Web Based Versus Mobile Agent Based Software Retrieval Systems: Performance Comparison

JoseMerseguer,JavierCampos,andEduardoMena

Dpto.deInformaticaeIngenieradeSistemas,UniversityofZaragoza,Spain

fjmerse,jcampos,emenag@posta .uniz ar.e s

Abstract. Formalmethodsarethemostsuitablewaytomodelandan-

alyzeseveralkindsofsoftwaresystems.However,conventionalmethods

aregainingplacedinSoftwareEngineeringeldbecausetheycanbeeas-

ilyappliedinall thestages ofthe software lifecycle. Thecombination

offormalandconventionalmethodscouldbeaninterestingapproachto

describeand analyzesomesoftwareaspects,asforexample, theperfor-

manceofthesystem.Inthisway,wemakeuseofasoftwareperformance

processenrichedwithformaltechniques.Theprocess hasas important

features that it usesUML as a design notation and it uses stochastic

Petrinetsasformalmodel.Petrinetsprovideaformalsemanticsforthe

system and a performance model, while UML supplies the framework

andtoolstodocumentthesystem.

Theprocessisusedinthisarticletopresentaperformancecomparison

betweensoftwareretrievalsystems.Nowadays,thereexistwebsitesthat

allowusersto retrieve andinstallsoftware inaneasyway.The perfor-

manceofthesesitesmaybepooriftheyareusedinwirelessnetworks;

thereasonis theinadequateuseofthenet resourcestheyneed.Inthis

article,we show that ifthis kindofsystemsare designed usingmobile

agenttechnologythepreviousproblemmightbeavoided.

Keywords: Softwareperformance engineering,stochastic Petri nets,Inter-

net,UML,wirelessnetworks,mobileagenttechnology.

1 Introduction

ThetasksofretrievingandinstallingsoftwareusingInternethavebecomecom-

monin thelast years. Therearesites (e.g., Tucows.com[3],Download.com[1],

and GameCenter.com [2])which permit to performthese tasks in aneasy and

friendlyway.Buttheprocessofselectingthesoftwarecouldbecomecostlyand

sometimes slow: the user must select a great number of categories until s/he

ndsthedesiredsoftware,speciallyifs/he isanaiveuser.Everytimetheuser

selects acategory,the website sends anew HTML page with the contents of

thecategory.Thistechniquemakesanintensiveuseofthenetworkconnection,

whichcanresult in apoor performance andexpensivecommunicationcost (in

awireless environment).Thus, itwouldbeinterestingto get resultsaboutthe

performanceofthiskindof systemstoexactlyknowhowcostlyisthisprocess.

?

ThisworkhasbeendevelopedwithintheprojectUZ00-TEC-03oftheUniversityof

A new technology, mobile agents [14], can aid to reduce the network con-

nection time. We recall in this papera software retrieval service belonging to

the ANTARCTICA system [9] which has been designed using mobile agents.

Theaimofthisserviceisto proposeanalternativemethod tothecurrentweb-

based software retrieval systems, called in this article \Tucows-like" systems.

TheANTARCTICAsystemhasbeenthoughtobeusedinwirelesscommunica-

tionenvironventswherenetspeedisaproblem,say800bytes/sec.isrealinGSM

networks.Therefore,itpromotesabetteruseofthenetresources,thussupply-

ingbetterperformanceresults.Inthefollowing,werefertotheANTARCTICA

softwareretrievalserviceasthe\ANTARCTICASRS".

Software performance engineering [16] proposes evaluating performance of

software systemsin the early stages of the developmentprocess. Thus, if per-

formance problems are detected, it will be easier and less expensive to take

theappropriatedesigndecisionstosolvethem.TheUniedModellingLanguage

(UML) [4] is the design notation that we will use to model software retrieval

systems.TheselectionofUMLismotivatedbecauseitiswidelyacceptedbythe

softwareengineering communityand it ts verywellin thesoftwarelife cycle.

Sinceweareinterestedinperformanceaspects,weuseUMLextendedwithper-

formance annotations. In general, software systems are complex systems even

more if they are distributed, so we advocate for the use of formal models to

obtain performance indices. But UML lacksof the necessaryformal semantics

to apply analyticaltechniques.Thus, weproposeto semi-automaticallyobtain

Petri nets (PNs) [13] from UML diagrams and to use them to obtain perfor-

manceindices.PNsareawidelyused formalmodelforconcurrentsystemsand

providedwithastochastictimeinterpretation,theyaresuitableforperformance

evaluation.Concretely,weusestochasticwell-formedcolourednets(SWNs)[5].

