Transcription

StudyonSelf- ‐HealingandSelf- ogeneousNetworksKevin amrail.com/frogswitch/turnouts.asp

Overview Introduction Background SoftwareDefinedNetworking Overview CurrentApplications Issues NewNetworkingLandscape IoT,VANET,SmartGrid Issues Solutions Self- ‐Healing Self- ‐Optimization Conclusion ctivity/facilities/futureinternet/

Introduction ProblemStatement Currentnetworkinginfrastructureiscomplex ManagingrequiresAutonomicSystems IBM Self- ‐ConfigurationSelf- ‐ProtectionSelf- ‐HealingSelf- ‐Optimization Explorationofnewnetworkingtechnologies. SoftwareDefinedNetworksbility feature of SDN can be used to achieve self-* attributesof the autonomic systems. The combination of autonomicsystem and SDN can be used to control and manage the network infrastructure. The purpose of both the technologiesis to overcome the growing complexity of the network. Theapplication of autonomic properties on SDN can unleash thetrue potential of future networks. The reliability of SDN canbe improved with the application of self-healing principle.SDN is described by ONF [1] as an architecture in whichthe control and data planes are decoupled, network intelligenceand state are logically centralized, and the underlyingImage: Ivan Pepelnjak, NIL Data Communications, TechTargetnetwork infrastructure is abstracted from om/feature/BGP- ‐essentials- ‐he- ‐protocol- ‐that- ‐makes- ‐the- ‐Figure 1 illustrates the layeredarchitectureofSDN.Internet- ‐workApplicationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDNestablithe alure oc[8] [9]the balink. Iinstallthe prwork ha modnumbesuch amay oing bostoringtionalbut bemore tIn pis notwhichpath snotificresultsAfter sentriesextra c

Introduction ProblemStatement Currentnetworkinginfrastructureiscomplex ManagingrequiresAutonomicSystems IBM Self- ‐ConfigurationSelf- ‐ProtectionSelf- ‐HealingSelf- ‐Optimization Explorationofnewnetworkingtechnologies. SoftwareDefinedNetworksbility feature of SDN can be used to achieve self-* attributesof the autonomic systems. The combination of autonomicsystem and SDN can be used to control and manage the network infrastructure. The purpose of both the technologiesis to overcome the growing complexity of the network. Theapplication of autonomic properties on SDN can unleash thetrue potential of future networks. The reliability of SDN canbe improved with the application of self-healing principle.SDN is described by ONF [1] as an architecture in whichthe control and data planes are decoupled, network intelligenceand state are logically centralized, and the underlyingImage: Ivan Pepelnjak, NIL Data Communications, TechTargetnetwork infrastructure is abstracted from om/feature/BGP- ‐essentials- ‐he- ‐protocol- ‐that- ‐makes- ‐the- ‐Figure 1 illustrates the layeredarchitectureofSDN.Internet- ‐workApplicationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDNestablithe alure oc[8] [9]the balink. Iinstallthe prwork ha modnumbesuch amay oing bostoringtionalbut bemore tIn pis notwhichpath snotificresultsAfter sentriesextra c

Background– TraditionalNetworks NetworkFlow SequenceofPackets SourcetoDestination RoutingProtocolsSwitchASwitchBSwitchCA- ‐ C LinkStateRoutingProtocolsA- ‐ C OSPF(Dijkstra Algorithm) ConnectivityGraphofRouters DistanceVectorRoutingProtocol RIP(Bellman- ‐FordAlgorithm)C– A NeighborGraphofRouters ExteriorGatewayProtocol BGP RoutingBetweenAutonomousSystemsC– A

Background– TraditionalNetworks TraditionalNetworkRouter MaintainsamappingbetweenDestinationAddressandPort EncapsulatesTwoFunctions Control Expertsconfigurerouter ProtocolsbuildtheRoutingTable DataTransfer UsestheRoutingTable ForwardsFlows (Packets)Image: Computer Desktop Encyclopedia, The Computer Language Company, nition.htm

Background– TraditionalNetworks TraditionalNetworkRouterDataLineDataLine MaintainsamappingbetweenDestinationAddressandPort EncapsulatesTwoFunctions ControlControl Expertsconfigurerouter ProtocolsbuildtheRoutingTable DataTransfer UsestheRoutingTable turnouts.asp

