WIXLIB

5G PPP Technology BoardEdge Computing for 5G NetworksWhile the beginning of Big Data can be set in the 90s, it is really in the last decade thatData explosion took place.The application of AI to Big Data increased the need for larger Data sets to train inferencemodels. Public cloud has played an instrumental role in this space, but the more the dataset grows, the more difficult is to move the data.That s why Dave McCrory in 2010 introduced the concept of “Data Gravity” 5. The ideais that data and applications are attracted to each other, similar to the attraction betweenobjects as explained by the Law of Gravity.ServicesLatencyDataThroughputAppsFigure 3: The Data Gravity concept introduced in 2010 by Dave McCroryIn mobile networks, Applications (Apps) run in smartphones, whereas Services run in theOperator s Core Network (IMS Services) or in Internet (commonly in Public clouds).Apps and Services are therefore very ‘far away’ from each other as perceived from a timepoint of view (e.g., typically more than 50-100 ms). This is because exchanged data haveto travel through a set of networking entities and devices (e.g., aggregation points, IProuters, Peering routers, Interconnection hubs). It is not uncommon that the links to thesedevices can get congested, and therefore it is impossible to guarantee any end-to-endQuality of Service (QoS) or throughput.In such an environment Edge Computing plays a key role as the enabling technology toshorten the distance between Users (Apps) and Services (Data) and enable guaranteedLatencies and Throughputs, as required by services and applications. These requirementshave become apparent especially with the digitization of Verticals such as Industry 4.0,Collaborative and Automated Driving, E-Health etc.61.2 Why is Edge Computing critical for 5G?The 5G Network is the most recent Mobile Network generation defined by 3GPP.Looking back at the evolution of Mobile Networks, before the introduction of a newgenerations it has always been a problem to predict which use cases would have been theones mostly valued by Users: 3G Networks were designed mainly for Voice (Circuit Switched) and limitedInternet browsing. However, Smartphones appearance in 2007 revealed Apps asthe main use case: people used to spend 90% of their mobile usage time withApps7.5 n-the-clouds/65G PPP, White paper, “Empowering Vertical Industries, Through 5G Networks”, VerticalsWhitePaper-2020-Final.pdf7 https://buildfire.com/app-statistics/Dissemination level: PublicPage 7 / 96

5G PPP Technology Board Edge Computing for 5G Networks4G Networks were designed for Data services, modelling Voice service as Data(VoLTE), while most of the traffic in 4G Networks is Video (Video will represent82% of all IP traffic in 2021)8.If the Telco Industry would have known that Video was to account for 80% of traffic,most probably the design of 4G Networks would have been different, e.g., introducingContent Delivery Network (CDN) in the architecture.The reality is that it is impossible to predict how users are going to drive the usage ofnewly introduced mobile networks. Therefore, for 5G Networks, 3GPP has taken aService Oriented approach, introducing new key concepts, such as Network Slicing, or aService Bus Architecture for Microservices, to offer the possibility to create a VirtualNetwork for a specific Service to deliver the best user experience to customers.The 5G Network value proposition relies on three pillars or capabilities, usually displayedlike in Figure 4, associated to most relevant use cases:5G Usage ScenariosEnhanced Mobile BroadbandGigabits in a second3D Video, UHD screensWork and play in the cloudSmart Home/Building. Augmented RealityIndustry AutomationVoiceMission critical applicationSmart CitySelf Driving CarMassive Machine TypeCommunicationsUltra-reliable and Low LatencyCommunicationsFigure 4: 5G Usage Scenarios (Source International Telecommunications Union9) Enhanced Mobile Broadband (eMBB): aims to service more densely populatedmetropolitan centers with downlink speeds approaching 1 Gbps (gigabits-persecond) indoors, and 300 Mbps (megabits-per-second) outdoors.Ultra-Reliable and Low Latency Communications (URLLC): addressescritical communications where bandwidth is not quite as important as speed specifically, an end-to-end (E2E) latency of 1 ms or less.8 traffic-will-be-video-by-2021-2017-6?IR 7-06-21.pdfDissemination level: PublicPage 8 / 96

5G PPP Technology Board Edge Computing for 5G NetworksMassive Machine Type Communications (mMTC): 5G enables an 1000Xincrease of devices connected to the Network, moving from 1K devices per Km2in 4G to 1M devices in 5G10.In order to deliver the above mentioned above value proposition, Edge Computing playsa fundamental role, as Compute resources are critical to enable those three capabilities tothe Network, so to be able to finally deliver a satisfactory E2E experience.Figure 5 elaborates on what the main enhancements to some key system capabilities are,when moving from a 4G network to a 5G one.eMBB: increasing Data transfer in Radiointerface is not enough. Content needs to becloser to customers in order to sustain highdata transfers rate with no congestion.URLLC: reducing Latency in Radio interfaceis not enough. We need to move Servicescloser to customers in order to deliver areduced and guaranteed E2E Latency.mMTC: increasing the number of connecteddevices to the network needs to beaccompanied by processing the signalling anddata from these devices at the edge of thenetwork to digest the volumes of informationgenerated by a huge number of Thingsconnected to the network.Figure 5: 5G capabilities vs. 4G capabilities (ITU-R11)Moving content, services and signalling processing closer to customers requires movingcompute resources closer to the devices consuming the content, running the Apps, orsending signalling coming from sensors. That is where Edge Computing not only meets5G, but allows it to fully deliver its promised enhancements: 5G cannot be conceived justas a set of focused technical enhancements, e.g., a new radio technology, but also as acompletely new paradigm for Mobile Networks, where Edge Computing plays asignificant role.1.3 Where is the Edge of the NetworkThere is no unique location, or range of locations, where Edge Computing must bedeployed. Edge nodes can be included in network routers, cell or radio towers, WiFi hotspots, DSL-boxes, and local data centers. As described in Section 1.1, Edge Computing10 Massive Machine-Type Communications: An Overview and Perspectives Towards 5G(https://www.fpz.unizg.hr/ikp/upload/RCITD 2015 /standardization/20170402/Documents/S2 4.%20Presentation ies%20can%20cope.pdfDissemination level: PublicPage 9 / 96