Ourgoalin thisworkistocomparetheperformanceofTucows-likesystems

withtheperformanceoftheANTARCTICASRSusinganalyticalsoftwareper-

formance techniques. The performance index under study for both systems is

thenetworktime,i.e.,howmuchtimethenetworkconnectionneedstobeopen

inordertoretrieveasoftwareproduct.Also,itwillbestudiedtheimpactofthe

intelligentagentsintheANTARCTICASRS.

Therest of the paperis organizedas follows. Insection 2, the process fol-

lowedtoevaluatesoftwareperformanceisdescribed.Insection3,aTucows-like

software retrieval system is modeled using UML and this model is translated

into SWNs. In section 4, the main characteristics of the ANTARCTICA SRS

are described. Insection 5, wepresentacomparison between performance g-

ures for both approaches. Finally, in section 6, some concluding remarks and

relatedwork arestressed.

2 The software performance process

\Thesoftwareperformanceengineering(SPE)isamethodforconstructingsoft-

ware systems to meet performance objectives. SPE augments others software

{time}

Do:examine

for download

{1K}

{1K}

{1K}

{1K}

wait

select_category(url)

^webServer.select_URL(url)

{20K..30K}

{1K}

{1K}

{1K}

{1K}

reply {file_size}

get(html_page)

^user.observe(html_page)

^user.succ_install

^webServer.download(url)

[^user.satisfied]select_sw(url) {1K}

[not_satisfied]

download(url)

^Browser.reply(file) WAIT

{1sg.} {1sg.}

{20K..30K}

{1K}

select_URL(url)

(d) (a)

(c) (b)

^Browser.get(html_page) Do:find_html_page Do:find_file

{1K}

{file_size}

wait for

wait

{1K}

[satisfied]^Browser.select_sw(url)

^Browser.select_category(url)

succ_install() html page

observe(html_page)

{1K}

WebServer

{20K..30K}

[satisfied]

WAIT {0.1}

get(html_page)

download(url)

succ_install() {1K}

{file_size}

{1K}

{1K}

{1K}

select_sw(url) {20K..30K}

observe(html_page)

{1K}

select_URL(url) select_category(url)

Browser

1..n

Fig.1.(a)AnnotatedsequencediagramfortheTucows-likesystem.AnnotatedSTDs

fortheTucows-likesystem:(b)user, (c)thewebserver,and(d)browser

brie y presenta software performance process, introduced in [11,10], that will

be applied in sections 3 and 4 to the software retrieval systems proposed in

the introduction. An interestingcharacteristic of this process is that it uses a

formal model (SWNs) to obtain performance indices. This formal model can

beobtainedsemi-automaticallyinsidethesoftwaredevelopmentprocess.Inthis

way, withoutmuch eort,theprocessallowstoobtainaperformance modelas

aby-productanditpreservesthebenetsofthesoftwaredesignmethodologies.

2.1 Modelingusing performance annotated UMLdiagrams

Theprocessbeginsbymodelingthesystemdynamicsinaconventionalway.The

UMLdynamicdiagrams[4]willguidetheprocess.Sequencediagrams andstate

transitiondiagrams(STDs)willbeusedinthispaper.Activitydiagrams arenot

used because we are not interested in modeling the actions performed by the

systemsthat wemodel.

Figure 1.a depictsan example of sequence diagram,that will be explained

lateron.Asequencediagramrepresentsmessagessentamongobjects.Messages

inthescopeofthemodeledsystem(forinstance,themessagessentbetweenthe

userand the browser in Figure 1.a). Messagessent among objectson dierent

computers,thosewhichtravelthroughthenet,willconsumetimeasafunction

ofthe messagesize andthenet speed(forinstance, in Figure1.athemessages

sentbetweenthebrowserand thewebserver). Each messagesize isannotated

inside braces. For instance, the selectURL message is labelled with f1 Kbyteg

in Figure 1.a. Itis also possibleto annotate thesize with arangein the UML

commonway(likethe getmessagewithlabelf20K..30Kgin Figure1.a). Ifthe

messagesize is unknown, theannotationis alabel representingaperformance

parameter(e.g., thereturnofthedownloadmessageisannotatedwiththelabel

flesizeg,alsoin Figure1.a).

Inasequencediagram,conditionsrepresentthepossibilityfortheassociated

messageto be dispatched. Annotations,also betweenbraces, expresstheevent

probability success associated with each condition (for instance, see probabil-

ity f0.1g associated with the condition [satised]in Figure 1.a). A range or a

probability(unknown)parameterareacceptedtooaseventprobabilitysuccess.

Inordertogetacompleteviewofthesystemdynamics,aSTDforeachclass

withrelevantdynamicbehaviourmustbedeveloped,asthosedepictedinFigures

1.b, 1.c and 1.d. To study performance aspects on a STD, two elements are

meaningful,theactivities andtheguards.Inthefollowing,webrie ycomment

theannotationsusedin thesekindofdiagrams.

