An Engineering Approach to Webbased Application Development

1
An Engineering Approach to Web-based Application
Development

• Cornelia Boldyreff
• Distributed Systems Engineering Group,
• RISE, University of Durham, UK
• www.dur.ac.uk/cornelia.boldyreff

2
Outline

• Web Site Development and Maintenance
• Web Metrics
• Web Site Evolution
• Web Management and Design Processes
• Web Site Engineering

3
Web Site Development and Maintenance

• Recognition of potential Web maintenance problem
  and very interesting case of software evolution
• Key paper - Measuring Readability and
  Maintainability of Hyperdocuments, Hatzimanikatis
  et al, JSM, 1995
• Possibility of working with Richard West and
  colleagues managing UK government Web sites

4
Initial Findings

• Uneven, largely poor, quality of authorship
• Badly structured web documents
• Difficulty in navigating web documents
• Hypertext structures rapidly become complex as
  the linking of nodes increases
• Lack of tools for managing Web development and
  maintenance

5
Further Findings

• Applications on the WWW are often distributed and
  therefore maintained by several
  authors/maintainers
• Maintenance of WWW applications relies on the
  error logs of each server, complaints from users,
  periodic checks by owners
• No general consensus of opinion, or standards, on
  what constitutes a good web application
  therefore assessment is difficult

6
Analyzing and Assessing Hyperdocuments

• Readability Factors
• Maintainability Factors
• Effectiveness Factors - Usability
• Evaluation process in practice needed
• High level factors needed to be related to low
  level measurable factors
• Lessons to be learnt from classical software
  engineering, especially software maintenance

7
Key Results

• Analysis and Assessment of Web documents GQM
  and ISO Product Quality Standard
• Requirements Studies for Web Site Development and
  Management Support
• Workbench developed - see www.dur.ac.uk/cornelia.b
  oldyreff/workbench

8
Solving the Broken Link Problem

• Major effort needed to maintain links and ensure
  their currency
• Existing solutions involved non-standard
  implementations, e.g. Hyper-G(Hyperwave)
• Guardian Agent solution implemented using
  standard http and CORBA - employing both eager
  and lazy link update - patented by British Telecom

9
BT Requirements for WWW Maintenance

• Link management - if documents are relocated,
  links should be redirected automatically to the
  new document location
• Version control - support re-configuration of
  existing sites with new up-dates
• Support of team working - several authors should
  be able to work on the same documents without
  interfering with each other

10
Guardian Agents

• Compatible with WWW
• Provide on-line maintenance
• Work in background - transparent to the user
• Flexible
• Scalable
• Configurable

11
Guardian Agents

• Guardian Agents analyze, monitor and record the
  following information
• Outgoing and incoming links
• Who accesses the web application
• Who changes the web application

12
Determining Success - Web Quality Metrics

• Apply Goal-Question-Metric
• Metrics derived from Software Metrics, e.g.
  Complexity measures
• Tailored Hypertext Metrics, e.g. tree impurity
• New Web Metrics, e.g. Brays html sincerity
• Both static and dynamic measures required

13
Software Evolution - Web Evolution

• Inspired by Lehman classic studies in Software
  Evolution
• Metrics here used to study changes over time
• Aim to understand and predict the web evolution
  process
• Support better, more controlled, web maintenance
  and management

14
Early Metrics

• Key paper - Measuring Readability and
  Maintainability of Hyperdocuments, Hatzimanikatis
  et al, JSM, 1995
• Also earlier work by Brown on maintaining
  long-life hypertext, Botafogo et al, and Garzotto
  et al on hypertext quality
• Brays paper, Measuring the Web, WWW5, 1996

15
Content Metrics

• Measures to effect editorial control, e.g. spell
  checking, grammar checking, content checking
  (refereeing)
• Measures to gauge readability, e.g. sentence
  lengths, vocabulary analysis, font analysis
• Measures to gauge usability, e.g. accessibility
  checking (Bobby - www.cast.org/bobby),
  colour-blindness checking

16
Structural Metrics

• Metrics derived from Software Metrics
• Per-document metrics LOC - lines of code, COM -
  lines of comments, NOM - number of modules
  (nodes), number of links (edges), counts of
  various html tags, MVG - McCabes Cyclomatic
  complexity within the documents (for local
  links), fan-in, fan-out, Henry-Kafura/Sheppard
  information flow measure - (fan-infan-out)2
• per-site metrics Number of documents, totals and
  means of the above measures

