Security Metrics in Industrial Control Systems · of the system, intended uses and unintended...

24
In: Colbert, E. and Kott, A. (eds.), "Cyber Security of Industrial Control Systems, Including SCADA Systems," Springer, NY, 2016 Security Metrics in Industrial Control Systems Zachary A. Collier 1 , Mahesh Panwar 2 , Alexander A. Ganin 3 , Alex Kott 4 , Igor Linkov 1* 1 US Army Engineer Research & Development Center, Concord, MA, USA 2 Contractor to US Army Engineer Research & Development Center, Concord, MA, USA 3 University of Virginia, Charlottesville, VA, USA 4 US Army Research Laboratory, Adelphi, MD, USA *Corresponding Author, [email protected] 1.1 Introduction Risk – the topic of the previous chapter – is the best known and perhaps the best studied example within a much broader class of cyber security metrics. However, risk is not the only possible cyber security metric. Other metrics such as resilience can exist and could be potentially very valuable to defenders of ICS systems. Often, metrics are defined as measurable properties of a system that quantify the degree to which objectives of the system are achieved. Metrics can provide cyber defenders of an ICS with critical insights regarding the system. Metrics are generally acquired by analyzing relevant attributes of that system. In terms of cyber security metrics, ICSs tend to have unique features: in many cases, these systems are older technologies that were designed for functionality rather than security. They are also extremely diverse systems that have different requirements and objectives. Therefore, metrics for ICSs must be tailored to a diverse group of systems with many features and perform many different functions. In this chapter, we first outline the general theory of performance metrics, and highlight examples from the cyber security domain and ICS in particular. We then focus on a particular example of a class of metrics that is different from the one we have considered in earlier chapters. Instead of risk, here we consider metrics of resilience. Resilience is defined by the National Academy of Sciences

Transcript of Security Metrics in Industrial Control Systems · of the system, intended uses and unintended...

Page 1: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

In: Colbert, E. and Kott, A. (eds.),  "Cyber Security of Industrial Control Systems, Including SCADA Systems,"  

Springer, NY, 2016 

Security Metrics in Industrial Control Systems 

ZacharyA.Collier1,MaheshPanwar2,AlexanderA.Ganin3,AlexKott4,IgorLinkov1*

1USArmyEngineerResearch&DevelopmentCenter,Concord,MA,USA

2ContractortoUSArmyEngineerResearch&DevelopmentCenter,Concord,MA,USA

3UniversityofVirginia,Charlottesville,VA,USA

4USArmyResearchLaboratory,Adelphi,MD,USA

*CorrespondingAuthor,[email protected]

1.1 Introduction 

Risk–thetopicofthepreviouschapter–isthebestknownandperhapsthebeststudiedexamplewithinamuchbroaderclassofcybersecuritymetrics.However,riskisnottheonlypossiblecybersecuritymetric.OthermetricssuchasresiliencecanexistandcouldbepotentiallyveryvaluabletodefendersofICSsystems.

Often,metricsaredefinedasmeasurablepropertiesofasystemthatquantifythedegreetowhichobjectivesofthesystemareachieved.MetricscanprovidecyberdefendersofanICSwithcriticalinsightsregardingthesystem.Metricsaregenerallyacquiredbyanalyzingrelevantattributesofthatsystem.

Intermsofcybersecuritymetrics,ICSstendtohaveuniquefeatures:inmanycases,thesesystemsareoldertechnologiesthatweredesignedforfunctionalityratherthansecurity.Theyarealsoextremelydiversesystemsthathavedifferentrequirementsandobjectives.Therefore,metricsforICSsmustbetailoredtoadiversegroupofsystemswithmanyfeaturesandperformmanydifferentfunctions.

Inthischapter,wefirstoutlinethegeneraltheoryofperformancemetrics,andhighlightexamplesfromthecybersecuritydomainandICSinparticular.Wethenfocusonaparticularexampleofaclassofmetricsthatisdifferentfromtheonewehaveconsideredinearlierchapters.Insteadofrisk,hereweconsidermetricsofresilience.ResilienceisdefinedbytheNationalAcademyofSciences

Page 2: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

2

(2012)as“Theabilitytoprepareandplanfor,absorb,recoverfrom,ormoresuccessfullyadapttoactualorpotentialadverseevents”.

Thischapterpresentstwoapproachesforthegenerationofmetricsbasedontheconceptofresilienceusingamatrix‐basedapproachandanetwork‐basedapproach.Finally,adiscussionofthebenefitsanddrawbacksofdifferentmethodsispresentedalongwithaprocessandtipsintendedtoaidindevisingeffectivemetrics.

1.2 Motivation 

UnderPresidentGeorgeW.Bush,theDepartmentofEnergyissuedbestpracticesforimprovedindustrialcontrolsystem(ICS)security(USDepartmentofEnergy,2002).Someoftheseincludetakingstepssuchas"disconnectunnecessaryconnectionstotheSCADAnetwork","establisharigorous,ongoingriskmanagementprocess"and"clearlyidentifycybersecurityrequirements."Additionally,ExecutiveOrder13636,signedbyPresidentBarackObamain2013,broughtforththeissueofcybersecurityandresilience,and

proposedthedevelopmentofarisk‐based“CybersecurityFramework”(EO13636,2013).TheframeworkwaspresentedbytheNationalInstituteofStandardsandTechnology(NIST)andoffersorganizationsguidanceonimplementingcybersecuritymeasures.

Despiteexistingguidelinesandframeworks,designingandmanagingforsecurityincyber‐enabledsystemsremainsdifficult.Thisisinlargepartduetothechallengesassociatedwiththemeasurementofsecurity.PfleegerandCunningham(2010)outlineninereasonswhymeasuringsecurityisadifficulttaskasitrelatestocybersecurityingeneral,butallofwhichalsoapplytothesecurityofICSdomain(Table1).

PfleegerandCunningham(2010)notethatonewaytoovercomethesechallengesistothoughtfullydevelopaclearsetofsecuritymetrics.Unfortunately,thislackofmetricshappenstobeoneofthegreatestbarrierstosuccessinimplementingICSsecurity.WhenICSswerefirstimplemented,"networksecuritywashardlyevenaconcern"(Igureetal,2006).Althougheffortsarebeingmadetodraftandenactcybersecuritymeasures,thatgaphasyettobeclosed,evenatatimeofgreaterrisk.

Page 3: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

3

Table1:ChallengeswithCybersecurityMeasurement(adaptedfromPfleeger&Cunningham,2010)

Challenge DescriptionWecan’ttestallsecurityrequirements

Itisnotpossibletoknowallpossibleconfigurationsandstatesofthesystem,intendedusesandunintendedmisusesfromusers,etc.

Environment,abstraction,andcontextaffectsecurity

Systemsarebuilttoevolveastheyprocessnewinformation,andnotallsystemchangesarederivedfrommalicioussources

Measurementandsecurityinteract

Knowledgeaboutasystem’svulnerabilitiesandsafeguardscanaffectthetypesoffurthersecuritymeasuresimplemented,aswellasmodifytherisksthatusersarewillingtotake

Nosystemstandsalone Systemsarenetworkedtointeractwithothercybersystemsandassets

Securityismultidimensional,emergent,andirreducible

Securityexistsatmultiplelevelsofsystemabstraction,andthesecurityofthewholesystemcannotbedeterminedfromthesecurityofthesumofitsparts

Theadversarychangestheenvironment

Developinganaccuratethreatlandscapeisdifficultduetoadaptiveadversarieswhocontinuallydevelopnovelattacks

Measurementisbothanexpectationandanorganizationalobjective

Differentorganizationswithdifferentmissionsandpreferencesplacedifferingvaluesonthebenefitsofsecurity

We’reoveroptimistic Userstendtounderestimatethelikelihoodthattheirsystemcouldbethetargetofattack

Weperceivegainsdifferentlythanlosses

Biasesininterpretingexpectedgainsandlossesbasedonproblemframingtendtoaffectrisktoleranceanddecisionmakingunderuncertaintyinpredictablebutirrationalways

1.3 Background on Resilience Metrics 

1.3.1 What Makes a Good Metric? 

Accordingtothemanagementadage,“whatgetsmeasuredgetsdone”.Assuch,well‐developedmetricscanassistanorganizationinreachingitsstrategicgoals(Marr,2010).

