CS 222
Database Management System
Spring 2010-11
Lecture 4
Database Design Theory
Korra Sathya Babu
Department of Computer Science
NIT Rourkela
1
Unit Overview
Database Design Problem
Functional Dependency
Axioms, Logical Implications of FDs, Redundant FDs, Closure, Equivalence
of FDS, extraneous attributes, covers
Decomposition
Redundancy and Anomaly
Rules of Decomposition, Test for Lossless Join and Dependency
Preservation
Normalization
Normal Forms, Multivalued Dependency, Join Dependency,
Denormalization
Modeling
year
title
Movies
name
Stars-In
address
Stars
name
length
filmType
Owns
Studios
address
Relational Database Design
Automatic mappings from E/R to relations may not produce the
best relational design possible
Suggested Design Strategy
Real-world to E/R model to Relational schema to Better relational schema
to Relational DBMS
Database designers sometimes go directly from Real-world to
Relational schema, in which case the relational design could be
really bad.
Many problems may arise if the design is not careful
Problems using bad design
Redundancy
Insertion Anomaly
Updation Anomaly
Deletion Anomaly
Functional Dependency
Definition
A functional dependency (FD) has the form XY, where X
and Y are sets of attributes in a relation R. Formally, XY
means that whenever two tuples in R agree on all the
attributes of X, they must also agree on all the attributes
of Y.
Movies (title, year, length, filmtype,studioName, starName)
FDs we can reasonable assert are: Title, year length ; Title,
year filmType; Title, year studioName
Trivial Dependency
Trivial: A fd A1A2An B is said to be trivial if B is one of
the As. ex. title year title
Nontrivial: atleast one of the Bs not among As. ex. title
year year length
Completely nontrivial: none of the Bs are part of As. ex.
title year length
7
Trivial Dependency
An FD A1 A2 . . .An -> B1 B2 . . .Bm is trivial if the Bs are a
subset of the As {B1,B2, . . . Bn} subset {A1,A2, . . . An}
Its non-trivial if at least one B is not among the As, i.e.,
{B1,B2, ... Bn} {A1,A2, ...An}
Its completely non-trivial if none of the Bs are among the As,
{B1,B2, ... Bn} Intersect {A1,A2, ...An} =
Trivial dependency rule: The FD A1 A2 . . .An B1 B2 . . .Bm
is equivalent to the FD A1 A2. . .An C1 C2 . . . Ck , where
the Cs are those Bs that are not As, i.e., {C1, C2, . ., Ck} =
{B1,B2, . ,Bm} {A1,A2,.. ,An}
Armstrong Axioms
Reflexivity
(If X Y, then XZ YZ for any Z)
Transitivity
Y, then X Y)
Augumentation
(If X
(If X Y and Y Z, then X Z)
Armstrong axioms are sound and complete
Sound
They generate only FDs in F+ when applied to a set of FDs
Complete
They when repeatedly applied, these rules will generate all
FDs in F+
9
More Inference Axioms
Finding the closure of an FD Set may be tedious. So
more rules may be derived from Armstrong Axioms
The Union Rule, Pseudotransitive Rule and
Decompostion Rule are Sound but not complete
Union Rule (If X Y and X Z, then X YZ)
Given xy, xz
Augument x to xy and y to xz
xx xy ; xy yz
x xy ; xy yz
x yz [using transitive Axiom; x
xy, xy
yz]
10
More Inference Axioms
Pseudotransitive Rule (If X
Y and YW Z, then XW Z)
Given xy, ywz
Augument w to xy
xw yw
xw z [using transitive Axiom; xw
Decomposition Rule (If X
yw, yw
z]
YZ then X Y and then X Z)
Lets prove from the back onwards
Assume xy and xz is given
Take x y and augument with x
xx yx
We already have xz, So replace x with z in determinee
xx yx
xyz
11
Logical Implications of FDs
Let F be the following set of functional dependencies:
{ABCD, BDE, CF, EG, AB}. Use Armstrongs axioms
to show that {AFG} is logically implied by F
12
The Satisfies Algorithm
Used to determine if a relation R satisfies or doesnt satisfy a
given FD: AB
Input: Relation R and an FD: A B
Output: TRUE if R satisfies A B, otherwise FALSE
The Satisfies Algorithm:
Step 1: Sort the tuples of the relation R on the attribute(s) A (determinant)
so that tuples with equal values under A are next to each other
Step 2: Check that tuples with equal values under A also have equal values
under attribute(s) B
Step 3: If any two tuples of R meet condition 1 but fail to meet condition 2
the output of the algorithm is FALSE. Otherwise, the relation satisfies the
Functional Dependency and the output of the algorithm is TRUE
In short the satisfies algorithm can be stated as: The relation R satisfies
the FD: AB if the following holds for every pair of tuples t1 and t2
in R, if t1.A = t2.A then t1.B = t2.B
13
Redundant FDs
Given a set F of FDs, a FD AB of F is said to be redundant
w.r.t the FDs of F if and only if AB can be derived from the
set of FDs F-{AB}
Eliminating Redundant FDs allows us to minimize the set of FDs
Membership Algorithm helps to determine the Redundant FDs.