Activitiesrepresenttasksperformedbytheobject,therefore,theyconsume

computationtimethatmustbeannotated.Theannotationwillbedonebetween

braces.Forexample,seeinFigure1.ctheboldlabelf1sec.gclosetotheactivity

ndle.Ifitisnecessary,minimumandmaximumvalueswill beannotated.

Guardsshowconditionsin atransition.Theymusthold inorder torethe

eventthat theylabel.Theprobabilityassociatedtothem,alreadyannotatedin

the sequence diagrams,are also indicated in STDs to gain readability. In the

sameway,thesize ofthemessagesmaybeannotatedoromitted.

2.2 Fromperformance annotated UMLdiagrams to SWNs

FromtheperformanceannotatedUMLdiagramsitwillbeinterestingtoobtain

performanceindicesforthesystem.But,aswesaid,UMLlacksofthenecessary

formalsemanticstoapplyquantitativeanalysistechniquestoobtainthem.Even

more,itisnotsuitabletospecifysomesystemaspects,forexampleconcurrency,

whichisfundamentalinperformanceevaluation.Therefore,theUML diagrams

willbetranslatedintoSWNs,whichallowtoobtainperformanceindices,taking

theproperdecisionsfortheunspeciedsystemrequirements.

The strategy to obtain the SWNs from the performance annotated UML

diagramsisasfollows.

1. DerivationofaSWNfromeachSTD.TheseSWNswillbecalledcomponent

nets.TheyrepresentthebehaviourforeachclasswiththeunderlyingSWN

formalsemantics.ToobtainacomponentnetfromaSTDseveralrulesmust

{ Each stateof theSTD isrepresented byaplace, with the samename,

intheSWN.

{ Foreachtransitionin theSTD,there willbeintheSWN:

(a) A transitionwith thesameastheeventwhichlabelsthe transition

in theSTD.

(b) Anarcfromtheplace(whichrepresentstheinitialstateintheSTD)

to thetransitionin theSWN(which representthetransitionin the

STD).

(c) An arcfromthetransitionintheSWN(whichrepresentthetransi-

tionintheSTD)totheplace(whichrepresentsthenalstateinthe

STD).

{ Guardsin theSTD will becomeimmediate transitionswith theassoci-

atedcorrespondingprobabilitiesfortheresolutionof con icts.

{ Activities inside a state of a STD are considering as time consuming,

sointheSWNtheywill be consideredastimedtransitions; therateof

the exponentiallydistributed service time of the timed transitions are

obtainedautomaticallyfrom thetimeannotationoftheSTD.

Asanexample,seeinFigure2thecomponentnetsobtainedfromtheSTDs

inFigure1.

2. ObtainingaSWNfor the system.Fromthecomponentnets,andguidedby

thesequencediagram/s,acomplete SWNforthesystemisobtained.Tran-

sitionsinthecomponentnetsthatrepresentthesamemessagearesynchro-

nizedifthe messagehaswait semantics;ontheother hand,ifthemessage

has no wait semantics the transitions are connected with an extra place,

modelingacommunicationbuer.

TheoutcomeSWNmodelsthebehaviourofthewholesystem.Figure3depicts

an example of a complete net obtained from the component nets in Figure 2

and from thesequence diagram in Figure 1.a. Theperformanceguresfor the

systemareobtainedbyanalyzingthecomplete SWNforthesystem.

3 Modeling the Tucows-like software retrieval system

There are avarietyof softwareretrievalsystems, asthe popularweb sites Tu-

cows.com [3], Download.com [1] or Gamecenter.com [2], that provide Internet

users withfacilities to retrieveand install software. These systems allowusers

to nd software in twodierentways,byusing akeyword-basedsearchengine

orbynavigatingthroughcategoriesespeciallydesignedtomakethetaskeasier.

Thesoftwarearchitectureofthesekindofsystemsforthenavigationfacility

isbasicallythesame,therefore,itispossibletomodelhowthesekindofsystems

work, making a number of assumptions, without loosing reality with respect

to performance aspects. We will refer to these kind of systems as Tucows-

like systems. Our intend is to evaluate performance indices for a Tucows-like

system in order to compare them with those obtained forthe ANTARCTICA

Thekeyword-basedsearchengineoershelp tothose usersthatknowsome

features of the wanted software. This search facility will not be considered in

this paper since it does not exist its counterpart in the ANTARCTICA SRS,

thereforenocomparisoncanbemade.Thenavigationfacilityconsistsofseveral

webpagesresidinginaserverandorganizedascategorieslinkedbetweenthem

in away that guides theuser to nd the software. For instance, a number of

thesesystemspresentaninitialwebpagewhere thecategoriescorrespondwith

dierentoperatingsystems,sayWindows2000,Windows95/98,LinuxorUnix

Themes.TheANTARCTICASRSoersamechanismtoretrievesoftwaresimilar

to thenavigationfacility, but itmakesuseof intelligentagentsto performthe

task,thereforeaperformancecomparisoncanbemadebetweenthetwosystems.