Reichertetal.(2007)definemetricsas“measurablepropertiesthatquantifythedegreeto

whichobjectiveshavebeenachieved”.Metricsprovidevitalinformationpertainingtoagivensystem,andaregenerallyacquiredbywayofanalyzingrelevantattributesofthatsystem.Someresearchersandpractitionersmakeadistinctionbetweenameasureanda

Page 4: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

4

metric(Blacketal.,2008,Linkovetal.,2013a),whereasothersmayrefertothemasperformancemeasures(Neelyetal.,1997),keyperformanceindicators(Marr,2010)orstrategicmeasures(Allen,2011).Forthepurposesofthischapter,thesearereferredtogenerallyasmetrics.

Whenusedefficiently,metricscanhelptoclarifyone’sunderstandingoftheprocessesofaparticularareaofasystem,andfromthere,provideinformationforexternalreviewandassisttowardsfurtherimprovement,amongotheroutputs(Marr,2010).Thiscanbedonebyestablishingbenchmarksforagivenmetric,wherethresholdsorrangescanbeestablished(Blacketal.,2008).Benchmarks,orstandards,helpformthebasisfordecisionmakingandtakingcorrectiveaction(Williamson,2006).

Acriticalelementinelicitingameaningfulmetricistogathertherelevantinformation

aboutone’ssystemandtoalignthatmetricwithmeasurablegoalsandstrategicobjectiveswhichliewithinthescopeofagivenprojectorthedomainofaparticularorganizationalstructure(Beasleyetal.2010,Neelyetal.1997).Thereisalsotheissueofscaleandadaptability.Smallerorganizationmayhavemetricsdealingwithrudimentarysecuritymeasures,butastheygrowlarger,thesemeasuresmayneedtobescaledappropriatelytodealwiththesecurityneededforalargerorganization(Blacketal.,2008).

Therearekeyelementsthatcontributetoproducingasuccessfulmetric.Metricsshouldbeactionable:theyarenotsimplyaboutmeasuringnumerousattributesofaproject;merelygatheringinformationwithoutagoalinmindwillnotprovideadiscerniblesolution(Marr,2010).Suchinformationinandofitselfwouldnotbesubstantialenoughtobeconsideredametric.Gatheringrelevantmetricsrequiresdelvingdeeperintotheissuesfacedbyagivensystemandaskingpertinentquestionswhichcanleadtoactionableimprovement.These

includequestionssuchas“Doesitlinktostrategy?Canitbequantified?Doesitdrivethe

rightbehavior?”(Eckerson,2009).Fromthese,onecanobtainmetricswhichcaninturninformactionableresults.Table2summarizesthedesirablecharacteristicsofmetricsingeneralterms,andapplytoalltypesofsystemsincludingICSs.

Page 5: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

5

Table2:CharacteristicsofGoodMetrics(adaptedfromMcKayetal.2012,KeeneyandGregory2005)Characteristic DescriptionRelevant MetricsaredirectlylinkedtodecisionmakinggoalsandobjectivesUnambiguous ConsequencesofalternativescanbeclearlymeasuredbymetricsDirect MetricsclearlyaddressanddescribeconsequencesofinterestOperational DataexistandareavailableforthemetricofinterestUnderstandable MetricscanbeunderstoodandcommunicatedeasilyComprehensive Thesetofmetricsaddressacompletesuiteofgoalsandconsequences

Metricsmaybedescribedasnatural,constructed,orproxy.Naturalmetricsdirectly

describeanobjectiveinunitsthatarestraightforward(e.g.,dollarsasametricfor“costs

associatedwithICSdowntime”).Constructedmetricsmaybeusedwhennaturalmetricsdonotexist(e.g.,scalesfrom1to10whereeachnumbercorrespondstoadefinedlevelofICSperformance),andusuallyincorporateexpertjudgment.Proxymetricscanbeusedtoindirectlymeasureanobjective(e.g.,thenumberofuserswithcertainadministrativeprivilegesasaproxyforaccess)(McKayetal.2012,KeeneyandGregory2005).

Therearedifferenttypesofinformationthatmetricsgaugeandtheprojectteamhastheresponsibilityofappropriatelyselectingandevaluatingthem.Thesecanbeseparatedintoquantitative,semi‐quantitativeandqualitativeapproaches.Quantitativemetricshavemeasurable,numericalvaluesattachedtothem.Semi‐quantitativemetricsarenotstrictlyquantifiablebutcanbecategorized.Qualitativemetricsprovidenon‐numericinformation,forexampleintheformofaesthetics.

1.3.2 Metrics for IT Systems  

AsdescribedaboveinTable1,cybersystemsprovideuniquechallenges.Inparticular,thecyberdomainextendsbeyondjusttheimmediatesystemandrequiresaholisticviewpoint,withmanydifferenttechnicalandhumanfactorstobeaccountedfor(Collieretal.,2014).Threatstothesystemarealsoconstantlyevolvingandgrowinginsophistication,andasaresult,thereisahighdegreeofadaptabilityrequiredinordertoremaincurrent.Duetotheconstantlyevolvingthreatspace,thereisoftenlittlehistoricaldataforpotentialthreats(Collieretal.,2014).

Page 6: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

6

Withcybermetrics,asignificantnumberofthemainissuesaretailoredtowardssecurityandresilience.TheDefenseScienceBoard(2013)arguesthateffectivecybermetricsshouldbebroadenoughtofitdifferenttypesofsystems,yetalsobepreciseenoughtodialdownintothespecificsofagivensystem.Thefollowingaresomeexamplesofcybersecuritymetricscurrentlyinuse.

TheCommonVulnerabilityScoringSystem(CVSS)wasintroducedtoprovidevariousorganizationswithactionableinformationinregardstoassessingITvulnerabilities(Melletal.,2007).CVSSgroupstheirmetricsintothreecategories,namelyBase,Temporal,andEnvironmentalmetrics.AfewofthesesecuritymetricsincludeCollateralDamagePotential,TargetDistribution,ReportConfidence,Exploitability,AccessComplexity,AccessVector,Authentication,IntegrityImpact,AvailabilityImpact,andConfidentialityImpact(Melletal.,2007).Therearegeneralscoringtipsforthewaythatvulnerabilitiesareassessed;vulnerabilitiesarenotscoredbasedoninteractionswithothervulnerabilities,rather,theyarescoredindependently.Themainmeasureofvulnerabilityisitsimpactonthekeyservice.Vulnerabilitiesarescoredaccordingtocommonlyusedprivileges,whichmightbeadefaultsettingincertainsituations.Ifavulnerabilitycanbeexploitedbymultipleexploits,itisscoredwiththeexploitthatwillpresentthemaximumimpact(Mell,etal.,2007).CVSSallowsvulnerabilityscorestobestandardized,andBasemetricsare

normalizedonascaleof0–10.TheycanbeoptionallyrefinedbyincludingvaluesfromTemporalandEnvironmentalmetrics.

