6-Design.ppt

Software Design
 Design activity begins with a set of requirements, and
maybe an architecture
 Design done before the system is implemented
 Design focuses on module view – i.e. what modules
should be in the system
 Module view may have easy or complex relationship
with the C&C view
 Design of a system is a blue print for implementation
 Often has two levels – high level (modules are
defined), and detailed design (logic specified)

Design…
 Design is a creative activity
 Goal: to create a plan to satisfy requirements
 Perhaps the most critical activity during system
development
 Design determines the major characteristics of a
system
 Has great impact on testing and maintenance
 Design document forms reference for later phases
 Design methodology – systematic approach for
creating a design

Design Concepts
 Design is correct, if it will satisfy all the
requirements and is consistent with
architecture
 Of the correct designs, we want best
design
 We focus on modularity as the main
criteria (besides correctness)

Modularity
 Modular system – in which modules can be built
separately and changes in one have minimum
impact on others
 Modularity supports independence of models
 Modularity enhances design clarity, eases
implementation
 Reduces cost of testing, debugging and
maintenance
 Cannot simply chop a program into modules to get
modularly
 Need some criteria for decomposition – coupling
and cohesion are such criteria

Coupling
 Independent modules: if one can function
completely without the presence of other
 Independence between modules is desirable
 Modules can be modified separately
 Can be implemented and tested separately
 Programming cost decreases
 In a system all modules cannot be independent
 Modules must cooperate with each other
 More connections between modules
 More dependent they are
 More knowledge about one module is required to
understand the other module.
 Coupling captures the notion of dependence

Coupling…
 Coupling between modules is the strength of
interconnections between modules
 In general, the more we must know about
module A in order to understand module B
the more closely connected is A to B
 "Highly coupled" modules are joined by
strong interconnection
 "Loosely coupled" modules have weak
interconnections

Coupling…
 Goal: modules as loosely coupled as possible
 Where possible, have independent modules
 Coupling is decided during high level design
 Cannot be reduced during implementation
 Coupling is inter-module concept
 Major factors influencing coupling
 Type of connection between modules
 Complexity of the interface
 Type of information flow between modules

Coupling – Type of connection
 Complexity and obscurity of interfaces increase
coupling
 Minimize the number of interfaces per module
 Minimize the complexity of each interface
 Coupling is minimized if
 Only defined entry of a module is used by others
 Information is passed exclusively through parameters
 Coupling increases if
 Indirect and obscure interface are used
 Internals of a module are directly used
 Shared variables employed for communication

Coupling – interface
complexity
 Coupling increases with complexity of
interfaces eg. number and complexity of
parms
 Interfaces are needed to support required
communication
 Often more than needed is used eg. passing
entire record when only a field is needed
 Keep the interface of a module as simple as
possible

Coupling – Type of Info flow
 Coupling depends on type of information flow
 Two kinds of information: data or control.
 Transfer of control information
 Action of module depends on the information
 Makes modules more difficult to understand
 Transfer of data information
 Module can be treated as input-output function
 Lowest coupling: interfaces with only data
communication
 Highest: hybrid interfaces

Coupling - Summary
Coupling Interface Type of Type of
complexity connections commu-
nication
Low Simple to module data
obvious by name
High complicated to internal Hybrid
obscure elements

Coupling in OO Systems
 In OO systems, basic modules are
classes, which are richer than fns
 OO Systems have three types of
coupling
 Interaction coupling
 Component coupling
 Inheritance coupling

Coupling in OO - Interaction
 Interaction coupling occurs due to
methods of a class invoking methods of
other classes
 Like calling of functions
 Worst form if methods directly access
internal parts of other methods
 Still bad if methods directly manipulate
variables of other classes
 Passing info through tmp vars is also bad

Coupling in OO …
 Least interaction coupling if methods
communicate directly with parameters
 With least number of parameters
 With least amount of info being passed
 With only data being passed
 I.e. methods should pass the least
amount of data, with least no of parms

Coupling in OO - Component
 Component coupling – when a class A has
variables of another class C
 A has instance vars of C
 A has some parms of type C
 A has a method with a local var of type C
 When A is coupled with C, it is coupled with
all subclasses of C as well
 Component coupling will generally imply the
presence of interaction coupling also

