Grid Computing Cluster – the Development and ... - Lim Lian Tze

Grid Computing ClusterThe Development and Integration ofGrid Services and ApplicationsGrid@USM2008/09 ReportEdited by Bahari Belaton and Lim Lian TzePlatform for Information & Communication Technology ResearchUniversiti Sains Malaysia

Grid Computing ClusterThe Development and Integration ofGrid Services and ApplicationsGrid@USM2008/09 ReportEdited by Bahari Belaton and Lim Lian TzePlatform for Information & Communication Technology ResearchUniversiti Sains Malaysia

© 2009 Grid@USMPlatform for Information & Communication Technology ResearchUniversiti Sains Malaysia11800 USM, Penang, MalaysiaAll rights reserved. No part of this publication may be reproduced, stored in aretrieval system, or transmitted, electronic or mechanical photocopying, recordingor otherwise without prior permission of the Publisher.Many of the designations used by manufacturers for products mentioned in thispublication are claimed as trademarks. The authors and the editors respect thecopyright and trademarks for all of these products. Where they are mentionedin the book, and Grid@USM was aware of a trademark claim, the designationshave been printed in capital letters or with initial capital letters.Perpustakaan Negara MalaysiaCataloguing-in-Publication DataGrid computing cluster : the development and integration of grid services andapplications / edited by Bahari Belaton, Lim Lian Tze.ISBN 978-983-3986-58-31. Computational grids (Computer systems). 2. Electronic data processing–Distributed Processing. I. Lim, Lian Tze.004.36Proof-read by Norliza Hani Md. Ghazali.Designed and typeset by Lim Lian Tze with LATEX.First printing, July 2009.Cover image by ©B S K (http://www.sxc.hu/profile/spekulator).

Image: The Petronas Towers, Kuala Lumpur. Photograph by ©Mahjabeen Mankani (http://www.sxc.hu/profile/mejmankani).PROJECT OVERVIEW

Image: It’s a wide, yet connected world. Illustration by ©B S K (http://www.sxc.hu/profile/spekulator).INTRODUCTIONGrid computing can provide easyand transparent access to high performancecomputing by aggregatinggeographically distributed resourcesto act as a single powerful resource.With grid computing technologies,users can access to any virtual resourcessuch as computers, special instruments,software applications anddatabases in a seamless manner. It isanalogous to the electric power grid,where users do not need to know thedetails of the technology. They simplyconnect to common interfaces andsubscribe only what they need. Gridcomputing has evolved from clustercomputing, meta-computing, utilitycomputing and today, a new term hasbeen coined as cloud computing.In order to achieve the objectiveof providing seamless access to highperformance resource and virtual services,the concepts of aggregation andmanagement are vital. Two typesof resources need to be aggregated,hardware and software. The managementaspect involves management oftechnical applications (resource management)and jobs (resource allocation).Resource monitoring monitorsthe grid resources and mitigate problemwhile resource allocation acceptsjob submission and ensures efficientand accurate delivery. Aggregationof grid services is classified into dataand applications. Distributed datamanagement (data placement, datafragmentation or replication and heterogeneousdata transformation) providesdata distribution transparencyfor various data sources.The grid application can be anykind of services that can reside anywhereand accessible from any place.Application distribution transparencyis a main objective in gluing differentservices together in grid environment.In order to manage the unstructured,distributed services, a good resourcediscovery framework has to bein place.Universiti Sains Malaysia (USM)is one of the first universitiesto embark on grid computing researchin Malaysia. In 2002, thefirst grid testbed was establishedthrough an e-Science project involvingUSM, Univeristi TeknologiMalaysia (UTM), and University ofMalaya (UM) in drug discovery andliquid crystal simulation. TheMalaysian Biogrid testbed has beenestablished in collaboration with theMalaysian Biotechnology and BioinformaticsNetwork, Universiti KebangsaanMalaysia (UKM), and SunMicrosystems. Through The PacificRim Applications and Grid MiddlewareAssembly (PRAGMA) and as partof the newly formed Malaysian ResearchNetwork (MyREN), USM is committedto taking Grid Computing Applicationsand Research in Malaysiato the next level.USM researchers have communicatedwith other PRAGMA participantsabout experiences in understandingpractice and policy issues concerningsharing resources and the establishmentof collaborations with otherPRAGMA institutions.In 2002, a research grant wasawarded for the project “The e-ScienceGrid: Development of Back-end GridEngine and Grid Infrastructure”. Twohigh performance clusters were assembled,each with 34 processors,18 GB RAM and 160 GB shared storage,running on a high-speed GigabitEthernet backbone switch. One clusteris located at USM and the other atUniversiti Teknologi Malaysia, KualaLumpur (UTM-KL). The e-Science GridPortal integrates both e-Science Gridcomponents (such as Grid ResourceMonitoring, Grid Resource Allocation,Grid Resource Metering, GridResource Prediction, etc) and e-ScienceGrid applications (such as IterativeSolver Agent, Molecular Simulationand Molecular Docking).2

Image: A simple network. Illustration by ©Svilen Mushkatov (http://www.sxc.hu/profile/svilen001).Grid@USM:DEVELOPING & INTEGRATINGGRID APPLICATIONS & SERVICESUSM’S R&D SUSTAINABILITYTo attain the level of excellence as aresearch university (RU), USM in recentyears has seriously engaged herselfin various efforts to formulateand chart her future research anddevelopment (R&D) directions. Thepinnacle of these efforts was translatedinto a 2 year strategic plan “USMResearch-Intensive University 2007–2009”.One of the crucial input elementsfor attaining and subsequently sustainingRU status is the creationof highly conducive research environmentand infrastructure. Gridcomputing is an example of suchinfrastructure, aiming at providingexcellent research facilities by ensuringgood computational horsepowerto support high impact domainsin the bio-sciences, physicalsciences, information and communicationtechnology (ICT), Environmentsciences, Education, Arts and others.Grid computing is by its naturehighly distributed geographically,consists of highly specialised equipment(storage, grid engines and managementtools), as well as expensive.It is as such an ideal example of common,core computational resourcesto be created and pooled among aspiringresearchers who require highperformance resources.Equally important, USM needs a focused,holistic and dedicated effort4

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsProf. Koh Hock Lye(PPSM)Grid Application to Wave Front Propagationand Containment of Vector-borne DiseasesSetting UpUSMGridAP Chan Huah Yong(PPSK)Dynamic Replica Managementin Data Grid EnvironmentAP Rahmat BudiartoProf. Sureswaran Ramadass(PPSK)iNet GridAP Bahari BelatonProject LeaderAP Tang Enya Kong(PPSK)Grid-enabled Blexisma2AP Phua Kia Kien(INFORMM)Using XGrid for CreatingRender-Farm for 3D AnimatorsDr Kamal Zuhairi Zamli(PPKEE)An Automated Java Testing Toolon the GridDr Vincent Khoo Kay Teong(PPSK)B2B StandardsComponent ModelingGrid middleware and enablers PPSK – School of Computer Sciences PPKEE – School of Electrical & Electronic EngineeringGrid applications PPSM – School of Mathematical Sciences INFORMM – Institute for Research in Molecular MedicineFigure 1: Grid@USM Project Team and Sub-projectsto build a sustainable human capitalin grid computing. We needresearch-based human capital in orderto achieve and maintain technical humanresource support with technicalskills, knowledge and experience toensure uptime of project deliverability,releasing academics to concentrateon activities to further the frontier ofknowledge.GRID COMPUTING AND USMUSM is no stranger to grid computing.We pioneered various grid computingresearch initiatives at both nationaland international levels, as exemplifiedby our past achievements:• USM led the first e-Science grid testbedin 2002, connecting USM–UTM–UKM in a project to simulate liquidcrystal and drug discovery.• USM also led the bio-grid test-bedengaging industry partners SunMicrosystems, Biotechnology andBioinformatics Network (BBN) andUKM.• USM launched its University Gridand International Collaboration initiativein May 2005 to provide awell-designed grid computing infrastructureamong its three campuses,pooling grid resources toprovide computational support toR&D activities.• USM participated in Malaysian ResearchNetwork (MyREN) which providedthe bandwidth to connectgrid resources within and outsideUSM.• USM is the first Malaysian memberin The Pacific Rim Applicationsand Grid Middleware Assembly(PRAGMA). The PRAGMA communityregularly uses USM’s computationalresources via the Auroracluster.• An MoU was signed between USM,Sun-APSTC (Asia-Pacific Science &Technology Center) and NationalUniversity of Singapore (NUS),where one of the proposals was toset up Asia-Pacific Science & TechnologyForum (APSTF) at USM.• USM was granted the honour tohost PRAGMA 15 in October 2008,a milestone in charting and leadingR&D in grid computing.Physically, pockets of grid infrastructuresin the form of clusteredcomputers, high-end grid enginesand other specific resources(e.g. domain-specific software) havebeen acquired and deployed withinUSM. The grid infrastructures aremainly concentrated at the School ofComputer Sciences (PPSK), the Centrefor Knowledge, Communication &Technology (PPKT) and the Laboratoryof Biocrystallography & StructuralBioinformatics (MBBS). The latest additionis the grid engine deployed atthe Collaborative µElectronic DesignExcellence Center (CEDEC) at the Engineeringcampus.GRID ACCESSIBILITY ANDHUMAN EXPERTISEDespite the various progresses made,particularly with respect to infrastructure,the real benefits of these computationalpowers to the scientists werenot yet fully realised. The penetrationrate among pure end-users isperhaps at the lower end – 10 % atmost. Clearly, with such costly investmentin the infrastructure and heavyparticipation from technologists (particularlycomputer scientists), a welldesigneddeployment plan to increaseaccessibility to the end-users is desperatelyneeded. In fact, PRAGMA wasestablished solely to handle this lastmileissue – making grid resourcesaccessible to researchers.Zooming into the technical level,there are two main barriers hamperingthe quick and large-scale acceptanceof grid computing among researchers.Firstly, the technology itself stillneeds further R&D efforts to makeit more user-friendly. The ultimatePROJECT OVERVIEW 5

Grid@USMaim here is to allow researchers withcomputation-intensive requirementsto simply plug their application intogrid technology and run it seamlesslywithout major user intervention. Ananalogy would be the seamless mannerwith which we plug electrical devicesinto power sockets and they willsimply function – without us knowingwhat actually goes on behind thepower socket.The second problem is relatedto the readiness of grid technology.There are stillmany open researchquestionsto address itsfour main domainareas: policyand governanceof gridresources; gridmiddleware andtool enablers;grid applications;infrastructureand security.It is withUSM’s R&D needsand sustainabilityin mindthat we set upGrid@USM, agrid computingresearch clusterto address the problems mentionedabove. Our two main driving principlesand objectives are:• Training of human capital in gridtechnology, and• Increasing the accessibility and utilisationof USM grid infrastructures.Grid@USM’s pilot project, led byAP Bahari Belaton and entitled GridComputing Cluster – the Developmentand Integration of Grid Services and Applications,was initiated under an RUresearch grant. This RU project involvesintegrated research couplingexpertise from different fields. Thetechnologists (mainly computer scientists)will continue to work andimprove on the previously mentionedgrid computing domain areas,while grid infrastructure accessibilityand utilisation is increasedby grid-enabling existing research applicationsby experts from other domains.This is directly related to ourthird objective:• To showcase these grid-enabled applicationsduring PRAGMA 15.Project Objectives❦ To increase accessability andutilization of USM Campus Gridinfrastructure❦ To train research-based humancapital in grid computing❦ To showcase potential gridapplications in PRAGMA 15The human capital training will focuson research-based skills, knowledgeand hands-on experience in gridcomputing for both technologists anddomain experts.THE IMPLEMENTATION PLANThe project can be viewed as havingtwo main domains:Grid middleware and tool enablers,concerning the administration andmanagement of grid resourcesand users. We allocated 1 / 3 of theoverall project resources to thisfundamental stage.Grid applications, where readilyavailable applications are gridenabledto run on existing gridinfrastructure available at USM.This accounts for 2 / 3 of the project.The former task is entrusted tothe Grid Computing Lab (GCL) ofPPSK. As for the latter aspect,a call for research proposals issuedby the ICT Research Platformelicited applications from researchersof various disciplines, ranging frommedicine, computer sciences to engineering.Seven seed applications andtheir principle investigatorswereidentified. Togetherwith thetask of settingup USM CampusGrid i.e. the gridinfrastructure forresearchers, wehave a total of8 sub-projects,the leaders ofwhich are alsomembers ofthe Grid@USMteam.To achievethe project aims,domain expertsin the respectivefields wereemployed asGraduate ResearchAssistants (GRAs) and/orResearch Assistants (RAs) for eachgrid application sub-project. TheseGRAs and RAs were given training toget acquainted with the grid enginesand infrastructure at USM. They thenadapted the domain-specific applicationsto run on the grid environmentunder guidance of the respective subprojectleaders. Finally, these gridapplications were deployed on USMCampus Grid, where they are accessiblevia various grid tool enablers. Detailsabout the research methodology ofeach sub-project are reported in laterchapters, as are the various activitiescarried out under this project.U@SGridMGrid@USM6 Grid@USM: DEVELOPING & INTEGRATING GRID APPLICATIONS & SERVICES

PROJECT MILESTONESPast events2007Planned eventsOctProject Kick-offPhase 1: Grid-enabling seedapplications for PRAGMA15Sub-project teamsequipped with gridenabledworkstationsSub-project teamshooked up toUSM Campus Grid2008FebMarIntroductory Workshopto Grid ComputingPRAGMA 14 in Taiwan(Promoting PRAGMA 15)Workshop on GridComputing Strategies& Security IssuesJunPRAGMA 15 in USMAugCampus Grid Portalwent liveAll sub-projects participatedin poster sessionsOct3 sub-projects selectedfor specialdemonstrations2009Phase 2: Completing gridservices projects to providecampus grid infrastructurefor more researchers andexternal partiesPRAGMA 16 in KoreaMarApr1st Knowledge GridMalaysia ForumUSM Campus Gridhardware upgradePublishing ProjectReportJulProject End: 30 Sep(Original plan)AugSepGrid-enabled Blender 3DAnimation WorkshopNovSouth China SeaTsunami Workshop 3Presence atsains@usm asservice provider2010U@SGridM7