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) SpearheadedbySunin1995 Designedtoallowsoftwareswitchingfornetworks. ndprotocols. DecouplesControlandData Allcontrollogicismovedintoacentralizedcontroller. HardwareisreplacedwithSoftware ionLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) EaseofConfiguration orwardingrules. Withouthavingtoconfigureindividualswitches. FundamentallynewArchitecture forwarded. neratesanewrule.ApplicationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) Features Fastconvergencetimeswhenpoweredon. Centralizedcontrollerprovidesfine- . Simplifiesnetworkdevices. Anydevicecannowbeanetworkdevice. Simplepacketforwarders.ApplicationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) CurrentApplications gtheirDataCenterNetworks(DCN)s ntedwithin5years(2019) zations.ApplicationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) MotivatingConcerns ExistingSDNsolutionsforDCNsassumeTCP,Anycast trafficthatislooselycorrelated. ExistingSDNworkassumesstatic,homogeneousdevices. SDNalsoposesreliabilityconcerns. ne,ortheDataPlane icationLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

SDN is described by ONF [1] as an architecture in wthe control and data planes are decoupled, network ingence and state are logically centralized, and the undernetwork infrastructure is abstracted from the applicatFigure 1 illustrates the layered architecture of SDN.SoftwareDefinedNetwork(SDN) Resilience toration Existingschemesaddbackuppathsforeachflowentry. Fornon- ‐DCNoperations,thisissimplynotpractical andwouldoverloadthecontroller. sandthecontrollermustinstallnewrulesoneachswitch. nLayerNetwork ApplicationsControlPlane LayerSDN controller/ NetworkOperating SystemOpenFlowDataPlaneLayerNetworkSwitchesImage: [2]Figure 1: Architecture of SDN

ModernNetworkingLandscape ModernNetworking Wireless Mobile Heterogeneous ,mmWave,802.11pVANET Applications VehicleCommunications ics/activity/facilities/futureinternet/

ModernNetworkingLandscape Applications InternetofThings(IoT) FieldbuiltonHeterogeneousWirelessDevices . Devicesdeployedinanuncoordinatedmanner. Multi- ‐ObjectiveOptimization QoS inaDCNfocusesonsingleoptimizations. QoS forIoT addsindelay,jitter,packetloss,throughput meenergy.co.uk/what- ‐smart- ‐home

SDNforModernNetworking Benefits Candifferentiateflowschedulingoverad- ‐hoc,heterogeneouspaths. nginterface. Vehicle- ‐Vehicleover802.11p Vehicle- ‐DoToverCellLTE /what- ‐smart- ‐home

ProblemswithSDNforIoT:Self- ‐Healing MotivatingApplication SmartGrid Packetflowscontrolrelaysatpowerstations. ce. tions,causingcascadefailures. 2003Blackout r,10millionpeopleaffected.Image: "Map of North America, blackout 2003" by Lokal Profil. Licensedunder CC BY-SA 2.5 via Commons https://commons.wikimedia.org/wiki/File:Map of North America, blackout 2003.svg#/media/File:Map of North America, blackout 2003.svg

culates the shortest path for each of the backup link. OSHmodule queries the network management modules to collect the network information. Based on the informationcollected, it calculates the optimal path for achieving theimproved recovery.ProblemswithSDNforIoT:Self- ‐Healing Problem Recoverytakestoolongusingrestorationtechnique. tiontechnique. SolutionTopologyNetworkPolicyDiscovery nResourceManagement Management ManagementSDN Controller ionBaseNotificationModuleDataPlane Treatallflowsfromtothesamelinkwithasinglerule. Pre- ‐allocateasinglebackuprule. licationMgmt.Control Plane RapidRecovery[2]LoadBalancingImage: [2]Figure 2: Proposed architecture of optimized selfhealing mechanism

culates the shortest path for each of the backup link. OSHmodule queries the network management modules to collect the network information. Based on the informationcollected, it calculates the optimal path for achieving theimproved recovery.ProblemswithSDNforIoT:Self- ‐Healing TopologyDiscoverymanagesroutingtopologyinplace. LoadBalancingmoduleestimatesthenetworkload. RoutingModulecalculatesshortestpathsperflow. OnFailureTopologyNetworkPolicyDiscovery nResourceManagement Management ManagementSDN Controller ionBaseDataPlane tworkSwitchManagementLevelApplicationMgmt.Control Plane OptimizedSelf- lPathValidatePathWRTQoSImage: [2]SendnewFlowRoutinginformation. Figure 2: Proposed architecture of optimized self-healing mechanism