Input : A set F of FDs and a particular FD of F that is being tested
Output: FD is Redundant or not
14
Redundant FDs
Assume F is a set of FDs with AB F
The Membership Algorithm:
Step 1: Remove temporarily AB from F and initialize the set of FDs G to F. ie. Set
G=F-{AB}. If G proceed to step 2; otherwise stop executing the algorithm
since AB is non redundant
Step 2: Initialize the set of attributes Ti (with i=1) with the set of attribute(s) A(the
determinant of the FD under consideration). ie. Set Ti = T1 = {A}. The set T1 is the
current Ti
Step 3: In the set G search for FDs XY such that all the attributes of the
determinant X are elements of the current set Ti. There are two possible outcomes
Step 3a: If such FD is found, add the attribute of Y (right hand side of FD) to set Ti
and form a new Set Ti + 1= Ti U Y. The Set Ti + 1 is the current Ti .
Check if all the attributes of B (the right hand side of FD under consideration) are
members of Ti + 1. If this is the case, stop executing algorithm becos the FD:AB is
redundant. If not all attributes of B are members of Ti + 1 , remove XY from G
and repeat step 3
Step 3a: If G= or there are no FDs in G that have all the attributes of its
determinant in the current Ti then AB is not redundant
15
Redundant FDs
Given the set F={x YW, XW Z, Z Y, XY Z}. Determine
if the FD XY Z is redundant in F
Eliminate redundant FDs from F={X Y, Y X, Y Z, Z
Y, X Z, Z X} using the Membership algorithm
Find the redundant FDs in the set F={X YZ, ZW P, P
Z, W XPQ, XYQ YW, WQ YZ}
16
Closure of FD Set
Definition
The set of all FDs implied by a given set F of FDs is called the closure of F,
and denoted as F+
Armstrong Axioms can be applied repeatedly to infer all FDs implied by a
set F of FDs
Given R = ABCD and F = {A B,
+
A C, CD A}. Compute F .
+
A F={ABC}
+
B F={B}
+
AB F={ABC}
+
AC F={ABC}
+
ABC F=
Given R = XYZ and F = {XY Z}.
+
Compute F .
+
F = {X X, Y Y, Z Z,
XY X, XY Y, XY XY, XY Z
XZ X, XZ Z, XZ XZ,
YZ Y, YZ Z, YZ YY,
XYZ X, XYZ Y, XYZ Z,
XYZ XY, XYZ XZ, XYZ YZ,
XYZ XYZ,}
Consider a relation with schema R(A,B,C,D) and FD's F={AB C, C D, D A}
+
Compute F
17
Attribute Closure
Finding all the attributes in the relation that the
current attribute can determine by using inference
axioms and given FD set. Its denoted by {A}+
Given FDs set F={X YZ, ZW P, P Z, W XPQ, XYQ YW,
WQ YZ}. Find the Closure of all the single attributes
18
Candidate Key
A unique minimal set of attribute(s) that determine
the set of other attributes in a relation
Two properties of key are unique and minimalism
A superkey is a set of attributes that has the
uniqueness property but is not necessarily minimal
If a relation has multiple keys, specify one to be the
primary key
Convention: in a relational schema, underline the
attributes of the primary key.
If a key has only one attribute A, we say
that A rather than {A} is a key.
19
Candidate Key
Given a relation R(ABC) and FDs set F={AB C, B D,
D B}. Find the candidate keys of the relation
Given a relation R(XYZWP) and FDs set F={Y Z, Z
Y, Z W, Y P}. Find the number of candidate keys
Consider a schema R={S,T,V,C,P,D} and F= {S T,
V SC, SD P}. Find keys for R
Given a relation R(XYZWP) and FDs set F={x Z, YZ
W, Z Y}. Find the number of candidate keys
20
Equivalence of set of FDs
Given two sets F and G of FDs defined over same
relational schema
A set of FDs S follows from a set of FDs T if every
relation instance that satisfies all the FDs in T also
satisfies all the FDs in S
A C follows from T = {A B, B C}.
