Network virtualization for future mobile networks General architecture and applications

Network Virtualization for Future Mobile NetworksGeneral Architecture and ApplicationsMarco HoffmannNSN ResearchNokia Siemens Networks GmbH & Co. KGMunich, Germanymarco.hoffmann@Markus StauferNSN ResearchNokia Siemens Networks GmbH & Co. KGMunich, Germanymarkus.staufer@Abstract—Based on the expected future requirements this paper describes a general network architecture enabled by network virtualization. This architecture consists of three major building blocks which we call virtualized physical resources, virtual resource manager and virtual network controller. Such an architecture will facilitate network sharing deployments, which might exist in the form of network consolidation or service specific networks. Furthermore the ability of our framework to combine control over various domains allows a resource optimization across IT and network infrastructure, multiple network layers and heterogeneous networks.Keywords: Network Virtualization, Control Plane, Business ModellingI.I NTRODUCTIONOne of the most apparent challenges for future mobile networks will be the handling of the predicted increase in mobile traffic volume. To make this traffic increase reality and to harvest the underlying business opportunity, this traffic growth has to be mastered without increasing network costs in a similar way, which in fact means that the cost per transported bit needs to be decreased significantly. Besides this race for higher capacity at lower costs, which rules the telecom industry already for quite a while, also the expected qualitative changes of future mobile traffic needs to be considered as well.Whereas mobile voice traffic will grow only very modest or will even stagnate in many regions, the vast majority of traffic growth will stem from mobile data traffic. Fueled by steadily improving mobile device capabilities and increasing adoption of portable computers (e.g. laptops or net books) the boundaries between fixed and mobile usage patterns will more and more blur. That is, users will demand to access applications in a defined quality independent from the location, the used device, and the available network. Together with the emerging cloud computing IT delivery model, which will lead to a shift of the application logic from the devices towards service platforms, this will impose strict requirements for mobile networks with respect to service level compliancy. Majority of the applications will be owned by third parties outside the walled garden of the traditional mobile operators.The capability of mobile networks to actively support the delivery of services in an efficient way, while at the same time ensuring highest service experience, will be one of the key assets of mobile operators in the future. Some of the resulting requirements for the future mobile network architecture will be the ability to master network heterogeneity, flexible deployment options supporting distributed gateways and service platforms, and the possibility to divide a physical network into several logical sub networks. The main rational of such logical sub networks (so-called virtual networks) is the ability to realize the cost efficiency, which is inherent in the sharing of networks, while at the same time being able to maintain isolation and operational independence between the virtual networks. These virtual networks could be used to build service specific logical networks, which are reserved and optimized for dedicated applications and/or users. Alternatively virtual networks could also facilitate network sharing between different mobile operators.Although details of future mobile network scenarios are naturally still uncertain, it is important to start already now to design their architecture such that they are flexible enough to cope with all the potentially emerging requirements of future mobile networks. We propose network virtualization as one of the main techniques towards the goal of efficient and flexible future mobile networks.In this paper we first briefly describe the state of the art regarding network virtualization in mobile networks. After the presentation of our network virtualization architecture, we describe applications, which highlight the benefit of network virtualization in further mobile networks. We conclude the paper with a summary, open issues and next steps.II.S TATE OF THE A RTHistorically early forms of network virtualization date back already until the beginning of the 1990s, when Asynchronous Transfer Mode (ATM) with its concept of virtual connections was established. More recent developments of transport separation using virtualization concepts include Virtual Local Area Networks (V-LAN), Multi-Protocol Label Switching (MPLS) or Optical Data Unit (ODU) switching. With respect to specific hardware elements server virtualization as well as virtual routers can be regarded as state of the art.The establishment of server virtualization and the accompanying virtualization of entire data center provided additional stimulus towards the virtualization of networks. In a virtual data center the traditional fixed one-to-one978-1-61284-955-3/11/$26.00 ©2011 IEEEcorrespondence between a software image and a physical server does no longer exist. The resulting complications in the network