Human exploration and development of space using XML database Space Wide Web Space Wide Web by adapt.docx

Human exploration and development of space using XML database Space Wide Web Space Wide Web by adapt.docx

《Human exploration and development of space using XML database Space Wide Web Space Wide Web by adapt.docx》由会员分享，可在线阅读，更多相关《Human exploration and development of space using XML database Space Wide Web Space Wide Web by adapt.docx（9页珍藏版）》请在冰豆网上搜索。

HumanexplorationanddevelopmentofspaceusingXMLdatabaseSpaceWideWebSpaceWideWebbyadapt

TheTLB（TranslationLookasideBuffer）missserviceshavebeenconcealedfromoperatingsystems,butsomenewRISCarchitecturesmanagetheTLBinsoftware.Sincesoftware-managedTLBsprovideflexibilitytoanoperatingsysteminpagetranslation,theyareconsideredanimportantfactorinthedesignofmicroprocessorsforopensystemenvironments.However,software-managedTLBssufferfromlargermisspenaltythanhardware-managedTLBs,sincetheyrequiremoreextracontextswitchingoverheadthanhardware-managedTLBs.

Thispaperintroducesanewtechniqueforreducingthemisspenaltyofsoftware-managedTLBsbyprefetchingnecessaryTLBentriesbeforebeingused.Thistechniqueisnotinherentlylimitedtospecificapplications.ThekeyofthisschemeistoperformtheprefetchoperationstoupdatetheTLBentriesbeforefirstaccessessothatTLBmissescanbeavoided.Usingtrace-drivensimulationandaquantitativeanalysis,theproposedschemeisevaluatedintermsofthemissrateandthetotalmisspenalty.OurresultsshowthattheproposedschemereducestheTLBmissratebyafactorof6%to77%duetoTLBcharacteristicsandpagesizes.Inaddition,itisfoundthatreducingthemissratebytheprefetchingschemereducesthetotalmisspenaltyandbustrafficsinsoftware-managedTLBs.

MostPrologmachineshavebeenbasedonspecializedarchitectures.Ourgoalistostartwithageneral-purposearchitectureanddetermineaminimalsetofextensionsforhigh-performancePrologexecution.Wehavedevelopedboththearchitectureandoptimizingcompilersimultaneously,drawingonresultsofpreviousimplementations.WefindthatmostProlog-specificoperationscanbedonesatisfactorilyinsoftware;however,thereisacrucialsetoffeaturesthatthearchitecturemustsupporttoachievethebestPrologperformance.Inthispaper,thecostsandbenefitsofspecialarchitecturalfeaturesandinstructionsareanalyzed.Inaddition,westudytherelationshipbetweenthestrengthofcompileroptimizationandthebenefitofspecializedhardware.WedemonstratethatourbasearchitecturecanbeextendedtoincludeexplicitsupportforPrologwithmodestincreaseinchiparea（13%）,andyetattainasignificantperformancebenefit（60–70%）.Experimentsusingoptimizedcodethatapproximatestheoutputoffutureoptimizingcompilersindicatethatspecialhardwaresupportcanstillprovideaperformancebenefitof30–35%.Themicroprocessordescribedhere,theVLSI-BAM,hasbeenfabricatedandincorporatedintoaworkingtestsystem.

Itiswellknownthatsoftwaremaintenanceandevolutionareexpensiveactivities,bothintermsofinvestedtimeandmoney.Reverseengineeringactivitiessupporttheobtainmentofabstractionsandviewsfromatargetsystemthatshouldhelptheengineerstomaintain,evolveandeventuallyre-engineerit.Twoimportanttaskspursuedbyreverseengineeringaredesignpatterndetectionandsoftwarearchitecturereconstruction,whosemainobjectivesaretheidentificationofthedesignpatternsthathavebeenusedintheimplementationofasystemaswellasthegenerationofviewsplacedatdifferentlevelsofabstractions,whichletthepractitionersfocusontheoverallarchitectureofthesystemwithoutworryingabouttheprogrammingdetailsithasbeenimplementedwith.

InthiscontextweproposeanEclipseplug-incalledMARPLE（MetricsandArchitectureReconstructionPlug-inforEclipse）,whichsupportsboththedetectionofdesignpatternsandsoftwarearchitecturereconstructionactivitiesthroughtheuseofbasicelementsandmetricsthataremechanicallyextractedfromthesourcecode.ThedevelopmentofthisplatformismainlybasedontheexploitationoftheEclipseframeworkandplug-insaswellasofdifferentJavalibrariesfordataaccessandgraphmanagementandvisualization.Inthispaperwefocusourattentiononthedesignpatterndetectionprocess.

Accesstosufficientresourcesisabarriertoscientificprogressformanyresearchersfacinglargecomputationalproblems.Gainingaccesstolarge-scaleresources（i.e.,university-wideorfederallysupportedcomputercenters）canbedifficult,giventheirlimitedavailability,particulararchitectures,andrequest/review/approvalcycles.Simultaneously,researchersoftenfindthemselveswithaccesstoworkstationsandolderclustersoverlookedbytheirownersinfavorofnewerhardware.SoftwaretotietheseresourcesintoacoherentGrid,however,hasbeenproblematic.Here,wedescribeourexperiencesbuildingaGridcomputingsystemtoconductalarge-scalesimulationstudyusing“borrowed”computingresourcesdistributedoverawidearea.Usingstandardsoftwarecomponents,wehaveproducedaGridcomputingsystemcapableofcouplingseveralhundredprocessorsspanningmultiplecontinentsandadministrativedomains.Webelievethatthissystemfillsanimportantnichebetweenacloselycoupledlocalsystemandaheavyweight,highlycustomizedwideareasystem.