Two sets of FDs S and T are equivalent if and only
if S follows from T, and T follows from S
S = {A B,B C,AC} and T={A B, B C} are
equivalent
These notions are useful in deriving new FDs from a
given set of FDs
21
Equivalence of set of FDs
Two sets of FDs F and G defined over same relation
schema are equivalent if
every FD in F can be inferred from G and
every FD in G can be inferred from F
F Covers G if every FD in G can be inferred from F
(ie if G+ is subset of F+)
F and G are equivalent if F covers G and G covers F
If G covers F and no proper subset H of G exist such
that H+ = G+ we say G is a non-redundant cover of
F
22
The non-redundant cover algorithm
The Non-Redundant Cover Algorithm:
Step 1: Initialize G to F. i.e. set G=F
Step 2: Test every FD of G for redundancy using the Membership
Algorithm until there are no more FDs of G to be tested
Step 3: The set G is a non-redundant cover of F
Note:
Given a set F, there may be more than one non-redundant cover since the
order in which the FDs are considered is irrelevant
Problem
Find the non-redundant cover G for the set
F={X YZ, ZW P, P Z, W XPQ, XYQ YW, WQ YZ}
23
Extraneous Attribute
Definition: A is extraneous in X Y if A can be removed from
the left side or right side of X Y without changing the
closure of F.
ex. Let G = {A B C, B C, A B D }
Attribute C is extraneous in the right side of A B C and
attribute B is extraneous in the left side of A B D
The set G = {A B C, B C, A B D } is neither leftreduced nor right-reduced. G1 = {A B C, B C, A D } is
left-reduced but not right-reduced, while G2 = (A B. B C,
A B D} is right-reduced but not left-reduced. The set G3 =
{A B, B C, A D} is left and right-reduced, hence
reduced
One need to eliminate the extraneous attribute
24
Left Reduced Algorithm
The Left-Reduced Algorithm:
Step 1: Initialize a set of FDs to F. i.e. set G=F
Step 2: For every A1,A2,,Ai,,An Y in G do step 3 until there are no
more FDS in G to which this step can be applied. The algorithm stops when
all FDs of G have executed step 3
Step 3: For each attribute A in the determinant of FD selected in the
i
previous step do step 4 until all attributes have been tested. After finishing
testing of all attributes of a particular FD repeat step 2
Step 4: Test if all attributes of Y (the RHS of the FD under consideration)
are elements of the closure of A1,A2,,An (notice that we have removed
attribute A from the determinant of the FD) with respect to the FDs of G. If
i
this is the case remove attribute Ai from the determinant of the FD
undergoing testing becos Ai is an extraneous left attribute. If not all
attributes of Y are elements of the closure of A1,A2,,An then attribute Ai is
not an extraneous left attribute and should remain in the determinant of
the FD under consideration
When the algorithm finishes the set G contains a left-reduced cover set of T
25
Extraneous Attribute
Remove any extraneous left attributes from
F={ABC, EC, DAEF, ABFBD}
Reduce the set F={XZ, XYWP, XYZWQ, XZR}
by removing extraneous left attribute
Reduce the set F={XWY, XWZ, ZY, XYZ} by
removing extraneous left attribute
Tip :
There is no need to consider FDs with determinant that consist of single
attribute
26
Canonical Cover
A set of FDs F is canonical if every FD in F is of the
form X A and F is left-reduced and nonredundant
Since a canonical set of FDs is non-redundant and
every FD has a single attribute on the right side, it
is right-reduced. Since it is also left-reduced, it is
reduced
Example:
The set F = {A B, A C, A D, A E, B I J} is a
canonical cover for G = {A B C E, A B D E, B I J}
27
Minimal Cover
A set of FDs is minimal if it satisfies the following
conditions
every dependency in F has a single attribute for its RHS
we cannot remove any dependency from F and have a set of
dependencies that is equivalent to F
we cannot replace any dependency XA in F with a dependency
YA, where Y proper-subset-of X (Y subset-of X) and still have a
set of dependencies that is equivalent to F
Every set of FDs has an equivalent minimal set
There can be several equivalent minimal sets
There is no simple algorithm for computing a
minimal set of FDs that is equivalent to a set F of
FDs
To synthesize a set of relations, we assume that we
start with a set of dependencies that is minimal set
28
Optimal Cover
We have been measuring our covers in terms of the
number of FDs they contain. We can also measure
them by the number of attribute symbols required
to express them. example. (A B C, CD E, A C
IJ> has size 10 under this measure
Defiition: A set of FDs F is optimal if there is no
equivalent set of FDs with fewer attribute symbols
than F
The set F = {EC D, AB E, E AB ) is an
optimal cover for G=(ABCD,AB E, E AB }.
Notice that G is reduced and minimum, but not
optimal.
29
Problems
Find canonical cover of
1.
2.
3.
4.
F={XZ, XYWP, XYZWQ, XZR}
F={XYW, XWZ, ZY, XYZ}
F={ABC, EC, DAEF, ABFBD}
G = {A C, A B C, C DI, CD I, EC AB, EICC}
Find minimal cover of the following FD sets
1.
2.
3.
4.
5.
6.
F={ABC, ABC, BC, AB}
F={AB, BA, BC, AC, CA}
F={ABC, AB, BA}
G = {ABCD E, E D, A B, AC D}
F={ABC, BAC, C AB}
G={ABDC, CBE, AD BF, BF}
30
Seminar change?
Compound Functional Dependency (CFD) and
Annular Covers
31
Summary
Good Design needs strategy
Armstrong Axioms are sound and complete
FDs are constraints
There may be a number of equivalent FD sets
FD sets may be minimized by checking the coverage
32