layer can be solved by means of virtual network interfaces and virtual distributed switches [1].Network sharing is considered as one of the main applications of network virtualization. Due to the high potential to save costs network sharing has been tackled especially in the domain of the Radio Access Network (RAN). Currently two main approaches are established in the market, i.e, Multi Operator RAN (MORAN) [3] and Multi Operator Core Network (MOCN) [2].Network virtualization was already investigated in the EU-funded project 4WARD [4]. Our approach is based on the business model (roles) that are defined in this project. We are extending the 4WARD concepts regarding concrete extensions of existing control plane protocols and interfaces for a layer-, vendor- and domain-independent provisioning, and operation of virtual networks.Albeit the above overview is not complete, it shows that platform and link virtualization and related applications exist in various flavors and realizations. However, so far no network architecture exists, which is consistently built on network virtualization and which fully covers the end to end aspect of the virtualized network resources.III.N ETWORK V IRTUALIZATION A RCHITECTURE The key concept of network virtualization towards flexibility is a consequent extension of the current network element centric design philosophy towards a more holistic view of the network. All network resources of a network element (e.g. the capability to handle and forward traffic data) will be made available in a granular and abstracted way to external control entities using a well-defined interface. This approach has a couple of advantages. First of all, the partitioning of the raw network element resources on the network elements allows the simultaneous existence of different control entities, each acting on isolated domains of the raw resources. Secondly, the well defined and open interface between raw network element resources and control entity facilitates the deployment of new and potentially proprietary control algorithms. Finally, the deployment of external control entities could take place as centralized elements. Those controllers are responsible for all network elements potentially even being part of previously separately controlled network domains. The centralization of the network state, which can be achieved in this way, is an important prerequisite for an efficient end to end control and optimization of the network.A future mobile network based on network virtualization as enabling technology will consist of three general building blocks (see Figure 1. ): a virtualized physical IT and network infrastructure that can be owned by one or more physical infrastructure providers, a central virtual resource manager that also can be represented by a hierarchy of different resource managers and one or more virtual network controllers. These building blocks will be connected via control plane interfaces that will be used for virtual network provisioning and operation. We propose a general architecture where each building block can be owned and controlled by different organizational entities.work Building Blocks1)Virtualized Physical ResourcesIn general Virtualized Physical Resources (VPR) could be virtual links and nodes on different network layers or other virtualized entities like storage or CPUs. These virtual resources are created and owned by physical infrastructure providers (PIPs). In general the resources will be part of different types of physical domains, e.g., IT and network domains. The IT domains could be server farms or cloud computing domains whereas the network domains could be mobile access or core networks or end customer domains like enterprise networks or even end user equipment.2)Virtual Resource ManagerThe Virtual Resource Manager (VRM) provides the linkage between the virtualized physical resources and virtual network controllers (VNCs). As a kind of mediation layer the tasks of the virtual resource manager are to ensure that each controller has access only to its dedicated slice of the virtual resources and that the portioned virtual network resources are exposed to the controller in a well-defined way. Furthermore, in case a larger network is divided into several domains, which are managed by separate virtual resource managers, the virtual resource managers need federation capabilities in order to maintain the end-to-end view to the virtual network controllers.The VRM is responsible for maintaining the mapping between the virtual network as seen by the virtual network controllers and the virtualized physical resources in the network elements. The mapping is based on the requested topology and requirements of the virtual network slices and the availability of the virtualized physical resources.The VRM gets virtual network requests including topology and network requirements from a VNC. Regarding to these requests the VRM combines virtual resources offered by the PIPs to the most cost efficient virtual network that fulfills the requested requirements. The VRM can act as a passive broker that offers only resources to the VNC that are propagated by the PIPs or as an active broker that requests virtual resources from the PIPs. In both cases two options exist, which depend on the business model and the requested provisioning time. The first