Software Engineering Research & Development at A.U.Th

Software Engineering Software Engineering Research & DevelopmentResearch & Development

at A.U.That A.U.Th

Ioannis StamelosIoannis StamelosPanagiotis KatsarosPanagiotis Katsaros

{stamelos, katsaros}@csd.auth.gr{stamelos, katsaros}@csd.auth.gr

18 Apr 200618 Apr 2006 SwedenSweden 22

Presentation OutlinePresentation Outline

• IntroductionIntroduction• AUTH / Dept of InformaticsAUTH / Dept of Informatics• Software Engineering Research Software Engineering Research

GroupGroup• Software Quality, Management, Software Quality, Management,

EducationEducation• Open Source Software Engineering / Open Source Software Engineering /

Extreme ProgrammingExtreme Programming


IntroductionIntroduction



City of ThessalonikiCity of Thessaloniki• Capital of Macedonia region, Capital of Macedonia region,

northern Greecenorthern Greece

• ~1 M inhabitants, 2~1 M inhabitants, 2ndnd greek city, greek city, major port of Balkansmajor port of Balkans

• 2300 years of continuous history, 2300 years of continuous history, joined modern Greece in 1912joined modern Greece in 1912



AUTH / Dept of InformaticsAUTH / Dept of Informatics


Aristotle University of Aristotle University of Thessaloniki, GreeceThessaloniki, Greece• First school (Philosophy) established in 1925First school (Philosophy) established in 1925

• Largest Greek UniversityLargest Greek University

– 45 departments45 departments

– > 60.000 students> 60.000 students

– > 3.000 staff / teachers> 3.000 staff / teachers

• www.auth.grwww.auth.gr


Department of InformaticsDepartment of Informatics

• School of Physical Science, plans to move to School of Physical Science, plans to move to Polytechnic SchoolPolytechnic School

• Established in 1993Established in 1993

• 22 academic staff members22 academic staff members

• ~700 students~700 students

• www.csd.auth.grwww.csd.auth.gr


4-years Degree Curriculum4-years Degree Curriculum• 8 semesters8 semesters

• 5 semesters with obligatory courses5 semesters with obligatory courses

• 3 semesters with options out of 4 directions:3 semesters with options out of 4 directions:– Information SystemsInformation Systems– Digital MediaDigital Media– Multimedia / Educational TechnologiesMultimedia / Educational Technologies– Networks / Architecture / CommunicationsNetworks / Architecture / Communications

• Thesis Project Thesis Project


Master CoursesMaster Courses• MSc in Informatics (4 directions)MSc in Informatics (4 directions)• MSc in Informatics & Management MSc in Informatics & Management

(with Dept. of Business Management)(with Dept. of Business Management)

PlusPlus

• MSc in Medical Informatics MSc in Medical Informatics


Five Research LabsFive Research Labs• Programming Languages and Software Programming Languages and Software

Engineering – PLASEEngineering – PLASE• Data Technologies and Engineering -DELABData Technologies and Engineering -DELAB• Information Processing and Artificial Information Processing and Artificial

IntelligenceIntelligence• Multimedia (Educational Technologies)Multimedia (Educational Technologies)• Networks, Communications and Networks, Communications and

ArchitectureArchitecture


Programming Languages and Programming Languages and Software Engineering – PLASESoftware Engineering – PLASE

• 6 staff members6 staff members• 10 research associates10 research associates• 20 PhD students20 PhD students• Major research areas:Major research areas:

– Intelligent SystemsIntelligent Systems– Web EngineeringWeb Engineering– Software EngineeringSoftware Engineering– MathematicsMathematics


Software Engineering Group Software Engineering Group (SWENG)(SWENG)• 3 ac. staff members:3 ac. staff members:

– Assistant Professors I. Stamelos, L. Angelis / PLASE Assistant Professors I. Stamelos, L. Angelis / PLASE – P. Katsaros / DELABP. Katsaros / DELAB

• 5 research associates5 research associates

• 10 PhD students (joined the group during last 4 10 PhD students (joined the group during last 4 years)years)

• http://sweng.csd.auth.grhttp://sweng.csd.auth.gr

http://sweng.csd.auth.gr/


SWENG: TeachingSWENG: Teaching• Dept. of Informatics / Hellenic Open Dept. of Informatics / Hellenic Open

University (ODL)University (ODL)– Mathematics, Graph TheoryMathematics, Graph Theory– Language Theory, CompilersLanguage Theory, Compilers– Introduction to Software EngineeringIntroduction to Software Engineering– Object Oriented Analysis / Java / C++Object Oriented Analysis / Java / C++