Inthissection,weapplythesoftwareperformanceprocesstothenavigation

facilityoftheTucows-likesystems.

3.1 Systemdescription and modelingassumptions

InaTucows-likesystem,theuser navigates,withthehelpofabrowser,through

dierentHTMLpages(representingsoftwarecategoriesanddescriptionsofcon-

cretepiecesofsoftware)untils/hendsapieceofsoftwarethatsatisesher/his

needs.Then,that pieceofsoftwareisdownloaded.

Itmustbeassumedthat theuserspendssometimereading theinformation

presented by the system. An exponentially distributed random variable with

parameter

examine (

examine

isobtainedastheinverseofthetimein seconds)

willbeusedtomodelseveralkindsofusers.

The number of HTML pages that the user must navigate until s/he nds

the software is diÆcult to estimate (it depends on her/his experience). The

probability that the user nds the software by selecting n categories models

dierentkindsofusers,fromnaiveusers,thosewhoneedtovisitmanycategories

to nd thesoftware, to expert users, those whond thesoftwarevisiting very

fewcategories.

WhenevertheuserrequestsanHTMLpage oraconcretepiece ofsoftware,

thewebservermustperformthecorrespondingactivitiestondthepageorthe

piece.The timeconsumed bythese activities will be modeled asexponentially

distributed randomvariableswithparameters

findHTML and

findFile .

Thebrowser, in the clientmachine, sendsmessages throughthe netto the

web server in the servermachine and vice versa. A exponentially distributed

randomvariablewithparameter

m

i

willmodelthetimespentbythemessagei

navigatingthroughthenet.Noticethatthemessagessentbetweentheuserand

thebrowserdonotconsumenetresources.

3.2 UML diagramswith performanceannotations

In this section we model the dynamic view of the Tucows-like system using

UML notation as proposed in [11]. Thesequence diagram in Figure 1.a shows

themessagessentamongtheobjectsinthesystemwiththepurposetoretrieve

distinguished,thosethat travelthroughthenet(sentbetweenthebrowserand

thewebserver)andthosethatdonot(sentbetweentheuserandthebrowser).

This feature will be relevant in the SWN model in order to associatetime to

transitions that represent messagessend through the net, taking into account

theassumptionsmadeintheprevioussection.

Thesequence diagram in Figure 1.a beginswith a selectcategory(url) mes-

sage, its size is f1 Kbyteg, sent by the user to the browser. It represents the

\click" performed by theuser in thebrowser to select a categoryin a HTML

page.Therestofthediagram describesin thesamewaythestepsexplainedin

theprevioussectionforselectingsoftware.

InordertogetacompletedescriptionoftheTucows-likesystemdynamicsand

itsload,weare goingtodeveloptheSTD foreachclass withrelevantdynamic

behaviour.

User state transition diagram. In Figure 1.b, the behaviour of a user is

represented.Theuserisinthewaitstateuntils/heactivatestheselectcategory

event. This event sets the user in the wait for HTML page state. The observe

event,sentbythebrowser,allowstheusertoperformtheexamineactivitythat

hasassociatedthelabelftimeg.Thislabelmodelsthetimethattheuserspends

readingtheHTMLpage.This activitywillbetranslatedin theSWNnetin an

exponentially distributed transition, and by modifying its rate dierent kinds

of users can be modeled, as wepointed outin the previous section. Once the

activityisperformedtwosituationscanarise:

{ IftherequestedsoftwareisnotpresentinthecurrentHTMLpagetheuser

returnstothewaitstate.

{ Inothercase,theusersendstheselectsw(url)messagetothebrowser,where

url means the direction where the software is located in the server, and

entersinthewaitfordownloadstate.Whenthebrowserfulllsthenecessary

activitiesto complete the download, it sends to the user the succ install()

messageandtheuserreturnstothewaitstate.

Browserstatetransitiondiagram.Figure1.dshowsthebrowser'sSTD.The

browserbehavesas aserverobject:itiswaiting foruser'srequests,represented

byselectcategoryandselectswevents.