Coupling in OO - Inheritance
 Inheritance coupling – two classes are
coupled if one is a subclass of other
 Worst form – when subclass modifies a
signature of a method or deletes a method
 Coupling is bad even when same signature
but a changed implementation
 Least, when subclass only adds instance vars
and methods but does not modify any

Cohesion
 Coupling characterized the inter-module bond
 Reduced by minimizing relationship between elts of
different modules
 Another method of achieving this is by maximizing
relationship between elts of same module
 Cohesion considers this relationship
 Interested in determining how closely the elements
of a module are related to each other
 In practice both are used

Cohesion…
 Cohesion of a module represents how tightly
bound are the elements of the module
 Gives a handle about whether the different
elements of a module belong together
 High cohesion is the goal
 Cohesion and coupling are interrelated
 Greater cohesion of modules, lower coupling
between module
 Correlation is not perfect.

Levels of Cohesion
 There are many levels of cohesion.
 Coincidental
 Logical
 Temporal
 Communicational
 Sequential
 Functional
 Coincidental is lowest, functional is highest
 Scale is not linear
 Functional is considered very strong

Determining Cohesion
 Describe the purpose of a module in a sentence
 Perform the following tests
1. If the sentence has to be a compound sentence, contains
more than one verbs, the module is probably performing
more than one function. Probably has sequential or
communicational cohesion.
2. If the sentence contains words relating to time, like "first",
"next", "after", "start" etc., the module probably has
sequential or temporal cohesion.

3. If the predicate of the sentence does not contain a
single specific object following the verb, the module
is probably logically cohesive. Eg "edit all data", while
"edit source data" may have functional cohesion.
4. Words like "initialize", "clean-up" often imply
temporal cohesion.
 Functionally cohesive module can always be
described by a simple statement

Cohesion in OO Systems
 In OO, different types of cohesion is possible
as classes are the modules
 Method cohesion
 Class cohesion
 Inheritance cohesion
 Method cohesion – why diff code elts are
together in a method
 Like cohesion in functional modules; highest form
is if each method implements a clearly defined
function with all elts contributing to implementing
this function

Cohesion in OO…
 Class cohesion – why diff attributes and
methods are together in a class
 A class should represent a single concept
with all elts contributing towards it
 Whenever multiple concepts encapsulated,
cohesion is not as high
 A symptom of multiple concepts – diff gps
of methods accessing diff subsets of
attributes

Cohesion in OO…
 Inheritance cohesion – focuses on why
classes are together in a hierarchy
 Two reasons for subclassing –
generalization-specialization and reuse
 Cohesion is higher if the hierarchy is for
providing generalization-specialization

Open-closed Principle
 Besides cohesion and coupling, open closed principle
also helps in achieving modularity
 Principle: A module should be open for extension but
closed for modification
 Behavior can be extended to accommodate new
requirements, but existing code is not modified
 I.e. allows addition of code, but not modification of existing
code
 Minimizes risk of having existing functionality stop working
due to changes – a very important consideration while
changing code
 Good for programmers as they like writing new code

Open-closed Principle…
 In OO this principle is satisfied by using
inheritance and polymorphism
 Inheritance allows creating a new class to
extend behavior without changing the original
class
 This can be used to support the open-closed
principle
 Consider example of a client object which
interacts with a printer object for printing

Example..
 Client directly calls methods on Printer1
 If another printer is to be allowed
 A new class Printer2 will be created
 But the client will have to be changed if it wants
to use Printer 2
 Alternative approach
 Have Printer1 a subclass of a general Printer
 For modification, add another subclass Printer 2
 Client does not need to be changed

Liskov’s Substitution Principle
 Principle: Program using object o1 of
base class C should remain unchanged
if o1 is replaced by an object of a
subclass of C
 If hierarchies follow this principle, the
open-closed principle gets supported

Summary
 Goal of designing is to find the best
possible correct design
 Modularity is the criteria for deciding
quality of the design
 Modularity enhanced by low coupling,
high cohesion, and following open-
closed principle