rules. This process is repeated for each of the disruptedtflow. When the switch receives the flow modification mes-f(TDFS,f TFM,f TUPDATE,f ) TPROP (2)T R TFD sage, all thematching rule from the flow table are modif 0fied in time (TUPDATE ). The propagation time (TPROP ) ofbthe failure notification message from switch to the controller(Our proposed schemefor RR is reactive in nature. Aftercontributes to the recovery process. The recovery time (TR )tfailure detection, thetakenaffectedswitchthebyflowreroutby thisschemehandlesis expressedequation2.tN!ProblemswithSDNforIoT:Self- ‐Healing ing without any controller intervention.Therefore, the timeN!uEvaluationofSelf- ‐Healingcomplexity of our proposedschemedepends protection(TDFS,f TFM,f TUPDATE,f) TPROP (2)T R TFDRRf 0 ModelEvaluationon the time a switch takes to detecta failure (TFD ) and theimeforstandardReactiveLinkOur Recoveryproposedschemereactivein nature. After 99%reductiontimeofbackupflowsneeded) forof RRtheisgroupentriesto changethe alivestatus T(TAS5FailDetectTaffectedime Propagationtime stored. which corresponds tothe failed link. According to Sharmadisruptedingany Sumoverallcontroller intervention.Therefore, the time Immediaterestorationofsaervicewithtakeswithoutet al combinationofscheme dependsproposed RR protectionapossiblysub- ‐optimalbackuppath. complexity of ourpto modify the alive-statusof n the timea switchtakesEntry.to detecta failure(TFD ) and the Optimalbackuppathispushedaftertime to change timetosendthemodificationmessage,) ofthegroup entriesathe isalivestatus (TAScalculatedbyequationtime(TR ) taken by our RR flowtable).which correspondsto thefailedlink. According to Sharmao3.et al [8], a switch takes an approximately 5.8 microsecondsotoTmodifyalive-status Tthe TAS of one Group Entry. The(3) recoveryrRFDtime (TR ) taken by our RR scheme is calculated by equationfl3. RecoveryTimeforRapidRecovery FailDetectTime Ttochange T AliveStatus(3)eT econds.

ProblemswithSDNforIoT:Self- ‐HealingFlow Table 0 for BRule VLN IDInstructions11Remove VLAN TAG 1 & forward packets to flow table 1. .*NForward packets to flow table 1Flow Table 0 for CRule VLN IDInstructions21Remove VLAN TAG 2 & forward packets to flow table 1. .*NForward packets to flow table 1Flow Table 1 for CFlow Table 1 for BIP srcIP Dst Ether typeInstructionsIP srcIP Dst Ether typeInstructionsRuleRule1 192.168.1.8 192.168.1.1 0x0800 Forward Packets to Group #11 192.168.1.1 192.168.1.8 0x0800 Forward Packets to Group #12 192.168.1.9 192.168.1.2 0x0800 Forward Packets to Group #12 192.168.1.2 192.168.1.9 0x0800 Forward Packets to Group #1.Group Table for BGroup Table for CGroup ID Group Type Action Buckets DescriptionGroup ID Group Type Action Buckets DescriptionB1:PrimaryB1:PrimaryOutput to port 3Output to port 1linklinkFlow2'sPacketHeaderFastFastPush VLAN tag1PushVLANtag1IPsrcIPDstFailover 2 and output to B2: BackupFailover 1 and output to B2: Backup192.168.1.2 192.168.1.9pathpathport 4port 2Flow22.2.Link ID # 2BC2(BC)1143 Link ID # 124(CB)DFlow 1Flow 1's Packet HeaderIP srcIP Dst192.168.1.1 192.168.1.8A14EFlow 3Flow Table # F23FFlow1,2,3Rule IP src IP Dst VLAN IDInstructions1**2Forward Packets to Port 22**1Forward Packets to Port 1-----------Image: [2]

