obtained by interchanging
the Y values in the original pair.

Multi-valued Dependency Inference rules

(Berri, Fagin, Howard...1977 ACM SIGMOD Proc.)
1. Reflexivity X -->> X
2. Augmentation If X -->> Y, then XZ -->> Y.
3. Transitivity If X -->>Y and Y -->> Z then X -->> (Z-Y).
4. Pseudo-transitivity If X -->> Y and YW -->> Z then
XW -->> (Z-YW).
(transitivity is a special case of pseudo-transitivity when W is null)
5. Union If X -->> Y and X -->> Z then X -->> YZ.
6. Decomposition If X -->> Y and X -->> Z,
then X -->> Y intersect Z and X -->> (Z-Y)
7. Complement If X -->> Y and Z=R-X-Y, then X -->> Z.
8. FD => MVD If X -> Y, then X -->> Y.
9. FD, MVD mix If X -->> Y and Y -> Z’ (where Z’ is contained
in Z, and Y and Z are disjoint), then X->Z’.

Why is 4NF useful?

Avoids certain update anomalies/inefficiencies.

1. delete anomaly - two independent facts get tied together unnaturally so there may be bad side
effects of certain deletes, e.g. in “skills required” the last record of a skill may be lost if employee
is temporarily not working on any projects).

2. update inefficiency - adding a new project in “skills required” requires insertions for many
records (rows) that to include all required skills for that new project. Likewise, loss of a project
requires many deletes.

3. 4NF maintains smaller pieces of information with less redundancy.

47

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Example of a ternary relationship (many-to-many-to-many) that can be

BCNF or 4NF depending on the semantics associated with it.
Table NF 2-way 3-way Nontrivial
name decomp.decomp. MVDs

skill_available BCNF yes yes 6

skill_required BCNF yes yes 2
skill_in_common 4NF no yes 0
Semantics and analysis of each relationship
skill_required—an employee must have all the required skills for a project to work on that project.

skill_required empno project skill Nontrivial MVDs

101 3 A project->>skill
101 3 B project->>empno
101 4 A
101 4 C
102 3 A
102 3 B
103 5 D
empno project empno skill project skill
101 3 101 A 3 A
101 4 101 B 3 B
102 3 101 C 4 A
103 5 102 A 4 C
102 B 5 D
103 D

2-way lossless join occurs when skill_required is projected over {empno, project} and {project, skill}.
Projection over {empno, project} and {empno, skill}, and over {empno, skill} and {project, skill},
however, are not lossless. 3-way lossless join occurs when skill_required is projected over {empno,
project}, {empno, skill}, {project, skill}.

48

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

skill_in_common—an employee must apply the intersection of available skills to the skills needed for
different projects. In other words if an employee has a certain skill and he or she works on a given
project that requires that skill, then he or she must provide that skill for that project (this is less restrictive
than skill_required because the employee need not supply all the required skills, but only those in
common).

skill_in_common empno project skill

101 3 A
101 3 B
101 4 A
101 4 B
102 3 A
102 3 B
103 3 A
103 4 A
103 5 A
103 5 C
empno project empno skill project skill
101 3 101 A 3 A
101 4 101 B 3 B
102 3 102 A 4 A
103 3 102 B 4 B
103 4 103 A 5 A
103 5 103 C 5 C
This has a 3-way lossless decomposition. There are no 2-way lossless decompositions and no MVDs,
thus the table is in 4NF.

49

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

V. Access Methods
Types of Queries
Query type 1: access all records of a given type
“Increase everyone’s salary by 10%”
access method: sequential processing

Query type 2: access at most one record

“Find the address of John Smith,
whose id number is 333-44-5555”
access methods: hashing, B + tree index

Query type 3: access a subset of records of a given type

“Find all employees who have C programming experience and over three years with the
company”
access method: secondary indexing (Oracle uses B+trees for this)

Sequential Access Methods

lra = n logical record accesses
sba = ceil(n/bf ) sequential block accesses
rba = 0 random block accesses

iotime = sba*Tsba + rba*Trba seconds

where Tsba is the average disk i/o service time for a sequential
block and Trba is the average disk i/o service time for a random block
access

Disk service time in a dedicated environment

sequential block access:

Tsba = rot/2 + bks/tr

where rot is the disk rotation time (for a full rotation),
bks is the block size in bytes (bf*record size), and
tr is the disk transfer rate in bytes per second.
Trba = seek(avg) + rot/2 + bks/tr
where seek(avg) is the average seek time over the extent of the file on disk

Disk service time in a shared environment

Tsba = Trba = seek(avg) + rot/2 + bks/tr
where seek(avg) is the average disk seek time over the extent of the entire disk.

50

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Batch processing of k sequentially stored records

read the transaction file:

lra = k where k = number of transaction records
sba = ceil(k/tfbf) where tfbf is the transaction file blocking factor

read the master file:

lra = n
sba = ceil(n/bf) where bf is the master file blocking factor

write a new master file:

lra = n + adds - deletes
sba = ceil((n+adds-deletes)/bf)
where adds is the number of records added or inserted,
and deletes is the number of records deleted.

