Intelligent Monitoring System for the Optimization

IntelligentMonitoringSystemfortheOptimizationoftheOperationofSystemsinResourceInsufficientEnvironmentGáborSamu,AnnamáriaR.Várkonyi-KóczyIntegratedIntelligentSystemsJapanese-HungarianLaboratoryDepartmentofMeasurementandInformationSystems,BudapestUniversityofTechnologyandEconomics,Magyartudósokkörútja2.,H-1117Budapest,Hungary,Abstract:Agentsoperatinginreal-timedomainhavetoencounterresourceinsufficiencyinsomecasesandapplicationsevenwithdeliberativeandawaredesign.Theresourcedemandsmayvarydependingontheoperationphases,changein/oftheenvironmentandespeciallyinalertconditionsandinsituationscausedbysuddenevents.Astronghardwarebasecanberequiredforoperatingindynamicallychangingenvironment.Somereal-timeagentsapplyingsophisticatedAIalgorithmshavetocomplywithtimingconstraintsandcontinuousoperationaswell,leavingthemtomakeacompromisebetweentheconsumptionofthecontinuousandsufficientcomputationtimeandoneormoreperformanceproperties.Anytimealgorithmsprovideatradeoffbetweenthetimeresourceandcomputationalperformance,namelythequalityoftheresults.Thiskindofbehaviourisfavorableinsomedomains,i.e.alessoptimaldecisioncanbebetterthanamissingone.Incaseofcomplexsystemsusuallyadecompositionisappliedduringthedesignprocess,thusfortheconstructionofausablesystemthecompilationoftheoperationalunits(modules)isalsoneeded.Theuseofanytimealgorithmsasoperationalunitsassumesacomplexsoftwareframe,anoperationsystemwhichcontainsaso-calledschedulercapableofdistributingthecomputationtimeamongtheanytimeandnon-anytimealgorithms.Thisschedulerwhichiscalledasmonitorinanytimesystemshastodealwiththetimingconsiderationsofmanyportionsoftheunderlyingoperationsystemandevenitsownrun-timecharacteristicstodeterminethetimeallocationswhiletrackingtheperformanceofthealgorithmsandthestateoftheenvironment.Thereisagapbetweentheimplementationoftheseanytimesystemsandthetheoreticalresultsasthecompilationandthemonitoringhave.Asurveyofanewcompilationtechnique,amonitoringscheme,andanimplementationapproacharecoveredbythispaper.Keywords:Anytimesystems,anytimecompilation,hierarchicalcompilation,activemonitoring,implementation1IntroductionThestateoftheartofthediagnosticssystemsappliedincontinuoustechnologiesisabletoquicklyidentifyrun-timefailuresarisinginthegiventechnologyandcanneutralizethembetweencertainlimits.Thisbehaviorassumesvariousinformationprocessingtaskssolvedbyacomputeratreal-timerespectivelyatawelldefinedresponsetime.Serioustimeand/ordatadeficiencyappearinginjustthecriticaloperationalphasesisunavoidableevenbyadeliberativedesign,causingmalfunctionsinthediagnosticssystem.Applyinggeneralizedanytimealgorithmsmayoffertheproperget-outfromtheproblemsabove.Theycanprovideresultswithlessbutstillacceptablequalitybyadaptingtotheavailabletimeandprocessingcapacitiesattheactualstateinordertosustaincontinuousoperationevenincriticalsituations.Systemspossessingthispropertyneedanintelligentmonitorasanadd-inwhichtriestooptimizetheoperationofthecompletesupervisedsystembyusingsensoryinformationaboutthecurrentstate.Complexsystemsusuallyconsistofsmallersubsystemswhicharecalledaselementarymodules.Forcomplexsystemsoptimizationmeansthedistributionofthetimeallocationsgiventotheabove-mentionedelementarymodules(computationalelements)astooptimizetheoverallperformanceofthewholesystem.Forthispurposespecialdatabasesareconstructedforthemonitorbycompilationmethodswhichdealwithspecialpropertiesoftheelementarymodules.Somepropertiesarecommonhencecompilationmethodscanbeacceleratedbyalgorithmsbasedontheconsequencesoftheseproperties.Inthispaperthetheoreticalbasicsofanewcompilationmethod,thehierarchicalcompilation,andtheactivemonitoringschemebasedonthisproposedtechniqueareintroducedbesidearoughdescriptionofanimplementationapproach.2AnytimecomputationIfanalgorithmcanbeexecutedfordifferentrunningtimesandalongertimeallocationcausesthealgorithmgivinganoutputofbetterqualityinsomeaspectthenitiscalledasananytimealgorithm.Anytimealgorithmsarecharacterizedbyrelationsinwhichtheexecutiontime,thequalityoftheinput(s)andoutput(s)maybeinvolved.Therelationsarerepresentedasone-,two-,andsometimesmultidimensionaltables.Theserelations(orrathermappings)arecalledperformanceprofiles.(See[1].)Qualityisthemeasureofthe“goodness”ofanygivenobjectinsomeaspectaccordingtothepossiblevaluesofthegivenobject.Qualitycanbedefinednotonlyforsimplenumbersbutalsoforcomplexandabstractstructures.E.g.apathgivenbyapath-planningalgorithmmaybequalifiedinseveralways.Onecandefineaqualitybasedonthelengthoftheactualpathbysimplyusingtheratiooftheminimalandtheactuallengths.(Minimallengthmaybeobtainedfrommaximalspeed.)Thequalityfunctionisamappingfromobjectvaluestoqualityvalues.Togetherwithit,the(timedependent)outputfunctionandameasurementmethod,performanceprofilescanbeconstructed.2.1ProfiletypesSeveraltypesofprofilescanbedefinedandconstructed.Themainaspectofdistinctionisthatwhichparameterispresentasapartoftheinputintherelation:inputqualityand/oroutputquality.Ifinputqualityisusedthentheprofilewillbeconditional.Theinput-outputrelationmay