TheCenterforInternetSecurity(CIS)hasalsoestablishedmetricsfororganizationstouse(CIS,2010).CIShasdividedtheirmetricsintosixcriticalbusinessfunctions.TheseareIncidentManagement,VulnerabilityManagement,PatchManagement,ConfigurationManagement,ChangeManagementandApplicationSecurity.Italsorecognizeshierarchiesandinterdependenciesofmetrics,forinstancecitingmanagementmetricsasbeingofprimaryimportancetoanorganization,whilenotingthatsomeofthosemetricsmaydependonthepriorimplementationoftechnicalmetrics(CIS,2010).SomeofthemetricsincludeCostofIncidentsandPatchPolicyCompliance.CostofIncidentsreferstoanumber

ofpotentiallosses,suchascustomerlistsortradesecretsundera“directloss”anda“cost

ofrestitution”,forexampleintheeventthatfinesareleviedduetoanincident.Thisismeasuredbythesummationofthenumericalvaluesofallthecostsassociatedwiththemetric.ExamplesrelatingtosecurityincludeMeanTimetoIncidentDiscovery,MeanTimeBetweenSecurityIncidentsandMeanTimetoIncidentRecovery(CIS,2010).Foranexampleofmeasurement,MeanTimetoIncidentDiscoverymeasuresthesummationofthe

Page 7: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

7

timebetweenincidentsanddiscoveriesofincidents,dividedbytotalnumberofincidentsrecoveredduringthosetimeframes(CIS,2010).

TheCybersecurityFrameworkdevelopedbyNISTstemmingfromEO13636wasreleasedinFebruary2014(NIST2014a).ThefinalCybersecurityFrameworkconsistsofa

FrameworkCore,whichpresentsasetoffive“concurrentandcontinuousFunctions–

Identify,Protect,Detect,Respond,Recover”(NIST2014).Thesefunctionsarethe“high‐

level,strategicviewofthelifecycleofanorganization’smanagementofcybersecurityrisk,”whichfeaturesubsequentcategoriesandsubcategoriesforthefunctions,relatingtooutcomesandactivities(NIST2014).Forexample,theRespondfunctionconsistsoffivecategories,amongwhichincludesMitigation.Mitigationisthenfurthersubdividedintometricsrelatedtocontaininganderadicatingincidents.TheFrameworkCoreisusedasa

scorecardofprogress–thecurrentguidancecallsforfirstdevelopinganorganization’s

CurrentProfile,whichconsistsofassignedscoresbasedontheorganization’sperformanceineachofthecategoriesandsubcategories.ThisCurrentProfileisthencomparedtoaTargetProfile,representingthedesiredstateoftheorganizationineachofthesamecategoriesandsubcategories.Theshortfallsbetweentheseprofilescanbeviewedasgaps

inanorganization’scyber‐riskmanagementcapabilitieswhichcaninformprioritizationofcorrectivemeasures(NIST2014;Collieretal.2014).

TheSoftwareEngineeringInstitute(SEI)atCarnegieMellonUniversitydevelopedaframeworkforassessingoperationalresiliencewhichfeaturesasetofTopTenStrategicMeasures,whichaimtobemappeddowntothelevelofspecificProcessAreameasures(AllenandCurtis,2011).UndertheheadingofHigh‐ValueServicesandAssets,oneofthemeasuresisrelatedtothepercentageofhigh‐valueservicesthatdonotsatisfytheirassignedresiliencerequirements(AllenandCurtis,2011).TheSEIframeworkalsocontainsalargeamountofresiliencemeasures,spanning26differentProcessAreas.Forexample,undertheProcessAreaofEnvironmentalControl,therearemeasuressuchasPercentageofFacilityAssetsthathavebeenInventoried,ElapsedTimeSincetheFacilityAssetInventorywasReviewed,andElapsedTimeSinceRiskAssessmentofFacilityAssets

Performed(AllenandCurtis,2011),wheretheterm“assets”appliestohigh‐valueservices.Thesearepresentedinatablewithtraceability,assigninganidentificationnumbertoeachmetricalongwiththeirapplicabilitytogoalswithintheProcessAreas.

Page 8: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

8

MITREproposedaframeworkentitledCyberResiliencyEngineeringFramework,which,

amongitsgoalsaimsto“motivateandcharacterizecyberresiliencymetrics”(Bodeau2011).TheframeworkcontainsfourCyberResiliencyGoals:Anticipate,Withstand,Recover,andEvolve.Thereareatotalofeightobjectiveswhichareasubsetofthegoals.ForexampleAnticipatehasthreeobjectives:Predict,Prevent,andPrepare(Bodeau,2011).Thishierarchycanbeusedtoinformandcategorizetheappropriateresiliencemetrics.Thesearemeanttobeperformedsimultaneously,andbeararesemblancetotheNISTframeworkmentionedearlier.

1.3.3 Metrics for ICS Networks 

Theabovemetricsweredevelopedfor“cyber”systemsgenerallyspeaking,notspecificallyforICSs,althoughtheycanbetailoredwithICSsinmind.ICSsinparticularareauniquecase;inmanysituations,thesesystemshaveoldermodels,andweredesignedforfunctionalityratherthansecurity(USDepartmentofEnergy,2002).Theyconstituteadiversegroupofsystemsthathavedifferentrequirementsfortheirvariousoperations(Pollet,2002).

SpecificallyasitrelatestoICSs,time,safetyandcontinuationofservicesareofgreatimportance,sincemanysystemsareinapositionwhereafailurecanresultinathreattohumanlives,environmentalsafety,orproductionoutput(Stouffer,2011).Sincetheserisksaredifferentthanthosefacedbyinformationtechnology(IT)systems,differentprioritiesarealsonecessary.Examplesofsomeuniqueconsiderationsincomparisontocybersecurityincludethelongerlifespanofsystemcomponents,physicallydifficulttoreachcomponents,andcontinuousavailabilityrequirements(Stouffer,2011).Additionally,thesesystemstypicallyoperateinseparatefieldsthancybersecurity,suchasinthegasandelectricindustries,andsometricsmustbeadaptedtofitthesedifferentorganizationalstructures(McIntyreetal.,2007).CriticalinfrastructuresarecommonforICSs,andasa

result“downtimeandhaltingofproductionareconsideredunacceptable”(McIntyreetal.,2007).

Page 9: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

9

Stoufferetal.(2011)comparethedifferencesbetweeninformationtechnology(IT)system

andICSs,focusingonthesafety‐criticalnatureofmanyICSnetworks.Forexample,“high

delayandjittermaybeacceptable”asaperformancerequirementforITsystems,whereasforICSs,itmaynotbeacceptable(Stouffer,2011).Thisisduetothefactthatthereisatime‐criticalnaturetoICSs,whereasforITsystemsthereishighthroughput,allowingfor

somejitter(Stouffer,2011).Similarly,forIT,“systemsaredesignedforusewithtypical

operatingsystems”andforICSs,thereare“differingandpossiblyproprietaryoperating

systems,oftenwithoutsecuritycapabilitiesbuiltin”.Therearealsoavailabilityrequirements,inthatsometimesanITstrategymayrequirerestartingorrebootingaprocess,somethingwhich,forICSprocesses,requiresmorecarefulplanningasunexpectedoutagesandquicklystoppingandstartingasystemarenotacceptablesolutions(Stouffer,2011).Withthesekeydifferencesbetweenthetwodomains,therearevaryinglevelsofadaptationneededinordertobegintheprocessofsecuringICSnetworks.

TheUSNationalSecurityAgency(NSA)draftedaframeworkforICSnetworks,focusingonpotentialimpactandlossrelatingtoanetworkcompromise(NSA,2010).Theysuggested

assigninglossmetricsincorporatingNIST’sframework:compromisespertainingtoConfidentiality,IntegrityandAvailabilityforeachnetworkasset(NSA,2010).A

Confidentialitycompromiseisdefinedasan“unauthorizedreleaseortheftofsensitive

information”e.g.theftofpasswords(NSA,2010).AnIntegritycompromiseisdefinedasan

“unauthorizedalterationormanipulationofdata”,e.g.manipulationofbillingdata(NSA,

2010).AnAvailabilitycompromiseisdefinedasa“lossofaccesstotheprimarymissionofa

networkedasset”e.g.deletionofimportantdatafromadatabase(NSA,2010).Thesemayalsobestreamlinedintoonemetric,usingthehighestvalue(e.g.ofLow,ModerateorHigh)amongthethreeareas.