51

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Random Access Methods

Hashing
Basic mechanism – transformation of a primary key directly to a physical address,
called a bucket (or indirectly via a logical address)

Collisions – handled by variations of chained overflow techniques

random access to a hashed file

lra = 1 + overflow(avg)
rba = 1 + overflow(avg)

insertion into a hashed file

lra = 1 + overflow(avg) + rewrite
rba = 1 + overflow(avg)
rba=1 for the rewrite

52

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Extendible Hashing
* number of buckets grow or contracts
* bucket splits when it becomes full
* collisions are resolved immediately, no long overflow chains
* primary key transformed to an entry in the Bucket Address Table (BAT),
typically in RAM
* BAT has pointers to disk buckets that hold the actual data
* Retrieve a single record = 1 rba (access the bucket in one step)
* Cost (service time) of I/O for updates, inserts, and deletes is the same as for B+-trees

53

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

B-trees and B+-trees

B-tree index basic characteristics
* each node contains p pointers and p-1 records
* each pointer at level i is for a data and pointer block at level i+1
* i=1 denotes the root level (single node or block)
* can be inefficient for searching because of the overhead in each search level

54

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

B+-tree index basic characteristics

* eliminates data pointers from all nodes except the leaf nodes
* each non-leaf index node has p pointers and p-1 key values
* each pointer at level i is for an index block (of key/pointer pairs) at level i+1
* each leaf index has a key value/pointer pair to point to the actual data block
(and record) containing that primary key value
* leaf index nodes can be logically connected via pointers for ordered sequence search
* hybrid method for efficient random access and sequential search

Example: B + -tree
To determine the order of a B+-tree, let us assume that the database has 500,000
records of 200 bytes each, the search key is 15 bytes, the tree and data pointers are 5
bytes, and the index node (and data block size) is 1024 bytes. For this configuration
we have non-leaf index node size = 1024 bytes = p*5 + (p-1)*15 bytes

p = floor((1024+15)/20) = floor(51.95) = 51
number of search key values in the leaf nodes = floor ((1024-5)/(15+5))=50
h = height of the B+-tree (number of index levels, including the leaf index nodes
n = number of records in the database (or file); all must be pointed at from the next to last level, h-1

ph-1(p-1) > n
(h-1)log p + log(p-1) > log n
(h-1)log p > log n-log(p-1)
h > 1 + (log n-log(p-1)) / log p
h > 1 + (log 500,000-log 50)/log 51 = 3.34, h=4 (nearest higher integer)
A good approximation can be made by assuming that the leaf index nodes are
implemented with p pointers and p key values:
ph > n
h log p > log n
h > log n/log p
In this case, the result above becomes h > 3.35 or h = 4.

55

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

B+-tree performance
read a single record (B+-tree) = h+1 rba

update a single record (B+-tree) = search cost + rewrite data block

= (h+1) rba + 1 rba

general update cost for insertion (B+-tree)

=search cost (i.e., h+1 reads)
+simple rewrite of data block and leaf index node pointing to the data
block (i.e., 2 rewrites)
+nos*(write of new split index node
+ rewrite of the index node pointer to the new index node)
+ nosb*(write of new split data block)

= (h+1) rba + 2 rba + nos*(2 rba) + nosb*(1 rba)

where nos is the number of index split node operations required and nosb is the
number of data split block operations required

general update cost for deletion (B+-tree)

= search cost (i.e., h+1 reads)
+ simple rewrite of data block and leaf index node pointing to the data
block (i.e., 2 rewrites)
+ noc*(rewrite of the node pointer to the remaining node)

= (h+1) rba + 2 rba + noc*(1 rba)

where noc is the number of consolidations of index nodes required.

As an example, consider the insertion of a node (with key value 77) to the B+-tree
shown in Fig. 6.6. This insertion requires a search (query) phase and an insertion
phase with one split node. The total insertion cost for height 3 is

insertion cost = (3 + 1) rba search cost + (2 rba) rewrite cost

+ 1 split *(2 rba rewrite cost)
= 8 rba

56

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

57

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Secondary Indexes
Basic characteristics of secondary indexes
* based on Boolean search criteria (AND, OR, NOT) of attributes that are not
the primary key

* attribute type index is level 1 (usually in RAM)

* attribute value index is level 2 (usually in RAM)

* accession list is level 3 (ordered list of pointers to blocks containing records

with the given attribute value)

* one accession list per attribute value; pointers have block address and record
offset typically

* accession lists can be merged to satisfy the intersection (AND) of records

that satisfy more than one condition

Boolean query cost (secondary index)

= search attribute type index + search attribute value index
+ search and merge m accession lists + access t target records

= (0 + 0 + sum of m accession list accesses) rba + t rba

= (sum of m accession list cost) rba + t rba

where m is the number of accession lists to be merged and t is the number of
target records to be accessed after the merge operation.

accession list cost (for accession list j) = ceil(pj/bfac) rba

where pj is the number of pointer entries in the jth accession list and bfac is
the blocking factor for all accession lists

bfac = block_size/pointer_size

* assume all accesses to the accession list are random due to dynamic re-allocation of
disk blocks

• use the 1% rule

(any variable affecting the result by less than 1% is ignored)

58

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

59

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Example: Mail Order Business