Mods are sent out to the switches for re-routing.B. Scenario 1: Fast Recovery for Smart Grid CommunicationsApplying this algorithm, multiple measurements have bexecutedto verify and analyse its effect. The experimental deThis scenario deals with enabling fast recovery after disturofthisscenariocan be conceived by means of Figure 2: Mbance of a communication link. Providing such functionality isreports respectively SV messages, both containing measuremof great importance for ensuring reliable operation of commuvalues, are transmitted from Server 1 to Client 1 using einication networks in critical environments such as substationsthe upper path (via Switch 2) or the lower path (via Switchof power systems. In particular, alternative routes through theDuring transmission, one of the active communication linetwork need to be established immediately, guaranteeing theis disconnected by a) physical disconnection of an interftransmission of monitoring and control traffic. Therefore aor b) by software command. Figure 4 (top) shows the oveproactive algorithm for calculating alternative paths throughrecovery times of SV and MMS traffic, in terms of a cumulathe network has been developed and integrated into the SDNdistributionfunction, considering both cases. In case ofController. The algorithm’shas been assessed2014 IEEEperformanceInternational Conferenceon Smart GridCommunicationsmessagesandlink disconnection by command, the mean doby measuring the duration of traffic interruption as well astime of transmission amounts to 87.16 ms (median: 85.17processing times at the controller and switches.whereas for physical link disconnection the mean delay increFirst, a brief description of the algorithm, which is applied fortoMMS360.64ms (median: 358.80ms). Hence,port Trafficstatus iven:MMS TrafficBackground Real-Time TrafficTrafficBackgroundReal-Timebythe OS inducesadditionalTotaldelayin the range of 210 to 305Background Data TransferTotal TrafficBackgroundData TransferTrafficA more complex behaviour can be observed for TCP-baSDN ControllerMMS reports due to reliability mechanisms, which applyknowledgements (ACKs) and retransmissions. AccordinglyRe-routing ofofcoveryRe-routingtime dependson the following TCP-specific parameSwitch 2BackgroundBackgroundwhicharesettoWindows7 default configuration: RetransDataTrafficData and RealServer1Client 1ms), acknowledgement frequesion Time-Out(RTO)(300Time Traffic(2packets)anddelayedacknowledgementtimer (50 ms). ThSwitch 1Switch 4fore, Figure 3 distinguishes three different cases, which moccur when a link is disconnected during TCP based commcation, explaining the effects encountered in FigureSwitch 2Switch 24.Switch 3MMS Case 1: In Case 1, the link is disconnected beforeSwitch 3Switchtransfer.3during the transmission window’s first packetControl NetworkServer 2packetwillnotbereceivedbytheclientandnoACKis issData pses.TMinimum Data RateFig. 2: Setup of the portswithrecoverytimesintherangGuarantee forLink Reservation forControl NetworkSmart Grid TrafficSmartbyGridTraffic300 ms after the link is disconnectedcommand.ProblemswithSDNforIoT:Self- ‐Healing Wouldbenefitfrom[2]Datarate [Mbps]20 40 60 80 100 0 NovelAdditionsforSelf- ‐Healing0 MultipleQoS Levels rallcurrentflows. 0150200Time [s] DedicatedflowsgettheirownFig. 5: Scenario 2a: Load Management at Switches 2 and 3links.to enqueue traffic flows using these queues, mapping the QoSrequirements of different traffic classes being equivalent toDatarate [Mbps]20 40 60 80 100 0 epost- ‐failureforbackuprouteinstallation.4240Datarate [Mbps]20 40 60 80 100 SmartGridApproach[5]Datarate [Mbps]20 40 60 80 100 EvaluationofSelf- ‐Healing050100150Time [s]200Fig. 6: Scenario2b: Link Reservation at Switch 3Images:[5]5) Sorting of Lower Priority Flows: Else, lower priorityflows are sorted into a list depending on their overlap with

ProblemswithSDNforIoT:Self- ‐Optimization Objective Reduceenergyusagefornetworkinfrastructure.[1] cted14%by2020. MotivatingApplication CampusNetwork rkinghardware,configuredtohandlepeaktraffic. Realtrafficoccursinpatterns. ities/futureinternet/