Whenaselectcategory eventarrivesrequesting aurl,the browser sendsto

thewebservertheselectURLmessageandwaitsforanewHTMLpage.When

thewebserverobtainsresults,ittriggersthegeteventattachingthenewHTML

page,whoseestimatedsizeisf20K..30Kg.Sincethismessageissentthroughthe

net,itwillbetranslatedin theSWNasanexponentiallydistributed transition

withrate

mg et

,aswepointedoutintheprevioussection.Afterthat,theHTML

pageisshownto theuser.

When a selectsw event arrives requesting a url that contains a piece of

softwareadownloadmessagewiththeurl issenttothewebserver.Thebrowser

waits forthereply messagethat containstherequestedlewith sizele size,it

willbetranslatedin theSWNin atransitionwithrate

m

r eply

.Finally,thele

isinstalled(succinstall).

wait S R

P3 R

P4 R P5 R

examine

observe not_satisfied

Browser.select_category

Browser.select_sw succ_install

<x>

<x>

<x> <x>

<x>

<x>

<x>

<x>

<x>

<x> <x> <x>

R:c request:c S:m

(a)

Fig.2. Component PNs for the Tucows-like system: (a) user, (b)browser, (c) web

server.

Web server state transition diagram. As thebrowser, thewebserverbe-

havesas aserverobject.It is waiting fora request (select URLand download)

from thebrowser. For eachrequest, thewebserverperformsthecorresponding

actionstoserveit(nd htmlpageandndle).Whentheactionsarecompleted,

it sends the corresponding message to the browser. Figure 1.c showsthe web

server'sstatetransitiondiagram.

3.3 Modelingthe system with SWNs

Inthissection, wedetailtheSWNmodelfortheTucows-likesystem.Thenets

havebeenobtainedbyapplyingthetranslationrules,givenin[11]andschemat-

icallyshown in section 2.In order to model situations that the Petri netscan

express but the UML notationcannot, the appropriatedecisionswill betaken

andcommented.

Figure2representsthecomponent nets,those obtainedfrom theSTDs, for

theTucows-likesystemandFigure3representsthenetforthewholesystem,ob-

tainedbysynchronizingthecomponentnets.Westartdescribingthecomponent

nets.

User component net. The number of tokens in the place wait models how

manyconcurrentuserssupportsthesystem.Thisparametercannotbemodeled

intheUMLdiagrams.

The ring of the transition named Browser.selectcategory models the dis-

patchto thebrowserofamessageto specializethecurrentHTMLpage,it will

P4 R

wait_WebServer P2

R P3 R

P5 R wait_Browser

S R

P10 R

buffer_download R P12 R

P13 R

P15 R wait_user

S R

P17 R

uffer_select_sw R

buffer_get R

P25 R

P28 buffer_reply R R P26 R

download_browser

select_URL get get_browser

examine

find_file find_html_page

reply select_category

select_sw_user not_satisfied

observe

Browser.reply download_webServer

select_sw_browser

succ_install

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x> <x> <x>

<x>

<x> <x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x> <x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

<x>

R:c request:c S:m

Fig.3.CompleteSWNfortheTucows-likesystem.

examine models the time spent by the user reading the information presented

in the new HTML page. After theend of the reading,a choice will determine

whether the useris satisedwith any ofthe productsshown (ringof theim-

mediate transitionBrowser.selectsw ),ornot(ringoftheimmediatetransition

notsatised).

Thering oftheimmediatetransition namedsuccinstallmodelsthearrival

of amessageto conrm that theretrieveof thesoftware hasbeensuccessfully

completed.

Browser componentnet.Thenumberoftokensintheplacewaitmodelshow

manyconcurrentbrowseraccesstothesystem.Thecolouristhesameasinthe

usercomponentnettoidentifyeachbrowserwithauser.Thisparametercannot

bemodeledintheUMLdiagrams.

Theringofthetransitionnamedselectcategorymodelsthearrivalofmes-

sages from the user requesting for a specialization of the category that s/he

hasexamined,thetransition isimmediatebecauseboth,thesenderandthere-

ceiver,arein theclientmachine.Therequestissenttothewebserverthrough

the net, therefore consuming time by ring the timed transition named web-

Server.select URL.The ringof transitionsgetand user.observemodels,respec-

tively,theobtainingoftheHTMLpagewithnewcategoriesanditsdispatchto

theuser.

Thering ofthe transition named selectsw modelsthe arrivalofmessages

from the user requesting a concrete piece of software. The request is sent to

the web server through the net by ring the timed transition named web-

Server.download.Thering ofthe timedtransition named replymodels theob-

taining of the le requested by previous transition. Finally, the ring of the

transitionnameduser.succ installmodelstheadvertisementtotheuserthatthe

Theselectcategoryandselectswtransitionswillbesynchronizedinthecom-

pletenetwiththetransitionsin theusercomponentnetwiththesamename.