PROJECT EXPENDITURESJanuary ’08 – May ’09Project allocation and expenditures until May 2009. All amounts are in RM.USM Expenditure Code & item Project Allocation Amount Spent Project Balance111 Salary & wages 465,600 274,885 190,715115 Other emoluments 0 4,272 −4,272221 Travel expenses & subsistence 60,000 52,456 7,544223 Communication & utilities 4,000 5 3,995224 Rental 10,000 0 10,000227 Research materials & supplies 20,400 11,226 9,174228 Maintenance & minor repair services 20,000 0 20,000229 Professional & other services 368,000 59,579 308,421335 Equipment 52,000 274,436 −222,436Total 1,000,000 676,859 323,141Salary & wages Other emoluments Travel expensesCommunication & utilities Rental Research materials & suppliesMaintenance & minor repair services Professional & other services EquipmentProject allocation (outer ring) and actual expenditure up till May ’09 (inner pie)U@SGridM8

Image: Eye on Malaysia by Titiwangsa Lake, Malaysia. Photography by ©Mee Lin Woon (http://www.sxc.hu/profile/meelin).SUB-PROJECTS

Image: Server room at DESY, Hamburg. Photograph by ©Christian Hoffmann (http://www.sxc.hu/profile/brcwcs).SETTING UPUSM CAMPUS GRIDGrid Computing LabSchool of Computer SciencesTEAM MEMBERSProject Leader : AP Chan Huah YongResearchers: Ang Sin Keat, Tan Chin Min, Kheoh Hooi Leng, Cheng Wai Khuen, Muhammad Muzzammil bin MohdSalahudinINTRODUCTIONIn order to increase the accessibilityof USM’s grid infrastructure, we at theGrid Computing Lab (GCL) embarkedon this effort to set up USM CampusGrid to link research groups in all 3Universiti Sains Malaysia (USM) campuses.USM Campus Grid also acts as atestbed for the sub-project teams’ gridenabledapplications. We welcome researchers,both in and outside USM,to utilize USM Campus Grid resources.In a nutshell, our objective is tointegrate clusters at the 7 sub-projectteam locations with GCL filtered availableresources and PPKT resources.USM CAMPUS GRIDRESOURCESCentrally, several machines were setup at GCL to serve all sub-project teammembers, to handle directory and servicequeries (Lightweight DirectoryAccess Protocol or LDAP), grid credentialmanagement, user authorisation,as well as for computational processingneeds.On the other hand, this project involves7 sub-project teams from differentresearch groups and centres ofUSM, spanning all 3 campuses in differentgeographical locations:Main campus (Minden, Penang)❦ Grid Computing Lab (GCL),School of Computer Sciences:Dynamic Replica Management inData Grid Environment.❦ Computer-aided TranslationUnit (UTMK), School of ComputerSciences: Grid-enabledBlexisma2.❦ School of Mathematical Sciences(PPSM): Grid Application to WaveFront Propagation and Containmentof Vector Borne Diseases.❦ Information Systems EngineeringResearch Group (ISE), Schoolof Computer Sciences: B2B StandardsComponent Modeling.❦ Network Research Group (NRG),School of Computer Sciences:iNet Grid.Transkrian Engineering Campus(Nibong Tebal, Penang)❦ School of Electrical & ElectronicEngineering (PPKEE): An AutomatedJava Testing Tool on theGrid.Health Campus(Kubang Kerian, Kelantan)❦ Institute for Research in MolecularMedicine (INFORMM): UsingXGrid for Creating Render-Farmfor 3D Animations.We supplied each sub-project teamwith a Dell Vostro 400 Mini TowerDesktop workstation, on which we installedand configured the Rocks Clusters4.2.1 (Cydonia) operating system,the Sun Grid Engine (SGE), the Globustoolkit, the MyProxy credential managementservice, the Ganglia moni-10

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsTable 1: Recent USM Campus Grid resourcesCluster No of CPU Physical Hosts Memory (each host) Storage (each host)Aurora 32 16 2 GB (master node)1 GB (compute node)80 GBGCLDatarm 20 5 2 GB 500 GB (master node)250 GB (4 compute)GCLMicro 40 5 8 GB 500 GBGCLHome 8 1 8 GB 5 TBLDAP server 4 1 4 GB 500 GBgclportal 4 1 4 GB 500 GBgclproxy 2 1 1 GB 160 GBgcldns 1 1 786 MB 160 GBmathvbd 4 1 2 GB 500 GBisitb2b 4 1 2 GB 500 GButmkblex 4 1 2 GB 500 GBengtest 4 1 2 GB 500 GBnav6mon 4 1 2 GB 500 GBComsics – – – –Table 2: Physical machines purchased under RU grantModel Quantity ParticularsDell Vostro 400 Mini Tower Desktop 7 A single desktop was distributed to each member in the campusgrid projectApple X-Grid Server CTO 1 Used for INFORMM’s projectSun SPARC Enterprise T1000 Server 3 Machines just arrived and still in installation progress.Dell Power Edge 1950 Rack Mount Server 1 Machine just arrived and still in installation progress.Dell Power Edge 1950 Rack Mount Server 3 Machines just arrived and still in installation progress.Dell Power Edge SC1435 Server 5 Set up as GCLMicro cluster.toring system, and tools for MessagePassing Interface (MPI) programming.Our research officers (ROs) set upthe machines on-site at each subprojectteam’s lab, connecting them toUSM Campus Grid such that each clustercan access resources on all other clustersin the grid. We also connectedthe USM Campus Grid to the MalaysianResearch Network (MyREN), thereforeallowing external parties to utilizeUSM grid resources as well – includingPRAGMA members. We also createduser accounts and grid credentialsfor the sub-project team membersduring this set up exercise.The current machine resources arelisted in Tables 1 and 2, while Figure1 shows a conceptual view of USMCampus Grid’s architecture.USM CAMPUS GRID USERSIn line with Grid@USM’s objective ofincreasing the accessibility and utilisationof USM grid infrastructures,we welcome researchers – both inand outside the School of ComputerSciences and even USM – to use USMCampus Grid for their computationalneeds. Table 3 lists current registeredUSM Campus Grid users and their researchprojects.SHARED HOME DIRECTORYEach sub-project team originallymaintained their data and applicationfiles on their own machines, andtherefore had to transfer files betweenmachines before submitting processingjobs on other clusters. We thereforeset up a shared user home directory,by using the network file system(NFS) protocol and the Lightweight DirectoryAccess Protocol (LDAP) serverto get the user’s details in setting upthe user profiles.With this shared home directory,USM Campus Grid users can log on toany node and have access to all theirfiles. They are therefore able to initiateand submit jobs without the hassleand overhead of transferring requiredfiles between nodes.CAMPUS GRID PORTALWe have built a web portal forUSM Campus Grid at http://gclportal.usmgrid.myren.net.my using theGridSphere platform. The Campus GridPortal acts as an information centrefor USM Campus Grid and Grid@USM’sprojects. Upon logging in, membersmay customize the portal by specifyinggroups that they wish to bein.We describe other important featuresof the Campus Grid Portal in thefollowing sections.PROJECT PROGRESSUPDATES WITH TRAC WIKIWe have set up a Trac Wiki platformfor the respective sub-project teamsto post information and progress updatesabout their respective applications.This wiki serves as a projectmonitoring mechanism, as well as aninvaluable archive for recording theirexperience using Grid technologies.SUB-PROJECTS 11

Grid@USMFigure 1: Simplified view of USM Campus Grid architectureAPPLICATION SHOWCASESWITH I-FRAMESub-project teams may showcase theirgrid-enabled applications in the CampusGrid Portal using a number ofmechanisms. In cases where a Webgraphical user interface (GUI) is availableand presumably hosted on thesub-project team’s node, this is just amatter of informing GCL researchersof the relevant URL. We would thenembed the web interface as an i-Frameportlet in the Campus Grid Portal. Inother words, the portal acts as a onestopshowcase of the sub-project gridenabledapplications.Each member can specify the membergroup they wish to be in, thuscontrolling the i-Frame applicationsthat will appear on their personalisedportal pages upon logging in. Thisalso provides an easier route for membersto make use of the results fromanother team’s application, i.e. viathe portlets and without having toknow the application’s original URL.Currently, the UTMK and ISE projectteams have incorporated Web GUIfront-ends of their applications intothe Campus Grid Portal as i-Frameportlets (Figure 2).JOB SUBMISSION PORTALJob submission portlets were deployedto provide a user-friendly GUIfor submitting jobs to the grid, as opposedto using the command line. Authorisedusers are only required toprovide their passwords and the locationof a executable file. At the clickof a button, the job is submitted toa designated cluster. The portal willnotify the user when the job is completed,and users may then downloadthe results. In particular, we have createda dedicated job submission portletto allow thePPSM team to submittheir computation jobs to the Auroracluster.SINGLE LOGONWe implemented a single log-on systemto provide transparent access toother grid resources in USM. Thesystem currently allows Campus GridPortal users to access the Centrefor Knowledge, Communication &12 SETTING UP USM CAMPUS GRID

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsUserAP Phua Kia KienLim Lian TzeProf. Koh Hock LyeDr Kamal Zuhairi ZamliProf. Sureswaran Ramadass, AP Rahmat BudiartoDr Vicent Khoo Kay TeongAtheerSupervisor: Dr Nur’Aini Abdul RashidAdnanSupervisor: Prof. Sureswaran RamadassSalah SalemSupervisor: AP Chan Huah YongAhmad AbusnainaSupervisor: AP Chan Huah YongDr Yoon Tiem Leong –Ng Kok FuSupervisor: AP Norhashidah Hj Mohd AliKoay Kok TeongSupervisor: AP Norhashidah Hj Mohd AliTable 3: USM Campus Grid UsersProjectUsing XGrid for Creating Render-Farm for 3D AnimatorsGrid-enabled Blexisma2Grid Application to Wave Front Propagation and Containment ofVector Borne DiseasesAn Automated Java Testing Tool on the GridiNet GridB2B Standards Component ModelingMSc (Mixed Mode)PhDMSc (Mixed Mode)MSc (Mixed Mode)PhD, Parallel Preconditioned Explicit Group Methods for Distributed MemoryMulticomputersMSc (Research), Parallel Numerical AlgorithmsFigure 2: An i-Frame portlet in Campus Grid PortalTechnology (PPKT)’s portal with a onetimesign-in, and vice versa.SUPPORT & MAINTENANCETo summarise, the GCL team acts asa support and maintenance team ofUSM Campus Grid and Grid@USM ’s RUproject. Among the tasks we havebeen (and are) carrying out include:• managing USM Campus Grid machineslocated centrally at GCL, includinghardware upgrades;• managing and monitoring USMCampus Grid services, includingmanagement of users, credentialsand computational resources;• managing the Campus Grid Portalas an information centre and jobsubmission portal;• handling project management andadministrative issues;• supporting sub-project teams ingrid-enabling applications, e.g.:Í installing Blender 3D onGCLDatarm and creating a samplePHP page for SGE job submissionfor the INFORMM team;Í preparing sample Fortran code,with and without enabling MPI,for the PPSM team;Í enabling database connectivitybetween Microsoft SQL2005 andLinux clients using FreeTDS,C/C++ library scripts and bindingwith ASP.NET (C#) for theISE team.U@SGridMSUB-PROJECTS 13

Image: Binary data everywhere. Illustration by ©Rodolfo Clix (http://www.sxc.hu/profile/clix).DYNAMIC REPLICA MANAGEMENTIN DATA GRID ENVIRONMENTGrid Computing LabSchool of Computer SciencesTEAM MEMBERSProject Leader : AP Chan Huah YongResearchers : Aloysius Indrayanto, Muhammad Muzzammil bin Mohd SalahudinContributors : Siamak Sarmady, Wong Jik Soon, Seow Kwen Jin, Zeinab Noorian, Ang Sin Keat, Cheng Wai KhuenINTRODUCTIONAccessing, managing, querying, andprogramming a database system arenormally dependent on the type ofthe database deployment, standards,conventions defined by the databasevendor, and the language that will beused to define the query. Currently,the Relational Database ManagementSystem (RDBMS) is one of the mostwidely used database system. Queryon these kind of databases are usuallydone by using a language namedStructured Query Language (SQL).RDBMS is currently implementedand supported by many different vendors.Every vendor has its ownmethod or convention on how usersshould access its database system.This convention is defined as a setof programming procedure calledApplication Programming Interface(API). These differences would causea developer (programmer) the needto learn many different API if he/shewants to use and integrate RDBMS deploymentsfrom multiple vendors.SQL is a standardized language.However, the standard set of SQL commandsnormally are not enough fora vendor. Hence, database vendorswould normally add non-standard extensionsto the SQL so that users canuse their RDBMS more efficiently. Thisextension would cause an administratoror a developer the need to learnmany non-standard SQL commands.In addition, some vendors also providea custom API which will need tobe used if a developer wants an evenmore efficient access to the RDBMS.Some vendors only provide customAPIs to access the RDBMS’ specific features(no extension to the SQL language).Learning many different APIs andnon-standard SQL extensions wouldnot be efficient for a developer. Timecould be wasted to factorize whatcan be done and what cannot bedone with a particular RDBMS product/deployment.Hence, interoperatingRDBMS from multiple vendorswould be a maintenance problem.Performing a custom distributedquery between two different RDBMSthat are installed in two differenthosts (computers) would be impossibledue to the fact that the vendorsuse their own proprietary protocolsfor communication betweentheir RDBMS servers and clients. Also,not all RDBMS vendors support distributedquery between two (or more)hosts. In addition, the cost of thesoftware license and deployment forthese kind of databases are expensive.A large database vendors, such asOracle and IBM DB2, provides supportsfor distributed query. Hence,their system is called DistributedDatabase System (DDS). However, thetwo systems may not cooperate wellwith each other. Hence, performinga distributed query between Oracleand DB2 would be difficult.The VDDBMS project aims to providea compatibility layer betweendifferent RDBMS implementations andthe end users. The main objective isto allow distributed queries betweentwo or more RDBMS from differentvendors with the same SQL and API.14