2014 IEEE 3rd International ConferenProblemswithSDNforIoT:Self- ‐Optimization DetermineMinimumSwitchestoPower NP- ‐Hard,butcanbeformulatedwithaMILP MultipleConstraints StrategicGreedyHeuristic desifneeded. ShortestShortestPathFirst(SPF) LongestShortestPathFirst(LPF) SmallestDemandFirst(SDF) HighestDemandFirst(HDF)Image: [1]Fig. 2.Campus network to be investigated

2014 IEEE 3rd International Conference on Cloud Networking (CloudNet)ProblemswithSDNforIoT:Self- ‐Optimization DetermineMinimumSwitchestoPower NP- ‐Hard,butcanbeformulatedwithaMILP MultipleConstraints StrategicGreedyHeuristic desifneeded. ShortestShortestPathFirst(SPF) LongestShortestPathFirst(LPF) SmallestDemandFirst(SDF) HighestDemandFirst(HDF)network to be investigatedFig. 3.Image: [1]Optimal network configuration for a low traffic utilization at night

or distribution switches have to be powered on, resulting in asharp increase in energy consumption.ProblemswithSDNforITheoT:Self- ‐Optimizationperformance of four heuristic algorithms is shown inB. Strategic Greedy Heuristic performanceFigure 7 and for 8 mesh and campus network, respectively. Evaluation LPFprovidesthegreatestenergysavings. panningtreeisproduced. edtobelaterconnected,resultinginsub- ‐optimalgraph.Fig. 7.Average energy savings for different strategies in a mesh networkFor the mesh network, LPFoutperformsthe other three byImage:[1]up to 5% more total energy saving. This clearly makes sense if

RESULTSrview Network Applicationp/down state of the network switches and their portso the asynchronous messages exchanged between theontroller and the switches. Data collected by thiselps the management layer to keep a global view ofThis section presents the scenario, the main elements of theproposed architecture (Figure 4) used in the testbed, andexperiments carried out in order to evidence how the QoE/QoSmanagement mechanism reacts after detecting nonconformities inthe service provided. An overview of the network topology usedin the experiments can be seen in Figure 5ProblemswithSDNforIoT:Self- ‐Optimization ObjectiveTeleconsultationte CalculationApplication EnhanceUserExperience.[3]stics Countermobilenetworking. QualityofExperience(QoE)of passive and active monitoring for measuring MotivatingApplicationwork metricsin different aggregationlevels. It alsoflow latency, errorrateand jitter by insertingSmartHomees in the network. MultipleQoE levels. DistanceLearning HealthProviderLinkthe management layer fortranslating the high level OnlineGamingcies into control rules. Thecontroller and the controlfor instance, routing) use those rules for calculatingformal andinformal carersWireless orWired ClientsTele-educationInternetthe available path(s), calculating the route(s) basedQoS isThecritical,however,usercontrol” of Mapping Rules.adequacyof a routexperienceisnowthe solicitation is perceivabledetermined ebythe networkimportantforuasser- ‐centric,by a set of performancemetrics, suchlatency.EthernetRemote server opHome Area NetworkGaming ServicesLegendOpenFlow MessagePhisycal ConextionDataAccessNetworkMetropolitan Area NetworkRemoteserverfor gamesService ProviderNetworkWide AreaNetworkping RulesFigure 5. Topology of the Network used in the Experiments[3]5.1 Description of the Image:ScenarioThe scenario consists of three sub-categories of services, provided

ProblemswithSDNforIoT:Self- ‐Optimization QoETC ectations. Needtomeethedonicandpragmaticneeds. Autonomically nce. MeasuredfromClientMachine ReportedtotheController UserExperienceInformationstoredinaKnowledgeBase back.User InterfaceQoE Aware Semantic EngineCapture Dimensions QoEQoE EvaluationQoS MeasurementsQoE OntologyQoS / QoEManagement Application(QoE Aware Engine, KB, Application Servers(eHealth, VoIP, VoD,Games, News.)LAYER MANAGEMENTNorthbound API (REST API)Rule BaseApplications ModulesMappingRules(Admission Control, Overview Network,Route Calculation, Statistical Counter)JavaAPIWuainilStandard SDNControllerLAYER CONTROLSouthbound API (OpenFlow Protocol API)Switchs OpenFlowLAYER DATAImage: [3]for QoE ManagementFigure 4. Functional Architecture4.1 Management Layer4Tve