Theassignmentofathreatmetricsateachpotentialattackvectorwassuggested,butspecificexampleswerenotprovided.Fivethreatsourceswereidentified:Insiders,TerroristsorActivists,HackersorCyber‐Criminals,Nation/StateSponsoredCyber‐WarfareandCompetitors(NSA,2010).Bothlossandthreatmetricscanberatedonaconstructedscale(Low,ModerateorHigh)andgivenanumericratingonasetscale.Itwasmentionedthattheimportantconsiderationistohaveascale,andthatthenumberofgraduationsin

Page 10: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

10

thescaleisnotimportant,solongastheconstructedscaleremainconsistent(e.g.apotentialforlossoflifewillrankasHigh)(NSA,2010).Combiningresultsofmetricswasalsodiscussedasapossibility.Asanexample,foragivenpointinthenetwork,aLossMetricisassignedascoreofHighontheconstructedscale(3)andaThreatmetricatthatsamenetworkpointisratedatModerate(2).Fromthis,onecanarriveatacompositepriorityvalue,whichissimplythesumofthosetwoscores.Othersuchpointscanbeevaluatedandthenprioritizedandranked(NSA,2010).Thescoringmethodologyisabasicexample,(andnottheonlymethod‐weighingmetricswaslistedasapossibility(NSA,2010))andmorerobustmethodscanbedevised.

BoyerandMcQueen(2008)devisedasetofideal‐basedtechnicalmetricsforcontrolsystems.Theyexaminedsevensecuritydimensionsandpresentanideal,orbestcasescenario,foreachofthem.TheidealsareSecurityGroupKnowledge,AttackGroupKnowledge,Access,Vulnerabilities,DamagePotential,Detection,andRecovery.FortheAccessdimension,theidealstatesthatthesystemisinaccessibletoattackgroups.ThesecuritydimensionofVulnerabilitieshasanidealstatingthatthesystemhasnovulnerabilities(BoyerandMcQueen,2008).Bytheverynatureofanideal,thesemaybeimpossibletoachieveandmaintainintherealworld.Butfromthem,metricsweredevisedthatcouldbestrepresenttherealizationoftheseideals.Underthevulnerabilitydimension,

themetricVulnerabilityExposureisdefinedas“thesumofknownandunpatched

vulnerabilities,eachmultipliedbytheirexposuretimeinterval.”Itwassuggestedthatthismetriccouldbebrokendownintoseparatemetricsfordifferentvulnerabilitycategories,aswellasincludingaprioritizationofvulnerabilities,citingCVSS.UndertheAccessdimension,thereisthemetricRootPrivilegeCount,whichisthecountofallpersonnelwith

keyprivileges,arguinginfavoroftheprincipleofleastprivilege,whichstatesthat“everyprogramandeveryprivilegeduserofthesystemshouldoperateusingtheleastamountof

privilegenecessarytocompletethejob”(Saltzer,1974).Thislogicalorderingofmetricswithinthescopeofidealscanbeofvaluetothosewishingtodevisetheirownsetofmetrics.

Theideal‐basedmetrics(BoyerandMcQueen,2008)alsoacknowledgethephysicalspaceofICSnetworks.ThemetricRogueChangeDays,whichisthenumberofchangestothesystemmultipliedbythenumberofdaysundetected,includesProgrammableLogicControllersandHuman‐MachineInterfacesandotherICSrelatedsystems.ComponentTestCount,ametricmeasuringthenumberofcontrolsystemcomponentswhichhavenotbeentestedisasimplemeasure,butofsignificanceduetonumerouscomponentsinuseinanICSsystem.

Page 11: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

11

Withintheideals,themetricofAttackSurface(definedbyManadhataandWing(2011)as

“thesubsetofthesystem’sresources(methods,channels,anddata)potentiallyusedin

attacksonthesystem”)wasdeterminedtonotbedevelopedenoughforrealworlduse.

BoyerandMcQueenfurtherarguethat“acrediblequantitativemeasureofsecurityriskis

notcurrentlyfeasible”(BoyerandMcQueen2008).Butwiththeinclusionofatheoreticalmetric,andaframeworkforsecurity,thisdemonstratesaforwardthinkingattitudethatcanbebuiltuponbythoseaimingtoestablishtheirownsecurityprotocols.ThisrepresentsimportantfutureworkfortheICSandsecuritycommunities.ComparisonsbetweentheNSAapproachandtheapproachoutlinedbyBoyerandMcQueenarepresentedinTable3.

Table3:ComparisonbetweenICSMetrics

NationalSecurityAgency(2010) BoyerandMcQueen(2008)

Focus LossandThreatfocusedMetrics(p.10,15)

Quantitativetechnicalmetrics(p.1),idealbased:attemptedtohavemetricsthatcouldstrivetowardidealscenarioswithinsevensecurityareas

Amount Threelossmetrics(pernetworkedasset),oneThreatmetric(perpotentialattackvector)

13totalmetrics(suggestedtotal:lessthan20)

AppliedorTheoretical

Suggestsdeployablemetrics Discussesbothdeployableandtheoreticalmetrics(p.10,11)

QuantitativeorQualitative

Semi‐qualitative(suggestsHigh,Medium,Low,withallowancefornumericattachmenttothesevalues)

Doesnotfocusonqualitativemetrics(p.1),butonquantitativemetrics

CombinationofMetrics

Presentsmethodtocombineresultsofmetricscoresforranking

Nocombinationofmetrics

ConsequenceConsiderations

LossMetricsarerelatedtoConfidentiality,Integrity,Availability

AcknowledgesthepurposeofsecurityisprotectionofConfidentiality,IntegrityandAvailability(p.4)

Page 12: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

12

ComplementaryresearchtometricsdevelopmentintheICSrealmiscurrentlybeingconducted.OnesucheffortistodevelopastandardizedtaxonomyofcyberattacksonSCADAsystems(Zhuetal.,2011).AcommonlanguagefordescribingattacksacrosssystemscanfacilitatethedevelopmentoffurtherthreatandvulnerabilitymetricsforICSs.Inaddition,thedevelopmentofanationaltestbedforSCADAsystemsisbeingdevelopedbytheDepartmentofEnergywhichwillenablethemodelingandsimulationofvariousthreatandvulnerabilityscenarios,whichwillallowresearcherstodevelopabetterunderstandingofwhatmetricsmayormaynotbeusefulinmonitoringandmanagementofthesesystems(USDepartmentofEnergy,2009).Anotherdevelopmentrelatedtometricsresearchistheinvestigationoftradeoffsbetweencertaincriticalmetrics.Oneexampleisbetweenoptimizingsystemperformancewithsystemsecurity,whereadditionalsecuritymeasuresmayresultinreducedperformance.Zeng&Chow(2012),developedanalgorithmictechniquetodeterminetheoptimaltradeoffbetweenthesetwometrics,andthemethodcanbeextendedtotradeoffsbetweenothermetricsaswell.

1.4 Approaches for ICS Metrics 

Whilevariousframeworksandsetsofmetricsexist,suchastheonesmentionedintheprevioussection,itcanbedifficultformanagersandsystemoperatorstodecidewhethertoadoptormodifyanexistingset,ortocreateanentirelynewsetofmetrics.Balancingthetradeoffsbetweengeneralizablemetricsandspecificsystem‐levelandcomponent‐levelmetricscanbechallenging(DefenseScienceBoard,2013).Thefollowingapproachesprovideastructuredwaytothinkaboutdevelopingmetrics,allowinguserstoleverageexistingmetricsbutalsoidentifygapswherenewmetricsmayneedtobecreated.Theuseofsuchstructuredandformalizedprocessesrequiresthethoughtfulanalysisofthesystemsbeingmeasured,butalsohowtheyrelatetothebroaderorganizationalcontext,suchasgoals,constraints,anddecisions(Marr,2010).Moreover,thedevelopmentofastandardizedlistofquestionsortopicshelpstosimplifytheprocessofdesigningametric.Thedevelopmentofmetricsshouldbeasmoothprocess,andsuchalistcanprovideinsight

intothe“behavioralimplications”ofthegivenmetrics(Neelyetal.1997).