Function Oriented Design and
Structured Design Methodology

Program Structure and Structure Charts
 Every program has a structure
 Structure Chart - graphic representation of structure
 SC represents modules and interconnections
 Each module is represented by a box
 If A invokes B, an arrow is drawn from A to B
 Arrows are labeled by data items
 Different types of modules in a SC
 Input, output, transform and coordinate modules
 A module may be a composite

Structure charts…
 SC shows the static structure, not the logic
 Different from flow charts
 Major decisions and loops can be shown
 Structure is decided during design
 Implementation does not change structure
 Structure effects maintainability
 SDM aims to control the structure

STRUCTURED DESIGN METHODOLOGY
 SDM views software as a transformation function that
converts given inputs to desired outputs
 The focus of SD is the transformation function
 Uses functional abstraction
 Goal of SDM: Specify functional modules and
connections
 Low coupling and high cohesion is the objective
Transformation
functions
Input Output

Steps in SD
1. Draw a DFD of the system
2. Identify most abstract inputs and most
abstract outputs
3. First level factoring
4. Factoring of input, output, transform
modules
5. Improving the structure

1. Data Flow Diagrams
 SD starts with a DFD to capture flow of data
in the proposed system
 DFD is an important representation; provides
a high level view of the system
 Emphasizes the flow of data through the
system
 Ignores procedural aspects
 (Purpose here is different from DFDs used in
requirements analysis, thought notation is the
same)

Drawing a DFG
 Start with identifying the inputs and outputs
 Work your way from inputs to outputs, or vice versa
 If stuck, reverse direction
 Ask: "What transformations will convert the inputs to
outputs"
 Never try to show control logic.
 If thinking about loops, if-then-else, start again
 Label each arrow carefully
 Make use of * and +, and show sufficient detail
 Ignore minor functions in the start
 For complex systems, make dfg hierarchical
 Never settle for the 1st dfg

Step 2 of SD Methodology
 Generally a system performs a basic function
 Often cannot be performed on inputs directly
 First inputs must be converted into a suitable
form
 Similarly for outputs - the outputs produced
 by main transforms need further processing
 Many transforms needed for processing
inputs and outputs
 Goal of step 2 is to separate such transforms
from the basic transform centers

Step 2…
 Most abstract inputs: data elements in dfg
that are furthest from the actual inputs, but
can still be considered as incoming
 These are logical data items for the
transformation
 May have little similarity with actual inputs.
 Often data items obtained after error
checking, formatting, data validation,
conversion etc.

Step 2…
 Travel from physical inputs towards outputs until data
can no longer be considered incoming
 Go as far as possible, without loosing the incoming
nature
 Similarly for most abstract outputs
 Represents a value judgment, but choice is often
obvious
 Bubbles between mai and mao: central transforms
 These transforms perform the basic transformation
 With mai and mao the central transforms can
concentrate on the transformation

Step 2…
 Problem View: Each system does some i/o
and some processing
 In many systems the i/o processing forms the
large part of the code
 This approach separates the different
functions
 subsystem primarily performing input
 subsystem primarily performing transformations
 subsystem primarily performing output
presentation

Example 1 – counting the no
of different words in a file

3. First Level Factoring
 First step towards a structure chart
 Specify a main module
 For each most abstract input data item, specify a
subordinate input module
 The purpose of these input modules is to deliver to
main the mai data items
 For each most abstract output data element, specify
an output module
 For each central transform, specify a subordinate
transform module
 Inputs and outputs of these transform modules are
specified in the DFD

 First level factoring is straight forward
 Main module is a coordinate module
 Some subordinates are responsible for delivering the
logical inputs
 These are passed to transform modules to get them
converted to logical outputs
 Output modules then consume them
 Divided the problem into three separate problems
 Each of the three diff. types of modules can be
designed separately
 These modules are independent

4. Factoring Input modules
 The transform that produced the mai data is treated
as the central transform
 Then repeat the process of first level factoring
 Input module being factored becomes the main
module
 A subordinate input module is created for each data
item coming in this new central transform
 A subordinate module is created for the new central
transform
 Generally there will be no output modules

 The new input modules are factored similarly Till the
physical inputs are reached
 Factoring of the output modules is symmetrical
 Subordinates - a transform and output modules
 Usually no input modules