Assume we have a file of 10,000,000 records of mail order customers for a large
commercial business. Customer records have attributes for customer name, customer
number, street address, city, state, zip code, phone number, employer, job title,
credit rating, date of last purchase, and total amount of purchases. Assume that the
record size is 250 bytes; block size is 5000 bytes (bf=20); and pointer size, including
record offset, is 5 bytes (bfac=1000). The query to be analyzed is “Find all customers
whose job title is ‘engineer’, city is ‘chicago’, and total amount of purchases is
greater than $1,000.” For each AND condition we have the following hit rates, that
is, records that satisfy each condition:
job title is ‘engineer’: 84,000 records
city is ‘chicago’: 210,000 records
total amount of purchases > $1000: 350,000 records
total number of target records that satisfy all three conditions = 750

query cost (inverted file)

= merge of 3 accession lists + access 750 target records

= [ceil(n1/bfac) + ceil(n2/bfac) + ceil(n3/bfac)] sba + 750 rba

= [ceil(84,000/1000) + ceil(210,000/1000) + ceil(350,000/1000] sba

+ 750 rba
= (84+210+350) sba + 750 rba
= 644 sba + 750 rba

If we assume Tsba is 10 milliseconds and Trba is 25 milliseconds, we obtain

query iotime (secondary index)
= 644 sba*10 ms + 750 rba*25 ms
= 25190 ms
= 25.19 sec (much more efficient than sequential scan, see below)

query iotime (sequential scan)

= ceil(n/bf) sba *Tsba
= ceil(10,000,000/20)*10 ms
= 5,000,000 ms
= 5000 sec

60

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Secondary Indexes using B + -trees

* used by Oracle and many others for non-unique indexes
* index nodes contain key/pointer pairs in the same way as a primary key index
using a B+-tree
* key at each level, leaf and non-leaf, is the concatenation of attributes used in
the query , e.g. jobtitle, city, total_purchases (as attributes of consumer)
* leaf node pointers are to the blocks containing records with the given combination
of attribute values indicated in the concatenated keys
* analysis of queries and updates for this type of index proceeds in the same way as
a primary key (unique) index, keeping in mind that the key formats are different
in the two cases

query iotime (B+tree secondary index) = rba*Trba

= [h + ceil(t/bfac) – 1 + t] * Trba

where h is the height of the B+tree index, bfac is the blocking factor for the accession
list (i.e. the number of pointer/key pairs in the leaf nodes in the B+tree), and t is the
number of target records in the table that satisfies all the conditions in the query.

61

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Denormalization
* motivation – poor performance by normalized databases

* search for potential denormalizations that avoid or minimize delete anomalies

To illustrate the effect of denormalization, let us assume that the table review is
associated with the tables employee and manager as the table that follows shows.
The extension of the review table, review-ext, is shown as a means of reducing
the number of joins required in the query shown below. This extension results in a
real denormalization, that is,
review_no -> emp_id -> emp_name, emp_address

with the side effects of add and update anomalies. However, the delete anomaly
cannot occur because the original data is redundant in the extended schema.

Original Tables and Process (Query)

Table Primary Key Nonkeys

employee emp_id emp_name, emp_address, mgr_id

manager mgr_id emp_name, emp_address
review review_no emp_id, mgr_id
Query:
For a given review number, display the employee name and address.
select e.emp_name, e.emp_addr
from employee as e, review as r
where r.review_no = ‘xxxx’
and e.emp_id = r.emp_id;

Extended table review_ext is not in 3NF

create table review_ext as
select r.review_no, e.emp_id, e.emp_name, e.emp_addr, e.mgr_id
from employee as e, review as r
where e.emp_id = r.emp_id;

total cost = [iotime(q) + iotime(u)]*cost(q) + volume(s)*cost(s)

where
cost(q) = unit cost per I/O second for query or update processes
cost(s) = unit cost per byte for stored data
iotime(q) = I/O service time (sec) for query processes
iotime(u) = I/O service time (sec) for update processes
volume(s) = total volume in bytes for stored data

62

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Table Denormalization Algorithm

1. Select the dominant processes based on such criteria as high frequency of
execution, high volume of data accessed, response time constraints, or explicit high
priority.

2. Define join tables, when appropriate, for the dominant processes.

3. Evaluate total cost for storage, query, and update for the database schema, with
and without the extended table, and determine which configuration minimizes total
cost.

4. Consider also the possibility of denormalization due to a join table and its side
effects. If a join table schema appears to have lower storage and processing cost and
insignificant side effects, then consider using that schema for physical design in
addition to the original candidate table schema. Otherwise use only the original
schema.

63

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Join Strategies
1. nested loop: complexity O(mn)
2. merge-join: complexity O(n log2 n)
3. indexed join: complexity O(n)
4. hash-join: complexity O(n)

where m and n are the rows of the two tables to be joined

Assume

* assigned_to table has 50,000 rows

* project table has 250 rows
* let the blocking factors for the assigned_to and project tables be 100 and 50,
respectively, and the block size is equal for the two tables.
the common join column is project_name.
High Selectivity Joins

select p.project_name, p.project_leader, a.emp_id

from project as p, assigned_to as a
where p.project_name = a.project_name;

Nested Loop Case 1: assigned_to is the outer loop table.

join cost = scan assigned_to once, scan project n times

= 50,000/100 + 50,000*250/50
= 500 + 250,000
= 250,500 sequential block accesses (sba)

