SOENG lecture 1

Introduction to Software Engineering

Lecture 1 - Background

Dr Jim Briggs, 24 September 1998

What is the course about?

How to produce

reliable,
large scale,
software systems
which satisfy user expectations
on time and
within budget

Large vs. small-scale programming

Typical software project sizes:

Type	People	Duration	Size
Trivial	1	1-4 wks	500L
Small	1	1-6 mths	1-2KL
Medium	2-20	1-3 yrs	5-100KL
Large	100-1000	4-5 yrs	1ML
Huge	2000-5000	5-10 yrs	1-10ML

K(M)L = Thousands (Millions) of delivered source lines of program

Note change in productivity - 6,000 lines pa to 200 lines pa.

Importance of LSP - Financial

In 1980 in US $47,900,000,000 ($47.9Bn) spent on computer systems including $40.2Bn on software (i.e approx. 80%)

In 1987, US Expenditure $50Bn pa on software and "maintenance"

In USA, computing was 5% of GDP in 1980, 8% in 1985, and estimated 12.5% in 1990 - same compound growth gives 100% in 2010.

USAF Advanced Logistics System (ALS) - abandoned after spending $217M as recognised that system would never work! - it was deemed cheaper to throw it away and start again

Software is becoming a dominant factor in the finances of many companies, and for most developed nations.

Importance of LSP - Social

Estimates for US in 1985 - 70% of workforce using computers (perhaps unwittingly) - 20% of workforce need to know how computer works.

Many Accidents (comp.risks)

cancer patient killed as computer ordered radiation therapy machine to administer lethal dose

syringe pump has failed giving overdose (are used to administer cyanide)

bank had to borrow $23.6Bn as software "bug" caused it to buy many times its actual assets

NASA mission to Mars lost due to missing comma in Fortran do-loop

Computers are becoming all pervasive and their errors (caused by software faults) can be catastrophic.

Problems of LSP - Economic/Managerial

One US Air Force Project for a "Command and Control" system.

Initial customer estimate $1,500,000
Winning contractor's estimate $400,000
Contractor's estimate after six months $700,000
Contractor's estimate after nine months $2,500,000
Project completion - 30 months at $3,700,000

This is an extreme, but not unique case. In 1988, a bank cancelled a project after spending $60M without fixing their problems!

Estimating the cost and duration of a major project is very difficult.
Software projects are hard to manage

LSP & Software Engineering

Software Engineering is the name given to the disciplined approach to producing large scale programs.

The goal of Software Engineering is the construction of systems in a cost-effective and timely manner so that they meet their initial requirements and they can be modified in a controlled fashion to satisfy changes in their requirements.

It is analogous to other engineering disciplines, although Software Engineering is less well developed.

We will use LSP to refer to the problem and SE to refer to the (emerging) solutions.

What is Software Engineering?

Definition similar to other engineering disciplines

Software Engineering is the application of science, mathematics, principles and tools by which computers are made useful to man by means of software, which comprises computer programs, associated procedures and documentation.

The science underlying software engineering is ill developed, but some is relevant (e.g. machine theory) and there is applicable mathematics, and there are many principles and tools (themselves computer programs) which can be used in software engineering.

Project Stages in SE

Typical set of Stages:

Initial Conception
Requirements Analysis
Specification
High Level Design (Architecture) - identification of modules
Low Level Design and Module Test
System Integration & Testing
Acceptance Testing
Use and "Maintenance"

Error Correction
Enhancement - adding facilities

Replacement by new System

There may be other stages, e.g. prototyping.

This may cover 25 or 30 years, and often maintenance is the dominant cost.

Problems of LSP - NOT Programming

Sample of US projects - distribution of effort

Project type	Analysis & design	Coding	Test & Integrate
C & C	35%	17%	48%
Space	34%	20%	46%
Operating System	33%	15%	50%
Scientific	44%	26%	30%
Business	44%	28%	28%

Problems mainly in:

establishing requirements
making different parts of the system (written by different people) work together
maintenance

Problems of LSP - Maintenance

typically over 50% of total effort is spent in maintenance - for some systems the figure may be nearer 80%

maintenance is a misnomer - evolution would be better

A goal of SE is to carry out development so as to make maintenance comparatively easy - in essence this means that, inter alia:

software must be easy to change (modular, etc.)
the design, especially design decisions, must be well documented
it must be possible to trace requirements through to implementation etc.

Problems of LSP - Productivity

productivity of professional programmers can vary by a factor of 10:1

ideally use small teams of highly skilled people - but how do we handle large projects?

productivity reduces with program size - some experimental evidence

effort = constant x size ^1.5

productivity = constant x size ^-1.5

productivity influenced by technology

productivity is largely independent of programming language (lines/day)

hence the emphasis on high level languages

Productivity varies with people, project size and development technology.

the nature of the project influences productivity
few interactions between parts (programmers) - 10,000 lines per man year
some interactions - 5,000 lines
many interactions - 1,500 lines
perfectly divisible project
effort constant
time inversely proportional to number of people
totally indivisible project (pregnancy)
time constant
effort proportional to number of people

Productivity depends on project type.

in general projects are not readily divisible
time spent communicating rises with the number of people - at some point it may exceed 100%!
Brooks' Law - "adding more people to a late project may make it even later"
it is impractical to increase the size of project teams by more than 30% pa (for large projects)
organisation of software projects is critical to success

Productivity depends on many factors, is difficult to predict, and difficult to control. The cost of communication is high - breaking down systems in to largely independent modules is critical to success in LSP.

Course plan
Schedule

Session	Subject
1	Introduction
2	Software engineering as a process
3	Characteristics of large-scale systems
4	Requirements and specifications 1
5	Requirements and specifications 2
6	The CORE requirements methodology
7	Software validation 1
8	Software validation 2
9	Software project management 1
10	Software project management 2
11	Revision session

Assessment

Exam during the semester 1 exam period.

Recommended books

To buy...

Ian Sommerville, Software engineering, Addison-Wesley (5^th edition 1996).

Alternative purchases...

Roger Pressman, Software engineering - a practitioner's approach, McGraw-Hill.

Supplementary reading...

Fred Brooks, The mythical man-month, Addison-Wesley.

Alan Davis, Software requirements, Prentice-Hall.

Alan Gillies, Software quality - theory and management, Chapman & Hall.