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 1: Current architecture of the VDDBMS systemMETHODOLOGYThe methodology applied by theproject to achieve its goal can bebriefly described as follow:1. An SQL parser has been implementedto parse the user-given SQLstatements into tokens (symbolsand data).2. A resource locator has been implementedto add references to thetokens to record the hosts (datasources) on which the SQL wouldneed to be executed.3. A token splitter has been implementedto split the tokens into subtokens.These steps are necessarybecause a data source does not haveto execute the complete tokens.4. An instruction sequencer has beenimplemented to execute to subtokensin the appropriate datasources in the correct sequence.5. A SQL generator has been implementedto translate back the subtokensinto SQL statements understandableby the target datasources.6. A post-processor has been implementedto aggregate back the subresultsfrom the data sources intothe final result.DESIGN & IMPLEMENTATIONThe system can be separated into fourmain modules:1. Core engine;2. Backend drivers for each RDBMSvendor to be supported;3. Language binding;4. Network server and client drivers.The system is mainly developedusing C++ in Linux. Figure 1 showsthe current architecture of the system.The core engine interacts with thebackend RDBMS using the specificdrivers. The core engine is workingtogether with resource catalogue andlock manager to process user requests.User can connect to the VDDBMS systemeither locally or remotely. Currentlylocal connection is supportedby native API in C++ and PHP. Remoteconnection is supported by networkAPI in C++, PHP, Java, and .NET.Figure 2 shows the modules in theVDDBMS system with their simplifieddataflow. In the figure the solid linesdefines communication path whichare done using API. The dashed linedefines communication path whichare done using network protocol.Most of the API communicationpaths are private. Only the C++ andPHP public interfaces are usable bythe user. Hence, changes in the internalAPI would not raise incompatibilityissue with user applications.The network (VDDBMS) server andthe client drivers (C++, Java, .NET)are completely independent. As longas the same protocol is used, changesin the server would not affect theclient. The protocol has been designedto be extensible. Hence, anewer server would be able to manageolder client driver implementations.Figure 3 shows the high level flowof the VDDBMS engine while Figure 4shows the detailed flow of the coreexecution engine.The main features of the currentsystem are:1. Data replication and horizontalfragmentation.2. Support for simple SQL INSERT,DELETE, UPDATE, andSELECT queries.3. Prepared statement for SQLINSERT.4. Native DATE/TIME field support.5. Stored procedure with supportfor looping, conditional statements,simple calculations, and atomic integeroperations.6. MySQL, PostgreSQL, Microsoft-SQL, and Oracle are supported asthe backend RDBMS (data sources).SUB-PROJECTS 15

Grid@USMFigure 2: Modules of the VDDBMS system and their simplified dataflowsFigure 3: Flow of the VDDBMS Engine16 DYNAMIC REPLICA MANAGEMENT IN DATA GRID ENVIRONMENT

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 4: Detailed Flow of the VDDBMS Execution Engine7. Plugin support (backend RDBMSdriver and user-implementedfunctionin stored procedure).8. C++ native API.9. PHP, Java, and .NET binding.10. Network server is implemented ontop of the core engine with userauthentication support.RESULTS & CONCLUSIONThe VDDBMS system is able to:• Display multiple backend RDBMSas if they are one single, virtualdatabase host.• Perform database and table managementsin multiple backendRDBMS.• Perform queries and updates inmultiple backend RDBMS.Compared with a standaloneRDBMS system such as MySQL, theVDDBMS system performance is about10 to 100 times slower. However, ithas the advantage in that it can performdistributed query on multipleRDBMS hosts from different vendors.FUTURE WORK• Distributed resource catalogue anddistributed lock manager.• Support for aggregation (bottomup)approach. In this mode, existingdatabase deployments can beintegrated as one virtual database.The user would still be able to usethe existing deployments as independentsystem as well as aggregatethem with the VDDBMS system.• Support for error recovery.• Performance optimization.• Integration of a more intelligent algorithmon top of the system forautomated data replica or fragmentmanagement using an already published/implementedwork.CONTRIBUTION TO THE OPENSOURCE COMMUNITYThis work has been published as anOpen Source Software (OSS) projectvia the SourceForge code repository.The project homepage can be accessedfrom the URL http://vddbms.sourceforge.net.PROJECT PUBLICATIONSAl-Mistarihi, H. H. E. & Chan, H.Y. (2009). On Fairness, OptimizingReplica Selection in Data Grids.IEEE Transactions on Parallel & DistributedSystems. (forthcoming).Al-Mistarihi, H. H. E. (2009). AData Grid Replica ManagementSystem with Local and GlobalMulti-Objective Optimization. PhDthesis, School of Computer Sciences,Universiti Sains Malaysia.Chan, H. Y. & Al-Mistarihi, H. H.E. (2008). Replica Management inData Grid. International Journal ofComputer Science and Network Security.Chan, H. Y. & Al-Mistarihi, H. H. E.(2008). Replica Optimization Servicein Data Grids. Sciences Publications.Indrayanto, A. (2009). Initial File-Placement in Data Grid Environmentusing Game Theory and FictitiousPlay. MSc thesis, School ofComputer Sciences, Universiti SainsMalaysia.Indrayanto, A. & Chan, H. Y. (2008).Application of Game Theory andFictitious Play in Data Placement.In Proceedings of the InternationalConference on Distributed Frameworks& Applications 2008 (DFmA 2008).Penang, Malaysia; pp. 79–83.U@SGridMSUB-PROJECTS 17

Image: Sun-lit spiral staircase. 3D illustration by ©Bertrand Benoit (http://www.bertrand-benoit.com/).Inset (on wall): INFORMM building model in Blender 3D.USING GRID TECHNOLOGY TOCREATE A RENDER-FARM FORBLENDER 3D ANIMATIONBioinformatics LabInstitute for Research in Molecular MedicineTEAM MEMBERSProject Leader : AP Phua Kia KenResearcher : Zafri MuhammadINTRODUCTIONGrid computing is becoming morepopular today as computers comeequipped with multi-core CPUs. Togetherwith innovative operating systems,multi-threaded and parallel processing,a.k.a. grid computing, has becomemore common, especially for applicationsthat require intensive computations.An attractive feature ofgrid computing is that it can utilizeidle CPUs connected together in a networkto create a virtual supercomputer.‘Xgrid’ is a distributed computingtechnology introduced by AppleInc. to create a grid network for parallelprocessing. Applications that canbe controlled via the command-lineinterface of the Unix operating systemshould be able to benefit from Xgrid.One such application is Blender 3D.Blender 3D was selected in this researchbecause it is an open source3D modelling and animation software,and more importantly because it hasan expansive architecture that canbe extended using Python programminglanguage. Blender 3D version2.46 is also capable of multi-threading,which enable one to tap into the multicorecapabilities of modern CPUs inaddition to the power of grid computing.Recently, several public-domain animationmovies have been producedthat showcase Blender 3D’s capabilityas a professional animation tool.However, the rendering time neededfor professional animations take considerabletime to complete using aconventional computer setup. To overcomethis limitation, we explore theuse of grid computing to create aRender-Farm that will reduce the burdenof rendering Blender 3D animations.To encourage end-users (3D animators)to use the Render-Farm, weexplored the use of PHP to developa web portal for easy access and retrievalof rendered results.OBJECTIVETo create a Render-Farm using Xgridtechnology for Blender 3D in order tospeed up the processor-intensive renderingand compression of animationfiles for Blender 3D animators.RESEARCH METHODOLOGYIn our grid environment, we used10 Apple iMac computers that wereavailable at the Bioinformatics Lab,18

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 1: System Architecture showing system flow for Blender 3D Render-Farm Portal developed at INFORMM, USM.Figure 2: Xgrid Admin showingthe combined CPU power of the gridobtained by networking 10 iMac2.0 GHz (Intel Core 2 Duo CPUs) ina cluster using Apple Xgrid.Institute for Research in MolecularMedicine (INFORMM), USM. All AppleiMac computers were equippedwith Mac OS 10.5, Intel Core 2 Duo2.0 GHz processors and 1 GB of RAM.These were connected through theUSM local area network (Figure 1). AnApple Xserve was used as a controller,and the Apple iMacs served as agentsin the grid. A total of 40 GHz CPUpower was successfully attained onthe grid cluster when all 10 iMac computersin the cluster were used (Figure2).Blender version 2.46 was downloadedfrom the Blender website(http://www.blender.org) and installedon all agents (see Figure 5(a)for screenshot). A Blender 3D Render-Farm portal was developed usingPHP and MySQL to provide easy accessfor users to submit their Blender3D rendering jobs (animation and texturesfiles; Figure 3(a)), and to retrievethe rendered results (individualframes) from the controller (Figure3(b)).Users can simply ‘submit’ theirjobs by logging-in to the Render-Farmportal; upload their animation andtexture files and fill in the start-frameand end-frame in the fields provided(Figure 3(a)). Once the job is submitted,the controller splits the job intomultiple tasks and then distributes itto any available agents on the grid.The submitted job contains a Job IDand is saved in the MySQL databaserunning on the server. During the renderingprocess, Xgrid AdministrationTool was used to log-in to the Xgridcontroller to monitor its activities (Figures4(a), (b) and (c)). The easy-to-useinterface provides a clean summaryof the cluster activities, such as theSUB-PROJECTS 19

Grid@USM(a) Job Submission Page. Users only need to upload their animation fileand textures. The render job is given a name and the ‘start-frame’ and‘end-frame’ fields are keyed in before clicking on the ‘Submit’ button tosend the job to the controller, which then splits and delegates the job tothe agents on network.(b) Results Page. When the job is completed, the agents return the resultsto the controller when they are stored in compressed form as a zip file.Users only need to click the ‘Download’ button to return the results fromthe controller.Figure 3: Screenshot of the Render-Farm portal(a) List of agents available in the grid(b) Jobs Status Monitor (Running jobs)(c) Jobs Status Monitor (Finished jobs)Figure 4: Xgrid Admin Tools20 USING GRID TECHNOLOGY TO CREATE A RENDER-FARM FOR BLENDER 3D ANIMATION

Grid Computing Cluster – the Development and Integration of Grid Services and Applications(a) Wireframe rendered image of INFORMM building at hd resolution (1920 × 1080 pixel) for Blender3D. Model took less than 1 second to render and save compressed as a JPEG image file.(b) Final rendered image from Blender 3D using radiocity ray-tracing. Image took 80 seconds to render1 still frame. 24 still frames are needed to produce 1 second of animation.Figure 5: Images rendered by Blender 3Djob status and how many jobs are currentlyrunning and/or pending, CPUusage, and much more.On completion of the renderingjob, the agents will send back the renderedresults (series of static imagefiles) to the controller. The controllerthen collects the images into a temporaryfolder and then proceeds to compresseach rendered file in the folder.The temporary folder is then packagedas a zipped file and stored inthe portal, ready for download. Theuser can then download the zippedfile to his terminal for post processing,such as QuickTime stitching togetherand temporal or spatial compressionto create a .mov file ready to play.RESULTSThe time periods taken for rendering10,000 frames of Blender 3D animationon the Render-Farm containing 1to 10 iMac computers were recorded.The result showed that there is ahuge difference in completion timebetween using one computer and theRender-Farm running on 10 agents.Xgrid with 10 computers (20 CPUs)running a job consisting of 10,000frames of Blender 3D animation took2340 seconds (39 minutes) comparedwith 20,160 seconds (336 minutes) ona single computer (2 CPUs). This representsa 9 fold increase in renderingspeed.SUB-PROJECTS 21