If a sequential block access requires an average of 10 ms, the total time required is 2505 seconds.
Nested Loop Case 2: project is the outer loop table.

join cost = scan project once, scan assigned_to m times

= 250/50 + 250*50,000/100
= 5 + 125,000
= 125,005 sequential block accesses (or 1250 seconds)

Note that this strategy does not take advantage of row order for these tables

Merge-Join Case 1: Both project and assigned_to are already ordered by

. project_name.

join cost = merge time (to scan both tables)

= 50,000/100 + 250/50
= 505 sequential block accesses (or 5.05 seconds)
Merge-Join Case 2: Only project is ordered by project_name.

join cost = sort time for assigned_to + merge time (to scan both sorted tables)
= (50,000*log 2 50,000)/100 + 50,000/100 + 250/50
= (50,000*16)/100 + 500 + 5
= 8505 sequential block accesses (or 85.05 seconds)

64

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Merge-Join Case 3: Neither project nor assigned_to are ordered by

. project_name.

join cost = sort time for both tables + merge time for both tables
= (50,000*log 2 50,000)/100 +(250*log 2 250)/50 + 50,000/100
+ 250/50
= 8000 + 40 + 500 + 5
= 8545 sequential block accesses (or 85.45 seconds)

We see that the sort phase of the merge-join strategy is the costliest component, but it still
significantly improves performance compared to the nested loop strategy.
Low Selectivity Joins

Let ntr=100 qualifying rows for the foreign key table (assigned_to) and let ntr=1 row for the
primary key table (project) in the example below. Assume h=2 for the unique index to project, Tsba
= 10 ms, and Trba = 40 ms.

select p.project_name, p.project_leader, a.emp_id

from project as p, assigned_to as a
where p.project_name = a.project_name
and p.project_name = 'financial analysis';

Indexed join Case 1: Scan foreign key table once and index to the primary key
join cost = scan the entire foreign key table (assigned_to)
+ index to the primary key table (project) qualifying row
= 50,000/100 sba + (h+1) rba
= 500 sba + 3 rba (or 5.12 seconds)
For the next case, assume the nonunique index height, hn = 3, index blocking factor
bfac = 500, with ntr = 100 target foreign key rows as given above.
Indexed join Case 2: Index to both the primary key table and the foreign key
Join cost = index to the primary key table + index to the foreign key table
= (h+1) rba + [hn + ceil(ntr/bfac) – 1 + ntr] rba
= 3 rba + [3 + 0 + 100] rba
= 106 rba (or 4.24 seconds)

Indexed join Case 3: Nonunique indexes required for both tables due to join on
two nonkeys.
Join cost = index to the first table + index to the second table
= [h1 + ceil(ntr1/bfac1) –1 + ntr1] rba
+ [h2 + ceil(ntr2/bfac2) –1 + ntr2] rba

Hash join Case 1:

join cost = scan first table (assigned_to) + scan second table (project)
+ access qualifying rows in the two tables
= 50,000/100 sba + 250/50 sba + 100 rba + 1 rba
= 505 sba + 101 rba (or 9.09 seconds)

In the hash join strategy, the table scans may only have to be done infrequently as
long as the hash file in RAM remains intact for a series of queries, so in Case 1
above, the incremental cost for the given query requires only 101 rba or 4.04
seconds.

65

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

VI. Database Distribution Strategies

Overview of Distributed Databases
Distributed database - a collection of multiple, logically interrelated databases distributed over a
computer network [OzVa91].

Distributed Database Management System (DDBMS) - a software system that permits the
management of a distributed database and makes the distribution transparent to the users. If
heterogeneous, it may allow transparent simultaneous access to data on multiple dissimilar systems.

Advantages

1. Improves performance, e.g. it saves communication costs and reduces query delays by providing data
at the sites where it is most frequently accessed.

2. Improves the reliability and availability of a system by providing alternate sites from where the
information can be accessed.

3. Increases the capacity of a system by increasing the number of sites where the data can be located.

4. Allows users to exercise control over their own data while allowing others to share some of the data
from other sites.

5. Helps solve more complex database problems.

Disadvantages

1. Increases the complexity of the system and introduces several technical as

well as management challenges especially when geographical and organizational boundaries are crossed.

2. Makes central control more difficult and raises several security issues because a data item stored at a
remote site can be always accessed by the users at the remote site.

3. Makes performance evaluation difficult because a process running at one node may impact the entire
network.

66

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

67

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Requirements of a Generalized DDBMS: Date’s 12 Rules

Rule 1. Local Autonomy. Local data is locally owned and managed, even when it is accessible by
a remote site. Security, integrity, and storage remain under control of the local system. Local users
should not be hampered when their system is part of a distributed system.

Rule 2. No Central Site. There must be no central point of failure or bottleneck. Therefore the
following must be distributed: dictionary management, query processing, concurrency control, and
recovery control.

Rule 3. Continuous Operation. The system should not require a shutdown to add or remove a
node from the network. User applications should not have to change when a new network is added,
provided they do not need information from the added node.

Rule 4. Location Independence (or Transparency). A common global user view of the
database should be supported so that users need not know where the data is located. This allows data to
be moved for performance considerations or in response to storage constraints without affecting the user
applications.