1.4.1 Cyber Resilience Matrix Example 

ThefirstmethodisbasedontheworkofLinkovetal.(2013a).Unliketraditionalrisk‐basedapproaches,thisapproachtakesaresilience‐centrictheme.Muchhasbeenwrittenelsewhereontherelativemeritsofaresilience‐focusedapproach(seeLinkovetal.,2013b,2014;Collieretal.2014;Roegeetal.2014;DiMaseetal.2015),butweshallbriefly

Page 13: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

13

summarizetheargumenthere.TraditionalriskassessmentbasedonthetripletformulationproposedbyKaplanandGarrick(1981)becomesdifficulttoimplementinthecybersecuritycontextduetotheinabilitytoframeandevaluatemultipledynamicthreatscenarios,quantifyvulnerabilityagainstadaptiveadversaries,andestimatethelong‐termandwidelydistributedconsequencesofasuccessfulattack.Insteadofmerelyhardeningthesystemagainstpotentialknownthreatsinarisk‐basedapproach,thesystemcanbemanagedfromtheperspectiveofresilience,whichincludestheabilityofoneormore

criticalsystemfunctionalitiestoquickly“bounceback”toacceptablelevelsofperformance.Asaresult,aresilientsystemcanwithstandandrecoverfromawidearrayofknownandunknownthreatsthroughprocessesoffeedback,adaptation,andlearning.

Followingthisthoughtprocess,Linkovetal.(2013a)establishedamatrix‐basedmethod.Ononeaxis,thestepsoftheeventmanagementcycleidentifiedasnecessaryforresiliencebytheNationalAcademyofSciences(2012)arelisted,andincludePlan/Prepare,Absorb,Recover,andAdapt.Notethattheabilitytoplan/prepareisrelevantbeforeanadverseevent,andtheothercapabilitiesarerelevantafterdisruption.OntheotheraxisarelistedthefourdomainsinwhichcomplexsystemsexistasidentifiedbyAlberts(2002),andincludePhysical,Information,Cognitive,andSocialdomains.ThePhysicaldomainreferstothephysicalresourcesandcapabilitiesofthesystem.TheInformationdomainreferstotheinformationanddatathatcharacterizethePhysicaldomain.TheCognitivedomaindescribestheuseoftheotherdomainsfordecisionmaking.Finally,theSocialdomainreferstotheorganizationalstructureandcommunicationsystemsfortransmittinginformationandmakingdecisions(Alberts2002).

Together,theseaxesformasetofcellsthatidentifyareaswhereactionscanbetakenin

specificdomainstoenhancethesystem’soverallabilitytoplanfor,andabsorb,recover,andadaptto,variousthreatsordisruptions(Figure1).Eachcellisdesignedtoanswerthe

question:“Howisthesystem’sabilityto[plan/preparefor,absorb,recoverfrom,adaptto]

acyberdisruptionimplementedinthe[physical,information,cognitive,social]domain?”(Linkovetal.2013a).

Page 14: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

14

Figure1:GenericResilienceMatrix

Aresultingsetof49metricsareproducedthatspanthevariouscellsofthematrix,andselectedmetricsareshowninTable4(seeLinkovetal.2013aforthecompletelist).Metricsaredrawnfromseveralsourcesandaremeanttobegeneralandnotnecessarily

comprehensive.Forexample,underAdaptandInformation,ametricstates“document

timebetweenproblemanddiscovery,discoveryandrecovery,”whichhasaparalleltothe

MeanTimetoIncidentDiscoverywithinSEI’sguidance.ThemetricsunderPlanandInformation,relatedtoidentifyinginternalandexternalsystemdependenciescanbecomparedtotheTemporalMetricofAccessComplexityfromCVSS,whichrelatestohoweasilyavulnerabilitycanbeexploited.ThemetricunderPrepareandSocialpresentsa

simpleyetimportantmessagethatholdstrueinalloftheframeworks:“establishacyber‐

awareculture.”

TheresiliencematrixapproachdescribedinLinkovetal.(2013a)hasseveralstrengthsinthatthemethodisrelativelysimpletouseandoncemetricshavebeengenerated,itcanserveasaplatformforamulti‐criteriadecisionaid(Collier&Linkov,2014).Ithasthepotentialtoserveasascorecardinordertocapturequalitativeinformationabouta

system’sresilience,andaidmanagersandtechnicalexpertsinidentifyinggapsinthe

system’ssecurity.However,theresiliencematrixdoesnotcapturetheexplicittemporalnatureofresilience(i.e.,mappingthecriticalfunctionalityovertime)orexplicitlymodelthesystemitself.Inthisregard,itcanbeviewedasahighlevelmanagementtoolthatcanbeusedtoidentifyasnapshotwheremoredetailedanalysesandmodelingcouldpotentiallybecarriedout.

Plan & 

Prepare

Absorb Recover Adapt

Physical

Information

Cognitive

Social

Page 15: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

15

Table4:SelectedCybersecurityMetricsDerivedfromtheResilienceMatrix(adaptedfromLinkovetal.,2013a).

Plan/Prepare Absorb Recover AdaptPhysical Implement

controls/sensorsforcriticalassetsandservices

Useredundantassetstocontinueservice

Investigateandrepairmalfunctioningcontrolsorsensors

Reviewassetandserviceconfigurationinresponsetorecentevent

Information Prepareplansforstorageandcontainmentofclassifiedorsensitiveinformation

Effectivelyandefficientlytransmitrelevantdatatoresponsiblestakeholders/decisionmakers

Reviewandcomparesystemsbeforeandaftertheevent

Documenttimebetweenproblemanddiscovery,discoveryandrecovery

Cognitive Understandperformancetrade‐offsoforganizationalgoals

Focuseffortonidentifiedcriticalassetsandservices

Establishdecisionmakingprotocolsoraidstoselectrecoveryoptions

Reviewmanagementresponseanddecisionmakingprocesses

Social Establishacyber‐awareculture

Locateandcontactidentifiedexpertsandresponsiblepersonnel

Determineliabilityfortheorganization

Evaluateemployeesresponsetoeventinordertodeterminepreparednessandcommunicationseffectiveness

1.4.2 Network Simulation Example 

Thesecondmethodisbasedonmodelingofcomplexcyberandothersystemsasinterconnectednetworks,whereafailureinonesectorcancascadetootherdependentnetworksandassets(Vespignani,2010).ThisisareasonableassumptionforICSnetworks;forexample,adisruptionoftheelectricalgridcandirectlyimpactdependentsectorssuchasthenetworkcontrollingICSdevicesleadingtoacascadeoffailuresasitisbelievedtohavehappenedduringtheItalianblackoutin2003(Buldyrevetal.,2010).Thusthe

Page 16: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

16

assessmentofthesecurityofasingleICSnetworkshouldbeviewedinthecontextofalargernetworkofinterdependentsystems.

Ganinetal.(2015)tookthisnetwork‐orientedviewindevelopingamethodologytoquantitativelyassesstheresilience(andthussecurity)ofnetworkedcybersystems.TheybuiltupontheNationalAcademyofSciences(2012)definitionofresilienceasasystempropertythatisinherentlytiedtoitsabilitytoplanfor,absorb,recoverfrom,andadapttoadverseevents.Inordertocapturethestateofthesystemtheauthorsproposetousetheconceptofcriticalfunctionalitydefinedasatime‐specificperformancefunctionofthe

systemconsideredandderivedbasedonthestakeholder’sinput.Forinstanceinthenetworkofpowerplants,thecriticalfunctionalitymightrepresentthetotaloperationalcapacity.Inthenetworkofcomputersitmightrepresentthefractionofserversandservicesavailable.Valuesofcriticalfunctionalityarerealnumbersfrom0to1.Otherkey

elementstoquantifyresiliencearethenetworkedsystem’stopologyanddynamics;therangeofpossibleadverseevents(forexample,acertaindamagetonodesofthenetwork);andthecontroltimeTC(thatisthetimerangeoverwhichtheperformanceofthesystemisevaluated).Thenthedependencyofthecriticalfunctionality(averagedoveralladverseevents)overtimeisbuilt.Ganinetal.(2015)refertothisdependencyastheresilienceprofile.Asitistypicallycomputationallyprohibitiveornotpossibleatall(incaseof

continuousvariablesdefiningnodes’states)toconsiderallthewaysanadverseeventcanhappen,itissuggestedtoutilizeasimulationbasedapproachwithMonte‐Carlosampling.