Web server componentnet.Thenumberoftokensin theplace waitmodels

howmanyconcurrentprocesseshasopenedthewebserverto attendbrowser's

requests.ThisparametercouldnotbemodeledintheUMLdiagrams.

TheringofthetimedtransitionnamedselectURLmodelsthearrivalofare-

motemessagetorequestforanewHTMLpage.Theringofthetimedtransition

nameddownloadmodelsthearrivalofaremotemessagetorequestforaconcrete

pieceofsoftware.Theringofthetimedtransitionnamedndhtmlpagemod-

elsthecompletionofthesearchforanewHTMLpage.Theringof thetimed

transition named ndle models the completion of the search for a requested

piece of software. Thering of thetimed transition namedBrowser.getmodels

thedispatch oftheHTMLpage tothebrowser. Finally, theringofthe timed

transitionnamedBrowser.replymodelsthedispatchoftheletothebrowser.

TheselectURL, download, Browser.get and Browser.reply transitionswill be

synchronizedinthecompletenetwiththetransitionsinthebrowsercomponent

netwiththesamename.

As we said before, the net for the whole system, depicted in Figure 3, is

obtainedbyapplyingtherulesgivenin[11].

4 The ANTARCTICA Software Retrieval Service

ThegoaloftheANTARCTICASRS[9]istoprovidemobilecomputeruserswith

aservicetoselectanddownloadsoftwarein aneasyandeÆcient way.EÆcient

becausethesystemoptimizesbatteryconsumptionandwirelesscommunication

costs.Itprovidesseveralinterestingfeatures:

{ Thesystemmanagestheknowledgeneededtoretrievesoftwarewithoutuser

intervention,usinganontology.

{ Thelocationandaccessmethod toremotesoftwareis transparenttousers.

{ Thereisa\catalog"browsingfeature tohelpuserinsoftwareselection.

{ Thesystem maintains up to date the information related to the available

software.

Thesoftwareperformanceprocessgivenin[11]wasappliedtotheANTARC-

TICASRSin[12,10].TheUMLdiagrams,theSWNscomponentsandtheSWN

net for the system were obtained. The SWN net for the system will be used

in thenextsectionto obtainperformanceresults.Itwill beinterestingtocom-

pare the performance of theANTARCTICA SRSwith theperformanceof the

Tucows-likesystem, in orderto obtainconclusions aboutthe impactof mobile

agenttechnologyin awirelessnetwork(lowspeed,costly,disconnections).

5 Performance results

TheresultspresentedinthissectionhavebeenobtainedfromthecompleteSWNs

which model the Tucows-like system (see Figure 3) and the ANTARCTICA

to be connected to the net, network time, in the presence of a user request.

Also, it is interesting to know how much intelligent the browser agent in the

ANTARCTICA SRS must be, to obtain the same or better results than the

Tucows-likesystem.

InordertoobtainthenetworktimeintheTucows-likesystem,thethrough-

putofthesuccinstalltransitionwillbecalculatedbycomputingthesteady-state

distributionof the isomorphic Continuous Time Markov Chain (CTMC) with

GreatSPN [6].Theinverseofthepreviousresultgivesthenetworktime.Inthe

ANTARCTICAsoftwaresystemthetargettransitionisselectswservice.

Tostudythenetworktime,wehavedevelopedatesttakingintoaccountthe

followingscenarios:

1. Totest theuser renement request, wehave consideredsix dierent possi-

bilities. A user requestinga mean of5, 10, 20, 30, 40 and 50 renements 1

(modelingdierentexpertiseoftheuser).

2. Twodierentkindsofusershavebeenconsidered:auserwhospends10sec.

tostudy the information presentedby thesystem (webpage orasoftware

catalog)andauserwhospends60sec.inthattask(modelingtheinformation

processingspeedoftheuser).

3. We have considered two cases for the net speed: 1 Kbytes/sec. and 5

Kbytes/sec.Byconsideringthese lowspeedvalueswewanttocomparethe

performance ofbothapproachesin awireless computingenvironment(real

GSMnetworkspeedisaround800bytes/sec).

FortheANTARCTICASRS,wehavealsoconsidered:

1. Abrowserwhichdoesnotneedtoaskforinformationtothesoftwareman-

agerthe 70%of the times that the user asks for a renement. When the

browser needs information, it requests the information by a remote proce-

durecall (RPC).