Rule 5. Fragmentation Independence (or Transparency). This allows tables to be split

among several sites, transparent to user applications. For example, we can store New York employee
records at the New York site and Boston employees at the Boston site, but allow the user to refer to the
separated data as EMPLOYEES, independent of their locations.

Rule 6. Replication Independence (or Transparency). This allows several copies of a table
(or portions thereof) to reside at different nodes. Query performance can be improved since applications
can work with a local copy instead of a remote one. Update performance, however, may be degraded due
to the additional copies. Availability can improve.

Rule 7. Distributed Query Processing. No central site should perform optimization; but the
submitting site, which receives the query from the user, should decide the overall strategy. Other
participants perform optimization at their own levels.

Rule 8. Distributed Transaction Processing. The system should process a transaction across
multiple databases exactly as if all of the data were local. Each node should be capable of acting as a
coordinator for distributed updates, and as a participant in other transactions. Concurrency control must
occur at the local level (Rule 2), but there must also be cooperation between individual systems to ensure
that a “global deadlock” does not occur.

Rule 9. Hardware Independence. The concept of a single database system must be presented
regardless of the underlying hardware used to implement the individual systems.

Rule 10. Operating System Independence. The concept of a single database system must be
presented regardless of the underlying operating systems used.

Rule 11. Network Independence. The distributed system must be capable of communicating over
a wide variety of networks, often different ones in the same configuration. Standard network protocols
must be adhered to.

Rule 12. DBMS Independence (Heterogeneity). The distributed system should be able to be
made up of individual sites running different database management systems.

68

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

What are the basic issues in the design and implementation of

distributed database systems?
* Data Distribution Strategies
- Fragmentation
- Data allocation
- Replication
- Network data directory distribution

* Query Processing and Optimization

* Distribution Transparency
- location, fragmentation, replication, update

* Integrity
- Transaction management
- Concurrency control
- Recovery and availability
- Integrity constraint checking

* Privacy and Security

- Database administrators

* Data Manipulation Languages

- SQL is the standard
- Forms coming into common use

69

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Modified Life Cycle for Data Distribution

IV. Data distribution (allocation). Create a data allocation schema that indicates
where each copy of each table is to be stored. The allocation schema defines at which site(s) a table is
located. A one-to-one mapping in the allocation schema results in non-redundancy, while a one-to-
many mapping defines a redundant distributed database.

Fragmentation.
Fragmentation is the process of taking subsets of rows and/or columns of tables as the smallest unit of
data to be sent across the network. Unfortunately, very few commercial systems have implemented this
feature, but we include a brief discussion for historical reasons. We could define a fragmentation
schema of the database based on dominant applications’ “select” predicates (set of conditions for
retrieval specified in a select statement).

Horizontal fragmentation partitions the rows of a global fragment into subsets. A fragment r1 is a
selection on the global fragment r using a predicate Pi, its qualification. The reconstruction of r is
obtained by taking the union of all fragments.

Vertical fragmentation subdivides the attributes of the global fragment into groups. The simplest
form of vertical fragmentation is decomposition. A unique row-id may be included in each fragment to
guarantee that the reconstruction through a join operation is possible.

Mixed fragmentation is the result of the successive application of both fragmentation techniques.

Rules for Fragmentation

1. Fragments are formed by the select predicates associated with

dominant database transactions. The predicates specify attribute
values used in the conjunctive (AND) and disjunctive (OR) form of
select commands, and rows (records) containing the same values form
fragments.

2. Fragments must be disjoint and their union must become the whole
fragment. Overlapping fragments are too difficult to analyze and
implement.
3. The largest fragment is the whole table. The smallest table is a
single record. Fragments should be designed to maintain a
balance between these extremes.

70

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Data Distribution
Data distribution defines the constraints under which data allocation strategies may operate. They are
determined by the system architecture and the available network database management software. The
four basic data distribution approaches are :

* Centralized
In the centralized database approach, all the data are located at a single site. The implementation of
this approach is simple. However, the size of the database is limited by the availability of the secondary
storage at the central site. Furthermore, the database may become unavailable from any of the remote
sites when communication failures occur, and the database system fails totally when the central site fails.

* Partitioned
In this approach, the database is partitioned by tables, and each table is assigned to a particular site.
This strategy is particularly appropriate where either local secondary storage is limited compared to the
database size, the reliability of the centralized database is not sufficient, or operating efficiencies can be
gained through the exploitation of the locality of references in database accesses.

* Replicated
The replicated data distribution strategy allocates a complete copy of the database to each site in the
network. This completely redundant distributed data strategy is particularly appropriate when reliability is
critical, the database is small, and update inefficiency can be tolerated.
* Hybrid
The hybrid data distribution strategy partitions the database into critical and non-critical tables.
Non-critical tables need only be stored once, while critical tables are duplicated as desired to meet the
required level of reliability.

71

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Distributed Database Requirements

Database Description

1. Conceptual schema (ER diagram)

2. Transactions: functions and data accessed

Configuration Information

1. Sources of data—where data can be located.

2. Sinks of data—where user transactions can be initiated and

data transferred.

3. Transaction rate (frequency) and volume (data flow).

4. Processing capability at each site—CPU and I/O capability

(speed).
5. Security—data ownership (who can update) and access
authorization (who can query) for each transaction.

6. Recovery—estimated frequency and volume of backup

operations.