Givenitsprofileinnormalizedtime(wheretimeTCistakentobe1),theresilienceofthenetworkcanbemeasuredastheareaunderthecurve(yellowregioninFigure2).Thisallowsmappingoftheresiliencetorealvaluesrangingbetween0and1.

Anotherimportantpropertyofthesystemisobtainedbyfindingtheminimumoftheaveragecriticalfunctionality.SomeresearchersrefertothisvalueasrobustnessM

(Cimellaroetal.,2010),whileLinkovetal.(2014)notethat1–Mcorrespondstothemeasureofrisk.

Page 17: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

17

Figure2:Ageneralizedresilienceprofile,whereasystem’sresilienceisequaltotheareabelowthe

criticalfunctionalitycurve(adaptedfromGaninetal.,2015).

IntheirpaperGaninetal.(2015)illustratedtheapproachonadirectedacyclicgraph.Eachlevelinthisgraphrepresentsasetofnodesfromcertaininfrastructuresystem(e.g.electricalgrid,computersetc.).Nodesofdifferentlevelsareconnectedbydirectedlinksrepresentingadependencyofthedestinationnodeonthesourcenode.Inthesimplestcaseanodeinacertainlevelrequiressupply(oradependencylink)fromanodeineachoftheupperlevelsanddoesnotdependonanynodesinthelowerlevels.Otherparametersofthe

modelincludenoderecoverytime(TR)–ameasureofhowquicklyanodecanreturntoan

activestateafterit’sbeeninactivatedasaresultofanadverseevent;redundancy(pm)–theprobabilitycontrollingthenumberofadditionalpotentialsupplylinksfromupperlevelstolowerlevels;andswitchingprobability(ps),controllingeaseofreplacementofadisruptedsupplylinkwithapotentialsupplylink.Theseparameterscouldbeextendedtoothersituationstoinformhowasystemmaydisplayresilientbehavior,andthusincreasingthesecurityofthesystemasawhole.

Theauthorsfoundthatthereisstrongsynergybetweenpmandps;increasingbothfactorstogetherproducesarapidincreaseinresilience,butincreasingonlyoneortheothervariablewillcausetheresiliencemetrictoplateau.Resilienceisstronglyaffectedbythetemporalswitchingtimefactor,TR.Thistemporalfactordeterminesthecharacteristicsoftherecoveryphaseandhasagreaterimpactonthecalculatedresiliencethandoesthepotentialincreaseinredundancy.Thisisparticularlytruewhentheswitchingprobabilitypsislow.Animportantlongtermchallengeistomodeladaptation,which,accordingtotheNationalAcademyofSciences,ispartoftheresponsecyclethatfollowsrestorationandincludesallactivitiesthatenablethesystemtobetterresistsimilaradverseeventsinthefuture.

Page 18: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

18

Ganinetal.(2015)notethatthemainadvantagesoftheapproachincludeitsapplicabilitytoanysystemthatcanberepresentedasasetofnetworks.Alsoboththeresilienceandtherobustnessofasystemaremetricizedusingarealvalueinrangebetween0and1(where1correspondstotheperfectresilienceorrobustness)makingcomparisonofresilienceofdifferentsystemseasy.Ontheotherhandmappingtheresiliencepropertyofasystemtoa

singlevaluenecessarilyshadowssomesystem’simportantcharacteristics(forinstance,therateofrecovery).Theresilienceprofilecouldbeusedasamoreholisticrepresentationof

thesystem’sresiliencenotingthateveninthatcaseonlytheaveragevalueofcriticalfunctionality(ateachtimestep)istakenintoaccount.Tofullydescribeasystemoneshouldconsiderthedistributionofthevalueofcriticalfunctionality(ateachtimestep)fordifferentinitialadverseevents.Finally,itisnotpossibletosimulatealladverseeventsfromtherangeusedtoestimateresilienceandtheapproachisMonte‐Carlobased.Itmeansthatinorderfortheresultstobereliablethenumberofsimulationsistypicallyrequiredtobeveryhigh.

1.5 Tips for Generating Metrics  

1.5.1 Generalized Metric Development Process 

ThefollowingprocesstowardsthedevelopmentofmetricsisadaptedbyMcKayetal.2012.

1. ObjectiveSetting:Articulateclear,specificgoals.Thisshouldbedoneinastructuredmanner.GregoryandKeeney(2002)outlineastructuredapproachtodothis.

a. Writedownalloftheconcernsthattheprojectteamfeelsisrelevant.b. Convertthoseconcernsintosuccinctverb‐objectgoals(e.g.,minimizedowntime).c. Next,theseshouldbeorganized,oftenhierarchically,separatinggoalswhich

representmeansfromthosewhichrepresentends.d. Finally,reviewandclarificationshouldbeconductedwiththeprojectteam.This

maybeaniterativeprocess.2. DevelopMetrics:Oncetheobjectivesareclearlyarticulatedandorganized,metricscanbe

formallydeveloped.a. Thefirststepistoselectabroadsetofmetrics,whichmaybeselectedfromexisting

listsorguidelines,orcreatedbyaprojectteamorsubjectmatterexpertsfortheparticularpurposeathand.ThisstepiswheretheResilienceMatrixcouldfacilitatemetricdevelopment.

b. Next,thissetofmetricsshouldbeevaluatedandscreenedtodeterminewhetheritmeetstheprojectobjectivesandthedegreetowhichthemetricsmeetthedesirablequalitiesofmetrics,explainedearlierinthischapter.Atthisstage,remainingmetricscanbeprioritized.

Page 19: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

19

c. Finallytheremainingmetricsshouldbedocumented,includingassumptionsandlimitations,andothersupportinginformation.

3. CombinationandComparison:Amethodshouldbedevelopedforhowthemetricswillultimatelybeusedtosupportdecisionmakinganddriveaction.Somemethodsinclude:

a. NarrativeDescription:Simpletechniqueswheretrade‐offsmaybesimplesuchaslistingevidenceorbestprofessionaljudgement.

b. ArithmeticCombination:Simplemathematicaltechniquesforcombiningdissimilarmetricssuchassimpleaggregationofmetricswithsimilarunits(e.g.,cost),convertingtosimilarunits(e.g.,monetization),ornormalizingtoasimilarscale(e.g.,0to1).

c. Multi‐CriteriaDecisionAnalysis:Amethodforweightingandscoringdissimilardecisioncriteriabasedontheirrelativeimportanceandperformancewithrespecttoanobjective.

d. InterdependentCombination:Forsystemsthatarecomplex,usuallyinvolvingintricateinternalrelationships,moreintensivemodelingeffortsmaybenecessary,suchasBayesiannetworksorothercomplexsystemsmodelingtechniques.

Theabove‐mentionedprocess,alongwithasolidmetricdevelopmentprocess,cangreatlyaidindevisingeffectivemetrics.Oftenitisnecessarytodevelopaconceptualmodelofthesysteminordertoidentifythefunctionalrelationshipsandcriticalelementsandprocesseswithinasystem.ThiscanbedoneusingaNetworkScienceapproachdescribedabove.

