A Dynamic Integrated Framework for Software Process Improvement

A Dynamic In eg a ed F amewo k o So wa e P ocess Imp o emen MERCEDES RUIZ [email p o ec ed] Depa men o Compu e Languages and Sys ems, Escuela Supe io de Ingenie ía, Uni e si y o Cádiz, Spain ISABEL RAMOS AND MIGUEL TORO {isabel. amos, m o o}@lsi.us.es Depa men o Compu e Languages and Sys ems, Escuela Técnica Supe io de Ingenie ía In o má ica Uni e si y o Se ille, Spain Abs ac . Cu en so wa e p ocess models (CMM, SPICE, e c.) s ongly ecommend he applica ion o s a is ical con ol and measu e guides o define, implemen , and e alua e he e ec s o di e en p ocess imp o emen s. Howe e , whils quan i a i e modeling has been widely used in o he fields, i has no been conside ed enough in he field o so wa e p ocess imp o emen . Du ing he las decade so wa e p ocess simula ion has been used o add ess a wide di e si y o managemen p oblems. Some o hese p oblems a e ela ed o s a egic managemen , echnology adop ion, unde s anding, aining and lea ning, and isk managemen , among o he s. In his wo k a dynamic in eg a ed amewo k o so wa e p ocess imp o emen is p esen ed. This amewo k combines adi ional es ima ion models wi h an in ensi e u iliza ion o dynamic simula ion models o he so wa e p ocess. The aim o his amewo k is o suppo a quali a i e and quan i a i e assessmen o so wa e p ocess imp o emen and decision making o achie e a highe so wa e de elopmen p ocess capabili y acco ding o he Capabili y Ma u i y Model. The concep s unde lying his amewo k ha e been implemen ed in a so wa e p ocess imp o emen ool ha has been used in a local so wa e o ganiza ion. The esul s ob ained and he lessons lea ned a e also p esen ed in his pape . Keywo ds: so wa e p ocess modelling and simula ion, p ocess imp o emen , p ocess ma u i y, dynamic models 1. In oduc ion O e he pas ew decades so wa e complexi y has significan ly inc eased in such a way ha so wa e has eplaced ha dwa e as ha ing he p incipal esponsibili y o much o he unc ionali y p o ided by cu en sys ems. This inc easing ole o so wa e, he p oblems ela ed o cos and schedule o e uns, and he cus ome pe cep ion o low p oduc quali y ha e changed he ocus o a en ion owa ds he ma u i y o so wa e de elopmen p ac ices. Al hough he so wa e indus y has ecei ed significan help by means o Compu e Aided So wa e Enginee ing (CASE) ools, new p og amming languages and app oaches, and mo e ad anced and complex machines, he e is a lack o p ocess analysis ools o o ganiza ions in e es ed in imp o ing hei p ocess pe o mance. Dynamic modeling and simula ion as p ocess imp o emen ools ha e been in en- si ely used in he manu ac u ing a ea. Cu en ly, so wa e p ocess modeling and simula ion a e gaining an inc easing in e es among esea che s and p ac i ione s as an app oach o analyze complex business and sol e policy ques ions. In p e ious wo k (Ruiz, Ramos, and To o, 2001) we a emp ed o apply a complex dynamic model o suppo p ocess imp o emen in a local so wa e de elopmen o ganiza ion. The non-exis ence o a his o ical da abase and o measu emen p ac- ices inside his o ganiza ion made i impossible, as he e we e no nume ical d i e s o supply he model pa ame e s and unc ions. Then, we decided o apply Ebe lein’s wo k (Ebe lein, 1989) abou unde s anding and simplifica ion o models o ob ain a educed dynamic model capable o ep oducing he so wa e p ocess dynamics, ye wi h less ini ial in o ma ion equi ed. Bu simula ion is only e ec i e i bo h he model and he da a used o d i e i accu a ely eflec he eal wo ld. Thus, he cons uc ion o he model i sel poin s o wha me ic da a mus be collec ed and helps as a clea guideline on wha o collec . In his pape an app oach is p oposed ha combines adi ional es ima ion ech- niques wi h Sys em Dynamics modeling. The aim o his combina ion is o build a amewo k o suppo a quali a i e and quan i a i e assessmen o so wa e p o- cess imp o emen