7. Integrity — referential integrity, concurrency control, journaling, overhead, etc.

Constraints

1. Network topology: Ethernet, token ring, ATM

2. Processing capability needed at each site.

3. Channel (link) transmission capacity.

4. Availability—related to mean-time-between-failures (MTBF) and

mean-time-to-repair (MTTR).
Objective Functions

1. Response time as a function of transaction size.

2. Total system cost—communications, local I/O, cpu time, disk space.

72

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

The General Data Allocation Problem

Given
1. the application system specifications
- A database global schema.
- A set of user transactions and their frequencies.
- Security, i.e. data ownership (who can update) and access authorization (who can query) for
each transaction.
- Recovery, estimated frequency and volume of backup operations.

2. The distributed system configuration and software:

- The network topology, network channel capacities, and network control mechanism.
- The site locations and their processing capacity (CPU and I/O processing).
- Sources of data (where data can be located), and sinks of data (where user transactions can be
initiated and data transferred).
- The transaction processing options and synchronization algorithms.
- The unit costs for data storage, local site processing, and communications.
Find

the allocation of programs and database tables to sites which minimizes C, the total cost:
C = Ccomm + Cproc + Cstor
where:
Ccomm = communications cost for message and data.
Cproc = site processing cost (CPU and I/O).
Cstor = storage cost for data and programs at sites.

subject to possible additional constraints on:

* Transaction response time which is the sum of communication delays, local processing,
and all resource queuing delays.

* Transaction availability which is the percentage of time the transaction executes with all
components available.

73

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

The Non-Redundant “Best Fit” Method

A general rule for data allocation states that data should be placed as close as possible to where it will be
used, and then load balancing should be considered to find a global optimization of system
performance.

The non-redundant “best fit” method determines the single most likely site to allocate a table based on
maximum benefit, where benefit is interpreted to mean total query and update references. In particular,
place table Ri at the site s* where the number of local query and update references by all the user
transactions are maximized.

Example
System Parameters
Avg. Service Time Avg. Service Time
Table Size Local Query(Update) Remote Query(Update)
R1 300 KBytes 100 ms (150 ms) 500 ms (600 ms)
R2 500 KBytes 150 ms (200 ms) 650 ms (700 ms)
R3 1.0 Mbytes 200 ms (250 ms) 1000 ms (1100 ms)
User transactions are described in terms of their frequency of occurrence, which tables they access, and
whether the accesses are reads or writes.

Transact Site(s) Frequency Table Accesses (Reads,Writes)

T1 S1,S4,S5 1 R1 (3 reads, 1 write), R2 (2 reads)

T2 S2,S4 2 R1 (2 reads), R3 (3 reads, 1 write)
T3 S3,S5 3 R2 (3 reads, 1 write), R3 (2 reads)

Security: User transactions T1,T2,T3 can either query or update (no

restrictions).
Sources of data: All sites S1 - S5.
Sinks of data (possible locations of transactions): All sites S1 - S5.

74

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Local Reference Computations

Our goal is to compute the number of local references to each table residing at each site, one by one.
The site that maximizes the local references to a given table is chosen as the site where that table should
reside.
Table Site Trans. T1(freq) T2(freq) T3(freq) Total local refs
R1 S1 3 read,1 write(1) 0 0 4
S2 0 2 read(2) 0 4
S3 0 0 0 0
S4 3 read,1 write(1) 2 read(2) 0 8 (max.)
S5 3 read,1 write(1) 0 0 4

R2 S1 2 read(1) 0 0 2
S2 0 0 0 0
S3 0 0 3 read,1 write(3) 12
S4 2 read(1) 0 0 2
S5 2 read(1) 0 3 read,1 write(3) 14 (max.)

R3 S1 0 0 0 0
S2 0 3 read,1 write(2) 0 8 (max.)
S3 0 0 2 read(3) 6
S4 0 3 read,1 write(2) 0 8 (max.)
S5 0 0 2 read(3) 6

Local references for each table at each of five possible sites.

Allocation Decision

Allocate R1 at site S4.

Allocate R2 at site S5.
Allocate R3 at either site S2 or S4
Additional information is needed to choose this allocation. For instance, if maximum availability
of data is a major consideration, then choose site S2 because site S4 already has table R1 allocated to it
and putting R3 there as well would decrease the potential availability of data should site S4 crash.

Advantages
- simple algorithm

Disadvantages
- number of local references may not accurately characterize time or cost (reads and writes
given equal weights)
- no insights regarding replication

75

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

76

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

The Redundant “All Beneficial Sites” Method

This method can be used for either the redundant or non-redundant case. It selects all sites for a table
allocation where the benefit is greater than the cost for one additional copy of that table. You are
assumed to start with zero copies.

The benefit for table R at site S is measured by the difference in elapsed time to do a remote query to
table R from site S (i.e. no replicated copy available locally) and a local query to table R at site S (i.e.
replicated copy available locally). Total benefit for table R at site S is the weighted sum of benefit for
each query times the frequency of queries.

The cost for table R at site S is the total elapsed time for all the local updates of table R, plus the total
elapsed time for all the remote updates for the given table at that site. Total cost for table R at site S is
weighted sum of cost for each update transaction times the frequency of update transactions.