• Master CoursesMaster Courses– Software Project ManagementSoftware Project Management– Enterprise Information Systems Enterprise Information Systems


SWENG: R&D ActivitiesSWENG: R&D Activities• Basic ResearchBasic Research

• Collaboration with other research teams in Greece and in Collaboration with other research teams in Greece and in EU:EU:– Politecnico di Torino (Prof. Morisio)Politecnico di Torino (Prof. Morisio)– Univ. LAMSADE / Paris (Prof. Tsoukias)Univ. LAMSADE / Paris (Prof. Tsoukias)– Univ. Juan Carlos / Madrid (Prof. Barahona)Univ. Juan Carlos / Madrid (Prof. Barahona)

• Funded Applied Research & DevelopmentFunded Applied Research & Development– Greek Public SectorGreek Public Sector– Greek Secretariat for Research and TechnologyGreek Secretariat for Research and Technology– Private CompaniesPrivate Companies– EUEU


SWENG: FundingSWENG: Funding• Software industry context:Software industry context:

– Public Sector still developing IT infrastructure (e.g. transportation)Public Sector still developing IT infrastructure (e.g. transportation)– Private Sector companies enhancing IT infrastructure (banks, utilities, Private Sector companies enhancing IT infrastructure (banks, utilities,

privatized public companies)privatized public companies)– Most large SW Companies mainly acting as vendors (e.g. of ERP systems)Most large SW Companies mainly acting as vendors (e.g. of ERP systems)– Various small but aggressive SW companies acting as developers Various small but aggressive SW companies acting as developers

• Funding: limited but growing fast Funding: limited but growing fast – (in last 5 years, while SWENG group was growing) ~0,7 M euros(in last 5 years, while SWENG group was growing) ~0,7 M euros– mostly from development projectsmostly from development projects– participated also to other PLASE projects (MATHIND, Go Net, …)participated also to other PLASE projects (MATHIND, Go Net, …)

• Now shifting to research projects as well: Now shifting to research projects as well: – obtained ~0,5 M, expecting another 0,6 M in years 2006 - 2007 (mainly obtained ~0,5 M, expecting another 0,6 M in years 2006 - 2007 (mainly

from EU, GSRT)from EU, GSRT)– see forthcoming research projects at see forthcoming research projects at Forthcoming Research ProjectsForthcoming Research Projects slide slide


Examples of current / Examples of current / completed projectscompleted projects

• Expert System for Software EvaluationExpert System for Software Evaluation• Automated Certification System (with Automated Certification System (with

data mining functionality)data mining functionality)• Enterprise Knowledge Management Enterprise Knowledge Management

SystemSystem• Specification of various public company Specification of various public company

information systems (e.g. ticketing) information systems (e.g. ticketing)


SWENG: Major Research SWENG: Major Research AreasAreas• Software QualitySoftware Quality

– Dependability, SecurityDependability, Security

• Software Cost EstimationSoftware Cost Estimation

• Software EvaluationSoftware Evaluation

• Open Source SoftwareOpen Source Software

• Extreme ProgrammingExtreme Programming

• Software EducationSoftware Education


Software Quality, Software Quality, Management, EducationManagement, Education


Software Quality (1)Software Quality (1)• Problems: Problems:

– how to quantify and predict / assess SW quality how to quantify and predict / assess SW quality attributesattributes

– many SW quality metrics /methods have been many SW quality metrics /methods have been proposed but have not been sufficiently validatedproposed but have not been sufficiently validated

• Need for Need for – novel, intuitively confirmed measures novel, intuitively confirmed measures – empirical evidence for assessing methods, tools, empirical evidence for assessing methods, tools,

approachesapproaches


Software Quality (2)Software Quality (2)• Formal experimentsFormal experiments

– Most powerful empirical investigation tool, producing Most powerful empirical investigation tool, producing generic conclusionsgeneric conclusions

• A typical formal experiment setting: A typical formal experiment setting: – Two groups of subjects are given the same object, Two groups of subjects are given the same object,

differing only in the controlled variable (e.g. design of the differing only in the controlled variable (e.g. design of the same system based on two different design approaches, same system based on two different design approaches, see next two slides) see next two slides)

– Subjects are asked to perform the same task Subjects are asked to perform the same task – All other potentially affecting variables are kept constant All other potentially affecting variables are kept constant

(e.g. group experience)(e.g. group experience)– The two groups’ performance is monitored accurately and The two groups’ performance is monitored accurately and

statistical tests are run to detect statistical difference statistical tests are run to detect statistical difference