Factoring Central Transforms
 Factoring i/o modules is straight forward if the DFD is
detailed
 No rules for factoring the transform modules
 Top-down refinement process can be used
 Goal: determine sub-transforms that will together
compose the transform
 Then repeat the process for newly found transforms
 Treat the transform as a problem in its own right
 Draw a data flow graph
 Then repeat the process of factoring
 Repeat this till atomic modules are reached

5. Improving Design through
Heuristics
 The above steps should not be followed blindly
 The structure obtained should be modified if needed
 Low coupling, high cohesion being the goal
 Design heuristics used to modify the initial design
 Design heuristics - A set of thumb rules that are
generally useful
 Module Size: Indication of module complexity
Carefully examine modules less than a few lines or
greater than about 100 lines
 Fan out and fan in
 A high fan out is not desired, should not be increased
beyond 5 or 6
 Fan in should be maximized

 Scope of effect of a module: the modules affected by
a decision inside the module
 Scope of control: All subordinates of the module
 Good thumb rule:
For each module scope of effect should be a subset
of scope of control
 Ideally a decision should only effect immediate
subordinates
 Moving up the decision, moving the module down
can be utilized to achieve this

Summary
 Structured design methodology is one way to create
modular design
 It partitions the system into input subsystems, output
subsystems & transform subsystems
 Idea: Many systems use a lot of code for handling
inputs & outputs
 SDM separates these concerns
 Then each of the subsystems is factored using the
DFD
 The design is finally documented & verified before
proceeding

Object Oriented Design and
UML

OO Concepts
 Information hiding – use encapsulation
to restrict external visibility
 OO encapsulates the data, provides
limited access, visibility
 Info hiding can be provided without OO
– is an old concept

OO Concepts…
 State retention – fns, procedures do not
retain state; an object is aware of its
past and maintains state
 Identity – each object can be identified
and treated as a distinct entity
 Behavior – state and services together
define the behavior of an object, or how
an object responds

OO Concepts..
 Messages – through which a sender obj
conveys to a target obj a request
 For requesting O1 must have – a handle
for O2, name of the op, info on ops that
O2 requires
 General format O2.method(args)

OO Concepts..
 Classes – a class is a stencil from which
objects are created; defines the structure and
services. A class has
 An interface which defines which parts of an
object can be accessed from outside
 Body that implements the operations
 Instance variables to hold object state
 Objects and classes are different; class is a
type, object is an instance
 State and identity is of objects

Relationship among objects
 An object has some capability – for other
services it interacts with other objects
 Some different ways for interaction:
1. Supplier object is global to client
2. Supplier obj is a parm to some op of the client
3. Supplier obj is part of the client obj
4. Supplier obj is locally declared in some op
 Relationship can be either aggregation
(whole-part relationship), or just client
server relationship

Inheritance
 Inheritance is unique to OO and not there in
function-oriented languages/models
 Inheritance by class B from class A is the
facility by which B implicitly gets the
attributes and ops of A as part of itself
 Attributes and methods of A are reused by B
 When B inherits from A, B is the subclass or
derived class and A is the base class or
superclass

Inheritance..
 A subclass B generally has a derived
part (inherited from A) and an
incremental part (is new)
 Hence, B needs to define only the
incremental part
 Creates an “is-a” relationship – objects
of type B are also objects of type A

Inheritance…
 The inheritance relationship between classes
forms a class hierarchy
 In models, hierarchy should represent the
natural relationships present in the problem
domain
 In a hierarchy, all the common features can
be accumulated in a superclass
 An existing class can be a specialization of an
existing general class – is also called
generalization-specialization relationships

Inheritance…
 Strict inheritance – a subclass takes all
features of parent class
 Only adds features to specialize it
 Non-strict: when some of the features
have been redefined
 Strict inheritance supports “is-a” cleanly
and has fewer side effects

Inheritance…
 Single inheritance – a subclass inherits
from only one superclass
 Class hierarchy is a tree
 Multiple inheritance – a class inherits
from more than one class
 Can cause runtime conflicts
 Repeated inheritance - a class inherits from
a class but from two separate paths