Example
Cost/Benefit Computations for “All Beneficial Sites”

Table Site Remote updates (local updates) No. of writes*freq*time Cost

R1 S1 T1 from S4 and S5 (T1 from S1) 2*1*600 ms +(1*1*150) 1350 ms
S2 T1 from S1, S4, S5 3*1*600 ms 1800 ms
S3 T1 from S1, S4, S5 3*1*600 ms 1800 ms
S4 T1 from S1 and S5 (T1 from S4) 2*1*600 ms +(1*1*150) 1350 ms
S5 T1 from S1 and S4 (T1 from S5) 2*1*600 ms +(1*1*150) 1350 ms

R2 S1 T3 from S3 and S5 2*3*700 ms 4200 ms

S2 T3 from S3 and S5 2*3*700 ms 4200 ms
S3 T3 from S5 (T3 from S3) 1*3*700 ms +(1*3*200) 2700 ms
S4 T3 from S3 and S5 2*3*700 ms 4200 ms
S5 T3 from S3 (T3 from S5) 1*3*700 ms +(1*3*200) 2700 ms

R3 S1 T2 from S2 and S4 2*2*1100 ms 4400 ms

S2 T2 from S4 (T2 from S2) 1*2*1100 ms +(1*2*250) 2700 ms
S3 T2 from S2 and S4 2*2*1100 ms 4400 ms
S4 T2 from S2 (T2 from S4) 1*2*1100 ms +(1*2*250) 2700 ms
S5 T2 from S2 and S4 2*2*1100 ms 4400 ms

Table Site Query (read) sources No. of reads*freq*(remote-local time) Benefit

R1 S1 T1 at S1 3*1*(500 - 100) 1200 ms
S2 T2 at S2 2*2*(500 - 100) 1600 ms
S3 None 0 0
S4 T1 and T2 at S4 (3*1 + 2*2)*(500 - 100) 2800 ms**
S5 T1 at S5 3*1*(500 - 100) 1200 ms

77

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

R2 S1 T1 at S1 2*1*(650 - 150) 1000 ms

S2 None 0 0
S3 T3 at S3 3*3*(650 - 150) 4500 ms**
S4 T1 at S4 2*1*(650 - 150) 1000 ms
S5 T1 and T3 at S5 (2*1 + 3*3)*(650 - 150) 5500 ms**

R3 S1 None 0 0
S2 T2 at S2 3*2*(1000 - 200) 4800 ms**
S3 T3 at S3 2*3*(1000 - 200) 4800 ms**
S4 T2 at S4 3*2*(1000 - 200) 4800 ms**
S5 T3 at S5 2*3*(1000 - 200) 4800 ms**

**sites where benefit > cost

Cost and benefit for each table located at five possible sites.

Advantages
- simple algorithm
- can be applied to either redundant or non-redundant case
- reads and writes given appropriate weights

Disadvantages
- global averages of query and update time may not be realistic
- network topology and protocols not taken into account

78

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

VII. Data Warehousing, OLAP, and Data Mining

Data warehouse – a large repository of historical data that can be integrated for decision support

OLTP Data Warehouse

Transaction oriented Subject oriented
Thousands of users Few users (typically under 100)
Small (MB up to several GB) Large (100s of GB up to several TB)
Current data Historical data
Normalized data (many tables, Denormalized data (few tables,
few columns per table) many columns per table)
Continuous updates Batch updates
Simple to complex queries Usually very complex queries

Table 9 1 Comparison between OLTP and Data Warehouse databases

79

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Core Requirements for Data Warehousing

1. DWs are organized around subject areas.

2. DWs should have some integration capability.

3. The data is considered to be nonvolatile and should be mass loaded.

4. Data tends to exist at multiple levels of granularity.

5. The DW should be flexible enough to meet changing requirements rapidly. .

6. The DW should have a capability for rewriting history, that is, allowing “what-if” analysis.

7. A usable DW user interface should be selected.

8. Data should be either centralized or distributed physically.

80

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Data Warehouse Life Cycle

I. Requirements analysis and specification
1.1 Analyze the end-user requirements and develop a requirements specification. This step follows
the practice used by conventional relational databases (see Chapter 1).

1.2 Define the DW architecture and do some initial capacity planning for servers and tools. Integrate
the servers, storage elements, and client tools.

1.3 Use enterprise data modeling

II. Logical database design

Design the enterprise DW schema and views.

Star schema is the most often used format –- good performance, ease of use

Fact table (one) – very large table containing numeric and/or non numeric attributes, including
the primary keys from the dimension tables; similar to intersection tables between entities with
many-to-many relationships

Dimension tables (several) - smaller tables containing mostly non numeric attributes; similar to
relational tables based on entities

Snowflake schema – similar to star schema, except dimension tables are normalized

Fact table family (constellation) – multiple fact tables interact with dimension tables

81

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

82

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Figure 9.6 Snowflake schema for the “order” data warehouse

III. Physical database design

3.1 Indexing (access methods)

join indexes – used to map dimension tables to the fact table efficiently

bit map indexes – used for low selectivity queries

3.2 View materialization – associated with aggregation of data by one or more dimensions
such as time or location

3.3 Partitioning – horizontal or vertical subsets of tables to enhance performance

83

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

IV. Data distribution

Define data placement, partitioning, and replication.