and decision making. The pu pose o his dynamic amewo k is o help o ganiza ions o achie e a highe so wa e de elopmen p ocess capabil- i y acco ding o he Capabili y Ma u i y Model (Paulk e al., 1993). The dynamic models buil inside his amewo k p o ide he capabili y o gaining insigh o e he whole li e cycle a di e en le els o abs ac ion. The le el o abs ac ion used in a ce ain o ganiza ion will depend on i s ma u i y le el. Fo ins ance, in a le el 1 o ga- niza ion he simula o can es ablish a baseline acco ding o adi ional es ima ion models om an ini ial es ima e o he size o he p ojec . Wi h his baseline, he so wa e manage can analyze he esul s ob ained wi h he simula ion o di e en p ocess imp o emen s and s udy he ou comes o o e o unde es ima ing cos o schedule. Du ing he simula ion me ic da a a e sa ed. These da a con o m o he SEI co e measu es (Ca le on e al., 1992) ecommenda ion, and a e mainly ela ed o cos , schedule and quali y. The s uc u e o he pape is as ollows. Sec ion 2 p o ides a b ie o e iew o he wo k conduc ed in he field o so wa e p ocess simula ion. In Sec ion 3, he jus ifi- ca ion ound o de elop he in eg a ed amewo k is p esen ed. Sec ion 4 desc ibes in de ail, he undamen al basis and s uc u e o his amewo k. The implemen a- ion and esul s ob ained when applying i inside a local o ganiza ion a e discussed in Sec ion 5. Finally, Sec ion 6 summa izes he pape and d aws he conclusions and lessons lea n . 2. So wa e p ocess simula ion Simula ion can be applied in many c i ical a eas in suppo o so wa e enginee ing. I enables one o add ess issues be o e hese issues become p oblems. Simula ion is mo e han jus a echnology, as i o ces one o hink in global e ms abou sys em beha io , and abou he ac ha sys ems a e mo e han he sum o hei compo- nen s (Ch is ie, 1999). A simula ion model is a compu a ional model ha ep esen s an abs ac ion o a simplified ep esen a ion o a complex dynamic sys em. Simula- ion models o e , as a main ad an age, he possibili y o expe imen ing wi h di e - en managemen decisions. Thus, i becomes possible o analyze he e ec o hose decisions in sys ems whe e he cos o isks o expe imen a ion make i un easible. Ano he impo an ac o is ha simula ion p o ides insigh s in o complex p ocess beha io which is no possible o analyze by means o s ochas ic models. Like many p ocesses, so wa e p ocesses can con ain mul iple eedback loops, such as hose associa ed wi h he co ec ion o de ec s. Delays esul ing om hese de ec s may ange om minu es o yea s. The esul ing complexi y makes i almos impossi- ble o men al analysis o p edic he consequences. The mos equen sou ces o complexi y in eal so wa e p ocesses a e: —Unce ain y. Some eal p ocesses a e cha ac e ized by a high deg ee o unce - ain y. Simula ion models make i possible o deal wi h his unce ain y as hey can ep esen i flexibly by means o pa ame e s and unc ions. —Dynamic beha io . Some p ocesses may ha e a ime dependen beha io . The e is no doub ha some so wa e p ocess a iables a y hei beha io as he ime cycle p og esses. Wi h a simula ion model i is possible o ep esen and o mal- ize he s uc u es and causal ela ionships ha dic a e he dynamic beha io o he sys em. —Feedback. In some sys ems he esul o a decision made in a ce ain momen can a ec hei u u e beha io . Fo example, in so wa e p ojec s he decision o educing he e o assigned o quali y assu ance ac i i ies has di e en e ec s o e he whole p og ess o hese p ojec s. Thus, he common objec i es o simula ion models consis o supplying mecha- nisms o expe imen , p edic , lea n, and answe ques ions such as: Wha i ? A so wa e p ocess simula ion model can be ocussed on ce ain aspec s o he so wa e p ocess o he o ganiza ion. I is impo an o bea in mind ha a simula- ion model cons i u es an abs ac ion o he eal sys em, and so i ep esen s only he pa s o he sys em ha ha e been in ended o be modeled. Fu he mo e, cu - en ly a ailable modeling ools such as i hink(High Pe o mance Sys ems, 2001), POWER-SIM(Powe Sim Co po a ion, 2001), and Vensim(Ven ana Sys ems, 2002) help o ep esen he so wa e de elopmen p ocess as a sys em o di e en- ial equa ions. This is a ema kable cha ac e is ic, as i makes i possible o o mal- ize and de elop a scien ific basis o so wa e p ocess modeling and imp o emen . Some no iceable applica ions o his dynamic app oach o model so wa e p ocess can be ound in (Kellne , Madachy, and Ra o, 1999). 