Software in Practice a series of four lectures on why software projects fail, and what you can do ab

1
Software in Practicea series of four lectures on
why software projects fail, and what you can do
about it

• Martyn Thomas
• Founder Praxis High Integrity Systems Ltd
• Visiting Professor of Software Engineering,
  Oxford University Computing Laboratory

2
Lecture 4 Principles of Software Engineering

• Dependability
• Theory and Practice the 30-year gap
• What we Know
• Strong Software Engineering

3
Software Dependabilityour highest priority
problem

• Dependability umbrella term for safety,
  reliability, security, trustworthiness,
  availability
• A dependable system is one known to have all its
  required properties.
• We need dependable processes as well as
  dependable systems
• to build systems on time and budget, and to be
  able to rely on what we deliver.

4
World dependability initiatives-evidence that
dependability is now high priority

• Microsoft
• trustworthiness initiative (11000 engineers)
• Palladium/Longhorn/Next Generation Secure
  Computing Base.
• Trusted Computing Platform Alliance
• Intels LaGrande technology
• IBMs Autonomic Computing initiative
• Critical Infrastructure Protection initiatives
• Dependable Systems Evolution identified as a
  computer science Grand Challenge
• DIRC

5
In a 50 year old discipline, there is a 30-year
gap between theory and practice!

• A study of program structure has revealed that
  programseven alternative programs for the same
  taskcan differ tremendously in their
  intellectual manageability. A number of rules
  have been discovered, violation of which will
  either seriously impair or totally destroy the
  intellectual manageability of the program.
  Dijkstra, 1972

6
Theory and Practice

• Theory
• The inherent difficulty of software design has
  been well understood for more than 30 years, and
  the problems and solutions have been taught to a
  generation of graduates.
• Practice
• The insight has been largely ignored by industry
  not because it has been tried and failed, but
  because it has been considered hard, and not
  tried. after G. K. Chesterton

7
Practice in Industry

• Despite much progress in software development
• most software development staff have
• little formal education in software engineering
• no good grounding in either the processes or the
  theories that underpin their professional work.
• A fascination with the latest fad or fashion
• managers who have even less understanding than
  their junior staff of the great complexity of the
  engineering tasks they are undertaking, and of
  the risks inherent in departing from what is
  known to work.
• most software staff are recruited for their
  knowledge of a programming language or a software
  package, not for any engineering knowledge.

8
Managers and engineering knowledge ...
9
Software Engineering

• Computer based applications are increasingly
  complex
• We choose to put that complexity into the
  software
• More and more software is business critical
• Quite often, software is safety critical
• Constructing complex and important systems is
  engineering

10
Engineering is ...

• Developing complex systems or objects ...
• using scientific principles ...
• and experience of successful systems and
  subsystems ..
• embedded in a mature process ...
• with effective quality and risk management
• Many of our current concerns would be familiar to
  the builders of the Great Pyramids

11
7 Lessons from the Past

• Complexity kills
• The cost of correcting an error rises steeply
  with time
• Change is inevitable
• Most development is documentation, not invention
• Reuse is better than originality
• Engineering is a process of risk management
• Optimisation is the last thing you should do

12
Complexity kills

• Writing simple programs for simple problems is
  easy
• Rigorous abstraction masters complexity
• Mathematical rigour has always been essential
  from LL1 grammars strong typing concurrency
  primitives and E/R models to formal
  specifications and proof
• Structural simplicity is key modularity,
  information hiding (Simula 67!), low coupling /
  high cohesion, problem-oriented structures (eg
  JSP/JSD), Object Orientation

13
The cost of correcting an error rises steeply
with time

• Possibly 10-fold with each lifecycle phase
• Specification errors are 100 times more costly
  than coding errors if both are found in unit
  testing
• A good development method provides strong VV at
  each phase - this is why formal methods are so
  cost-effective
• Costs increase with time, even when the product
  is in service

14
Change is inevitable

• Every business system encapsulates a business
  process - which has to change to stay competitive
• Every successful product needs new versions with
  new features
• The lifetime cost of most software is 5-10 times
  the development cost
• So maintenance and upgrade are the most costly
  steps in the lifecycle
• We therefore need to focus on methods, tools and
  system architectures that
• preserve structural integrity
• limit the growth in complexity

15
Most development is documentation, not invention

• In all engineering, the main output is
  documentation
• aeronautical engineers say that an aircraft is
  not ready to fly until the documentation weighs
  more than the aircraft
• For software, the documentation must support the
  maintenance/update phase
• and vice-versa!
• Updating code before updating specification or
  design documents is vandalism - it reduces (and
  ultimately destroys) the value of the product
• Writing good code from good documentation is easy
  - the reverse can be impossible

16
Reuse is better than originality

• Engineers build systems from reliable subsystems
• They use successful designs from the past
• A new design is almost always wrong
• even if it isnt, building justified confidence
  is very expensive
• Studying successful systems of the past is the
  basis for building successful systems in the
  future
• Design for reuse
• components that provide a single function or
  closely related group of functions
• clear, precise specification
• secure version management

17
Engineering requires the management of risk

• We cannot know everything we need to, when we
  start
• Unacceptable risks must be identified and
  eliminated early
• The residual risks have to be managed.
• plan to do work to avoid or accommodate them
• the cost multiplied by the probability must be
  planned in the budget for the project
• revised costs and probabilities are needed for
  every project review
• The forecast cost to complete is a three-valued
  function
• if all risks materialise
• engineering judgement
• if no risks materialise

18
Optimisation is the last thing you should do

• Jacksons law
• Jacksons second law (for experts only)
• This doesnt mean being wasteful of resources
• a good engineer should know how to work
  economically
• an engineer is someone who can do for twopence
  what any fool can do for a shilling

19
Conclusions

• The professional software engineer is someone who
• has a good grounding in computer science,
  mathematics, and human factors
• understands and can work with mature engineering
  processes involving teamwork, risk management,
  and quality assurance
• understands measurement as a basis for quality
  improvement
• has studied successful systems of the past, and
  understands their design principles
• understands and works within the limits of their
  own competence
• keeps up to date by reading the journals and
  continuous professional development

20
What we know Specifications

• Form must mirror the structure of the real world
  requirement. Principles of Program Design,
  Jackson, AP 1975
• Notation must support unambiguous statement of
  requirements and reasoning about properties. Only
  then can you understand the system, or the impact
  of changes.
• Mathematical foundations are essential - Alan
  Turing knew this in the 1940s.