Inheritance and Polymorphism
 Inheritance brings polymorphism, i.e. an
object can be of different types
 An object of type B is also an object of type A
 Hence an object has a static type and a
dynamic type
 Implications on type checking
 Also brings dynamic binding of operations which
allows writing of general code where operations
do different things depending on the type

Unified Modeling Language
(UML) and Modeling
 UML is a graphical notation useful for
OO analysis and design
 Allows representing various aspects of
the system
 Various notations are used to build
different models for the system
 OOAD methodologies use UML to
represent the models they create

Modeling
 Modeling is used in many disciplines –
architecture, aircraft building, …
 A model is a simplification of reality
 “All models are wrong, some are useful”
 A good model includes those elts that
have broad effect and omits minor elts
 A model of a system is not the system!

Why build models?
 Models help us visualize a system
 Help specify the system structure
 Gives us a template that can guide the
construction
 Document the decisions taken and their
rationale

Modeling
 Every complex system requires multiple
models, representing diff aspects
 These models are related but can be
studied in isolation
 Eg. Arch view, electrical view, plumbing
view of a building
 Model can be structural, or behavioral

Views in an UML
 A use case view
 A design view
 A process view
 Implementation view
 Deployment view
 We will focus primarily on models for design –
class diagram, interaction diagram, etc.

Class Diagrams
 Classes are the basic building blocks of
an OO system as classes are the
implementation units also
 Class diagram is the central piece in an
OO design. It specifies
 Classes in the system
 Association between classes
 Subtype, supertype relationship

Class Diagram…
 Class itself represented as a box with
name, attributes, and methods
 There are conventions for naming
 If a class is an interface, this can be
specified by <<interface>> stereotype
 Properties of attr/methods can be
specified by tags between { }

Generalization-Specialization
 This relationship leads to class hierarchy
 Can be captured in a class diagram
 Arrows coming from the subclass to the
superclass with head touching super
 Allows multiple subclasses
 If specialization is done on the basis of
some discriminator, arrow can be labeled

Association/aggregation
 Classes have other relationships
 Association: when objects of a class need
services from other objects
 Shown by a line joining classes
 Multiplicity can be represented
 Aggregation: when an object is composed of
other objects
 Captures part-whole relationship
 Shown with a diamond connecting classes

Example – association/aggregation

Interaction Diagrams
 Class diagram represent static structure of
the system (classes and their rel)
 Do not model the behavior of system
 Behavioral view – shows how objects interact
for performing actions (typically a use case)
 Interaction is between objects, not classes
 Interaction diagram in two styles
 Collaboration diagram
 Sequence diagram
 Two are equivalent in power

Sequence Diagram
 Objects participating in an interaction are
shown at the top
 For each object a vertical bar represents its
lifeline
 Message from an object to another,
represented as a labeled arrow
 If message sent under some condition, it can
be specified in bracket
 Time increases downwards, ordering of
events is captured

Collaboration diagram
 Also shows how objects interact
 Instead of timeline, this diagram looks
more like a state diagram
 Ordering of messages captured by
numbering them
 Is equivalent to sequence diagram in
modeling power

Example – collaboration diag

Other Diagrams
 Class diagram and interaction diagrams
most commonly used during design
 There are other diagrams used to build
different types of models

Other Diagrams
 Instead of objects/classes, can represent
components, packages, subsystems
 These are useful for developing architecture
structures
 UML is extensible – can model a new but
similar concept by using stereotypes (by
adding <<name>>)
 Tagged values can be used to specify
additional properties, e.g. private, readonly..
 Notes can be added

Design using UML
 Many OOAD methodologies have been
proposed
 They provide some guidelines on the steps to
be performed
 Basic goal is to identify classes, understand
their behavior, and relationships
 Different UML models are used for this
 Often UML is used, methodologies are not
followed strictly

Design using UML
 Basic steps
 Identify classes, attributes, and operations from
use cases
 Define relationships between classes
 Make dynamic models for key use cases and use
them to refine class diagrams
 Make a functional model and use it to refine the
classes
 Optimize and package
 Class diagrams play the central role; class
defn gets refined as we proceed