2. Thesizeof thecatalog obtainedbythebrowseris50Kbytes.

Figure 4 shows network time (in minutes) forthe Tucows-like systemand the

ANTARCTICASRSin dierentscenarios.Concretelyin Figure4.bwecanob-

servethatwhenthenetspeedis1kbyte/sec.,theuserisnaiveandperforms50

renements (the worst case),then theANTARCTICA SRS is almost thirteen

seconds faster than the Tucows-like system. The same results are obtained if

theuser isexpert,see Figure4.a. However,when thenetspeedisincreasedto

5kbyte/sec.(seeFigure 4.d),thedierencesdecrease,and ifthe userperforms

morethanthirtyrenementstheANTARCTICASRSbehavesworse.Inconclu-

sion, we cansaythat theANTARCTICA approach behavesmuch better than

a Tucows-like system for low network speed. Dierences between the two ap-

proachesbecomelesssignicantforahighernetworkspeed.Takingthisanalysis

asbasiswecouldestimatewhich approachisbetterforagivensituation.

About the intelligenceof the ANTARCTICA SRS browser agent,Figure 5

give us interesting results. This gure shows the same scenarios than Figure

4.a and 4.b, but varying the intelligence of the ANTARCTICA SRS browser,

1

Wemeanbyrenementa\click" inaTucows-likesystemandacatalogrenement

(a) (b)

(c) (d)

0 5 10 15 20 25 30 35 40

refinements

mi nutes

TUCOWS 4,28779693 7,891414141 15,09661836 22,16312057 29,55082742 37,28560776 ANTARCTICA 5,062778453 7,208765859 11,49425287 15,69365976 20,08032129 24,72799209

5 10 20 30 40 50 0

10 20 30 40 50 60 70 80 90

refinements

mi nutes

TUCOWS 9,331840239 17,1998624 32,93807642 48,30917874 64,35006435 81,30081301 ANTARCTICA 10,10713564 16,51800463 29,29115407 41,8760469 55,00550055 68,87052342

5 10 20 30 40 50

mi nutes

0 2 4 6 8 10 12 14 16 18

refinements

TUCOWS 1,782912566 3,242542153 6,191183754 9,082652134 12,10360687 16,35590448 ANTARCTICA 1,868041545 3,06710833 5,462689828 7,813720894 10,26904909 12,85016705

5 10 20 30 40 50 0

10 20 30 40 50 60 70

refinements

minutes

TUCOWS 6,811061163 12,5502008 24,01536984 35,23608175 49,75124378 56,88282139 ANTARCTICA 8,304268394 12,37317496 23,24500232 39,40110323 52,24660397 59,31198102

5 10 20 30 40 50

Fig.4. Network time for dierent scenarios: (a) and (b)represent a net speed of 1

Kbyte/sec,(c)and (d)representanetspeedof5Kbyte/sec.,(a)and(c)representa

\userdelay"of10sec.,(b)and(d)representa\userdelay"of60sec.Theintelligence

oftheANTARCTICA'sbrowserhasbeensetto70%.

(a) (b)

Fig.5.(a)and(b)representthesamescenariosthanFigures4.aand4.b,respectively,

butvaryingtheintelligenceoftheANTARCTICASRSbrowser.

from a browser that needs to ask for information the 100% of the times to

a browser that needs to ask for information the 0% of the times. When the

intelligenceofthebrowserislessthan40(itdoesnotneedtoaskforinformation

the40%ofthetimes)theANTARCTICASRSbehavesworsethantheTucows-

likesystem.However,whentheintelligenceofthebrowserdoesnotneedtoask

forinformationthe40%ofthetimesormore,thenANTARCTICASRSobtains

similaror betterresultsthanaTucows-likesystem.

6 Conclusions and Related work

In this paperwehave compared the performance between a classicalsoftware

retrieval system (the so-called Tucows-like system) and another one proposed

using mobileagents(the ANTARCTICA SRS). Thecomparisonhas been per-

formedbyapplyingtoeachofthemasoftwareperformanceevaluationprocess,

whichhas as amajor advantagethat it isintegrated in theearly stagesof the

softwarelifecycle.

Wewouldliketostressthefollowingpoints:

{ ThecombinationofaUMLperformanceextensionandSWNsisexpressive

enoughtomodelcomplexdistributed softwaresystemseventakingintoac-

countdierenttechnologies.Itmustberemarkedthataperformanceformal

model (SWN) can be obtained semi-automatically from the UML perfor-

manceannotateddiagramsinthecontextofthesoftwarelife cycle.

{ Dierentscenarioscanbetestedeasilybyusingthisprocesswithoutinvest-

ingtimein implementingprototypes.

{ As a result of our tests, we can aÆrm that the ANTARCTICA SRS be-

havesbetter than Tucows-likesystem when thenet speed is slow. So, the

ANTARCTICASRSisappropriateforwirelessenvironments.

{ Asafuture work,weplan togivesemanticsto theUML diagramsinterms

ofPetrinets.