Grid@USMTable 1: Reliability of Xgrid Render-FarmTime Taken (Seconds) To Complete Rendering 1000 FramesNo. of nodes1 2 3 4 5 6 7 8 9 10❵❵❵❵❵❵❵❵❵❵ Run No.❵1 2435.00 1306.90 963.42 663.78 569.78 435.71 422.64 355.13 327.98 305.872 2436.00 1336.00 963.35 663.33 569.40 436.00 421.00 354.00 327.66 303.003 2435.00 1306.90 963.42 663.78 560.78 435.71 422.64 355.23 327.98 305.874 2435.00 1306.90 963.42 663.78 569.78 435.71 333.00 355.13 327.98 305.275 2435.00 1306.90 963.33 660.00 564.00 432.00 421.00 350.95 321.22 301.17coefficient of variation (%) 1.848 3.061 4.152 0.002 0.007 0.003 0.002 0.005 0.009 0.006Figure 6: Blender 3D rendering times using Xgrid and increasing number of agentsDISCUSSION &CONCLUSIONIn this project, we constructed a gridnetwork of 10 iMac computers usingApple Xgrid middleware to create aRender-Farm for Blender 3D animation.A web portal was also developedusing PHP to enable easy submissionof render jobs to the gridand to retrieve the results of therendering. To study the efficiencyof the grid, we recorded the timetaken to render 1000, 5000, and 10,000frames of Blender 3D animation witha fixed number of agents in the gridfor each study. The results showedthat when the number of agents wereincreased from 1 to 10, the renderingtimes were reduced by 86.78 %,86.68 % and 88.39 % for 1000, 5000and 10,000 frames, respectively. Onaverage the 10-node Render-Farm wasable to complete the rendering job byas much as 9 times faster than theconventional single computer setup.As the workload increases, the efficiencyalso increases, although largerjobs also increase the time taken toreturn the results to the controller.To study the reliability (reproducibility)of the Xgrid Render-Farm,we repeated 5 times each run acrossa fixed number of agents and foundthat the coefficient of variation oftimes taken for completing of eachrendering job was less than 5 % (Table1). We conclude that Xgrid canbe used to create a reliable low-costRender-Farm to accelerate CPU intensivetasks, saving time and cost.FUTURE WORKFollowing the PRAGMA 15 exhibitionand showcase, we have receivedmany queries regarding our Blender3D Render-Farm portal. We arepreparing to organise a workshop onBlender 3D and grid computing.PROJECT PUBLICATIONSPhua, K. K. & Wong, V. C. (2007).Developing A Compelling 3D Animationand Multimedia PresentationUsing Blender for INFORMM’sOpening Ceremony. Malaysian Journalof Medical Sciences, 14(Supplement1).Zafri, M., Sanjay, K. C., & Phua, K.K. (2009). Implementation Methodsfor Estimating Haplotypes withGRID Computing Technology. InCompendium of Abstracts for the 14thNational Conference on Medical andHealth Sciences (NCMHS).Zafri, M., Fadhilah, N. K., & Phua,K. K. (2008). Using Xgrid to ImproveFASTA Efficiency for Alignmentof Multiple DNA and ProteinSequences. Malaysian Journal of MedicalSciences, 15(Supplement 1).U@SGridM22 USING GRID TECHNOLOGY TO CREATE A RENDER-FARM FOR BLENDER 3D ANIMATION

Image: S/Cultura Luminosa bookshelves.Designed by ©Bruno Petronzi (http://www.brunopetronzi.it/).GRID-ENABLEDBLEXISMA2Computer-aided Translation UnitSchool of Computer SciencesTEAM MEMBERSProject Leader : AP Tang Enya KongResearchersCollaborator: Lim Lian Tze, Ye Hong Hoe: Dr Didier Schwab,Groupe d’Étude en Traduction Automatique/Traitement Automatisé des Langues et de la Parole (GETALP),Laboratoire d’Informatique de Grenoble, Université Pierre Mendès France (Grenoble 2), FranceNATURAL LANGUAGEPROCESSING &GRID COMPUTINGEfforts to create computer programsthat can understand human languages,or natural languages, has along tradition dating back to the inventionof the computer. Much workis still on-going in the field of NaturalLanguage Processing (NLP) today.On one hand, there are still manypertinent research questions as naturallanguage is intricate, ambiguous,dynamic and continuously evolving.On the other hand, novel needs forNLP applications are discovered everydayas people from different communitiesinteract in this increasinglyglobalised world, across national, linguisticand cultural borders. Machinetranslation, automatic summarisation,intelligent search and targeted advertisingare some examples that springto mind. Suffice to say that NLP is anexciting research field to be in.NLP systems are notorious for beingresource-hungry. Intricate grammaticalstructures, large vocabularies,the complex interactions betweenthese linguistic layers, as well as thesheer amount of documents to beprocessed, all call for high processingpower, speed and storage. Distributedprocessing in grid environmentshave proved to be a helpful solution:the Tsujii Lab at the Universityof Tokyo had successfully performeddeep grammatical parsing of the entireMEDLINE corpus 1 in 8 days on pc1 MEDLINE is an online database of 11 million citations and abstracts from health and medical journals and other news sources.It contains about 10 GB of uncompressed data.23

Grid@USMwhole#2entity#1273 nodes 7 nodes 5 nodes 6 nodes148699 7553medical building#1medical institution#1medical man#1hypernymhypernymhypernymhospital#1 hospital#2doctor#18 64hypernym hypernymsurgeon#1 1 allergist#1 2has instance3Hippocrates#1Figure 1: Discrete relations and non-discrete neighbourhoods of lexical items and their meanings. Hypernymy (the ‘is-a’relation) links surgeon to doctor etc, but does not link hospital to surgeon nor doctor directly. Instead, all lexical items related tothe medical profession are within nearby neighbourhood of each other by the conceptual vector model.clusters consisting of 340 CPUs. Closerto home, we had previously deployedour machine translation system on agrid cluster for load balancing.An NLP application would needto understand – or seen to be capableof understanding – ‘meaning’ asconveyed by natural language discourse,if it was to perform anythingdeemed ‘intelligent’. While the exactdefinition of ‘meaning’ is still debatedamong linguists and philosophers, weattempted to create a simple computationalrepresentation of word (lexicalitem) meanings. This model is used byBlexisma2, our distributed multi-agentsystem, to construct a semantic lexicaldatabase and to perform semanticprocessing of natural language texts.REPRESENTING LEXICALMEANING WITH COMPUTERSWe designed a Semantic Lexical Base(SLB) to store computer representationsof word meanings. The meaningof a word or lexical item can bemodelled by its relations to other lexicalitems. For example, the Englishword heavy is used to convey the intensityof rain – we do not say bigrain or even strong rain in English.In addition, rain intuitively calls tomind other words like snow, sunny,wet, water. We model such explicitlinguistic relations with discrete linksbetween lexical items (and betweentheir meanings), and implicit ‘conceptualneighbourhood’ with mathematicalvectors. Essentially, the conceptualvector model projects words and theirmeanings onto a spherical space, inwhich meanings with similar ‘conceptualthemes’ (e.g. water, weather) arelocated close to each other. Figure 1illustrates how this hybrid representationscheme captures multiple facetsof lexical item meanings.The SLB is designed to handle polysemy,i.e. that a surface lexical form(e.g. mouse) can have multiple meanings(a small, furry animal; a computeraccessory; a timid person; etc).In other words, the SLB will store arecord for each acknowledged meaning.One of the key aims of Blexisma2is to automatically acquire as manymeaning records (called acceptions)as possible from multiple sources,including dictionaries, terminologylists and Web articles. We chose touse multiple learning sources becauseno single source can claim exhaustivecoverage of a language’s lexicon.Moreover, if the information or definitiongiven by a particular source isbadly written to the point of introducingnoise, the negative effect can beautomatically reduced when data ofbetter quality from another source isprocessed. Therefore, we also store arecord (called a lexie) for each meaningentry acquired from a differentsource. The overall SLB organisationis shown in Figure 2.Blexisma2 is expected to run continuouslyand iteratively, so that newlycoinedterms or new usages of existingwords will be captured in theSLB as they appear. The quality ofacception and lexie objects will alsobe refined with each iterative cycle,as each refined record would ‘feed’positive inputs into the learning processof other lexical meanings. Weimplemented the SLB as a PostgreSQLdatabase, hosted on utmkblex in theUSM Campus Grid.BLEXISMA2 AGENTSBlexisma2 is a distributed multi-agentsystem, implemented with the Mad-Kit platform (http://www.madkit.net). Each Blexisma2 agent has a specificrole or function to perform. Theymay communicate with each other bysending messages, managed by theMadKit kernel. An agent may requestthe help of other agents to completeits task. Conversely, an agent may24 GRID-ENABLED BLEXISMA2

Grid Computing Cluster – the Development and Integration of Grid Services and Applicationslexical item acceptions lexiesPWN:any of numeroussmall rodents. . .PWN:person who is quietor timidPWN:mousea hand-operatedelectronic device. . .Wikipedia:in computing, apointing device. . .Wikipedia:a small animal. . .Wikitionary:any small rodent. . .Figure 2: Schematic organisation of lexical items, acceptions and lexies in the SLB. Lexies are mined from various sources,and are then aligned or clustered into acceptions.respond to other agents’ requests bysupplying information after performingits own processing, or setting offfurther chains of agent interactions.The first implemented agents inour Blexisma2 system prototype attemptsto compute acception and lexieobjects for English lexical items, basedon Princeton WordNet (PWN), a widecoverage,freely available online lexicaldatabase for the English language.PWN Analyser agents would requestinput data from Lexicon Dispenseragents, whose role is to interface withthe SLB on behalf of all other agents.After performing their computations,the Analysers would pass their resultsto the Dispensers to be storedin the SLB. (See Lim and Schwab2008; Schwab and Lim 2008 for detailsof the computation method). Wedeployed instances of these agents onthe USM Campus Grid using the Globusjob submission toolkit (Figure 3), thusfreeing up utmkblex to devote itsresources to the running and maintenanceof the SLB PostgreSQL server.RESULTS AND APPLICATIONWe were able to process as many as1,902,080 lexical items in 5 days, averagingaround 263 items per minute.Performance-wise, we noted bottleneckeffects i.e. hitting the performanceceiling at around 8 nodes dueto frequent connections, high I/O andoverall heavy load on the SLB, whichwe aim to overcome in the future.To demonstrate how the SLB mightbe used, we also implemented asimple Sense-Tagger and a SyntacticParser. The Sense-Tagger takes an Englishsentence as input, and requeststhe Syntactic Parser to analyse thegrammatical properties of the wordsin the sentence. It then uses semanticdata from the SLB to identify themost likely meanings of each word.For instance, when given the inputs(1) Spica is a star in the Virgo constellation.(2) She is the star actress in our play.(3) A school of fish swam past.(4) The school’s classrooms are spacious.the Sense-Tagger identifies star as acelestial body in (1), and as an actorplaying a principal role in (2). Schoolin (3) is tagged as a large group offish; and as an education institutionin (4). This capability to discern betweendifferent meanings of a wordin different contexts is important inhelping other NLP applications provideaccurate results.WHAT’S NEXT?While the Blexisma2 prototype catersonly for the English language, we areSUB-PROJECTS 25

Grid@USMPWNAnalyserPWNAnalyserPWNAnalyserPWNAnalyserRemote HostLexiconDispenserRemote HostLexiconDispenserLexiconDispenserRemote HostLexiconDispenserSense-TaggerSyntacticParserLexiconDispenserSLB on utmkblexPWNAnalyserRemote HostFigure 3: Blexisma2 agents are deployed on USM Campus Grid nodes using the Globus job submission toolkit.(a) Different meanings of star in context(b) Different meanings of school in contextFigure 4: SLB data generated by Blexisma2 agents used to determine most probable meanings of ambiguous lexicalitem. A Web interface to the Sense-Tagger is available at http://utmkblex.usmgrid.myren.net.my:8080/Blexisma2Servlets/CVSenseTagger.interested to improve its design toa multilingual setting, where morecomplicated cross-lingual phenomenawill have to be considered. We alsohope to improve the agent communicationmechanisms to reduce latency.Apart from creating moreagents that implement different learningalgorithms and heuristics, we arealso planning agents responsible forother NLP tasks such as deep parsing,detecting named entities, etc. In otherwords, we hope to have agents of variousresponsibilities so that they canbe ‘mixed-and-matched’ to constructnew NLP applications, such as thosementioned in the introduction.On the performance issues, wemay attempt several solutions for theSLB bottleneck problem mentionedearlier:• hardware (RAM) upgrades,• further optimisation of PostgreSQLserver settings,• database connection pooling,• load balancing,• parallelising queries.PROJECT PUBLICATIONSLim, L. T. & Schwab, D. (2008). Limitsof Lexical Semantic Relatednesswith Ontology-based ConceptualVectors. In Proceedings of the 5th InternationalWorkshop on Natural LanguageProcessing and Cognitive Science(NLPCS’08). Barcelona, Spain;pp. 153–158.Schwab, D. & Lim, L. T. (2008). Blexisma2:a Distributed Agent Frameworkfor Constructing a SemanticLexical Database based on ConceptualVectors. In Proceedingsof the International Conference onDistributed Frameworks & Applications2008 (DFmA 2008). Penang,Malaysia; pp. 102–110.U@SGridM26 GRID-ENABLED BLEXISMA2

Image: A business handshake. Photograph by ©FOTOCROMO (http://www.sxc.hu/profile/FOTOCROMO).B2B STANDARDSCOMPONENTMODELINGInformation Systems Engineering Research GroupSchool of Computer SciencesTEAM MEMBERSProject Leader : Dr Vincent Khoo Kay TeongResearchers : Ting Tin Tin, Rinki Yadav, Johnson Foong, Kor Chan HockINTRODUCTIONThis report is about the research, design,development and implementationof two B2B communications modelson a USM networked non-Gridarchitecture, and also an attemptedimplementation of the models on aMyREN networked Grid architecture.B2B INTEGRATIONSTANDARDSA major challenge in integrating thebusiness processes among the tradingpartners is the enhancement ofthe documents interchange processes.Traditional business-to-business (B2B)process integration is based on a‘Push’ model through which documentsare pushed from the sendersto the receivers, as in e-mailing andelectronic data interchange. However,as the volume of document increases,the ‘Push’ model suffers from diminishingdata quality, which could bedue to the low personalizability inthe communication subject. Pushinga large volume of pre-defined standarddocuments among the tradersusually results in data redundancy.In addition, low personalizability incommunication channel also incurs ahigher cost and longer time in developingand deploying the infrastructure.In this research, we explored the‘Pull-only’ model and the ‘Push andPull’ model to increase the personalizabilityand thereby promote a morepervasive use of the B2B integrationstandards.‘PULL-ONLY’ MODELThe ‘Pull-only’ model is a style of communicationin which the receiver initiatesthe data request to be respondedby the sender accordingly. The processthat involves the sender is automatedby the Web services. Afterthe receiver has initiated the informationrequest, the stored authenticationinformation will be sent by theWeb services to the sender’s serverfor authentication, and connecting theWeb services to the sender’s databaseview. Once it has been authenticated,the sender’s database view will allowthe receiver to ‘pull’ data through theWeb services. The raw data will betransformed into a XML format messagebased on the predefined standardschema. Once the XML messageis validated, it can be transformedback into raw data and inserted intothe receiver’s database automatically.The purpose of the transformationof raw data into XML message is toensure that the message conforms tothe standard schema. This will easethe data insertion process.‘PUSH AND PULL’ MODELA ‘Push and Pull’ model is a style ofcommunication in which the senderpushes the data to the receiver butthe amount of data received is filteredby a standard schema pre-defined bythe receiver (pull). The receiver controlswhat to receive through the filterwhich does not require the receiver tomanually select what to be inserted27