17
Web Evolution Studies

• Structural metrics used
• Method
• develop and apply measurements over time
• identify patterns of change and develop models of
  change
• test theories against more case studies
• Fixed intervals of times used instead of release
  dates as in classical software evolution studies

18
WWW as Software-in-the-large

• Web software evolves and degrades like
  conventional software - all forms of change
  occur. Perfective, adaptive, corrective, and
  preventative maintenance is needed
• Speculative or Pre-emptive maintenance is also
  needed (e.g. Link Checking)
• Maintenance may introduce errors - regression
  testing
• A large manpower commitment is required

19
WWW as Software-in-the-small

• Hypertext is represented as a graph - compare the
  call graph of traditional programs
• Links are edges Modules (files) are nodes.
• Web pages may contain embedded software, e.g.
  cgi, Java, VRML, etc
• Users execute links
• Different kinds of links images, frames, etc
• Referential/organizational links

20
Web Sites Chosen for Study

• S.T.A. - 20 files (very small)
• Cartercopters - 60 files (small)
• Durham University (part) - 641 (medium)
• Sunderland University (part) - 438 (medium)
• BBC - thousands
• BT - thousands

21
SiteSeer and parsley
1 per site

• Web spider
• parsley built using UNIX lex and yacc
• problems of robot exclusion and perturbation
• other tools used to view and manipulate graphs
  produced

SiteSeer
parsley
parsley
1 per document
22
Early results from studies

• Simple metrics collected easily
• Large quantities of data to analyze
• Index documents easily identifiable -
  characterized by high fan-out (and fan-in) -
  exhibit very high coupling
• Observed that the overall rate of change of a
  site increased with the size of the site -
  smallest site exhibited no changes during first
  study
• Growth but never shrinkage!

23
Further Results

• Recurrent structures emerged from studies of site
  graphs - indices, tours, picture galleries can
  all be identified. Possible basis for further
  research on web design patterns.
• High link density (links per document) is
  strongly related to the probability of additional
  links being added to the document per unit time.

24
Future Web Metrics Research

• Research on hypertext and web metrics is in its
  infancy. Existing metrics require better
  interpretation (e.g. mapping from low level
  measures to high level quality factors).
• A unified measurement programme combining
  structural metrics with content metrics,
  especially usability, is needed.

25
Web Site Classification

• Web site evolution studies used a 2-D
  classification site size and nature of site
  ownership
• Others have used size and development focus
  (long-lived, short-lived, one-off applications),
  site type educational, commercial and
  institutional
• Some metrics may be more relevant to certain
  classes of sites, so determining a web site
  classification is a useful step before further
  metrics research.

26
Proposed Dimensions

• Size
• Domain
• Purpose
• Functionality
• Technology

• Age
• Rate of Change
• Evolution Strategy

27
Observations

• It seems likely that the maintenance process
  evolves with time and as the site grows, e.g. no
  changes, minor corrections, managed re-design,
  multi-developer site, multiple data base driven
  site.
• Evolution is closely related to usability as
  following Lehmans first law, it can be
  anticipated that a large web application must
  undergo continuous change or it will become
  progressively less useful.

28
Further Observations

• Measuring and modeling the WWW allows us to study
  its structure and contents, and determine quality
  factors operationally.
• Classification of Web Sites can give us a better
  insight into appropriate measures to study their
  evolution. One size does not fit all!
• Studying Web sites and how they change over time,
  gives us insights in web design and maintenance
  processes and their possible improvement.

29
Web Management and Design Processes

• Early process models developed based on UK
  government web site management practice
• Hypermedia/Web Development methods and models -
  surveyed and classified
• Metrics used here to guide research on process
  models for Web Site Engineering

30
Web Process Models

• Durham Workbench model for Web Maintenance - 1995
• UK government - CCTA Management Model 1996
• Survey of Hypermedia Design Methods - 1998
• Lowe and Halls Hypermedia Development Process
  Model - 1999

31
Early Maintenance Process Model
32
CCTA Management Model
Web page Development
Customer Awareness
Train in Web page development
External Organization
Database of HTML files
Transfer checked Web pages
Get customer for CCTA
CCTA Customer
CGIS Team
Mirror database of HTML files
Develop Web Pages
Check the page
Guidelines for Web page presentation
CGIS Team
33
Hypermedia/Web Development models and methods
Survey