Thr is the throughput; Jt is the jitter and Plr is the packet loss rate.Coefficients α, β, γ, and ε are calculated particularly for each case.ProblemswithSDNforIoT:Self- ‐OptimizationWith all information captured about the QoE dimensions, pluscation Serversth, VoIP, VoD,mes, News.)user and network status information, the semantic engine learnsabout the user’s experience using a service and is able to provide QoEinformationof(QoE MeanOpinionScoreMOS)degradation to the network controller. Table 2illustratesexamplesthe KB Jrecordsvideo acketLossstreaming.Rates.dard SDNntrollerotocol API) le 2. KB instances used for QoE LearningUSERUSER CONTEXTCONTENTQoSAPPLICATIONQoS esolutionDelayJiiterPLR iceTalkshowMPEG41280X72020ms15ms01232003,0Image: [3]4.1.4 Module for Analyzing and Verifying QoE

210051015202530354045505560Time (second)satisfaction. Thpoll presentedliterature. Concperspective demultidimensiondimensions, coan interdiscipproposal is loliterature, by bprojected for thTo provide anwe propose aperceived by tusing ontologieare mapped anImage: [3]engine proposexperience in tdetect QoE denetwork controtested in an SDA usage scenpresented andthroughput metan alternate rbandwidth avaicomponents oProblemswithSDNforIoT:Self- ‐OptimizationFigure 8. Throughput adaptation policy, after changealternative routeider and IPTVservice Backgrounde network and service providers arewtheasoservicenlyentity. The providertrafficprovidesice access network.inducedonthefinaltest.s provided, an applicationserver wasation server wasconfiguredinaLinux Flowratesrethe server, videos andgamesbweremadelimitedasedonuld watch and play over the network. Anomaintainused to collect dataMOSabout tpatients’vitalacceptabledata are requestedat any time by theuring or after teleconsultation.levels.machine, the QoE/QoS managingred with a semantic engine, a KBnetwork adaptation policies base. Thees REST API to communicate with theTele-educationTeleconsultationGame on DemandThroughput (Mbps) QoE Evaluation10Thefourth experiment was to test the throughput y, with flowrate ResultslimitersandQoE/QoSwithoutalternateroute. In this8experiment,only route 1 was kept to test the system behavior to7accommodatebest effort flows and QoS guarantee flows. QoS6guaranteesof 4,5Mbps, 2,5Mbps and 2Mbps were established,5respectivelyfor Teleconsultation, Tele-education, and GoD and43withbandwidth saturation. Up to 10 seconds, the throughput was2keptin all sessions. But those new sessions compromised the1bandwidthuntilwas0 10 20 30 40of50 active60 70 80 90flows,100 110 120 130140 150 the160 170 problem180 190 200 210 220230 240 detected250 260 270 280 290and300Time (second)fixed, as illustrated in Figure 9.10results of measurements without9 Figure 6. ThroughputThroughput Adaptation Policy - Limiters Flow Rates8QoE/QoS mechanism7Tele-educationTeleconsultationGame on Demandheavy games downloadsUpdate eHealth softwareThesecond experiment was to evaluate the same flow, without65andafter enabling the QoS management mechanism. The flows4withQoS guarantee correspond to 5Mbps for Teleconsultation, 332Mbpsfor Tele-education and 2Mbps for GoD. Up to 90 secondsall1 flows compete for the total bandwidth available, and at 900seconds,mechanism,05after10activation1520of the25 QoS30 control35404550 each55 flow60Time (second)receives its web portion, as can be seen in Figure 7.Figure 9. Throughput adaptation policy, after change limitingTrhoughput (Mbps)P-Link with 10Mbps capacity was used,ble / PON service.10

ProposedIoT Architecture ProposedarchitecturetoaddressbasicIoT concerns[4] Task- ‐ResourceMatchingModule eling. ServiceSolutionSpecificationModule poseds

application of autonomic properties on SDN can unleash the true potential of future networks. The reliability of SDN can be improved with the application of self-healing principle. SDN is described by ONF [1] as an architecture in which the control and data planes are decoupled, network intelli-