1.5.2 Best Practices in Metric Development and Validation 

Validationofmetricsisanoftenoverlookedaspectofthemetricdevelopmentprocess.Neelyetal.(1997)providesomequestionstoaskregardingwhethertheoutputfromthemetricsisappropriate,specificallywhetherthemetricshaveaspecificpurpose,arebasedonanexplicitformulaand/ordatasource,andareobjectiveandnotbasedsolelyonopinion(Neelyetal.,1997).Similarly,Eckerson(2009)laysoutaseriesofquestionsthatcanserveasaqualitycheckondevelopedmetrics,toensurethattheyareofhighquality:

•Doesitlinktostrategy?•Canitbequantified?•Doesitdrivetherightbehavior?•Isitunderstandable?•Isitactionable?•Doesthedataexist?

Regardingthenumberofmetricsnecessary,itisn’tnecessarilythequantityofmetricsthatconstituteasuccessfulimplementation,butwhetherthesemetricsarecollectivelycomprehensiveenoughtoaddresseverythingdeemedimportant(McKayetal.2012).Eckerson(2009)recommendsthatasetofmetricsbesparse,sincewithalimitednumber

Page 20: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

20

ofmetricsitiseasiertoanalyzehowmetric‐levelchangesdrivetheperformanceinthesystem,aswellasthepracticalfactthatgathering,synthesizing,andpresentingmultipledatastreamsoftentakesquitesometime.Moregranular,process‐levelmetricsmaystillberequiredhowever,andEckerson(2009)proposesaMAD(monitor,analyze,drill)frameworkforpresentingdifferentlevelsofresolutiontodifferentusersofthatinformation.

Anotherongoingelementofvalidationistraceability,asevidencedintheframeworkpresentedbyNeelyetal.(1997),whichincludesalistofinformation(knownastheperformancemeasurerecordsheet)suchashowoftendataistobecollected,andby

whom,aswellasimportantquestionssuchas“whoactsonthedata?”and“whatdothey

do?”.Ifthesequestionsareconsideredandansweredastheneedarises,itisknownwhoisresponsibleformakingthemeasurementandwhatactionsaretobetakenasaresult.Thiscanrevealinsightintothemetricandhowtheyaremeasuredandbeingutilized,notjustforthecurrentprojectbutforfuturereference.Anitemonthelistaskswhatthemetric

“relatesto.”Thiscanassistinenteringthemindsetofapproachingmetricswithaninterconnectedandgoal‐orientedviewpoint.

Othervalidation‐relatedeffortsincludestandardizingmethodsforICSmetricdevelopmentandimplementation,aswellasinstitutionalizingaclearmeanstointegratemetricswithdecisionanalytictoolstosupporttheriskmanagementprocess.Finally,giventhedynamicnatureofcyberthreats,periodicreviewandupdatingofICSmetricsshouldbeconductedtokeepabreastofthelatestdevelopmentsinthefield.

1.6 Conclusions 

Despiteexistingguidelinesandframeworks,designingandmanagingforsecurityincyber‐enabledsystemsremainsdifficult.Thisisinlargepartduetothechallengesassociatedwiththemeasurementofsecurity.Acriticalelementinelicitingameaningfulmetricisingatheringtherelevantinformationaboutone’ssystemandaligningthatmetricwithmeasurablegoalsandstrategicobjectives.ForICSs,time,safetyandcontinuationofservicesfactorconsiderablyintooverallgoals,sincemanysystemsareinapositionwhereafailurecanresultinathreattohumanlives,environmentalsafety,orproductionoutput.Oftenitisnecessarytodevelopaconceptualmodelofthesystemordevelopastandardizedlistofquestionsortopicshelpstoidentifycriticalprocesselements,thefunctionalrelationshipsandcriticalelementsandprocesseswithinasystem.Inthischapter,wediscussindetailtwoapproachesforthegenerationofbroadlyapplicablesecurityandresiliencemetricsandtheirintegrationtoquantifysystemresilience.Thefirstmethodisasemi‐quantitativeapproachinwhichthestagesoftheeventmanagementcycle(plan/prepare,

Page 21: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

21

absorb,recover,andadapt)areappliedacrossfourrelevantdomains(physical,information,cognitive,social),formingamatrixofpotentialsecuritymetrics.SecondisaquantitativeapproachbasedonNetworkScience,inwhichfeaturessuchasnetworktopologiescanbemodeledtoassessthemagnitudeandresponsivenessofthecriticalfunctionalitiesofnetworkedsystems.Validationofmetricsisanoftenoverlookedaspectofthemetricdevelopmentprocess;howeveraseriesofquestionscanserveasaqualitycheckondevelopedmetrics,toensurethattheyareofhighquality.

PermissionwasgrantedbytheUSACEChiefofEngineerstopublishthismaterial.TheviewsandopinionsexpressedinthispaperarethoseoftheindividualauthorsandnotthoseoftheUSArmy,orothersponsororganizations.

1.7 References 

Alberts,D.S.(2002)Informationagetransformation,gettingtoa21stcenturymilitary.Washington,

DC:DODCommandandControlResearchProgram,Retrievedfromhttp://www.dtic.mil/get‐tr‐doc/pdf?AD=ADA457904.

Allen,J.,&Curtis,P.(2011)MeasuresforManagingOperationalResilience.Pittsburgh,PA:Software

EngineeringInstitute,CarnegieMellonUniversity,Retrievedfromhttp://www.sei.cmu.edu/reports/11tr019.pdf

Beasley,M.S.,Branson,B.C.,&Hancock,B.V.(2010)BuildingKeyRiskIndicatorstoStrengthen

EnterpriseRiskManagement.Durham,NC:TheCommitteeofSponsoringOrganizationsoftheTreadwayCommission(COSO).

Bodeau,D.,&Graubart,R.(2011)MITRECyberResiliencyEngineeringFramework,MTR110237.

Bedford,MA:MITRECorporation,Retrievedfromhttp://www.mitre.org/sites/default/files/pdf/11_4436.pdf

Boyer,W.,&McQueen,M.(2008)IdealBasedCyberSecurityTechnicalMetricsforControlSystems.

Retrievedfromhttp://www.if.uidaho.edu/~amm/faculty/Ideal%20Based%20Cyber%20Security%20Technical%20Metrics%20for%20Control%20Systems.pdf

Black,P.,Scarfone,K.,&Souppaya,M.(2008)Cybersecuritymetricsandmeasures.In:Voeller,J.G.

(Ed.),HandbookofScienceandTechnologyforHomelandSecurity,Vol5,Hoboken,NJ:JohnWileyandSons,Inc.

Buldyrev,S.V.,Parshani,R.,Paul,G.,Stanley,H.E.,&Havlin,S.(2010)Catastrophiccascadeof

failuresininterdependentnetworks.Nature464,1025‐1028CIS(TheCenterforInternetSecurity)(2010)TheCISSecurityMetricsv1.1.0.EastGreenbush,NY:

TheCenterforInternetSecurity,Retrievedfromhttps://benchmarks.cisecurity.org/tools2/metrics/CIS_Security_Metrics_v1.1.0.pdf

Cimellaro,G.P.,Reinhorn,A.M.,&Bruneau,M.(2010)Frameworkforanalyticalquantificationof

disasterresilience.EngineeringStructures,32,3639–3649

Page 22: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

22

Collier,Z.A.,Linkov,I.,DiMase,D.,Walters,S.,Tehranipoor,M.,&Lambert,J.H.(2014)Cybersecurity

standards:managingriskandcreatingresilience.Computer,47(9),70‐76Collier,Z.A.,&Linkov,I.(2014)Decisionmakingforresiliencewithinthecontextofnetworkcentric

operations.19thInternationalCommandandControlResearchandTechnologySymposium(ICCRTS),16‐19June,Alexandria,VA,USA.