Software Quality (3)Software Quality (3)• Currently investigating quality assessment based Currently investigating quality assessment based

on Heuristicson Heuristics- A Heuristic is a good practice (e.g. avoidance of ‘God’ - A Heuristic is a good practice (e.g. avoidance of ‘God’

classes in OO design, see previous slide)classes in OO design, see previous slide)– Developed tools assessing compliance with Riel’s set of Developed tools assessing compliance with Riel’s set of

object-oriented heuristics (based on C++/Java code, object-oriented heuristics (based on C++/Java code, UML class diagrams)UML class diagrams)

• Design of new methods / models for quality Design of new methods / models for quality predictionprediction– Software Defect Prediction Using Regression via Software Defect Prediction Using Regression via

ClassificationClassification (AICSSA ’06)(AICSSA ’06)


Decision tree for defect Decision tree for defect prediction (for a specific data prediction (for a specific data set)set)


Software Cost EstimationSoftware Cost Estimation• The problem: predict the human effort for developing a systemThe problem: predict the human effort for developing a system

– Expert judgmentExpert judgment– Analogy based method, project is compared with past projects, Analogy based method, project is compared with past projects,

project cost is proportional to cost of ‘similar’ projectsproject cost is proportional to cost of ‘similar’ projects– Algorithmic / ad hoc models, a set of equations are used (e.g. Algorithmic / ad hoc models, a set of equations are used (e.g.

the COCOMO models)the COCOMO models)

• Issues:Issues:– Wrong, inaccurate, missing past projects dataWrong, inaccurate, missing past projects data– Uncertainty about new projectUncertainty about new project– Many categorical values (e.g. type of system, language used)Many categorical values (e.g. type of system, language used)– Method accuracyMethod accuracy– Calibration with local dataCalibration with local data– Method suitability to local situation (e.g. local project data may Method suitability to local situation (e.g. local project data may

be not available or projects are not measured accordingly)be not available or projects are not measured accordingly)


Software Cost Estimation (1)Software Cost Estimation (1)• Interval estimation Interval estimation

– A range of values is producedA range of values is produced– More realistic approach w.r.t to a point estimateMore realistic approach w.r.t to a point estimate

• Interval incorporates model errorInterval incorporates model error• Interval accounts for estimation data inaccuracy Interval accounts for estimation data inaccuracy

– May be easily transformed to a point estimate May be easily transformed to a point estimate (e.g. take the mean value of the interval)(e.g. take the mean value of the interval)

• A number of techniques has been proposedA number of techniques has been proposed


Software Cost Estimation (2)Software Cost Estimation (2)• Improvement of Analogy Based Estimation (Bootstrap Improvement of Analogy Based Estimation (Bootstrap

sampling method)sampling method)– To calibrate method parameters (e.g. number of similar projects To calibrate method parameters (e.g. number of similar projects

used for the estimation)used for the estimation)– To produce interval estimatesTo produce interval estimates

• Project portfolio estimationProject portfolio estimation– Combination of interval estimates from portfolio project membersCombination of interval estimates from portfolio project members

• Categorical, multinomial, ordinal regressionCategorical, multinomial, ordinal regression– Regression models that provide interval estimatesRegression models that provide interval estimates– Intervals are often predefined by managersIntervals are often predefined by managers– Produced more accurate results than previously proposed Produced more accurate results than previously proposed

methods (e.g. OLS) methods (e.g. OLS)


BRACE: a tool for advanced BRACE: a tool for advanced estimation by analogyestimation by analogyScreenshot of Screenshot of

BRACE BRACE showing the showing the results of a results of a calibration calibration sessionsession


BRACE: interval estimate BRACE: interval estimate generationgenerationNew project is New project is

estimated to estimated to demand an demand an effort of 2000-effort of 2000-2500 mh (prob. 2500 mh (prob. 65,1%), 2500-65,1%), 2500-3000 (prob. 3000 (prob. 23,1%), etc.23,1%), etc.