21
Abstraction

• The two most important characteristics of a
  specification notation are (1) that it should
  permit problem-oriented abstractions to be
  expressed
• and (2) that it should have rigorous semantics
  so that specifications can be analysed for
  anomalies and to explore system properties.
• In this connection it might be worthwhile to
  point out that the purpose of abstracting is not
  to be vague, but to create a new semantic level
  in which one can be absolutely precise. Dijkstra
  1972

22
Realism in specification

• we must confine ourselves to the design and
  implementation of intellectually manageable
  programs. If someone fears that this
  restriction is so severe that we cannot live with
  it, I can reassure him the class of
  intellectually manageable programs is still
  sufficiently rich to contain very many realistic
  programs for any problem capable of algorithmic
  solution. Dijkstra 1972
• Modern formal specification methods support
  problem-oriented abstractions and hugely extend
  the range of systems that are intellectually
  manageable. But they are infrequently used!

23
What we know Engineering

• Engineering is the application of science, within
  mature processes, to build practical systems,
  cost-effectively.
• For software engineers, this means applying
  computer science, within mature processes for
  planning, risk and quality management, formal
  reviews, configuration management, and the rest
  of ISO 9001/CMM level 3
• All engineers use mathematics to model and
  analyse their systems.

24
What we know Testing

• Testing shows the presence, not the absence, of
  bugs. Dijkstra 1972
• One can construct convincing proofs quite
  readily of the ultimate futility of exhaustive
  testing of a program and even of testing by
  sampling. So how can one proceed? The role of
  testing, in theory, is to establish the base
  propositions of an inductive proof. You should
  convince yourself, or other people, as firmly as
  possible, that if the program works a certain
  number of times on specified data, then it will
  always work on any data. This can be done by an
  inductive approach to the proof. Hoare 1969

25
What we know Evidence

• If you cannot reason about the impact of a
  change, the change must invalidate all
  pre-existing evidence of dependability.

26
Software Death

• Software dies when a typical fix introduces more
  errors than it corrects
• If your average error rate is 1 error/50 LoC,
  your software dies when the size of your average
  bug fix exceeds 50 LoC.
• Debugging maximises the number of bugs remaining,
  for a given reliability.
• IBM data showed that typical bugs only recurred
  every 100 user-years!

27
The State of the Art in Software

• Annual cost to US economy of poor quality
  software 60B source US NIST Report
  7007.011, May 2002.
• Typical industrial / commercial software
  development
• 6-30 faults delivered / 1000 lines of software
• 1M lines 6000-30,000 faults on delivery
• source Pfleeger Hatton, IEEE Computer, pp33-42,
  February 1997.

28
What we know Languages

• I will start with a strong statement of opinion.
  I think that any significant advance in the
  programming art is sure to involve very extensive
  automated analyses of programs. Doing
  thorough analyses of programs is a big job. It
  requires a programming language which is
  susceptible to analysis. I think other
  programming languages will head either to the
  junk pile or to the repair shop for overhaul, or
  they will not be effective tools for the
  production of large programs. E S Lowry (IBM)
  Nato Conference, Rome, October 1969.
• Such languages and tools exist
  http//www.sparkada.com

29
Experience with Strong Software Engineering

• Software development based on formal
  specification, strong static analysis, ISO 9001
  mature processes
• 0.1 - 1 faults / 1000 lines of software
• 1M lines 100 - 1000 faults on delivery instead
  of 6000 - 30000
• 100-fold improvement at no extra development
  cost!
• sources
• Pfleeger Hatton, IEEE Computer, pp33-42,
  February 1997
• Amey, http//www.sparkada.com/downloads/Mar2002Am
  ey.pdf

30
Formal methods can reduce risk, save money, and
deliver successful projects

• Z is more efficient at finding faults than the
  most efficient test phase
• http//www.sparkada.com/downloads/sholis.pdf
• compelling evidence that development methods
  that focus on bug prevention rather than bug
  detection can both raise quality and save time
  and money.
• http//www.sparkada.com/downloads/Mar2002Amey.pdf

31
But get the basics right first

• First ISO 9001 and/or CMM Level 3
• now you are under control
• ThenFormal methods for requirements
• now you really understand what you are trying to
  do
• Then ...Implementation using well-defined
  languages with formal annotations
• for example, SPARK

32
Two reasons for using Formal Methods

• Formal methods reduce development risks and save
  development budget
• FMs are cheaper
• Formal methods lead to far fewer delivered bugs
• delivered software is much higher quality
• the extra quality is free
• support costs are much reduced
• you can even guarantee the software

33
Summary

• We can build complex, dependable systems - but
  only if we always use strong software
  engineering.
• System and software engineering must be rigorous,
  disciplined, conservative and evolutionary,
  learning from what has worked dependably in the
  past. Formal methods are fundamental to strong
  software engineering.
• It will take numerate, motivated graduates to
  change things for the better you.

34
Questions?