Grid@USMFigure 1: B2B Web Services Integration FrameworkFigure 2: ‘Pull’ Model Mechanisminto their database during the datainsertion process.ENTERPRISE GRIDARCHITECTURE FORCOLLABORATIVE B2BINFORMATIONINTERCHANGEThe Grid infrastructure layer supportsa collaborative database sharingmiddleware. All the virtualdatabases sharable among the partnersare maintained at a centralizedphysical database. The grid middlewareused in this implementationcontrols and manages the retrievalof the databases. The ‘Push’ and‘Pull’ modeling in Education ServiceMashup (EdSeM) helps the partners tointerchange confidential informationmore effectively. EdSeM is the partnerof choice for business partnersthat are interested in building longtermrelationship among the educationalinstitutions worldwide. EdSeMhas been set up to deploy, leverageand integrate many different typesof emerging technologies includingthe Web 2.0 content extraction, agentbased,and decision-support technologies;knowledge dissemination andknowledge representation technologies.The EdSeM infrastructure complementsthe middleware by sharingthe information interchange Web servicesacross the network. Every partnercan use the services to interchangeconfidential information.RESULTS & APPLICATIONInfoX is a data mapping tool thatmaps raw data among the files in adatabase. It uses the ‘Push and Pull’models for transforming data betweenthe trading partners. Each time a tradingpartner needs to send or receive28 B2B STANDARDS COMPONENT MODELING

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 3: ‘Push and Pull’ Model MechanismFigure 4: Enterprise Grid ArchitectureSUB-PROJECTS 29

Grid@USMFigure 5: InfoX ArchitectureFigure 6: InfoX Infrastructure Incorporating Grid Architecturedata, it has to be connected to theInfoX Web Services Server to requestfor the Web service to perform therequired action. All the raw data willbe transformed into the XML formatto ease the data transformation process.The XML format message willbe stored in the InfoX server whichis sharable throughout the networkthrough a grid middleware. The gridmiddleware controls and manages theretrieval of the database. The managementis transparent to the users.The end users only need to query thedesired information through the middleware.The grid services will subsequentlyretrieve the data from variousdatabases based on a data catalog.The distribution of data at variousdatabases is carried out by the gridservices.By deploying the InfoX system ina grid environment, it is able to send45 records per second on an average.However, it was observed that some340 records per second on an averagecould be achieved in a non-gridenvironment.In order to deploy Infox in a gridenvironment through a grid middleware,both the InfoX and the middlewaremust be located in a grid MyRENnetwork. Due to the incompatibilityof InfoX in the MyREN network,it takes a much longer time to processthe data and eventually affectsthe response time.We have also implemented a selfserviceevaluation system for the potentialeducational institutions to estimatethe possible time required forthe development and deployment ofan industry-wide standardised architecture.WHAT’S NEXT?As part of the future work, InfoXand the grid middleware have to beenhanced from various perspectives.First of all, the InfoX system musthave the ability to compress the databefore sending to the grid middlewarefor further processing. It ishoped that the response time especiallywhen sending large files willeventually be enhanced. From thegrid system perspective, we need tofurther investigate the existence of abotteneck in the MyREN network.PROJECT PUBLICATIONSTing, T. T. & Khoo, V. K. T. (2009).B2B Standardized Information InterchangeChallenges – A Study onStandardization versus Personalization.In Proceedings of the 2009 InternationalConference on AdvancedManagement Science (ICAMS 2009);pp. 244–248.Ting, T. T. & Khoo, V. K. T. (2008).Personalizable Information InterchangeArchitecture for EducationalInstitutions. In Proceedingsof the 3rd International Conference one-Commerce with Focus on DevelopingCountries (ECDC’08). Isfahan, Iran.U@SGridM30 B2B STANDARDS COMPONENT MODELING

Image: A programmer hard at work. Photograph by ©Milen Yakimov (http://www.sxc.hu/profile/todorov40).AN AUTOMATED JAVATESTING TOOL ON THE GRIDSoftware Engineering Research GroupSchool of Electrical & Electronic EngineeringTEAM MEMBERSProject Leader : Dr Kamal Zuhairi ZamliResearchers : Dr Nor Ashidi Mat Isa, Mohammed Issam Younis, Saidatul Khatimah binti SaidINTRODUCTIONNowadays, we are increasingly dependenton software to assist as wellas facilitate our daily chores. In fact,whenever possible, most hardware implementationis now being replacedby the software counterpart. Fromthe washing machine controllers, mobilephone applications to the sophisticatedairplane control systems, thegrowing dependent on software canbe attributed to a number of factors.Unlike hardware, software does notwear out. Thus, the use of softwarecan also help to control maintenancecost. Additionally, software is alsomalleable and can be easily changedas the need arises.Covering as much as 40 to 50 %of the total software developmentcosts, testing can be considered oneof the most important activities forsoftware validation and verification.Lack of testing can lead to disastrousconsequences including loss of31

Grid@USMFigure 1: Option dialog in Microsoft ExcelFigure 2: G_MIPOG Graphical User Interfacedata, fortunes and even lives. Manycombinations of possible input parameters,hardware/software environments,and system conditions needto be tested and verified against forconformance based on the system’sspecification. Often, this results intocombinatorial explosion problem.As illustration, consider the optiondialog in Microsoft Excel software(see Figure 1). Even if onlythe View tab option is considered,there are already 20 possible configurationsto be tested. With the exceptionof Gridlines color which takes56 possible values, each configurationcan take two values (i.e. checked orunchecked). Here, there are 2 20 × 56(i.e. 58,720,256) combinations of testcases to be evaluated. Assuming thatit takes 5 minute for one test case,then it would require nearly 559 yearsfor a complete test of the View tab option.As highlighted above, given limitedtime and resources, exhaustivetesting is next to impossible. Obviously,there is a need for a systematicstrategy in order to reduce the testdata into manageable ones.Earlier work has indicated thatpairwise testing (i.e. based on 2-wayinteraction of variables) can be effectiveto sample the test data appropriately.While such conclusion maybe true for some systems, it cannotbe generalized to all software systemfaults (i.e. especially when there aresignificant interactions between variables).This argument is supportedby the fact that software applicationsgrew tremendously in size from kilobytesto terabytes in the last 5 years.Driven by the advancement of technologiesand market demands, the neteffect of this massive software growthcan often lead to intertwined dependencybetween parameters involved,thus, justifying the need to considerfor high interaction strength (i.e. oftenindicated as t).Although desirable, the consider-32 AN AUTOMATED JAVA TESTING TOOL ON THE GRID

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsTable 1: Results For 7 to 10 5-Valued Parameters in 6-Way Testing# Parameters 7 8 9 10Test Case Size 15,625 28,125 40,146 45,168Time (MIPOG) (Single Computer) 118.703 485.047 1637.097 4657.457Time (G_MIPOG) (6-Computers) 55.234 183.184 626.797 1552.125Speedup 2.149 2.648 2.838 3.203Table 2: Results using 1 to 11 Computers for 10 10-valued Parameters in 4-way Testing# Computers Time Speedup Test Case Size1 (MIPOG) 77,882.16 12 (G_MIPOG) 43,680.404 1.7833 (G_MIPOG) 31,140.4 2.5014 (G_MIPOG) 24,089.749 3.2335 (G_MIPOG) 17,150.883 4.5416 (G_MIPOG) 15,193.554 5.126 27,3067 (G_MIPOG) 14,928.537 5.2178 (G_MIPOG) 14,789.624 5.2669 (G_MIPOG) 12,928.645 6.02410 (G_MIPOG) 9570.184 8.13811 (G_MIPOG) 9352.967 8.327ation for higher interaction strength(e.g. from t = 3 onwards) can be problematic.When the parameter interactioncoverage t increases (i.e. as t > 2),the number of t-way test set also increasesexponentially. For example,consider a system with 10 parameters,where each parameter has 5 values.There are 1125 2-way tuples (or pairs),15,000 3-way tuples, 131,250 4-way tuples,787,500 5-way tuples, 3,281,2506-way tuples, 9,375,000 7-way tuples,17,578,125 8-way tuples, 19,531,250 9-way tuples, and 9,765,625 10-way tuples.From this example, it is evidentthat for large system with manyparameters, considering higher ordert-way test data invites massively notoriouscomputation.Facing these challenges, our researchinvestigates a new strategy andits implementation, called G_MIPOG.Here, G_MIPOG (see Figure 2) originatesfrom a well known sequentialt-way strategy called IPOG. Unlike itspredecessor strategy (IPOG), G_MIPOGpermits harnessing of grid based architecture,allowing the support forhigher t than that of IPOG.OUR EXPERIENCE WITH THEG_MIPOG STRATEGYAs the first step to the development ofG_MIPOG, we investigate an intermediatestrategy, called MIPOG 2 , basedon the modification of IPOG. Our experimentalresult demonstrates thatthe control and data dependencies inMIPOG can be removed to permit distributedas well as parallel processing.As part of our research, we intendto investigate whether there isa speedup gain from distributingMIPOG in G_MIPOG. Three experimentsare applied to both MIPOGand G_MIPOG in order to gauge thespeedup. Here, the speedup is definedas ratio of the time taken bysingle computer to the time taken bymultiple computers. The experimentalgoals are to investigate:• the speedup as the number of parametersincreases;• the speedup as the number of computernodes increases;• the speedup against the increase ofparameter coverage;• the comparison in terms of test sizeagainst other strategies.To achieve the first goal, we applyMIPOG, and G_MIPOG (consists of6 computers) to 5-valued parameters,and vary the number of parametersfrom 7 to 10, and fix t = 6. The resultsare demonstrated in Table 1. Toachieve the second goal, we have fixedt = 4, with 10 10-valued parameters.Then, we determine the speedup using2 to 11 computers. The results areshown in Table 2. To achieve the thirdgoal, we apply G_MIPOG to the TrafficCollision Avoidance System (TCAS)module. Here, the TCAS module isan aircraft collision avoidance systemfrom the Federal Aviation Administrationwhich has been used as casestudy in other related works. TheTCAS module has twelve parameters:seven parameters have 2 values, twoparameters have 3 values, one parameterhas 4 values, and two parametershave 10 values. In this case, we use2 computers, and vary t from 2 to 11to determine the speedup, as givenin Table 3. Finally, to achieve the lastgoal, we compare the test size generatedby G_MIPOG and other strategiesfor the aforementioned TCAS module.The results are tabulated in Table 4.We have darkened the entries to showthe best result. Entry with ‘NS’ refersto as not supported by the implementation.DISCUSSIONIn Table 1, we note that the speedupincreases linearly as the number of parametersincreases. Here, extra overheadis added for the fifth parametersdue to the need to start andshut down the corresponding threads.Referring to Table 2, we note thespeedup gain also increases quadraticallyas the number of values increases.Extrapolating and performingcurve fitting of the results from Ta-2 The details of MIPOG can be found at http://gclportal.usmgrid.myren.net.my/trac/engpjtstSUB-PROJECTS 33