Software Cost Estimation (3)Software Cost Estimation (3)• How to deal with missing data in your How to deal with missing data in your

historical data base?historical data base?• Imputation methods: replace missing Imputation methods: replace missing

valuesvalues• Proposed Multiple Logistic Regression as an Proposed Multiple Logistic Regression as an

imputation method for missing categorical imputation method for missing categorical valuesvalues

• MLR found to perform better w.r.t. to MLR found to perform better w.r.t. to previously proposed methodspreviously proposed methods


Software Cost Estimation (4)Software Cost Estimation (4)• Comparative assessment of Comparative assessment of

proposed methods (focusing on AI proposed methods (focusing on AI methods)methods)– Association Rules, BBN, CARTAssociation Rules, BBN, CART– AR are found to be the most accurate / AR are found to be the most accurate /

promising approachpromising approach


Example Association Rule Example Association Rule for Productivity Estimationfor Productivity EstimationAssociation RuleAssociation RuleSupportSupport ConfidenceConfidence Rule BodyRule Body Rule Head Rule Head 6.36.3 66.666.6 [ACAP_H]+[DATA_N][ACAP_H]+[DATA_N] ==> ==> [PROD_4][PROD_4]

Interpreted as follows: Interpreted as follows: When the programmers’ analysis capability When the programmers’ analysis capability is high and the database size is nominal then is high and the database size is nominal then the productivity is likely to be in the fourth the productivity is likely to be in the fourth category (100<PROD<160 KLOC/MM)category (100<PROD<160 KLOC/MM)


Software process modeling Software process modeling with iterative Bayesian Belief with iterative Bayesian Belief NetworksNetworks


Software EvaluationSoftware Evaluation• Application of MCDA (Multiple Criteria Application of MCDA (Multiple Criteria

Decision Aid) on software problem Decision Aid) on software problem situationssituations


Profile evaluation of software Profile evaluation of software artifacts / processesartifacts / processes

User defines profiles as sets of attribute User defines profiles as sets of attribute values, then product / process is assessed values, then product / process is assessed through the Electre method to decide its through the Electre method to decide its profile value profile value

g1

g2

g3

gm-1

gm

Categ. 1 Categ. 2 Categ. p-1 Categ. p Categ. p+1

b1bp-1 bp


Software Engineering Software Engineering EducationEducation• Investigation of novel approaches in Investigation of novel approaches in

SE education through formal SE education through formal experimentsexperiments– Distance learning of UML (sequence Distance learning of UML (sequence

diagrams)diagrams)– Use of a lesson sheet for teaching Use of a lesson sheet for teaching

UsabilityUsability


Open Source Software Open Source Software Engineering / Extreme Engineering / Extreme ProgrammingProgramming


Free / Open Source Software Free / Open Source Software (1)(1)• Revolutionary software developmentRevolutionary software development• Issues:Issues:

– What is the quality of F/OSS products?What is the quality of F/OSS products?– How can we assess an F/OSS project?How can we assess an F/OSS project?

• Is it active?Is it active?• Is it interesting?Is it interesting?

– How can we predict an F/OSS project?How can we predict an F/OSS project?•How will it grow in terms of code size, How will it grow in terms of code size,

number of programmers?number of programmers?


Free / Open Source Software Free / Open Source Software (2)(2)• F/OSS product quality assessmentF/OSS product quality assessment

• F/OSS process modelingF/OSS process modeling– Dynamic models for predicting F/OSS project Dynamic models for predicting F/OSS project

behaviorbehavior

• Knowledge sharing and learning in F/OSSKnowledge sharing and learning in F/OSS


F/OSS product quality F/OSS product quality assessmentassessment• Case StudiesCase Studies

– Measured and assessed 100 LINUX applications: Measured and assessed 100 LINUX applications: applications’ code found to be of at least equal quality applications’ code found to be of at least equal quality w.r.t. to closed source codew.r.t. to closed source code

– Measured and assessed specific software systems’ quality Measured and assessed specific software systems’ quality (e.g. assessed COMPIERE code, an Open Source ERP (e.g. assessed COMPIERE code, an Open Source ERP system, found to be of excellent quality) system, found to be of excellent quality)

• Quality assessment horizontal studiesQuality assessment horizontal studies– Measured, assessed 5 F/OSS projects and compared with Measured, assessed 5 F/OSS projects and compared with

closed source counterparts using Maintainability Index (SEI)closed source counterparts using Maintainability Index (SEI)– F/OSS systems found to exhibit superior maintainability, F/OSS systems found to exhibit superior maintainability,

however F/OSS maintainability deteriorated in similar way however F/OSS maintainability deteriorated in similar way as in closed source as in closed source


F/OSS maintainability F/OSS maintainability evolutionevolution

PrC vs. PrD vs. PrE

01020304050607080

1 2 3 4 5 6 7 8 9 10 11 12 13 14

successive versionsM

I

PrC

PrD

PrE

OSS PrA vs. CSS PrA

01020304050607080

1 2 3 4 5 6 7 8 9 10 11 12 13

successive versions

MI OSS PrA

CSS PrA


F/OSS process modelingF/OSS process modeling• Dynamical F/OSS process modelDynamical F/OSS process model• Based on programmer motivationBased on programmer motivation• Elaborates on programmer tasks Elaborates on programmer tasks

