Comparing Designs

1
Comparing Designs

• By Chris McCall

2
Comparing Designs

• A decision-making method for deciding between
  many designs for a single specification
• Provides a tangible basis for choosing one design
  over another
• Method
• Review all available designs
• Decide on important characteristics
• Assign priorities to each characteristic (1-5)
• Assign scores to each design for each
  characteristic
• Find winning design by multiplying priority by
  score for each characteristic and adding them up

3
Example Key Word in Context (KWIC)
Specification

• An index system that accepts an ordered set of
  lines
• A line is an ordered set of words
• A word is an ordered set of characters
• Any line can be circularly shifted by removing
  a word from the front of the line and appending
  it to the back
• Outputs an alphabetical list of all circular
  shifts of all lines

4
KWIC Design 1 Shared Data

• Master program calls parts in sequence
• Data is central, computation is modularized
• Difficult to change because changing data affects
  all components and isnt very reusable

5
KWIC Design 2 Abstract Data

• Data not centrally stored
• Processing still decomposed and controlled by a
  Master control
• Less dependencies and easier to change than
  previous example

6
KWIC Design 3 Implicit Invocation

• Data accessed abstractly
• Computation isnt explicit, event-driven
• Easy to reuse because computation relies on events

7
KWIC Design 4 Pipe-and-filter

• Easy to control order in which computation is
  done, as opposed to previous example
• Highly reusable design (filters run
  independently)
• Limited design, hard to make interactive

8
Next Step Choose Important Characteristics
Assign Priorities

• For our example, important characteristics
  priorities
• Easy to reuse -- high 5
• Easy to change data representation -- high 4
• Easy to change function -- med 3
• Good performance -- med 3
• Easy to change algorithm -- low 1
• Other possible characteristics
• Modularity
• Testability
• Maintainability
• Easy to use
• etc

9
Last step Assign scores for each design and
calculate winner
37 57 40 55

• To calculate a score, multiply priority by score
  for each characteristic of a design
• Example Abstract Data Type (12) (45) (31)
  (34) (54) 57
• Highest score indicates the design that best fits
  your needs

10
Criticism

• Priorities, design ratings, and the choice of
  attributes are all subjective
• Best design might not be the most realistic to
  implement
• Everyone has different needs and perspectives
  (customer vs. developer)
• Not all designs are as simple or easy to compare
  as the example shown

11
The End

• Sources
• Software Engineering Theory and Practice (3rd Ed)
• http//en.wikipedia.org/wiki/Software_design_patte
  rn