5G PPP Technology BoardEdge Computing for 5G Networksis the concept of placing computing resources closer to users locations. Almost anydevice with computational power that is near or at the user’s location can act as an EdgeComputing device, as long as it can process a computational workload.CentralizedCloudComputingScale (PoPs)Internet / CloudInfrastructureTypical Latencytens 20 msHundreds thousands 10-20 msMillions 5 ms10s ofmillions 1 msTelecom InfrastructureEDGECOMPUTING Customer PremiseDeviceComputingCustomer devices (incown computingresources)Figure 6: Edge Computing LocationEdge Computing is typically placed between users Devices and Centralized computingdatacenters whether they are Public clouds or Telco Cloud facilities.Device computing resources are hard to manage because of their heterogeneity and thenetwork environment where they are connected to (typically LAN environments).We can mention several Edge Computing deployment examples that help us to identifydifferent Edge Computing Locations: On Premise: Companies deploying 4G/5G Private Networks deploy a fullNetwork Core in the premise infrastructure connected to business applications 12 RAN/Base station: some companies are deploying infrastructure collocated withRAN in the streets, using Cabinets / MiniDatacenters (e.g., see Figure 75GCity/Vapor.io13)Figure 7: Vapor.io Edge module and 5GCity Multifunctional Post Central Offices (COs): COs are at the Cloud Service Provider (CSP) networkedge, which serves as the aggregation point for fixed and mobile traffic to andfrom end user. All traffic is aggregated to the CO, which creates a bottleneck12 ory-56.html13 1/Dissemination level: PublicPage 10 / 96

5G PPP Technology BoardEdge Computing for 5G Networksthat can cause a variety of problems. Throughput and latency suffer greatly inthe traditional access network, essentially cancelling out much of the gainfrom technologies such as optical line transfer (OLT) and fiber-to-the-home(FTTH), and 5G networks.To address this issue an ongoing transformation has been initiated. A promisingsolution is to deploy a virtualized, distributed network at the Edge. Central OfficeRe-Architected as a Datacenter by CORD 14 and followed by OPNFV15 and otherprojects, have started a process where the economies of a data center and theagility of Software Defined Network (SDN) applied with cloud design andnetwork disaggregation principles will tackle the aforementioned problems.Traditional Central Office Transition to Virtualized Central OfficeMobileOTT ernetCore DataCenterBusinessProprietary hardware appliances arereplaced by servers for control anddata planes. Hardware Acceleration inadded as a means of meetingcustomers service expectations (speed,latency, jitter) economically.Traditional CO Virtual COGGSN/SGSNBBUIMSvBNG/vEPC/vFWControl PlaneS-GWData PlaneMVCFigure 8: Virtualization of the CO principles: Cloud and Network Disaggregation Private Datacenters: Telcos and other companies are deploying PrivateDatacenters to host Edge Computing infrastructure. This approach requires theseDatacenters to be interconnected with Mobile Network Aggregation Point ofPresences (POP) to get traffic from users.Hyperscalers Edge Locations: Public cloud companies define their own Edgelocations. The AWS Edge solution is called AWS Cloudfront, and is typicallydeployed in one or two physical points per country in Europe16. The Azuresolution for Edge is Azure CDN, mainly for content distribution, and is similarlydistributed as the AWS Cloudfront17.While Edge can be located in different locations, they are not exclusive, and there can beseveral Edge locations used in a network deployment.The term Fog Computing as defined by the National Institute of Standards andTechnology18, states that Fog Computing is a layered model for enabling ubiquitous14 https://www.opennetworking.org/cord/15 2017/09/OPNFV VCO Oct17.pdf16 https://aws.amazon.com/cloudfront/features/17 .pdfDissemination level: PublicPage 11 / 96

5G PPP Technology BoardEdge Computing for 5G Networksaccess to a shared continuum of scalable computing resources. The model facilitates thedeployment of distributed, latency-aware applications and services, and consists of fognodes (physical or virtual), residing between smart end-devices and centralized (cloud)services.1.4 How does the Edge look like?An Edge Computing infrastructure may be implemented in many different ways,depending on several parameters. It can go from a Raspberry Pi device to a several racksDatacenter footprint. Different Industry initiatives such as ONF, Broadband Forum andOPNFV have come up with similar architectures for Edge Computing infrastructure to bedeployed at a CO level. The ONF design is called CORD, the Broadband Forum (BBF)design is called Cloud CO, and the OPNFV, from the Linux Foundation, is called teNodesVNF1VNF3VNF2VNF4LeafSwitchNetworkI/OFigure 9: ONF CORD & Broadband Forum ArchitectureIn these architectures, the Edge Computing infrastructure is composed of: Compute nodes: these are the servers where Compute lo

Computing to other technologies such as Cloud federation (connecting the Edge Cloud to other Clouds), End to End Slicing (where Edge Compute resources are part of some Network Slice), Radio Access Network (in particular the Open RAN model that can leverage Edge Computing resources), Inter Edge Border connectivity (to show how Edge