option is that the VRM has already a lot of virtual resources collected based on assumptions of incoming requests to be able to react fast on a virtual network request. The second option is that it starts collecting after a request was sent by the VNC. The VRM is responsible for the optimal calculation of e2e connectivity with the required e2e quality of service (QoS) or service level agreements (SLA) parameters. This is very important for multi-domain scenarios where different PIPs offer virtual resources.3)Virtual Network ControllerThe VNC requests virtual networks with special requirements from the VRM. The VRM provides access to the virtual network resources. The VNC controls the virtual network resources via a standardized or open interface. It……ServiceControlService ControlService ControlService Specific NetworksNetworkStandardized Interface…Router Network SliceSwitchDWDMAccess PointServerguration Aggregation and Slicing of Resources VNCcan control and upload software on the virtual resources if this was requested. One VNC can control one virtual network for one service, one virtual network for some services with the same service type and one virtual network for different services with different service types. Different VNCs can be used by different Virtual Network Operators to configure their isolated virtual networks.B. Control PlaneWhile the architecture presented in this paper has similarities to other frameworks currently introduced (e.g. [5]), one of the critical steps towards a broad applicability in real network deployments will be the methods and interfaces to establish and manage the different virtual networksTo set up a virtual network we will use two different types of control planes. The first one is a horizontal control plane thatwill be used independently in each of the three building blocks.Additionally, we will use a vertical control plane that enables the interaction between the building blocks to set up and operate virtual networks in a dynamic and automatic way. We propose an extension of existing GMPLS control plane protocols and the extension of existing Path Computation Element (PCE) concepts. A detailed description of the necessary protocols and extensions is not in the scope of this paper. Here will focus on applications of network virtualization. Figure 1. Building Blocks of Network Virtualization ArchitectureIV. A PPLICATIONS OF NETWORK VIRTUALIZATION Before entering the detailed description of various applications the three main high-level properties of network virtualization shall be summarized. These are:• Partitioning of forwarding plane to allow controlfrom several independent control planes.• Open interface between forwarding and control planes facilitating the development of new proprietary control planes.• Possibility to centralize control plane (and thus network state), wherever beneficial.Keeping these points in mind applications of network virtualization can be split into two main categories. The first category is network sharing, or in other words, the accommodation of several logical networks on one physical network.The second category is the possibility to build combined controllers, which are able to execute control over domains, which traditionally have been controlled separately.Both categories will be discussed in detail in the following two subchapters, starting with network sharing.A. Network SharingTwo different reasons for network sharing have to be distinguished. In the first case, i.e., network consolidation, the target is to combine separate physical networks on one physical network infrastructure (see Figure 2. ). In case this physical substrate supports network virtualization, each of thepreviously separate networks could be implemented as avirtual network, which in this way could still be controlled independently. A typical scenario, in which network consolidation is used, is the sharing of Radio Access Networks (RAN) among several operators as mentioned in Chapter II [1]. While these current approaches [2][3], which are getting more and more traction in the market, do not rely on a formal method of network virtualization, network virtualization could further improve RAN sharing by providing the involved operators a more independent control over their share of the common physical network.Figure 2. Network consolidationFigure 3. Network slicingThe second variant of network sharing, network slicing, is illustrated in Figure 3. In this case the starting point is a single network, which is used to deliver several different services.This is close to the over-the-top model of the current Internet.With the help of network virtualization this network can bepartitioned into several different networks, which can be controlled independently. Ideally each of the resulting virtualnetworks could be controlled such, that it is optimally suited todeliver a specific service, e.g. an interactive real-time application with very high requirements with respect to delay.In this way one physical network could be split into a multitude of service specific networks.Although the motivation of the two variants of network sharing, i.e. network consolidation and network slicing, is exactly opposite, the expected benefit is based on the same rational of network virtualization: Providing of high cost efficiency due to usage of a shared physical substrate, while at the same time keeping a high isolation and operational independence between the resulting logical (i.e. virtual) sub-networks.B. Combined ControlThe second big category of applications of network virtualization is the combined control of network domains, which are traditionally controlled separately. In doing so three different directions can be distinguished, the End to end (E2e) dimension, multi-layer dimension, and the heterogeneity dimension. All three dimensions will be discussed in the following starting with the E2e dimension. 