Grid@USMTable 3: Results Using t = 2 to 11 for TCAS Modulet-Way Time (MIPOG) Time(G_MIPOG) Speedup Test Case Size2 0.156 0.292 0.534 1003 0.609 0.547 1.113 4004 3.234 2.578 1.2544 12655 36.797 29.125 1.263 41966 301.128 214.091 1.406 10,8517 1772.407 1086.7 1.631 26,0618 10,242.09 5839.276 1.754 56,7429 36,284.7 19,124.78 1.897 120,36110 41,481.672 21,832.46 1.9 201,60111 14,939.891 7821.932 1.91 230,400Table 4: Comparative Test Size Results Using TCAS Module for t = 2 to 12t G_MIPOG IPOG IPOG_D IPOF1 IPOF2 ITCH Jenny TConfig TVGII2 100 100 130 100 100 120 106 108 1013 400 400 480 402 427 2388 411 472 4344 1265 1361 2522 1352 1644 1484 1527 1476 15995 4196 4219 5306 4290 5018 NS 4680 NS 47736 10,851 10,919 14,480 11,234 13,310 NS 11,608 NS NS7 26,061 NS NS NS NS NS 27,630 NS NS8 56,742 NS NS NS NS NS 58,865 NS NS9 120,361 NS NS NS NS NS NS NS NS10 201,601 NS NS NS NS NS NS NS NS11 230,400 NS NS NS NS NS NS NS NS12 460,800 NS NS NS NS NS NS NS NSble 3, we observe that the speedup increaseslogarithmically as the strengthof coverage increases. In this case,there is also no speedup gain forthis strategy when t = 2, possiblydue to the overhead required for creation,synchronization, and deletionof threads for small degree of interaction.Concerning comparison withother strategies in Table 4, G_MIPOG,IPOG, IPOF1, and IPOF2 gave the mostoptimum test size at t = 2. At t = 3,both G_MIPOG and IPOG gave the mostoptimum test size. For all other cases,G_MIPOG always outperforms otherstrategies. Putting G_MIPOG aside,only Jenny can go more than t > 6for the TCAS module. TVG implementationdo not give any result whent > 5 whilst TConfig does not produceany result when t > 4. ITCH cannot support t > 4. We also note thatin the case of ITCH, the test size fort = 3 is greater than that of t = 4.CONCLUSION &FUTURE WORKIn short, our experimental resultsdemonstrate that G_MIPOG scales wellagainst its sequential strategy (MIPOG)with the increase of the computersas computational nodes. Moreover,G_MIPOG produces minimal test set ascompared with other existing strategies(whilst keeping the test set ofMIPOG the same). Finally, G_MIPOG isthe only strategy that supports hight even against all other IPOG variants(i.e. IPOG_D, IPOF1, and IPOF2).As t-way testing approach is still ininfancy within the software engineeringcommunity, novel research andfindings are highly welcomed. Thework discussed here is indeed significantto facilitate the testing activitiesby systematically minimizing the testdata for test consideration. As humanare more and more dependent towardsoftware, this work presents a smallleap toward the betterment of the wayengineers do testing.As a part of our future work, weare trying to integrate test executionand reporting along with parallelizingthe data structure for storing the interactionelement using tuples-spacetechnology.PROJECT PUBLICATIONSYounis, M. I., Zamli, K. Z., Klaib, M.F. J., Che Soh, Z. H., Che Abdullah,S. A., & Mat Isa, N. A. (2009).Assessing IRPS as an Efficient PairwiseTest Data Generation Strategy.International Journal of Advanced IntelligenceParadigms. (forthcoming).Younis, M. I., Zamli, K. Z., & MatIsa, N. (2008). IRPS – An EfficientTest Data Generation Strategy forPairwise Testing. Lecture Notes inArtificial Intelligence, 5177.Younis, M. I., Zamli, K. Z., & MatIsa, N. A. (2008). A Strategy forGrid Based t-Way Test Data Generation.In Proceedings of the InternationalConference on DistributedFrameworks & Applications (DFmA2008). Penang, Malaysia; pp. 73–78.Zamli, K. Z., Che Abdullah, S. A., Younis,M. I., & Che Soh, Z. H. (2009).Software Testing Module. (forthcoming).OpenUniversity Malaysia andPearson Publishing.U@SGridM34 AN AUTOMATED JAVA TESTING TOOL ON THE GRID

Image: Ethernet crossover cable. Photograph by ©Elia Schmidt (http://The-Condor.deviantart.com/)INET-GRIDNetwork Research GroupSchool of Computer SciencesTEAM MEMBERSProject Leader : Prof. Sureswaran RamadassResearchers : AP Rahmat Budiarto, AP Chan Huah Yong, Dr Ahmed M. ManasrahGRID COMPUTINGA Grid environment is a complexglobally distributed system that involveslarge sets of diverse, geographicallydistributed components used fora numerous type of applications. Thecomponents discussed here includeall of the software and hardware servicesand resources needed by applications.Due to the diversity of thecomponents and enormous numberof users, the Grid environment hasbecome more vulnerable to fault, failureand excessive tools. Thus, a suitablemechanism is needed to monitorthe components, their usage and detectingconditions that may lead tobottlenecks, faults or failures. Gridmonitoring is a potential platform towardsa robust, reliable and efficientenvironment in networking wherebyit is crucial to maintain network reliabilityduring a peak environment.In conclusion, a combination betweennetwork monitoring and grid monitoringare required in order to enhancenetwork reliability and performances.Network monitoring in Grid environmentplays an important role innetwork management and is essentialin Grid utilization. According to theGlobal Grid Forum schema, the managementof network infrastructure observationsis divided into three mainactivities: their production, using networkmonitoring tools, their publication,by way of powerful databases,and their utilization by administrationand workflow analysis tools.We are embarking on an intelligentreal time grid monitoring systemcalled iNet-Grid. The purpose ofthis research is to enable an intelligenttool to assist network and system administratorsintelligently by providinginformation for preventive measuresto be taken to minimize damages dueto system or network down time thatcan be very costly. Real-time networkanalysis helps in detecting and resolvingnetwork faults and performanceproblems in a minimum amount oftime. This system has the power to analyzemulti-topology, multi-protocolnetworks, automatically.TECHNICAL OBJECTIVETo prove the ability of the proposedwork to be integrated with any opensource application for grid monitoring(Ganglia) and thus, building a platformthat allow all other grid monitoringsystem to be integrated togetherfor better monitoring results. For example,the platform allows a networkmonitoring solution and a grid monitoringsolution to work hand in handto provide network monitoring informationand therefore, accurate gridmonitoring results.GENERAL BENEFITSFigure 1 depicts the general benefitsfrom monitoring grid environments.1. To understand the performance,identify problems and to tune agrid system for better overall performance,2. Fault detection and recovery mechanisms;to determine if there areany units within the grid environmentthat are not functioning properly,and thus, to decide whetherto restart a component or redirecta service requests elsewhere,3. For debugging purposes,4. Resource utilization,5. Management decisions,6. Performance utilization,7. Accounting,8. Security.INET-GRIDiNet-Grid was built on top of EclipsePlatform as a Rich Client Platform35

Grid@USMFigure 1: Grid monitoring benefits(RCP) application that gives the flexibilityin deploying the end solutioninto various/any grid environmentsregardless of the grid architecture orthe operating system. The RCP providesthe power of plugins that canextend the system functionality easilyand quickly without refactoringthe whole core component from thescratch. Moreover, these plugins alsoprovide extension points for other pluginsto be integrated or add/enhanceany existing service (plugin) such asinterfacing to Ganglia and any othermonitoring tools.ARCHITECTUREiNet-Grid is one of the core modulesof iNet-Enterprise Suite and plays arole as the centralized grid monitoringmodule that retrieves and managesall the nodes information withina cluster send by a grid monitoringagent. Only one grid agent needsto be installed in a Linux-based machinewithin a cluster equipped withany grid monitoring solution such asGanglia. This monitoring agent willlook into Ganglia database, retrievethe data and send it to the centralmonitoring Server a specific definedtime interval (15 seconds by default).Figure 2 best visualizes the iNet-Grid’sarchitecture followed by its monitoringagent architecture in Figure 3.Furthermore, iNet-Grid consists ofthree components: grid agent, griddata manager and grid data viewer.A grid agent is installed in one machinefor each cluster. It collects andprocesses data loaded from the Gangliadatabase and sends it to griddata manager which resides in iNet-Server. Figure 4 portrays iNet-Grid’sfunctional architecture.The monitoring information consistsof the number of all incoming/outgoingpackets, CPU speed,CPU number, CPU usage load andhard disk used/free size. On theother hand, the network monitoringinformation consists of; bandwidthutilization, viruses/worms infectingor propagating within the network,the applications that are consumingbandwidth, traffic analyzers, etc. Thevalues are read from Ganglia databasewhich kept in round robin database(RRD) format files in the location /usr/share/ganglia/rrdtool/file/libs. Belowis the list of the files:• boottime.rrd • bytes_out.rrd• cpu_idle.rrd • cpu_num.rrd• cpu_system.rrd • cpu_wio.rrd• disk_total.rrd • load_five.rrd• mem_buffers.rrd • mem_free.rrd• mem_total.rrd • pkts_in.rrd• proc_run.rrd • swap_free.rrd• bytes_in.rrd • cpu_aidle.rrd• cpu_nice.rrd • cpu_speed.rrd• cpu_user.rrd • disk_free.rrd• load_fifteen.rrd • load_one.rrd• mem_cached.rrd • mem_shared.rrd• part_max_used.rrd • pkts_out.rrd• proc_total.rrd • swap_total.rrdCONNECTIONFUNCTIONALITYThe grid monitoring agent sends a requestto the central monitoring server.The central monitoring server thenchecks/verifies whether the requestis a valid request or not (the monitoringagent will verify his identity usinga unique ID that is pre-defined in thedatabase), if it is valid, the server willgive the authorization to the monitoringagent and allows him to sendthe data so it can be used for realtimeor historical monitoring information.When the data reaches tothe server, it will be stored into thedatabase. “Grid” table is where thedata is stored (Table 1 shows the tabledata dictionary).iNet-Console (which acts as aviewer) will send a grid data requestto the central monitoring server requestingfor grid monitoring information.Finally, the received grid informationwill be displayed to the viewer.The format of the data sent to theclient is as shown in Figure 5.RESULTSThe implementation of iNet-Grid prototypehas been done. We tested itwith 5 clusters and 90 nodes in total.From our evaluation we can concludethe effectiveness of the system. Thescreenshots taken during the prototypeare shown in Figure 6.EFFICIENT GRIDMONITORINGGrid monitoring is considered efficientif it is a combination of networkand grid monitoring, the networkmonitoring in term of bandwidth,anomaly, security, user misusage/misdealingwith the networkresources which obviously will resultin slow network, thus, minimizingthe grid productivity specially if it requirescommunication between differentparties over the network. On theother hand, grid monitoring will ensurethat the Grid resources are usedand managed wisely and efficiently36 INET-GRID

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 2: Grid monitoring general architectureFigure 3: Grid agent architectureTable 1: Data dictionary of the “Grid” tableColumn Name Comment Column Name CommentNode_name Holds the CPU ID value Time Holds the time valueBoot_time Holds the boot time value Cpu_idle Holds the CPU idle valueCpu_system Holds the CPU system value Disk_total Holds the disk total valueMem_buffers Holds the memory buffers value Mem_total Holds the memory total valueProc_run Holds the number of running processes Bytes_in Holds the bytes in valueCpu_nice Holds the CPU nice value Cpu_user Holds the CPU user valueLoad_fifteen Holds the load fifteen value Mem_cached Holds the cached memory valuePart_max_used Holds the part max used value Proc_total Holds the total processes valueBytes_out Holds the bytes out value Cpu_num Holds the number of CPUs valueCpu_wio Holds the CPU I/O value Load_five Holds the load five valueMem_free Holds the free memory value Pkts_in Holds the packets in valueSwap_free Holds the swap free value Cpu_aidle Holds the CPU idle valueCpu_speed Holds the CPU speed value Disk_free Holds the free disk valueLoad_one Holds the load one value Mem_shared Holds the shared memory valuePkts_out Holds the packets out value Swap_total Holds the total swap valueby keeping an eye on each individualCPU available in the grid and alertingwhen certain threshold is triggered.This is illustrated in Figure 7.The Grid and network monitoringis one tool that all modules are pluggableto each other on top of one coreengine that manages the communicationbetween the different availablemodules controlled by XML manifest.This XML manifest makes it easierto add/remove/enhance any existingfunctionality. As shown in Figure 8,each module is working/operatinghand in hand with others. The coreengine is hardwired with a circularbuffer using iNetmon technologyto support high speed network withhigh amount of traffic efficiently witha minimal packet loss. iNet-Grid is thecore of the Grid monitoring where ithas another pluggable interface for existinggrid monitoring tool integration(such as Ganglia) without any specialSUB-PROJECTS 37

Grid@USMFigure 5: Data format sent to the clientFigure 4: Functional architecture of iNet-GridFigure 6: Snapshots of the systemFigure 7: Efficient Grid Environment Monitoring Features38 INET-GRID

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 8: Efficient grid environment monitoring components and architectureFigure 9: Grid environmentmonitoring and deploymentconfiguration accept that for Ganglia(client) which is assumed to be readyand functioning as required. iNet-Grid client is installed with Gangliaclient to avoid installing the wholeGanglia package. Instead, the entiresystem will provide as a whole:1. Application usage statistics,2. Worm activity alerts,3. Bandwidth usage,4. Top users/usage,5. Nodes status,6. Disk usage,7. Packets per second,8. Cluster statistics,9. CPU load,10. RAM usage.DEPLOYMENTAs we described earlier, the gridnetworkwill have an intelligent agentfor the network part that is responsiblefor any network activity handin hand with a grid intelligent agentthat is in charge of each grid environmentresources. Figure 9 portrays thecombination between network monitoringand grid monitoring as well asillustrating the deployment plan.CONCLUSION &FUTURE WORKResource monitoring is very importantfor Grid environment. Many maturesystems exist today. However,none of them provides complete solution,which covers all requirementsneeded for efficient grid monitoring.With this we propose grid monitoringarchitecture namely iNet-Grid. Wefocus on following issues: integratingexisting systems on top of Ganglia,providing component for the managementof monitoring component anddesigning iNet Console for the GUI. Deploymentof the described monitoringsystems has already been successfullyaccomplished and the preliminary resultshave shown positive outcomes.PROJECT PUBLICATIONSAhmed, M. (2009). A Real TimeDistributed Network Monitoringand Security Monitoring Platform(RTDNMS). PhD thesis, School ofComputer Sciences, Universiti SainsMalaysia.Ahmed, M. & Talib, N. A. (2008).iNet-Grid: A Real-Time Grid Monitoringand Troubleshooting System.In Proceedings of the InternationalConference on Distributed Frameworks& Applications 2008 (DFmA 2008).Penang, Malaysia; pp. 68–72.U@SGridMSUB-PROJECTS 39