ArticleOutline

1.Introduction

2.Scientificcontext

3.Implementation

3.1.Systemconstraints

3.2.Generaldesignofthegridsystem

3.3.Systemrequirements

3.3.1.Operatingsystem

3.3.2.Client

3.3.3.Server

3.3.4.Account

3.4.Systemprocesses

3.4.1.userlevelprocesses

3.4.2.grid_clientprocesses

3.4.3.projectprocesses:

executedonceperinvocationbygrid_clientprocess

3.5.Basicfeatures

3.5.1.Client–servercommunications

3.5.2.Authentication

3.5.3.Architecturespecificbinaries

3.5.4.Client-sidesecurity

3.5.5.Server-sidesecurity

3.5.6.Systemmonitoring

3.5.7.Errorhandling

4.Performanceconsiderations

5.Futurework

5.1.Securecommunications

5.2.SQLtransactionsupport

5.3.Alittlelanguage

5.4.Validitychecking

6.Conclusions

Acknowledgements

References

Vitae

ThispaperdescribesthearchitectureofthefirstimplementationoftheIn-VIGOgrid-computingsystem.ThearchitectureisdesignedtosupportcomputationaltoolsforengineeringandscienceresearchInVirtualInformationGridOrganizations（asopposedtoinvivoorinvitroexperimentalresearch）.AnovelaspectofIn-VIGOistheextensiveuseofvirtualizationtechnology,emergingstandardsforgrid-computingandotherInternetmiddleware.InthecontextofIn-VIGO,virtualizationdenotestheabilityofresourcestosupportmultiplexing,manifoldingandpolymorphism（i.e.tosimultaneouslyappearasmultipleresourceswithpossiblydifferentfunctionalities）.Virtualizationtechnologiesareavailableoremergingforalltheresourcesneededtoconstructvirtualgridswhichwouldideallyinherittheabovementionedproperties.Inparticular,thesetechnologiesenablethecreationofdynamicpoolsofvirtualresourcesthatcanbeaggregatedon-demandforapplication-specificuser-specificgrid-computing.Thischangeinparadigmfrombuildinggridsoutofphysicalresourcestoconstructingvirtualgridshasmanyadvantagesbutalsorequiresnewthinkingonhowtoarchitect,manageandoptimizethenecessarymiddleware.ThispaperreviewsthemotivationforIn-VIGOapproach,discussesthetechnologiesused,describesanearlyarchitectureforIn-VIGOthatrepresentsafirststeptowardstheendgoalofbuildingvirtualinformationgrids,andreportsonfirstexperienceswiththeIn-VIGOsoftwareunderdevelopment.

ArticleOutline

1.Introduction

2.TheIn-VIGOconcept

3.VirtualizationinIn-VIGO

3.1.Virtualdataandthevirtualfilesystem

3.2.Virtualmachines

3.3.Virtualapplications

3.4.Virtualnetworks

3.5.Virtualuserinterfaces

4.ThearchitectureofIn-VIGO

4.1.Thevirtualapplication

4.2.Thevirtualfilesystem

4.3.Theresourcemanager

4.4.Theuserinterfacemanager

4.5.Theglobalinformationsystem

4.6.Theusermanager

5.Implementation

6.Conclusions

Acknowledgements

References

Vitae

Leveragingcostmatrixstructureforhardwareimplementationofstereodisparitycomputationusingdynamicprogramming  OriginalResearchArticle

ComputerVisionandImageUnderstanding

ArticleOutline

1.Introduction

2.Relatedworks

2.1.Designpatterndetection

2.2.Softwarearchitecturereconstruction

2.3.Concludingremarks

3.AnoverviewonMARPLE

3.1.Theinformationdetectorenginemodule

3.2.TheJoinermodule

3.3.Theclassifiermodule

3.4.Thesoftwarearchitecturereconstructionmodule

3.5.DistributedMARPLE

4.ExperimentalresultsforDPD

4.1.Resultsfortheinformationdetectorenginemodule

4.2.ResultsfortheJoinermodule

4.3.Resultsfortheclassifiermodule

4.3.1.Comparisonwithothertools

4.3.2.Resultsonotherdesignpatterns

4.4.ResultsfortheSARmodule

5.Conclusionsandfutureworks

Acknowledgements

References

Atoolfordesignpatterndetectionandsoftwarearchitecturereconstruction  OriginalResearchArticle

InformationSciences

packageofLinuxscriptsfortheparallelizationofMonteCarlosimulations  OriginalResearchArticle

ComputerPhysicsCommunications

Despitethefactthatfastcomputersarenowadaysavailableatlowcost,therearemanysituationswhereobtainingareasonablylowstatisticaluncertaintyinaMonteCarlo（MC）simulationinvolvesaprohibitivelylargeamountoftime.Thislimitationcanbeovercomebyhavingrecoursetoparallelcomputing.Mosttoolsdesignedtofacilitatethisapproachrequiremodificationofthesourcecodeandtheinstallationofadditionalsoftware,whichmaybeinconvenientforsomeusers.Wepresentasetoftools,namedclonEasy,thatimplementaparallelizationschemeofaMCsimulationthatisfreefromthesedrawbacks.InclonEasy,whichisdesignedtorununderLinux,asetof“clone”CPUsisgovernedbya“master”computerbytakingadvantageofthecapabilitiesoftheSe