(coding, debugging), modules, defects (coding, debugging), modules, defects • Calibrated on typical F/OSS projects Calibrated on typical F/OSS projects

(e.g. Apache, Gnome)(e.g. Apache, Gnome)• Predicting curves for size (LOCs), Predicting curves for size (LOCs),

number of programmersnumber of programmers


Knowledge sharing and Knowledge sharing and learning in F/OSSlearning in F/OSS• Modeling of knowledge sharing in F/OSSModeling of knowledge sharing in F/OSS

– Knowledge sharing modelKnowledge sharing model– Mining F/OSS mail lists to determine knowledge Mining F/OSS mail lists to determine knowledge

brokersbrokers• Learning mechanisms in F/OSSLearning mechanisms in F/OSS

– Learning in F/OSS occurs following the Learning in F/OSS occurs following the constructivist model constructivist model

– Ultimate goal: use of F/OSS projects in SE Ultimate goal: use of F/OSS projects in SE teachingteaching

– Ran a pilot study: 15 students practiced Ran a pilot study: 15 students practiced software testing by participating in F/OSS software testing by participating in F/OSS projects, obtained promising resultsprojects, obtained promising results


F/OSS Knowledge Sharing F/OSS Knowledge Sharing Model Model Knowledge is Knowledge is

generated, generated, shared and shared and acquired by acquired by traversing traversing various paths various paths of the graph of the graph model in many model in many different waysdifferent ways


F/OSS Mailing List Social F/OSS Mailing List Social NetworkNetwork


Extreme ProgrammingExtreme Programming• Surveying Agile Methods / XP spread in GreeceSurveying Agile Methods / XP spread in Greece

– SW SMEs use XP methods without knowing it!SW SMEs use XP methods without knowing it!

• Investigation of human factorsInvestigation of human factors– Developer personality and temperament when Developer personality and temperament when

forming a pair in pair programmingforming a pair in pair programming– Conducted formal experiments with students Conducted formal experiments with students – Found that mixed personalities and temperaments Found that mixed personalities and temperaments

produce better resultsproduce better results


Forthcoming Funded Research Forthcoming Funded Research ProjectsProjects• SQO – OSS (IST/EU), Software quality observatory for Open SQO – OSS (IST/EU), Software quality observatory for Open

Source, a system that will continuously monitor and Source, a system that will continuously monitor and evaluate OSS productsevaluate OSS products

• FLOSSMETRICS (IST/EU), studies and models for OSS FLOSSMETRICS (IST/EU), studies and models for OSS process managementprocess management

• F/OSS in education (MINERVA / EU), exploitation of OSS in F/OSS in education (MINERVA / EU), exploitation of OSS in informatics educationinformatics education

• DIERGASIA (means PROCESS in Greek) (TELETEL / GSRT), DIERGASIA (means PROCESS in Greek) (TELETEL / GSRT), prediction models and procedures for advanced software prediction models and procedures for advanced software quality assurance in a TLC systems company quality assurance in a TLC systems company

Also bidding for funding for various development projects Also bidding for funding for various development projects (training in basic computer skills using ODL, portal (training in basic computer skills using ODL, portal development for technology museums, …) development for technology museums, …)


Current / Future Research Current / Future Research topicstopics• F/OSS quality assessment through the use of heuristics F/OSS quality assessment through the use of heuristics • Social Networks in F/OSSSocial Networks in F/OSS

– Indicators for community decay in F/OSS projects: when is the project Indicators for community decay in F/OSS projects: when is the project endangered because of mentors / key developers abandonment? endangered because of mentors / key developers abandonment?

• Software ArchitectureSoftware Architecture– Exploit architectural tactics in component specification to support Exploit architectural tactics in component specification to support

reusability (ICSR’06, Torino)reusability (ICSR’06, Torino)• Management Anti-PatternsManagement Anti-Patterns

– Modeling with BBNModeling with BBN– XP management anti-patternsXP management anti-patterns

• Intellectual Software Capital ValuationIntellectual Software Capital Valuation– How much is a software artifact (component, method, piece of How much is a software artifact (component, method, piece of

knowledge) worth? knowledge) worth? • Investigate XP through experiments with professionals Investigate XP through experiments with professionals • Introduction of F/OSS projects in SE educationIntroduction of F/OSS projects in SE education


Tactic driven processTactic driven process

Software Engineering Research & Development at A.U.Th

Documents

research projects

research teams

informatics management

staff teacherswww

academic staff members

aggressive sw companies

modern greece

northern greece