Concerning related work, in [8] stochastic Petri nets are obtained from UML

diagramswithperformanceevaluationpurpose.Inouropinion,itisnotclearly

stated the translation process, and it is not clear how the performance model

is obtained. Therefore, it has not been possible to test their techniquein our

examples to compare their solution with ours. In [17] an approach to perfor-

mance prediction for the real time eld is presented. It uses ROOM [15] as a

developmentmethodandalayeredqueuingnetwork asaperformancemodel,a

prototypetoolhasbeendevelopedto assisttheprocess.A parallelwork toour

softwareperformance processusing UML and SWNsis beingdeveloped in [7].

However,weclaimthatforthecaseofparallelanddistributedsystems,asthose

presentedinthispaper,Petrinetsimprovestheadequacyofqueuenetwork,since

theyprovidegeneralsynchronizationmechanismsandvalidation techniques for

logicalproperties.

Acknowledgements:WewouldliketothankLauraRecaldeforallthehelpful

References

1. CNET Inc.,1999. http://www.download.com.

2. CNET Inc.,1999. http://www.gamecenter.com.

3. Tucows.cominc.,1999.http://www.tucows.com.

4. G.Booch,I.Jacobson,andJ.Rumbaugh,OMGUniedModelingLanguagespec-

ication,June1999,version1.3.

5. G.Chiola,C.Dutheillet,G.Franceschinis,andS.Haddad,Stochasticwell-formed

coloured nets for symmetric modelling applications, IEEETransactions onCom-

puters42(1993),no.11,1343{1360.

6. G.Chiola,G.Franceschinis,R.Gaeta,andM.Ribaudo,GreatSPN1.7:GRaphical

EditorandAnalyzerforTimedandStochasticPetriNets,PerformanceEvaluation

24(1995),47{68.

7. V. Cortellesaand R.Mirandola, Deriving a queueing networkbased performance

model from UML diagrams, Proceedings of the Second International Workshop

onSoftware andPerformance (WOSP2000)(Ottawa,Canada),ACM,September

2000,pp.58{69.

8. P.KingandR.Pooley,UsingUMLtoderivestochasticPetrinetsmodels,Proceed-

ingsoftheFifteenthAnnual UKPerformance EngineeringWorkshop(J.Bradley

andN.Davies,eds.),DepartmentofComputerScience,UniversityofBristol,July

1999,pp.45{56.

9. E. Mena, A. Illarramendi, and A. Go~ni, A software retrieval service based on

knowledge-driven agents, Cooperative Information Systems CoopIS'2000 (Eliat,

Israel),OpherEtzion,PeterScheuermanneditors.LectureNotesinComputerSci-

ence,(LNCS)Vol.1901,Springer,September2000,pp.174{185.

10. J.Merseguer,J.Campos,andE.Mena,Evaluatingperformance onmobileagents

software design, Actas de las VIII Jornadas de Concurrencia (Cuenca, Spain)

(DiegoCazorla,ed.),UniversidaddeCastilla-laMancha,June2000,pp.291{307.

11. J Merseguer,J Campos,and E. Mena, Performance evaluation for thedesign of

agent-basedsystems:A Petrinetapproach,ProceedingsoftheWorkshoponSoft-

wareEngineeringandPetriNets,withinthe21stInternationalConferenceonAp-

plicationandTheory ofPetriNets (Aarhus,Denmark)(MauroPezzeandSolM.

Shatz,eds.),UniversityofAarhus,June2000, pp.1{20.

12. J. Merseguer, J. Campos, and E. Mena, A performance engineering case study:

Softwareretrieval system,PerformanceEngineering.StateoftheArtandCurrent

Trends (R. Dumke, C. Rautenstrauch, A. Schmietendorf, and A. Scholz, eds.),

LectureNotesinComputerScience,Springer-Verlag,Heidelberg,2001,Toappear.

13. T.Murata, Petri nets: Properties, analysis, andapplications,Proceedings of the

IEEE77(1989),no.4,541{580.

14. E.PitouraandG.Samaras,Datamanagementformobilecomputing,KluwerAca-

demicPublishers,1998.

15. B. Selic, G.Guleckson, and P.Ward, Real-time object-oriented modeling, Wiley,

1994.

16. C.U.Smith,Performance engineeringofsoftwaresystems,TheSeiSeriesinSoft-

wareEngineering,Addison{Wesley,1990.

17. M. Woodside, C.Hrischuck, B. Selic, and S.Bayarov, A wide band approach to

integratingperformancepredictionintoasoftwaredesignenvironment,Proceedings

ofthe1stInternationalWorkshoponSoftwarePerformance(WOSP'98),1998.