Image: Spread of diseases can be modeled mathematically. Illustration by ©Kroma Kromalsky (http://www.sxc.hu/profile/krominator).GRID APPLICATION TO WAVE FRONTPROPAGATION AND CONTAINMENTOF VECTOR BORNE DISEASESSchool of Mathematical SciencesTEAM MEMBERSProject Leader : Prof. Koh Hock LyeResearchers : Dr Teh Su Yean, Tan Kah BeeINTRODUCTIONVector borne diseases such asdengue, Yellow fever, malaria,West Nile Encephalitis (WNE),Japanese Encephalitis (JE), St. LouisEncephalitis (SLE) and WesternEquine Encephalitis (WEE) pose seriouspublic health hazard worldwide.Nearly half of the world populationis infected with at least one type ofvector borne diseases (WHO 2000).Approximately 1.4 million lives arelost each year due to vector-borne diseases.For most of these vector-bornediseases, vaccines are currently notavailable and are unlikely to be availablein the near future. Therefore,the focus is diverted to controllingthe spread of a disease by controllingthe vector of that disease. For thispurpose, a mathematical model isdeveloped by the team to study themechanisms involved in the invasionand persistence of vector borne diseases.Through model simulations,qualitative and quantitative assessmentscan be made to determinethe factors affecting the dispersal ofvector population and the transmissionof the disease from the vector tohuman. Further, control methods canbe experimented numerically on itseffectiveness before implementation.CLIMATE CHANGE IMPACTON DENGUEDengue fever is the world fastestgrowing vector borne disease whichis currently endemic or intermittentlyepidemic in many tropical andsubtropical regions. Current estimatessuggest that up to 50 milliondengue cases occur annually, including500,000 cases of dengue haemorrhagicfever (WHO 2001). Denguevirus is transmitted by arthropods ofthe species Aedes aegypti, a mosquitofound throughout the world wherehot and humid climate is predominant.The transmission of the denguevirus has only one epidemiologicalcycle linking the human host and femaleAedes aegypti mosquito. Fourmajor serotypes of dengue virus havebeen identified. A person who recoversfrom an infection is immune toonly that serotype and may becomesecondarily infected with a differentserotype virus (Atkinson et al. 2007).In the absence of vaccine, the effortsto control dengue disease are focusedon controlling the mosquito vector.Further, global warming may increasethe outbreak risk of dengue diseaseassociated with weather or precipitationpattern modifications (Schaeffer,Mondet, and Touzeau 2008). Climatechange may be irreversible, and is predictedto become more extreme in thefuture (IPCC 2001). To examine theglobal-scale relationships between climate,Aedes aegypti populations anddengue disease, grid technology is appropriate,where a grid will simulateeach local habitat.APPLICATION OF GRIDTECHNOLOGYWNE is a vector borne disease withmore than one epidemiological cyclelinking the human host, the mosquitovector and another mammal such asbirds and pigs. WNE started in New40

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 1: Conceptual computation gridFigure 2: Compartment model for Dengue disease transmission∂WS∂t∂WI∂t= DIFF × ∂2 WS∂x 2= DIFF × ∂2 WI∂x 2− v × ∂WS(∂x + CONV × A × 1 − WN )− ALPHAW × WS − DISW × BITR × WS × HICCWHN− v × ∂WI∂x(∂A∂t = BETA × WN × 1 − ACCAHI− ALPHAW × WI + DISW × BITR × WS ×HN)− ALPHAA × A − CONV × A∂HS= ALPHAH × HN − ALPHAH × HS − DISH × BITR × HS × WI∂tWN∂HI∂t∂HR= REMVH × HI − ALPHAH × HR∂t∂HD∂t= DISH × BITR × HS × WI − REMVH × HI − ALPHAH × HI − DEADH × HIWN= DEADH × HI (1)York in northeast USA in 1990 butquickly spread to the southwest ina matter of 3 to 4 years. The localscale of mosquito dispersion cannotaccount for this speed with which thedisease spread. The spread was indeedenhanced by infected birds thatmay fly over large distances of tensof kilometers over a season or a fewmonths. This pattern of dispersionmay be conceptualized in Figure 1,in which uneven patches of distributions(squares) spreading over a rectangleof 3000 km by 4000 km are usedto represent WNE distribution on thescale of continental USA. Over a localpatch (a square) of the order ofa few km, the PDE for mosquito willbe solved by finite difference methodwith mesh size of the order of 0.1 km,each patch requiring a total numberof meshes of the order of 10 to 100thousands nodes.For large system simulations, normalcomputing power offered by theregular PC would not be adequate.We therefore use the high power computingsuch as parallel or grid computingto provide the needed power.In this project, we develop grid technologyfor general applications towave front propagation simulations,with particular reference to vectorborne disease dispersal and transmission.The grid application developedhas been extended to tsunami wavepropagation and will be extended toother areas.DEER MODELDengue and Encephalitis EradicationRoutines (DEER) is a numerical simulationmodel developed in this projectto simulate the dynamics of vectorborne diseases transmission. DEER isdeveloped to simulate the distributionof Aedes aegypti mosquito populationand the dynamics of dengue diseasetransmission, which has only one epidemiologicalcycle linking the humanhosts and the vector mosquitoes.DEER is based upon a mass balanceequation for two phases of mosquitolifecycle: the winged female mosquito(represented by W) and the aquaticform (represented by A), which includeseggs, larvae and pupae. Fordisease transmission, the winged femalemosquitoes (W) are divided intotwo different classes, susceptible wing(WS) and infected wing (WI). Humansare another state variables andit has 4 different sub-variables, susceptiblehuman (HS), infected human(HI), recovered human (HR) anddead human (HD). The compartmentmodel for dengue disease transmissionis illustrated in Figure 2. Theequations governing the spatial andtemporal evolution of the disease inDEER are given in (1).Let the mortality rate of themosquitoes and aquatic forms berespectively ALPHAW (day −1 ) andALPHAA (day −1 ). The specific rateof development of the aquatic formsinto wing mosquitoes will be CONV(day −1 ). The rate of oviposition byfemale mosquito is BETA (day −1 ).We will consider the Aedes aegyptidispersal as the result of a ran-SUB-PROJECTS 41

Grid@USMFigure 3: Compartment model for West Nile EncephalitisFigure 4: Graphical user interface (GUI) for DEERdom flying movement representedby a diffusion process with coefficientDIFF (km 2 /day) and the constantvelocity flux of the wind advectionis v (km/day). The densityof wing mosquitoes and aquaticforms is limited by carrying capacityof wing mosquitoes CCW (mosquito)and aquatic forms CCA (mosquito).Wing female mosquitoes become infectedwhen they bite infected humans.The transmission probabilityfrom infected human hosts tosusceptible vector mosquitoes is DIS.Similarly, humans get dengue virusinfections from the bite of infectedwing female mosquitoes. DISH isthe transmission probability from infectedvector mosquitoes to susceptiblehuman hosts. REMVH is recoveryrate of human (day −1 ). DEADH isthe mortality rate of human causedby the virus. ALPHAH is the naturalmortality rate and birth rate of human(day −1 ). BITR is biting rate ofmosquito (day −1 ). AHOST is numberof alternative hosts available asblood sources. HN is the total humanpopulation (HN = HS + HI + HR).DEER is further enhanced to enablethe simulation of the transmission ofother vector borne diseases with morethan one epidemiological cycle suchas WNE and JE. WNE disease transmissioninvolves another mammal likebirds or horse in the epidemiologicalcycle linking human host andmosquito. To enable simulation ofWNE transmission, bird is included asa state variable in our model. Similarto the human state variable, it hasfour sub-variables such as susceptiblebirds (BS), infected birds (BI), recoveredbirds (BR) and dead birds (BD).Figure 3 shows the life cycle of WestNile Virus Disease that mainly circulatesbetween mosquitoes and birds.The virus is transmitted to human byinfected mosquito bites. Humans arethe end host in the life cycle of theencephalitis virus.GRID APPLICATIONMPI commands are incorporated intothe DEER algorithm so that some routinescan be performed in parallel. Wehave successfully run DEER using ourUSM Campus Grid with 22 nodes, theresults of which are presented anddemonstrated in PRAGMA 15 and 16.GRAPHICAL USER INTERFACEA web-based application will enableeasy access to job submission and resultsretrieval by users. The user cansubmit jobs using a web browser runningeither from Windows, Linux orMac OS X operating systems. Webbasedapplication enables simple retrievalof results and reduces datastorage size. The Microsoft .NetFramework is a platform that providestools and technologies we needto build Networked Applications aswell as Distributed Web Services andWeb Applications. For creating a userfriendlymodel for the web services,we develop the GUI for DEER modelin C#.NET. Figure 4 shows the simpleGUI panel for DEER.AN EXAMPLE APPLICATIONFor the purpose of presentation, astudy domain with 400 × 400 gridcells is used for the simulations. Thespatial and temporal distribution ofAedes aegypti mosquito population isshown in Figure 5. Model simulationbegins with a small concentrationof wing mosquito. The wingmosquitoes spread out to locationswith water containers, which providesites for the mosquitoes to lay eggs.A small number of mosquitoes arefrequently transported by vehicles toother locations. Here, we investigatethe effectiveness of insecticide sprayingon controlling the dispersal ofmosquitoes. We assume that insecticideis applied to kill the mosquitoesat a certain time. The amount of insecticidesapplied is then gradually reducedto zero. Upon application of insecticide,the density of mosquito populationdecreases with their dispersalcontained. However, the density anddispersal of the mosquitoes eventuallyreturn to the pre-treatment level.Figure 6 shows the physical computationaltime consumed subject to the42 GRID APPLICATION TO WAVE FRONT PROPAGATION AND CONTAINMENT OF VECTOR BORNE DISEASES

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsFigure 5: Mosquito Population DistributionFigure 6: Computational time using different number of nodesFigure 7: Memory required for different research areaFigure 8: Dengue disease transmissionSUB-PROJECTS 43

Grid@USMTable 1: DEER computational time on a single PC and cluster with 10 and 20 nodesSegmentsComputational timeSingle PC 10 nodes 20 nodesCalculation 6 hrs 50 mins 40 minsPrinting 3 hrs 60 mins 20 minsTotal 9 hrs 1 hr 50 mins 1 hrvarious numbers of nodes used. Thereduction of time used becomes saturatedafter 12 nodes. This is becausethe execution speed of the programis limited by the system input andoutput latencies such as file accessand message passing over network.Figure 7 shows the memory storageneeded for several sizes of study domain.The memory size needed fordata storage increases exponentiallyas the calculation domain increases.To examine the global-scale relationshipsamong climate, Aedes aegyptipopulations and transmissionof dengue disease, a large study domaintypically in terms of thousandsof kilometers is needed. This incursexcessive demands on the computationaltime and storage size as thecomputational grid size of the studydomain must be small enough to reflectthe local distribution and flightreach of Aedes aegypti, which is onlya few hundred meters (Reiter et al.1995). Therefore, grid technology isused to speed up the computationaltime and to reduce storage size. Thisis achieved by dividing and allocatingsections of program computation toseveral computers.REGIONAL DENGUESIMULATION BY GRIDA global scale of dengue disease transmissionis considered in this studyto demonstrate the capability of gridcomputing in reducing the computationalcost of simulating global transmissionof dengue. Table 1 showsthe computational time used to run aglobal case of dengue transmission fora single PC and clusters with 10 and20 nodes. A single simulation takesabout 9 hours to run on a single normalPC of Intel® Pentium® 1.60 GHzand 768 MB RAM. Using a clusterwith 20 nodes reduces the computationaltime to about half an hour;a total time reduction of more than80 %.DIRECTION OFFUTURE RESEARCHDEER is undergoing continuous enhancementsto upgrade its capabilityin simulating the dynamics of denguedisease transmission. Further, DEER isundergoing revisions for applicationon the PRAGMA Grid network. Wehope that this project would stimulateactive research and collaborationin this region. Further we have successfullyextended grid technology totsunami simulations, which typicallyrequire large computational resources,made available by Grid technology.PROJECT PUBLICATIONSKew, L. M., Teh, S. Y., & Koh, H. L.(2009). Optimization of TsunamiModel TUNA by Grid Technology.In Proceedings of the 5th Asian MathematicalConference (AMC). KualaLumpur, Malaysia.Koh, H. L., Lee, H. L., Teh, S. Y.,& Izani, A. (2009). Dengue andTsunami Modeling: Application ofGrid Technology. In Proceedingsof 2nd Regional Conference on Ecologicaland Environmental Modeling(ECOMOD 2007). Penang, Malaysia;pp. 22–28.Tan, K. B., Teh, S. Y., Koh, H. L.,Sui, L. L., Bahari, B., & Izani, A.(2009). Modeling West Nile Viruswith Grid Technology. In Proceedingsof the 16th Pacific-Rim ApplicationAnd Grid Middleware Assembly(PRAGMA 16). Daejeon, Korea.Teh, S. Y., Kew, L. M., & Koh, H.(2008). Application of Grid Computingin Modeling Tsunami andDengue. In ICTP Advanced Schoolin High Performance and GRID Computing.Trieste, Italy.REFERENCESAtkinson, M. P., Su, Z., Alphey, N.,Alphey, L., Coleman, P. G., & Weln,L. M. (2007). Analyzing the Controlof Mosquito-borne Diseases bya Dominant Lethal Genetic System.Proceedings of the NationalAcademy of Sciences of the UnitedStates of America (PNAS), 104(22),pp. 9540–9545.Intergovernmental Panel for ClimateChange. (2001). Summary for policymakers climate change 2001: the scientificbasis. Cambridge UniversityPress.Reiter, P., Amador, M. A., Anderson,R. A., & Clark, G. G. (1995). ShortReport: Dispersal of Aedes aegyptiin Urban Area after Blood Feedingas Demonstrated by Rubidium-Marked Eggs. American Journal ofTropical Medicine and Hygiene, 52,pp. 177–179.Schaeffer, B., Mondet, B., & Touzeau,S. (2008). Using a Climate-Dependent Model to PredictMosquito Abundance: Applicationto Aedes (Stegomyia) africanusand Aedes (Diceromyia) furcifer(Diptera: Culicidae). Infection, Geneticsand Evolution, 8, pp. 422–432.World Health Organization. (2001).Chiang Mai Declaration onDengue/Dengue HaemorrhagicFever. Wkly Epidemiol Rec,pp. 29–30.World Health Organization. (2000).World Health Report 2000 – HealthSystems: Improving Performance.World Health Organization.U@SGridM44 GRID APPLICATION TO WAVE FRONT PROPAGATION AND CONTAINMENT OF VECTOR BORNE DISEASES

