Accordingtothelatest”2023CloudStatusSurveyReport”releasedbyFlexera,amongthe750companiesinterviewed,36%ofthecompaniessaidthattheircloudcostexpenditureexceededexpectations,andanother9%oftheircloudcostsseriouslyexceededtheirexpectations,strong>Enterprisesurgentlyneedeffectivemeanstoreducecloudcosts:
▲Picture1
Atthesametime,Amongtheseenterprises,asmanyas87%usemulti-cloud.Theadvantageofmulti-cloudarchitectureisthatitcaneasilyprovidehigh-availabilitybusinessdeployment,localizeddeploymentthatmeetssecuritycompliance,andpubliccloudelasticity.andothercapabilities,butifthereisalackofcorrespondingcostmanagement,itwilleasilyleadtoanincreaseincloudcosts.
▲Picture2
Inordertosolvethecostproblemundermulti-cloudandmulti-cluster,Karmadatooktheleadinproposingandimplementinganewcross-clusterHPA(FederatedHPA)thatsupportsmulti-indicatorsandmulti-strategies.Withthecombinationofdatacenter+publiccloud,localdatacenterresourcesareusedfirstforbusiness.Whenlocalresourcesareinsufficient,theunlimitedelasticityofthepubliccloudcanbeusedtousecloudresourcesondemand,therebysavingcloudcosts.
2.5%offcloudserver!HUAWEICLOUDoffershighcashbackuponplacinganorder!
FederatedHPA
KarmadaFederatedHPAcanautomaticallyscaleservicesbasedonCPU/Memoryutilization,andcanalsoscaleservicesbasedonvariouscustomindicators.AnexampleofitsYAMLconfigurationis:
apiVersion:autoscaling.karmada.io/v1alpha1 kind:FederatedHPA metadata: name:nginx spec: scaleTargetRef: apiVersion:apps/v1 kind:Deployment name:nginx minReplicas:1 maxReplicas:10 metrics: -type:Resource resource: name:cpu target: type:Utilization averageUtilization:10
ByusingFederatedHPA,applicationcross-clusterelasticitycanbeachieved.Asshowninthefigurebelow,theapplicationisdeployedinthecluster1cluster.Whenthetrafficpeakarrives,theapplicationcanbeautomaticallyexpandedinthecluster1clusterfirst.Whentheresourceofcluster1islimited,theapplicationcanautomaticallyexpandExpandcapacityinthecluster2cluster.
▲Picture3
Ofcourse,KarmadaFederatedHPAbringsnotonlycross-clusterelasticscaling,butalsothefollowingcoreadvantages:
1.Foramulti-clusterbusiness,eachclusterhascorrespondingHPAresourcestoscalethebusiness.However,itisinefficienttomanagetheseHPAconfigurationsseparately,andusingKarmadaFederatedHPAcanuniformlyconfigurethescalingofmulti-clusterservicesandsimplifytheprocess.
2.Foramulti-clusterbusinessusingKarmadaFederatedHPA,thenumberofinstanceswillchangeastheloadchanges.Forthesenewlyaddedorreducedinstances,userswanttoscaledifferentlyindifferentclusters,suchasaccordingtotheproportionofavailableresources,staticweightproportion,priority,etc.KarmadaFederatedHPAcanalsomeetthedemandsofsuchmulti-clusterdifferentialscaling.
3.Foramulti-clusterbusinessthatusesKarmadaFederatedHPA,whenaclustercannotberesilientduetoafailure,Karmadawillberesilientinothernormalclusters,soastosolvethesinglepointoffailureproblem.
Unifiedelasticscalingconfigurationtoimprovemanagementefficiency
Inthetraditionaldeploymentmethod,ifuserswanttoconfigureautoscalinginmultipleclusterstomatchtheloadofbusinessrequests,theyneedtomanagetheHPAintheclustersonebyone,whichiscumbersomeanderror-prone,asshowninthefollowingfigure:
▲Picture4
UsingKarmadeFederatedHPAcanrealizetheunifiedconfigurationofelasticscalingofmulti-clusterservices.Inthecaseofalargenumberofclusters,itcangreatlyimproveefficiency,asshowninthefollowingfigure:
▲Picture5
ThroughasingleFederatedHPAobject,Karmadawillautomaticallymonitorthebusinessloadofmultipleclusters,scaleindifferentclustersaccordingtotheconfiguredstrategy,andfinallymatchthebusinessloadofmulti-clusterservices.
2.5%offcloudserver!HUAWEICLOUDoffershighcashbackuponplacinganorder!
Priorityexpansionoflow-costclusterbusinesstoreducecloudcosts
Formultipleclustersdeployedinthesamebusiness,theremaybecostdifferences.UserscanuseFederatedHPAtoprioritizethebusinessexpansionoflower-costclustersandachievelowercloudcostconsumption.Forexample,thecostofusinglocaldatacenterclustersislower,andpubliccloudvendorsThecostoftheprovidedhostingclusterishigher,therefore,usersaremorewillingtoexpandtheirbusinessinthelocaldatacenter.
Belowwegiveanexampleofprioritizingtheexpansionoflocalclusterservices:
apiVersion:autoscaling.karmada.io/v1alpha1 kind:FederatedHPA metadata: name:nginx spec: scaleTargetRef: apiVersion:apps/v1 kind:Deployment name:nginx minReplicas:1 maxReplicas:10 metrics: -type:Resource resource: name:cpu target: type:Utilization averageUtilization:80 --- apiVersion:policy.karmada.io/v1alpha1 kind:PropagationPolicy metadata: name:nginx spec: resourceSelectors: -apiVersion:apps/v1 kind:Deployment name:nginx placement: clusterAffinities: -affinityName:local-cluster clusterNames: -local-cluster1 -affinityName:cloud-cluster clusterNames: -local-cluster1 -huawei-cluster1 replicaScheduling: replicaDivisionPreference:Weighted replicaSchedulingType:Divided weightPreference: dynamicWeight:AvailableReplicas
TheabovePropagationPolicyisconfiguredwithtwoclustergroups,thelocalclustergroup(local-cluster)andthecloudclustergroup(cloud-cluster).WhenKarmadaexpandsthebusiness,itwillfirsttrytoexpandthebusinessinthelocalclustergroup,ifitfails(lackofresources),itwillcontinuetoexpandthebusinessoftheclustergrouponthecloud,sothatwhenthelocalclusterresourcesaresufficient,thebusinessofthelocalclusterwillbeexpandedfirst,andlowercloudcostconsumptionwillbeachieved.
Summary
FederatedHPAprovidesuserswithcross-clusterelasticscalingcapabilities,combinedwithrichPropagationPolicy/ClusterPropagationPolicyschedulingpolicies,itcanmeetdifferentcross-clusterscalingscenarios.
Karmadawillcontinuetoexploremorecross-clusterscalingscenarios,includingtimedfederatedHPAanddistributedmulti-clusterHPA.Ifyouhaveanyinterestingideas,youarewelcometodiscussandsharethemintheKarmadacommunity.
2.5%offcloudserver!HUAWEICLOUDoffershighcashbackuponplacinganorder!