3. Jus ifica ion Al hough adi ional me hods o so wa e p ojec managemen ha e e ealed hei weaknesses du ing he las decades, he e is common ag eemen ha hey a e s ill use ul. We hink ha i is impo an o in eg a e adi ional me hods and p ocess simula ion unde a common app oach, in o de o ob ain a aluable ool o design p ocess imp o emen s and e alua e hei e ec s. Rod igues and Bowe s (1996), d aw he conclusions ob ained a e ha ing compa ed bo h app oaches, and poin ou he necessi y o in eg a ion. T adi ional and dynamic app oaches ha e simila objec i es, ye he pe spec i e unde which hey wo k is comple ely di e en . T a- di ional me hods a e no mally based on a Top-down decomposi ion o he so wa e p ojec , while he dynamic me hod can be cha ac e ized by he agg ega ion p ocess i is ocussed on, acco ding o which, some ea u es o a p ojec a e joined oge he unde a simula ion model. In acco dance wi h wha has been said, i can be deduced ha dynamic models a e sui able o deal wi h p oblems placed a he s a egic le el, while adi ional me hods a e use ul a he ope a ional le el o so wa e p ojec s. Wi h he de elopmen o a Dynamic In eg a ed F amewo k o So wa e P o- cess Imp o emen (DIFSPI) we o e a me hodology and wo king en i onmen o join bo h app oaches and o allow p ojec manage s and membe s o he So - wa e Enginee ing Imp o emen G oup (SEIG) o design and e alua e new p ocess imp o emen s. One o he main objec i es o DIFSPI is o suppo he e olu ion o he ma u i y le el o an o ganiza ion acco ding o he Capabili y Ma u i y Model (Paulk e al., 1993). The p ocess o design and de elopmen o bo h he amewo k and he dynamic models ha in eg a e i , allows one o define a me ics collec ion p og am. These me ics a e necessa y o bo h ini ialize and alida e he dynamic models. This me ics collec ion is no only use ul o he dynamic models; i also se es as an in aluable oppo uni y o ob ain a eal knowledge o he s a e o he so wa e p ocesses inside an o ganiza ion. This knowledge is essen ial be o e ack- ling any p ocess imp o emen . 4. DIFSPI de elopmen 4.1. Concep ual app oach Using simula ion o p ocess imp o emen in conjunc ion wi h CMM is no a new idea. As a ma e o ac , (Ch is ie, 1999) sugges s ha CMM is an excellen inc e- men al amewo k o gain expe ience h ough p ocess simula ion. Ne e heless, he e is a lack o a dynamic amewo k capable o assessmen in he achie emen o highe p ocess ma u i y. One o he main ea u es o DIFSPI is ha his assessmen is p o ided no only by using he associa ed final ool, bu du ing he de elopmen o he whole dynamic amewo k. The eason o his is ha he benefi s ha can be ob ained wi h he u iliza ion o dynamic models inside an o ganiza ion, a e di ec ly ela ed o he knowledge and he empi ical in o ma ion he o ganiza ion has abou i s p ocesses. Figu e 1 illus a es his idea. I shows he exis ing causal ela ionships among he ma u i y le el o he o ganiza ion, he u iliza ion o dynamic models and he benefi s ob ained. The posi i e eedback loop comes o illus a e he causal ela ionship ha ein- o ces he me ics collec ion inside he o ganiza ion. The me ics collec ed will be used o calib a e and ini ialize he dynamic models. Lowe ma u i y o ganiza ions a e cha ac e ized by he absence o me ic p og ams and his o ical da abases. In his case, i is necessa y o begin by iden i ying he gene al p ocesses and he in o - ma ion ha has o be collec ed abou hem. The ques ions o wha o collec , a (+) Ma u i y Le el Cos Es ima e P ocess Imp o emen P edic ion Dynamic Model Me ics Collec ion Da a + + + + + + Figu e 1. Causal ela ionships de i ed om he de elopmen and u iliza ion o dynamic models. wha equency, and wi h wha accu acy ha e o be answe ed a his momen . The design p ocess o dynamic models helps o come o a solu ion o hese ques ions. When de eloping a dynamic model i is equi ed o know: a) wha is in ended o be modeled, b) he scope o he model, and c) wha beha io s need o be ana- lyzed. Once he model is de eloped, i needs o be ini ialized wi h a se o ini ial condi ions in o de o execu e he uns and ob ain he simula ed beha io s. These ini ial condi ions cus omize he model o he p ojec and o he o ganiza ion o be simula ed and hey a e e ec i ely implemen ed by a se o ini ial pa ame e s. These pa ame e s ha ule he e olu ion o he model uns answe p ecisely he o me ques ion o wha da a collec : hose da a equi ed o ini ialize and alida e he model will be he main componen s o he me ics collec ion p og am. Once he componen s o he me ics collec ion p og am ha e been defined i can be implemen ed inside he o ganiza ion. This p ocess will lead o he achie emen o a his o ical da abase. The da a ga he ed can hen be used o simula e and empi - ically alida e he dynamic model. When he dynamic model has been alida ed, he esul s o i s uns can be used o gene a e a simula ed da abase; wi h his da abase i is possible o pe o m p ocess imp o emen analyses. An inc ease in he complexi y o he ac ions in ended o be analyzed will di ec ly lead o an inc ease in he com- plexi y o he dynamic model equi ed and, he e o e, o a new me ics collec ion p og am o he new simula ion modules. The bo om hal o Figu e 1 illus a es he e ec s de i ed om he u iliza ion o dynamic models in he con ex o p ocess imp o emen . Using dynamic mod- els which ha e been designed and calib a ed acco ding o an o ganiza ion’s da a p o ides h ee impo an benefi s. Fi s ly, he da a o he simula ion uns can be used o p edic he u u e e olu ion o he p ojec . The g aphical ep esen a ions o hese da a show he e olu ion o he p ojec om a se o ini ial condi ions (which ha e been es ablished by he ini ializa ion pa ame e s). By analyzing hese g aphics, o ganiza ions wi h a low le el o ma u i y can ob ain a use ul quali a- i e knowledge abou he e olu ion o he p ojec . As he ma u i y le el o he o ganiza ion inc eases, he knowledge abou i s p ocesses is also highe and he simula ion uns can be used as eal quan i a i e es ima es. These es ima es help o p edic he u u e e olu ion o he p ojec wi h an accu acy ha is in ima ely ela ed o he unce ain y o he ini ial pa ame e s. Secondly, i becomes possible o define and expe imen wi h di e en p ocess imp o emen s by analyzing he di - e en simula ion uns. This capabili y helps in he decision-making p ocess, as only he imp o emen s which ga e he bes esul s will be implemen ed. Mo eo e , one o he mos ema kable hings he e is ha hese expe imen s a e pe o med wi h no cos and isk o he o ganiza ion as hey use he simula ion o scena ios. Thi dly, he simula ion model can also be used o p edic he cos o he p ojec ; his cos can be e e ed o as he o e all cos , o o a hie a chical decomposi ion o he o al cos , as o ins ance, he cos o quali y o e ision ac i i ies. These h ee benefi s a e he main ac o s ha lead o he achie emen o a highe ma u i y le el inside an o ganiza ion acco ding o CMM. 4.2. DIFSPI s uc u e P ojec managemen is composed o ac i i ies which a e in ima ely in e ela ed in he sense ha a ce ain ac ion pe o med o e a de e mined a ea will possibly a ec o he a eas. Fo ins ance, a ime delay will always a ec he cos o he p ojec bu i may o may no a ec he mo ale o he de elopmen eam, o he quali y o he p oduc . The in e ac ions among he di e en a eas o p ojec managemen a e so s ong ha on some occasions he h oughpu o one o hem can only be achie ed by educing he h