Steve Mattingley

1
Steve Mattingley

• October 22, 2002
• Paper Presentation

2
A Vector-Based Approach to Software Size
Measurement and Effort Estimation

• T.E Hastings and A.S.M Sajeev
• IEEE Transactions on Software Engineering
• Vol 27, NO. 4, April 2001
• IEEECS Log Number 107066

3
Paper Overview

• Introduction
• Software Size Measurement
• Software Specification
• Vector Size Measure
• Vector Prediction Models
• Formal Validation
• Empirical Validation
• Discussion

4
Introduction

• Software size is difficult to define and measure
• SLOC and Function Points are two common methods
• Vector Size Measure (VSM) and Vector Prediction
  Model (VPM) using Algebraic Specification
  Language (ASL)

5
Software Size Measurement

• Measure Processes, Products, and Resources
• We are interested in internal attributes of
  software (independent from environment).
• Size is a function of length, functionality, and
  complexity (problem complexity).

6
Software Size Specification

• To accurately predict software size one must have
  an accurate software specification.
• Algebraic approach to software specification is
  based on Abstract Data Types (ADT)
• ADT F R
• An ADT is a set of functions and Rules

7
Software Size Specification

• ASL used to describe software. Example
• deposit (Money, Account) Account
  balance(Account) balance(Account) add Money
  post condition
• deposit an amount of Money into an Account such
  that the new account equals the sum of the
  original account and the new money.

8
Vector Size Measure

• Rule set for determining functionality from
  number of Functional Ops in ALS Fa ?OPF
• Rule set for determining complexity from number
  of Complex Ops in ALS Ca ?OPC
• Length is the sum of functionality and
  complexity. La Fa Ca

9
Vector Size Measure

• Vector representation ltF, Cgt (Functionality and
  Complexity)
• Normal Vector mathematics apply to solve for
  magnitude and gradient of vector.
• Magnitude provides a single measure which
  balances functionality and complexity.
• Gradient provides a ration of problem complexity
  vs. functionality
• Vectors may be scaled with real world units and
  scale factors.

10
Vector Prediction Models

• Used instead of Productivity Based Models, Linear
  Regression, or COCOMO.
• VPM is a cost model which takes the magnitude and
  gradient from the VSM and uses a multivariate
  regression model.
• VPM is based on having a defined and repeatable
  software process in place.

11
Vector Prediction Models

• Estimated Effort E ambgz
• m vector magnitude, g vector gradient
• a,b,z are coefficients which are determined
  empirically for the organization and software
  type.

12
Formal Validation

• Authors propose 4 rules which must be followed
  for ADT, ASL and VSM.
• The rules ensure Formal Validation of the Vector
  approach.

13
Empirical Validation

• The Vector based approach was validated
  empirically using a pilot study of 8 industry
  projects.
• For the various projects Requirements were
  gathered, product attributes were gathered,
  process and resource attributes were gathered,
  and ASLs were developed for each project.

14
Empirical Validation

• A measurement tool then produced measurements
  from the project ASLs.
• A function point count, MKII FP count, Linear
  Regression, VPM , and COCOMO 2 analysis were all
  performed.
• Results were then compared and tested against
  defined test criteria.

15
Empirical Validation
16
Empirical Validation
17
Authors Discussion

• VPM provides a means of classifying software
  systems (based on functionality, complexity
  vectors)
• Less inputs then COCOMO
• ?20 percent within actualize (Better then
  alternatives)
• Limited by having to specify using ASL
• Limited by having a bottoms up approach

18
My discussion

• Interesting approach
• Takes much of the guess work out of estimating a
  well specified system.
• Seems like an enormous amount of work to specify
  large systems at that detail.
• Would require a lot of maintenance to ASLs as
  system requirements change

19
Questions

• ???