1) E2e ControlTarget of the E2e Control is to put all parts of a network, which are used to deliver a service to a user, under the control of one common entity, in our proposed architecture the VRM will combine the resources and VNC will control it. Currently this is not the case. For instance the network and the data center, which hosts a service delivery platform, are controlled separately.Figure 4. Combined Control of IT and network (ITN) resourcesAnother example is the mobile packet core, where currently the 3GPP network elements and the interjacent transport network are controlled separately. Both examples will be explained in the following paragraphs.The most important application of a combined control of IT and network resources can be seen in the area of cloud computing. Within the cloud computing IT delivery model a user consumes services, which are provided by a data center, via a wide area network. These services can be either software applications, like customer relationship applications, or naked infrastructure services, like pure server capacity. The first category is usually called Software as a Service (SaaS), whereas the second one is known as Infrastructure as a Service (IaaS).Currently the cloud computing market is very much driven by so-called over-the-top players, like Amazon with its Elastic Compute Cloud offering [6], which only handle the data center part, and rely that the connectivity is provided by the normal public Internet. In case the service consumed from a cloud are really business critical, the offered connectivity from the plain Internet might not be sufficient. Therefore, the network part has to be considered as an integral part of a cloud framework and needs to be provisioned with the required service level as well.Furthermore, a future cloud will consist not only of a single data center, but of several distributed data centers. A main driver behind this development is the need for so-called hybrid clouds. In case of a hybrid cloud a part of the cloud might be a private cloud, which resides still within the data center of the cloud user (e.g. an enterprise customer), whereas the other part is a public cloud, which is operated by a third party cloud provider.With the help of network virtualization and a combined IT and network control it will be possible to make those individual clouds look like a single cloud and to ensure that the user is always connected to this cloud with the necessary service level. The resulting combination of virtual data center resources (like processing or storage) and the connections to and between the data center done by the VRM and controlled by the VNC leads to the concept of a Virtual Private Cloud [7].The second example discussed in the context of combined E2e control is the application of network virtualization to optimize the Evolved Packet Core (EPC). Within a typical EPC architecture a base station is connected to several gateways (GW), one of which is acting as the serving GW for a user.Typically, the process of gateway selection is based on two criteria. The first one is the position of the user and the second one the load of a GW. That is, the movement of a user as well as a capacity bottleneck in the gateway might trigger a reselection of the gateway. This process is completely unaware of the network, which is used to connect base stations and gateways.With the help of the concept of network virtualization, it will become possible to build a VRM and VNC, which include the functionality of the 3GPP Mobility Management Entity (MME) as well as control functionality of the transport network. In this way it will be possible to design mobility schemes and gateway selections schemes, which take into account both, 3GPP elements and the transport network and thus allow a joint resource optimization.The importance of such an E2e controlled EPC gets even clearer, if one takes into account that the evolved packet core of future mobile networks most likely will be less structured and more flexible than current 3G deployments. A 3G network deployment is rather structured consisting of central office, local office and cell sites. In the future the deployment of the EPC might become more flexible with a strong likelihood that the packet core will be more distributed than in today’s deployments. Furthermore, the topology of the network might be more complex and meshed enabling new possibilities to interconnect local offices or cell sites. Drivers for these architectural shifts are increasing over-all traffic volumes, increase of point to point traffic, and the need for local break out in order to increase service quality and network efficiency. 