V. Database implementation, monitoring, and modification

5.1 Connect the data sources using gateways, ODBC drivers, etc.

5.2 Design and implement scripts for data extraction, cleaning, transformation, load, and refresh.

84

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

5.3 Populate the repository with the schema and view definitions, scripts, and other metadata.

5.4 Design and implement end-user applications. Rollout the DW and applications.

85

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

On-Line Analytical Processing (OLAP)

Common Features of Multidimensional Databases (MDD)

1. Dimensions –- perspectives or entities about the real world

2. Hypercubes –- basic structure for multidimensional databases

3. Hierarchies –- certain dimensions are hierarchical in nature

4. Formulas –- derived data values can be defined by formulas (sum, average, etc.)

5. Links – links are needed to connect hypercubes and their data sources

OLAP Logical Design

Step 1 – Analyze the end-user requirements and environment

Step 2 – Define cubes, dimensions, hierarchies, and links (high level)

Step 3 – Define dimension members (low level)

Step 4 – Define aggregations and other formulas (derived data)

Aggregation Issues

1. Which data to aggregate

2. How to store aggregate data

3. When to pre-aggregate derived data

• Pre-aggregate nothing

• Pre-aggregate nothing, but save the materialized view (dynamic)

• Pre-aggregate everything (static)

• Pre-aggregate selectively, based on known statistics

86

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Example of aggregation: 3-dimensional problem

Dimensions Product Region Time-period

Levels all-products world-wide 10-year

product-type country 5-year
-------------- state 3-year
product-name city year
-------- quarter
store month
week
--------
day
No. of aggregate
levels 2 4 7

Potential variables: quantity sold, quota, gross revenue

Potential further dimensions: customer, salesperson, type-of-sale

Subset of the 3-dimensional problem

Dimensions Product Region Time-period

Levels product-type state month

-------------- ----- -------
product-name store day

No. of aggregate 1 1 1
levels

Statistics on aggregates and hierarchies of aggregates

Number of dimensions = d = 3
Number of possible views = 2d-1 = 7
1. product
2. region
3. time-period
4. product, region
5. product, time-period
6. region, time-period
7. product, region, time-period

Number of aggregate levels for dimension i = ni

Number of one-way aggregates

= Σ i=1,d ni = 1 + 1 + 1 = 3 for the subset

= Σ i=1,d ni = 2 + 4 + 7 = 13 for the full set

87

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Number of two-way aggregates

= Σ i=1,d-1 Σ j>i,d ni n j = 1*1 + 1*1 + 1*1 = 3 for the subset

= Σ i=1,d-1 Σ j>i,d ni n j = 2*4 + 2*7 + 4*7 = 50 for the full set

Number of three-way aggregates

= Σ i=1,d-2 Σ j>i,d-1 Σ k>j,d ninjnk = 1*1*1 = 1 for the subset

= Σ i=1,d-2 Σ j>i,d-1 Σ k>j,d ninjnk = 2*4*7 = 56 for the full set

Number of d-way aggregates (in general)

= n1n2n3 ............. n d

Total number of aggregates

= 3 + 3 + 1 = 7 for the subset
= 12 + 50 + 56 = 118 for the full set

Subset configuration

One-way aggregates
Product-type totals by store by day
State totals by product-name by day
Monthly totals by product-name by store

Two-way aggregates
Product-type totals by state totals by day
Product-type totals by month totals by store
State totals by monthly totals by product-name

Three-way aggregates
Product-type totals by monthly totals by state totals

88

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Number of Records in Aggregates

Dimensions Product Region Time-period

Levels all-products = 1 world-wide = 1 10-year = 1
product-type = 210 country = 100 5-year = 2
--------------------- state = 4,500 3-year = 3
product-name = 4,050 city = 15,000 year = 1*10
-------------- quarter = 4*10
store = 25,000 month = 12*10
week = 52*10
---------------
day = 365*10

Number of records (rows) in each1-way aggregate = individual cell = nr ij

for level i and dimension j

Number of records (rows) in each 2-way aggregate = nr ij nr km

for levels i,k and dimensions j,m

Number of records (rows) in each d-way aggregate = nr i1nr k2............ nr nd

for levels i,k,.....,n and dimensions 1,2,.....,d

Subset configuration Full set configuration

Smallest 3-way
aggregate fact table 210*4500*120 1*1*1 = 1 record
= 113,400,000 records

Largest 3-way
aggregate fact table 210*4500*120 210*15000*520
= 113,400,000 records = 1,638,000,000 records

Number of fact records

in the data warehouse 4050*25000*3650 4050*25000*3650
(including nulls) = 369,562,500,000 = 369,562,500,000

89

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

90

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

91

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

92

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

Data Mining
Definition – data mining is the activity of sifting through large files and databases to discover useful,
nonobvious, and often unexpected trends and relationships

The Knowledge Discovery in Databases (KDD) Process

1. Data selection and cleaning

2. Data transformation and reduction

3. Data mining

4. Interpretation and evaluation

5. Taking action on the discovery

Data Mining Methods

1. Predictive modeling

2. Database segmentation

3. Data summarization and link analysis

4. Dependency analysis or modeling

5. Change and deviation analysis

6. Optimization searching

93

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

94

Downloaded by Hot Boy ([email protected])

 lOMoARcPSD|56591169

95

Downloaded by Hot Boy ([email protected])