Restaurant example: Initial classes

Restaurant example: a seq diag

Detailed Design
 HLD does not specify module logic; this is done during
detailed design
 One way to communicate the logic design: use natural
language
 Is imprecise and can lead to misunderstanding
 Other extreme is to use a formal language
 Generally a semi-formal language is used – has formal
outer structures but informal inside

Logic/Algorithm Design
 Once the functional module (function or
methods in a class) are specified, the
algo to implement it is designed
 Various techniques possible for
designing algorithm – in algos course
 Stepwise refinements technique is
useful here

State Modeling of Classes
 Dynamic model to represent behavior of
an individual object or a system
 Shows the states of an object and
transitions between them
 Helps understand the object – focus
only on the important logical states
 State diagrams can be very useful for
automated and systematic testing

Design Verification
 Main objective: does the design
implement the requirements
 Analysis for performance, efficiency, etc
may also be done
 If formal languages used for design
representation, tools can help
 Design reviews remain the most
common approach for verification

Background
 Basic purpose to provide a quantitative
evaluation of the design (so the final product
can be better)
 Size is always a metric – after design it can
be more accurately estimated
 Number of modules and estimated size of each is
one approach
 Complexity is another metric of interest – will
discuss a few metrics

Network Metrics
 Focus on structure chart; a good SC is
considered as one with each module having
one caller (reduces coupling)
 The more the SC deviates from a tree, the
more impure it is
Graph impurity = n – e – 1
n – nodes, e- edges in the graph
 Impurity of 0 means tree; as this no
increases, the impurity increases

Stability Metrics
 Stability tries to capture the impact of a
change on the design
 Higher the stability, the better it is
 Stability of a module – the number of
assumptions made by other modules about
this module
 Depends on module interface, global data the
module uses
 Are known after design

Information Flow Metrics
 Complexity of a module is viewed as
depending on intra-module complexity
 Intramodule estimated by module size
and the information flowing
 Size in LOC
 Inflow – info flowing in the module
 Outflow – info flowing out of the module
 Dc = size * (inflow * outflow)2

Information flow metrics…
 (inflow * outflow) represents total
combination of inputs and outputs
 Its square reps interconnection between
the modules
 Size represents the internal complexity
of the module
 Product represents the total complexity

Identifying error-prone
modules
 Uses avg complexity of modules and std
dev to identify error prone and complex
modules:
Error prone: If Dc > avg complexity +
std_dev
Complex: If avg complexity < Dc < avg +
std dev
Normal: Otherwise

Complexity metrics for OO
 Weighted methods per class
 Complexity of a class depends on no of
classes and their complexity
 Suppose complexity of methods is c1, c2..;
by some functional complexity metric
WMC = Σ ci
 Large WMC might mean that the class is
more fault-prone

OO Metrics…
 Depth of Inheritance Tree
 DIT of C is depth from the root class
 Length of the path from root to C
 DIT is significant in predicting fault
proneness
 Number of Children
 Immediate no of subclasses of C
 Gives a sense of reuse

OO Metrics…
 Coupling between classes
 No of classes to which this class is coupled
 Two classes are coupled if methods of one
use methods or attr of other
 Can be determined from code
 (There are indirect forms of coupling that
cannot be statically determined)

Metrics…
 Response for a class
 The total no of methods that can be invoked from
this class
 Captures the strength of connections
 Lack of cohesion in methods
 Two methods form a cohesive pair if they access
some common vars (form a non-cohesive pair if
no common var)
 LCOM is the number of method pairs that are non-
cohesive – the no of cohesive pairs

Metrics with detailed design
 When logic is known, internal
complexity metrics can be determined
 We will cover all detailed design based
metrics along with code metrics

Summary
 Design for a system is a plan for a solution –
want correct and modular
 Cohesion and coupling key concepts to
ensure modularity
 Structure charts and structured design
methodology can be used for function-
oriented design
 UML can be used for OO design
 Various complexity metrics exist to evaluate a
design complexity

6-Design.ppt

More Related Content

Similar to 6-Design.ppt (20)

More from DuraisamySubramaniam1 (7)

Recently uploaded (20)

6-Design.ppt