oughpu o ano he . A clea example o his beha io can be ound in he equen p ac ice o educing he quali y, o he numbe o equi emen s o be implemen ed in a ce ain e sion o he p oduc wi h he aim o mee ing he ime o cos es ima es. Dynamic models help o unde s and he in eg a ed na u e o p ojec managemen , as hey desc ibe i h ough di e en p ocesses, s uc u es, and main in e ela ion- ships. In he amewo k p oposed he e, p ojec managemen is conside ed as a se o dynamic in e ela ed p ocesses. P ojec s a e composed o p ocesses. Each p ocess is composed o a se ies o ac i i ies designed o he achie emen o an objec i e (Paulk e al., 1993). F om a gene al poin o iew, i could be said ha p ojec s a e composed o p ocesses which all in o one o he ollowing ca ego ies: —Managemen p ocess. This ca ego y collec s all hose p ocesses ela ed o he desc ip ion, o ganiza ion, and con ol o he p ojec . —Enginee ing p ocess. All hose p ocesses ela ed o he specifica ion and de el- opmen ac i i ies o he so wa e p oduc a e collec ed in his ca ego y. Bo h ca ego ies in e ac du ing he ime cycle o he p ojec as Figu e 2 shows. F om an ini ial plan pe o med by he p ojec managemen p ocesses, enginee ing p ocesses begin o be execu ed. Using he in o ma ion ga he ed abou he p og ess o his second g oup o p ocesses, p ojec managemen p ocesses de e mine he modifica ions ha need o be made o he plan in o de o achie e he p ojec objec i es. The DIFSPI p oposed ollows his same classifica ion and is s uc u ed P L A N P R O G R E S S Managemen p ocesses Managemen p ocesses Enginee ing p ocesses Enginee ing p ocesses Figu e 2. Classifica ion o p ocesses o so wa e de elopmen . o a end o p ojec managemen and enginee ing p ocesses. In bo h le els, he u iliza ion o dynamic models o simula e eal p ocesses and o define and de elop a his o ical da abase, will be he main ea u e. 4.2.1. Enginee ing p ocesses in he DIFSPI. On his le el he dynamic models sim- ula e he li e cycle o he so wa e p oduc . The benefi s ha simula ion p o ides a his le el a e he ollowing: — To build a model i is necessa y o imp o e he knowledge one has abou he so wa e de elopmen p ocess, as i is equi ed o es ablish he limi s and scope o hose eal beha io s o be modeled and simula ed. — The pa ame e s equi ed by he model and he ables which de e mine i s ime beha io will cons i u e he main elemen s o a me ics collec ion p og am o define a his o ical da abase. — The e ec i e applica ion o his me ics p og am will eed he da abase. The his o ical da a ga he ed will help assess in he alida ion and calib a ion o he model. — The dynamic model will finally simula e he so wa e p ocesses wi h he knowl- edge and he ma u i y ha he o ganiza ion has a he momen . — The u iliza ion o he dynamic model allows he es ablishmen o a baseline o he p ojec , he in es iga ion o possible imp o emen s, and he de elopmen o a his o ical da abase which can be ed ei he by eal o simula ed da a. The dynamic models o his le el a DIFSPI should ollow he le els o isibili y and knowledge o he enginee ing p ocesses ha o ganiza ions ha e a each ma u- i y le el. I is ob ious ha he complexi y o he dynamic model used in le el 1 o ganiza ions canno be he same as ha o he models capable o simula ing he enginee ing p ocesses o , o ins ance, le el 4 o ganiza ions. 4.2.2. P ojec managemen p ocesses in he DIFSPI. Managemen p ocesses a e di ided in o wo main ca ego ies: —Plan. This g oups he p ocesses de o ed o he design o he ini ial plan and he equi ed modifica ions when he p og ess epo s indica e he appea ance o p oblems. The models o his g oup in eg a e adi ional es ima ion and planning echniques oge he wi h dynamic ones. —Con ol. In his g oup all he models designed o he moni o ing and acking ac i i ies a e ga he ed. These models will also ha e he esponsibili y o de e - mining he co ec i e ac ions o he p ojec plan. The e o e, he simula ion o p ocess imp o emen s will be