DefenseScienceBoard(2013)Taskforcereport:resilientmilitarysystemsandtheadvancedcyber

threat.Washington,DC:OfficeoftheUnderSecretaryofDefenseforAcquisition,Technology,andLogistics,Retrievedfromhttp://www.acq.osd.mil/dsb/reports/ResilientMilitarySystems.CyberThreat.pdf

DiMase,D.,Collier,Z.A.,Heffner,K.,&Linkov,I.(2015)Systemsengineeringframeworkforcyber

physicalsecurityandresilience.EnvironmentSystems&Decisions,35(2),291‐300.Eckerson,W.W.(2009)PerformanceManagementStrategies:HowtoCreateandDeployEffective

Metrics.TDWIBestPracticesReport.Renton,WA:TheDataWarehousingInstitute.Retrievedfromhttps://tdwi.org/research/2009/01/bpr‐1q‐performance‐management‐strategies.aspx

ExecutiveOrderNo.13636,ImprovingCriticalInfrastructureCybersecurity,Retrievedfrom

http://www.gpo.gov/fdsys/pkg/FR‐2013‐02‐19/pdf/2013‐03915.pdfGanin,A.A.,Massaro,E.,Gutfraind,A.,Steen,N.,Keisler,J.M.,Kott,A.,Mangoubi,R.,&Linkov,I.

(2015)Resilientcomplexsystemsandnetworks:concepts,design,andanalysis.NatureScientificReports,submitted

Gregory,R.S.,&Keeney,R.L.(2002)Makingsmarterenvironmentalmanagementdecisions.Journal

oftheAmericanWaterResourcesAssociation.38(6):1601‐1612.Kaplan,S.,&Garrick,B.J.(1981)OntheQuantitativeDefinitionofRisk.RiskAnalysis,1(1),11–27Keeney,R.L.,&Gregory,R.S.(2005)Selectingattributestomeasuretheachievementofobjectives.

OperationsResearch53(1),1‐11Igure,V.,Laughter,S.,&Williams,R.(2006)SecurityIssuesinSCADANetworks.Computersand

Society,25(7),498‐506Linkov,I.,Eisenberg,D.A.,Plourde,K.,Seager,T.P.,Allen,J.,&Kott,A.(2013a)Resiliencemetricsfor

cybersystems.EnvironmentSystems&Decisions,33(4),471‐476Linkov,I.,Eisenberg,D.A.,Bates,M.E.,Chang,D.,Convertino,M.,Allen,J.H.,Flynn,S.E.,&Seager,T.P.

(2013b)Measurableresilienceforactionablepolicy.EnvironmentalScience&Technology,47(18),10108–10110

Linkov,I.,Bridges,T.,Creutzig,F.,Decker,J.,Fox‐Lent,C.,Kröger,W.,Lambert,J.H.,Levermann,A.,

Montreuil,B.,Nathwani,J.,Nyer,R.,Renn,O.,Scharte,B.,Scheffler,A.,Schreurs,M.,&Thiel‐Clemen,T.(2014)ChangingtheResilienceParadigm.NatureClimateChange,4,407–409

Page 23: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

23

Manadhata,P.K.,&Wing,J.M.(2011)AnAttackSurfaceMetric.IEEETransactionsOnSoftwareEngineering,37(3),371–386.

Marr,B.(2010)HowtodesignKeyPerformanceIndicators.MiltonKeynes,UnitedKingdom:The

AdvancedPerformanceInstitute.Retrievedfromwww.ap‐institute.comMcIntyre,A.,Becker,B.,&Halbgewachs,R.(2007).SecurityMetricsforProcessControlSystems.

SAND2007‐2070P.Albuquerque,NM:SandiaNationalLaboratories,U.S.DepartmentofEnergy.

McKay,S.K.,Linkov,I.,Fischenich,J.C.,Miller,S.J.,&ValverdeJr,L.J.(2012)EcosystemRestoration

ObjectivesandMetrics,ERDCTN‐EMRRP‐EBA‐12‐16.Vicksburg,MS:U.S.ArmyEngineerResearchandDevelopmentCenter

Mell,P.,Scarfone,K.,&Romanosky,S.(2007)ACompleteGuidetotheCommonVulnerabilityScoring

SystemVersion2.0.Morrisville,NC:ForumforIncidentResponseandSecurityTeams.Retrievedfromhttps://www.first.org/cvss/cvss‐guide.pdf

NationalSecurityAgency(NSA).(2010).AFrameworkforAssessingandImprovingtheSecurity

PostureofIndustrialControlSystems(ICS).Retrievedfromhttps://www.nsa.gov/ia/_files/ics/ics_fact_sheet.pdf

Neely,A.,Richards,H.,Mills,J.,Platts,K.,&Bourne,M.(1997)Designingperformancemeasures:a

structuredapproach.InternationalJournalofOperations&ProductionManagement,17(11),1131‐1152

NIST(2014)FrameworkforImprovingCriticalInfrastructureCyberSecurity.Version1.0.

Gaithersburg,MD:NationalInstituteofStandardsandTechnology.Retrievedfromhttp://www.nist.gov/cyberframework/upload/cybersecurity‐framework‐021214‐final.pdfPfleeger,S.L.,&Cunningham,R.K.(2010)Whymeasuringsecurityishard.IEEESecurity&Privacy,

8(4),46‐54Pollet,J.(2002)DevelopingaSolidSCADAStrategy.Sicon/02–SensorsforIndustryConference.

Houston,Texas,USA.19‐21,November2002.Reichert,P.,Borsuk,M.,Hostmann,M.,Schweizer,S.,Sporri,C.,Tockner,K.,&Truffer,B.(2007)

Conceptsofdecisionsupportforriverrehabilitation.EnvironmentalModelingandSoftware22:188‐201.

Roege,P.E.,Collier,Z.A.,Mancillas,J.,McDonagh,J.A.,&Linkov,I.(2014)Metricsforenergy

resilience.EnergyPolicy,72(1),249–256Saltzer,J.H.(1974)ProtectionandthecontrolofinformationsharinginMultics.Communicationsof

theACM,17(7):388‐402.Stouffer,K.,Falco,J.,&Scarfone,K.(2011)GuidetoIndustrialControlSystems(ICS)Security.Special

Publication800‐82.Gaithersburg,MD:NationalInstituteofStandards.Retrievedfromhttp://csrc.nist.gov/publications/nistpubs/800‐82/SP800‐82‐final.pdf

Page 24: Security Metrics in Industrial Control Systems · of the system, intended uses and unintended misuses from users, etc. Environment, abstraction, and context affect security Systems

24

USDepartmentofEnergy(2002)21StepstoImproveCyberSecurityofSCADANetworks.Washington,DC:USDepartmentofEnergy.Retrievedfromhttp://energy.gov/sites/prod/files/oeprod/DocumentsandMedia/21_Steps_‐_SCADA.pdf

USDepartmentofEnergy(2009)NationalSCADATestBed:Enhancingcontrolsystemssecurityinthe

energysector.Washington,DC:USDepartmentofEnergy.Retrievedfromhttp://energy.gov/sites/prod/files/oeprod/DocumentsandMedia/NSTB_Fact_Sheet_FINAL_09‐16‐09.pdf

Vespignani,A.(2010)Complexnetworks:thefragilityofinterdependency.Nature,464(7291),984–

985Williamson,R.M.(2006)Whatgetsmeasuredgetsdone:areyoumeasuringwhatreallymatters?

Columbus,NC:StrategicWorkSystems,Inc.Retrievedfromwww.swspitcrew.comZeng,W.,&Chow,M.Y.(2012)OptimalTradeoffBetweenPerformanceandSecurityinNetworked

ControlSystemsBasedonCoevolutionaryAlgorithms.IEEETransactionsonIndustrialElectronics,59(7):3016‐3025.

Zhu,B.,Joseph,A.,&Sastry,S.(2011).AtaxonomyofcyberattacksonSCADAsystems.InInternetof

things(iThings/CPSCom),2011internationalconferenceoncyber,physicalandsocialcomputing(pp.380‐388).