Image: Fireworks. Photograph by ©G Schouten de Jel (http://www.sxc.hu/profile/gabriel77).PROJECT ACTIVITIES

Image: A conference room. Photograph by ©Justyna Furmanczyk (http://www.sxc.hu/profile/just4you).ORGANISED EVENTSAPPLE CONTENT CREATION WORKSHOPThis workshop was jointly organised by the School of Computer Sciences (PPSK) and OpenSOSSdn. Bhd. on 26–28 February 2008 at PPSK. The objective of the workshop is to learn how technologyengages students in exploring their natural curiosity of the world while learning the core curriculum.Mr Ahmad Shaharuddin Amin b. Sahar, the facilitator, guided the participants from familiarisingthemselves with a Mac to using applications in the iLife ’08 suite.INTRODUCTORY WORKSHOP TO GRID COMPUTINGAP Chan Huah Yong organised and facilitated this one-day workshop on 1–2 March, 2008 at PPSK,assisted by researchers from the Grid Computing Lab (GCL). All ROs and RAs from the sub-projectteams were invited to get a first taste of Grid computing technologies, especially from an end-userpoint perspective. The workshop included an introductory talk on Grid computing and hands-onsessions on grid credentials, the Globus toolkit and the GridSphere portal framework.WORKSHOP ON GRID COMPUTINGSTRATEGIES & SECURITY ISSUESHeld in Pulau Perhentian on 2–4 July, 2008, thisworkshop focusing on grid security and monitoring.During a co-located session, ROs and RAsfrom the respective sub-project teams updatedeach other on their progress, and received trainingon the Trac Wiki system. We would later use Tracto post information and progress updates on theirrespective sub-projects in one centralized repository.46

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsTHE 2ND INTEL MULTICORE WORKSHOPThis event was co-organised with Intel and held in Bangunan Eureka on 22–25 July 2008. AP ChanHuah Yong hosted a workshop including a hands-on session there.PRAGMA INSTITUTE, PRAGMA 15 & DFMA 2008The PRAGMA 15 Workshop was held and hosted byUSM on 22–24 October, 2008, while PRAGMA Institutemet at the same venue earlier on the 21 & 22. Allsub-projects were selected for poster exhibition, whileAP Chan Huah Yong, AP Phua Kia Kien and Prof. KohHock Lye’s projects were chosen for showcase demonstrations.In addition, The International Conference on Distributed Frameworks & Applications2008 (DFmA 2008) was also held in conjunction on the 21 & 22. DFmA 2008 accepted 4 papers fromGrid@USM members.MICROSOFT HIGH PERFORMANCECOMPUTING (HPC) WORKSHOPThis 3-day workshop was held at the School of ComputerSciences, USM on 8–10 April 2009, organised by AP ChanHuah Yong. He was also joined by Mr Ang Sin Keat and MsKheoh Hooi Leng.Participants were introduced to the infrastructure of WindowsHigh Performance Computing (HPC) Server 2008 infrastructure,as well as deploying and administrating a WindowsHPC Server 2008 Cluster. Topics on the second day includednetwork configuration, diagnostics, performance tuning andscheduling jobs. The third and final day saw the participantsdeveloping C/C++ MPI applications with Visual Studio andMPI.NET.U@SGridMPROJECT ACTIVITIES 47

Image: a conference room. Photograph by ©Ann Petersen (http://www.sxc.hu/profile/Lilja).OTHER EVENTSPRAGMA 14AP Bahari Belaton was present at PRAGMA 14 heldin March 2008 in Taichung, Taiwan. He and DatoProf. Muhd. Idiris, then USM’s Deputy Vice Chancellorfor Research and Innovation, were invitedas co-chairs and took the opportunity to promoteand raise awareness about PRAGMA 15, which wasto be held in USM in October the same year.COURSE ON ENABLING GRIDS FORE-SCIENCEMr Aloysius Indrayanto and Ms Kheoh Hooi Lengattended this training course at MIMOS Berhad inKuala Lumpur on 10–14 December, 2007.PBS TRAINING AND GRIDING OFUNIVERSITIESMr Tan Choo Jun and Ms Tan Chin Min participatedin this training held at International IslamicUniversity Malaysia (IIUM) on 5–7 May 2008.ICTP ADVANCED SCHOOL IN HIGHPERFORMANCE AND GRIDCOMPUTINGDr Teh Su Yean attended this course at the AbdusSalam International Centre for Theoretical Physics(ICTP), Trieste, Italy on 3–14 Nov 2008.48

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsTHE 1ST KNOWLEDGE GRID MALAYSIA FORUM 2009This event was held on 18-19 March, 2009 at the IIUM. AP Chan Huah Yong was invitedto conduct a hands-on session entitled “USM Grid (Globus)”. Three USM ROs i.e. MrAng Sin Keat, Ms Tan Kah Bee and Ms Sui Lee Lin also attended the two-day forum.SCOPE ICT CAREER FAIRThis event was held in conjunction with the SCoPE ICTExpo on 18 March 2009 at Suntech Penang Cybercity.Grid@USM members exhibited 6 posters about theirsub-projects at rented booths at the fair, including:❦ AP Chan Huah Yong• Virtual Distributed Database Management System• Mobile Desktop Grid• Campus Grid Portal❦ Dr Vincent Khoo Kay Teong• Enterprise Grid Architecture for CollaborativeB2B Information Interchange❦ Prof. Koh Hock Lye• In-house Tsunami Propagation Model TUNA-M2: Application of Grid• Grid Application: Modelling Mosquitoes DistributionROs attending the event include Mr (Mr) JohnsonFoong, Mr Kor Chan Hock, Mr Aloysius Indrayantoand Mr Tan Choo Jun.PRAGMA 16 & IDRICPRAGMA 16 was held on 23–26 March 2009 at DCC, Deajon, Korea, in conjunction with theInternational Workshop on Infectious Disease Researches in Cyberinfrastructure (IDRiC).4 posters from USM participated in PRAGMA 16:❦ Prof. Koh Hock Lye• Modeling Dengue Infection for Malaysia• Modelling West Nile Virus with Grid Technology❦ AP Chan Huah Yong – Virtual Distributed Database Management System❦ Mr Mohd. Azam Osman – Mobile Desktop GridU@SGridMPROJECT ACTIVITIES 49

Image: Books at a library. Photograph by ©Carlo Lazzeri (http://www.sxc.hu/profile/mvttley).PUBLICATIONSJOURNAL ARTICLESPhua, K. K. & Wong, V. C. (2007).Developing A Compelling 3DAnimation and Multimedia PresentationUsing Blender for IN-FORMM’s Opening Ceremony.Malaysian Journal of Medical Sciences,14(Supplement 1).Al-Mistarihi, H. H. E. & Chan, H. Y.(2009). On Fairness, OptimizingReplica Selection in Data Grids.IEEE Transactions on Parallel &Distributed Systems. (forthcoming).Chan, H. Y. & Al-Mistarihi, H. H. E.(2008). Replica Management inData Grid. International Journalof Computer Science and NetworkSecurity.Chan, H. Y. & Al-Mistarihi, H. H.E. (2008). Replica OptimizationService in Data Grids. SciencesPublications.Younis, M. I., Zamli, K. Z., Klaib, M.F. J., Che Soh, Z. H., Che Abdullah,S. A., & Mat Isa, N. A. (2009).Assessing IRPS as an EfficientPairwise Test Data GenerationStrategy. International Journal ofAdvanced Intelligence Paradigms.(forthcoming).Younis, M. I., Zamli, K. Z., & MatIsa, N. (2008). IRPS – An EfficientTest Data Generation Strategy forPairwise Testing. Lecture Notes inArtificial Intelligence, 5177.Zafri, M., Fadhilah, N. K., & Phua,K. K. (2008). Using Xgrid to ImproveFASTA Efficiency for Alignmentof Multiple DNA and ProteinSequences. Malaysian Journalof Medical Sciences, 15(Supplement1).CONFERENCE & WORKSHOPPROCEEDINGSZafri, M., Sanjay, K. C., & Phua, K.K. (2009). Implementation Methodsfor Estimating Haplotypeswith GRID Computing Technology.In Compendium of Abstractsfor the 14th National Conferenceon Medical and Health Sciences(NCMHS).Ahmed, M. & Talib, N. A. (2008).iNet-Grid: A Real-Time Grid Monitoringand Troubleshooting System.In Proceedings of the InternationalConference on DistributedFrameworks & Applications 2008(DFmA 2008). Penang, Malaysia;pp. 68–72.Indrayanto, A. & Chan, H. Y. (2008).Application of Game Theoryand Fictitious Play in Data Placement.In Proceedings of the InternationalConference on DistributedFrameworks & Applications 2008(DFmA 2008). Penang, Malaysia;pp. 79–83.Kew, L. M., Teh, S. Y., & Koh, H. L.(2009). Optimization of TsunamiModel TUNA by Grid Technology.In Proceedings of the 5th AsianMathematical Conference (AMC).Kuala Lumpur, Malaysia.Koh, H. L., Lee, H. L., Teh, S. Y., &Izani, A. (2009). Dengue andTsunami Modeling: Applicationof Grid Technology. In Proceedingsof 2nd Regional Conference onEcological and Environmental Modeling(ECOMOD 2007). Penang,Malaysia; pp. 22–28.50

Grid Computing Cluster – the Development and Integration of Grid Services and ApplicationsLim, L. T. & Schwab, D. (2008).Limits of Lexical Semantic Relatednesswith Ontology-basedConceptual Vectors. In Proceedingsof the 5th International Workshopon Natural Language Processingand Cognitive Science (NLPCS’08).Barcelona, Spain; pp. 153–158.Schwab, D. & Lim, L. T. (2008).Blexisma2: a Distributed AgentFramework for Constructing a SemanticLexical Database based onConceptual Vectors. In Proceedingsof the International Conference onDistributed Frameworks & Applications2008 (DFmA 2008). Penang,Malaysia; pp. 102–110.Tan, K. B., Teh, S. Y., Koh, H. L., Sui,L. L., Bahari, B., & Izani, A. (2009).Modeling West Nile Virus withGrid Technology. In Proceedingsof the 16th Pacific-Rim ApplicationAnd Grid Middleware Assembly(PRAGMA 16). Daejeon, Korea.Teh, S. Y., Kew, L. M., & Koh, H.(2008). Application of Grid Computingin Modeling Tsunami andDengue. In ICTP Advanced Schoolin High Performance and GRIDComputing. Trieste, Italy.Ting, T. T. & Khoo, V. K. T. (2009).B2B Standardized Information InterchangeChallenges – A Studyon Standardization versus Personalization.In Proceedings ofthe 2009 International Conferenceon Advanced Management Science(ICAMS 2009); pp. 244–248.Ting, T. T. & Khoo, V. K. T. (2008).Personalizable Information InterchangeArchitecture for EducationalInstitutions. In Proceedings ofthe 3rd International Conference one-Commerce with Focus on DevelopingCountries (ECDC’08). Isfahan,Iran.Younis, M. I., Zamli, K. Z., & MatIsa, N. A. (2008). A Strategyfor Grid Based t-Way Test DataGeneration. In Proceedings of theInternational Conference on DistributedFrameworks & Applications(DFmA 2008). Penang, Malaysia;pp. 73–78.THESESAhmed, M. (2009). A Real TimeDistributed Network Monitoringand Security Monitoring Platform(RTDNMS). PhD thesis, Schoolof Computer Sciences, UniversitiSains Malaysia.Al-Mistarihi, H. H. E. (2009). AData Grid Replica ManagementSystem with Local and GlobalMulti-Objective Optimization. PhDthesis, School of Computer Sciences,Universiti Sains Malaysia.Indrayanto, A. (2009). Initial File-Placement in Data Grid Environmentusing Game Theory andFictitious Play. MSc thesis, Schoolof Computer Sciences, UniversitiSains Malaysia.BOOKSZamli, K. Z., Che Abdullah, S. A.,Younis, M. I., & Che Soh, Z. H.(2009). Software Testing Module.(forthcoming). OpenUniversityMalaysia and Pearson Publishing.U@SGridMImage: A huge pile of books. Photograph by ©Sanja Gjenero(http://www.sxc.hu/profile/lusi).PROJECT ACTIVITIES 51

Grid@USMGrid computing can provide easy and transparent access tohigh performance computing by aggregating geographicallydistributed resources to act as a single powerful resource.With grid computing technologies, users can access to anyvirtual resources such as computers, special instruments, softwareapplications and databases in a seamless manner.Grid computing is by its nature highly distributed geographically,consists of highly specialised equipment (storage, gridengines and management tools), as well as expensive. It isas such an ideal example of common, core computational resourcesto be created and pooled among aspiring researcherswho require high performance resources. Equally important,Universiti Sains Malaysia (USM) as a research university (RU)needs a focused, holistic and dedicated effort to build a sustainablehuman capital in grid computing.It is with USM’s R&D needs and sustainability in mind thatGrid@USM, a grid computing research cluster to address theproblems mentioned above, came into being. Grid@USM’spilot project, led by AP Bahari Belaton and entitled Grid ComputingCluster – the Development and Integration of Grid Servicesand Applications, was initiated under an RU research grant.This book records our achievements thus far, in the form oftechnical reports on 8 sub-projects. These sub-projects includeboth grid middleware and tool enablers, as well asapplications from various domains, spanning grid infrastructuresetup, grid monitoring, virtual distributed databases, 3Drendering, natural language processing, B2B standards componentmodeling, software testing and disease modeling.Platform for Information & Communication Technology ResearchUniversiti Sains Malaysia, 11800 USM, Penang, MalaysiaISBN 978-983-3986-58-39 789833 986583