• Design and Development stages from different
  methods
• abstracted Hypermedia/Web Development stages
• Relationship hierarchy of various methods

34
Method Relationship Hierarchy
Hypermedia Design Methods
EORM Lange
HDM Garzotto
OOHM Hendrix
STDT Bichler
HM-Data Maurer
RMM Isakowitz
Semi-formal methods
OOHDM Schwabe
Web Architect Takahashi
Extended RMM
Database Design Techniques
Object Oriented Techniques
35
Stages from Methods
36
Hypermedia Development stages
Feasibility Requirements Analysis
Requirements document
Conceptual Design
Conceptual design diagrams e.g. ER diagrams or
classes, etc
Navigational Design
Nodes and links, navigational diagrams, e.g.
Slices
User Interface Design
Interface and screen diagrams
Conversion Implementation
Finished Product
Testing Maintenance
37
Lowe and Halls Development Process Models

• found in their book - Hypermedia the Web An
  Engineering Approach
• Based on Traditional Software Engineering process
  models the Waterfall, Prototyping, Spiral Model.
• Fullest model includes project planning, risk
  management and project management with overall
  system architecture, system design and
  application partitioning.

38
Web Process Improvement and CMM

• Process models, methods, plus metrics provide the
  basis for Web Process Improvement
• Companies need guidance on how best to improve
  their existing Web development and maintenance
  practices
• Web Usability Engineering is a good starting
  point as it is key when a company is trying to
  attract and keep its Web site customers

39
Web Site Engineering projects

• Web-SEM project - Establishing Effective Web Site
  Evaluation metrics - building on earlier web
  metrics and evolution studies combined with web
  usability and quality metrics
• Small projects on web site re-engineering, clone
  detection, web site re-use (web-in-a-box), web
  site life cycle and automating recurrent
  maintenance activities, metrics for web quality
  determination.

40
Improving Basic Web Site and Web Product Design

• Popularising the concept of web engineering -
  taking a systematic and disciplined approach to
  engineering web applications among small and
  medium enterprises - Business Informatics project
• Working with the CACDP to migrate all their
  products and services to the WWW, helping them to
  develop a well-defined engineering approach with
  systematic reuse along classic SPI/CMM lines

41
Web Site Engineering

• Applying and adapting classical Software
  Engineering models, methods and tools to the
  engineering of web-based applications, e.g. web
  pages, web sites, web applications in general.
• Special case of Distributed System Engineering
• Closely related to Software Engineering but also
  recognizing important differences, e.g. periodic
  nature of web maintenance, pre-emptive nature of
  maintenance, faster rates of change, larger
  scales of deployment/usage than in classical
  software.

42
Towards Web Site Engineering

• Well developed models of Web development and
  maintenance processes
• Web Software Quality Determination
• Models and Laws of Web-based Software Evolution
• Support for distributed developers - Computer
  Supported Co-operative Working applied to Web
  Site Engineering

43
Key Points

• Web developers and maintainers can learn from
  Software Engineering
• Web metrics can help to evaluate, to describe,
  and to develop new approaches to web engineering
  processes and products
• Software Engineering can provide a foundation for
  Web Engineering
• Web Evolution studies provide guidance

44
Software Engineering
Distributed System Engineering
Software Metrics and SPI
Human-Computer Interaction
Hypermedia (Graph) Theory
Usability Metrics
Hypertext Metrics
Hyper-media Design
Web-based Software Engineering
45
References

• Warren et al, The Evolution of Websites, IWPC99.
• Warren et al, Characterising Evolution in Web
  Sites Some Case Studies, WSE99.
• Boldyreff et al, Web-SEM Project Establishing
  Web Site Evaluation Metrics, WSE2000
• Boldyreff et al, Establishing a Measurement
  Programme for the World Wide Web, SAINT01
• Boldyreff et al, Towards the Engineering of
  Commercial Web Based Applications, SSGRR01

• Kyaws technical report - Survey of Hypermedia
  Design Methods, CS-3-98
• Lavery, Designing Web Site Usability, and other
  links on www.dur.ac.uk/janet.lavery
• see Donkins webpages at www.dur.ac.uk/j.m.donkin
• Boldyreff and Kewish, Reverse Engineering to
  Achieve Maintainable WWW Sites, WCRE01