o eno mous impo ance. 4.3. Elabo a ion o he dynamic models The app oach ollowed in he cons uc ion o he dynamic models is based on wo undamen al p inciples: — The p inciple o ex ensibili y o dynamic models. Acco ding o his p inciple, di e en dynamic modules a e joined o an ini ial and basic dynamic model. This ini ial model models he undamen al beha io o a so wa e p ojec . Each one o he dynamic modules models each one o he key p ocess a eas which con o m he s ep o e ol e o he nex le el o ma u i y. These modules can be ei he “enabled” o “disabled” acco ding o he objec i es o he p ojec manage o he membe s o he SEIG. — The p inciple o agg ega ion/decomposi ion o asks acco ding o he le el o abs ac ion equi ed o he model. Two le els o agg ega ion/decomposi ion a e used: •Ho izon al agg ega ion/decomposi ion acco ding o which di e en sequen ial asks a e agg ega ed in o a unique ask wi h a unique schedule. •Ve ical agg ega ion/decomposi ion acco ding o which di e en and indi id- ual, bu in e ela ed and pa allel asks a e conside ed as a unique ask wi h a unique schedule. The defini ion o he igh le el o agg ega ion and/o decomposi ion o he asks mainly a ec s he modeling o he enginee ing ac i i ies and p incipally depends on he ma u i y le el o he p ocess in ended o be simula ed. To define he ini ial dynamic model he common eedback loops among he so - wa e p ojec s we e aken in o accoun . The objec i e o his app oach was ying o achie e a gene ic model and a oid modeling ce ain beha io s o conc e e o ganiza- ions which migh limi he flexibili y o he DIFSPI. To ini ialize he unc ions and pa ame e s o he ini ial model, da a o igina ing om his o ical da abases collec ed in he a ailable li e a u e we e used (Pu nam, 1992). By eplica ing some o he equa ions o he ini ial model i is possible o model he p og ess o highe ma u i y le els. Figu es 3 and 4 illus a e he o me idea o basic modeling and s uc u e epli- ca ion. The ini ial model can be used o simula e so wa e p ojec s de eloped in o ganiza ions p og essing o le el 2. Figu e 3 uses Sys em Dynamics no a ion o illus a e he componen s de eloped o model he so wa e de elopmen ac i i y. The numbe o asks o be de eloped is de e mined om an ini ial es ima e o e ision a e INITIAL SIZE de elopmen a e quali y QUALITY Pendin g asks Accomplished asks Re ision pending asks Figu e 3. Basic dynamic module o so wa e de elopmen modeling. he size o he p ojec . These pending asks become accomplished asks acco ding o he de elopmen a e. Du ing his p ocess, e o s can be commi ed. Thus, in acco dance o he desi ed quali y objec i e o he p ojec , he quali y a e and he e ision a e a e de e mined. These wo a es go e n he numbe o asks ha a e e ised. To model he p og ess o le el 3, he model will make use o a ho izon al decomposi ion, c ea ing as many subs uc u es as phases o ac i i ies a e p esen in he ask b eakdown s uc u e o he p ojec (analysis, design, code and es , in ou case). Acco ding o his app oach, each ime a comple e model o some pa o i is eplica ed, i will be necessa y o define he new fixing mechanisms (dynamic mod- ules) o he new s uc u es. These mechanisms e ec i ely implemen he p inciple o agg ega ion/decomposi ion p e iously men ioned. The eplica ion o s uc u es also p o ides he possibili y o eplica ing he modules ela ed o he p ojec man- agemen p ocesses. This eplica ion is especially use ul o high ma u i y le el o ga- niza ions, which will be able o es ablish p ocess imp o emen p ac ices o each ce ain ac i i y o he li e cycle. In Figu e 4, he componen s o he dynamic model o le el 3 o ganiza ions a e shown. Each one o he ou boxes labelled wi h he name o one phase o he p ojec is, in ac , a comple e dynamic module iden ical o ha shown in Figu e 3. The new ea u es added o he eplica ed s uc u e define he coupling s uc u e Figu e 4. Replica ion o he basic dynamic module o model highe ma u i y p ocesses.