2) Multi-layer ControlAnother subject, where the capabilities of network virtualization to provide a combined control mechanism might help, is the area of multi-layer control.A typical transport network consists of several layers, for instance a purely L1 optical layer, a L2 switching layer and finally the IP/routing layer. In a classical network these networks have separate control planes.As a general rule the costs per transported bit are the higher the higher the layer is. Therefore, a network should be designed such that the traffic should be kept on the lowest layers as far as possible. This has lead to the fact that during network planning so-called multi-layer optimizations are carried out, which, depending on the traffic forecast, allow deploying a cost optimized network [8].A further optimization could be achieved, if the optimization of the resources over multiple layers could also be part of the normal network operations. This could be done with a combined external control entity, which is able to simultaneously get information from all the layers and to control the forwarding engines on all layers. Please note that the unified control plane framework of GMPLS and the path computation element concept [9] are a first step to this idea. The simultaneous vendor-independent control of different network layers, which is sometime referred to as convergence, is also one of the drivers behind the OpenFlow virtualization approach of the Stanford CleanSlate program [10].3) Control of Heterogeneous NetworksFinally the third dimension of combined control, which is the exploitation of network heterogeneity, should be discussed as well.Very often within a certain area several physical networks are available, which serve the same purpose, but are controlled independently. With the help of a network virtualization concept it will be possible to put those domains under one control (see Figure 5. )One example, which currently is discussed most intensively in the framework of network heterogeneity, is the co-existence of Wifi and 3GPP access networks. The main driver for this kind of Fixed Mobile Convergence (FMC) is the different cost structure of both networks. Since the cost per bit are typically much higher in a 3G network, offloading cellular traffic through a Wifi network is an appropriate means to save network costs.Figure 5. Smart ControlIn general, combined control of heterogeneous networks could consist of “always best connected” strategies, common hand-over management, or even an aggregation of different networks [11]. An example of application of a virtualization framework used to tackle heterogeneous networks is the OpenRoads project [12], which is part of the Stanford CleanSlate program.V.C ONCLUSIONIn this paper we described a general architecture enabled by network virtualization that can handle the increasing and changing traffic demands of the future. Network virtualization is a key technology to realize service specific networks, network sharing between different operators and a combined control like the e2e control of IT and network infrastructure. The next steps would be to specify detailed control plane protocol extensions for the communication between the different building blocks and to investigate different multi-domain scenarios like the application of network virtualization in the framework of cloud computing. The specified protocols and interfaces will be implemented in a simulator to validate the concept.A CKNOWLEDGMENTThis work has been performed in the framework of the G-Lab subproject COMCON and it is partly funded by the German BMBF (ID 01BK0915).R EFERENCES[1] Cisco, Cisco VN-Link: Virtualization-Aware Networking, White Paper,/en/US/solutions/collateral/ns340/ns517/ns224/ns892/ns894/white_paper_c11525307_ps9902_Products_White_Paper.html , 2009.[2] 3GPP TS 23.251 V2.0.0 (2004-06): Network Sharing; Architecture andFunctional Description, Rel. 6. 3rd Generation Partnership Project, 2004. [3] Press release (23.05.2001): Nokia launches Multi-Operator RadioAccess Network for controlled 3G network sharing. Nokia, 2001.[4] S. Baucke et al., “Virtualization Approach: Concept”, Deliverable D-3.1.1, EU-funded Project 4WARD (FP 7), 2010.[5] R. Scherwood, et al. “FlowVisor: A Network Virtualization Layer”,OpenFlow Technical Report, OPENFLOW-TR-2009-1, 2009. [6] /ec2/.[7] T. Wood, et al. “CloudNet: A Platform for Optimized WAN Migrationof Virtual Machines”, University of Massachusetts, Technical Report 2010-002, /~twood/pubs/cloudnet-tr.pdf, 2010.[8] M. Scheffel, et al., “Optimal Routing and Grooming for Multilayer Net-works with Transponders and Muxponders”, Global Telecommunica-tions Conference, 2006. IEEE, pp. 1-6, November 2006.[9] A.R. Rubuyat, “Path Computation Element in GMPLS Enabled Multi-Layer Networks”, Masters Degree Project, KTH Electrical Engineering, Stockholm, Sweden, 2006.[10] S. Das, et al., “Packet and Circuit Network Convergence withOpenFlow”, Optical Fiber Communication (OFC), collocated National Fiber Optic Engineers Conference, pp. 1-3, March 2010.[11] K. Evensen, et al., “A Network-Layer Proxy for Bandwidth Aggregationand Reduction of IP Packet Reordering”, Local Computer Networks, IEEE 34th Conference, pp. 585-592, Zürich, October 2009.[12] K. Yap, et al., “Blueprint for Introducing Innovation into the WirelessNetworks we use every day”, Technical Report, OPENFLOW-TR-2009-3, /downloads/technicalreports/openflow-tr-2009-3-openflow-wireless.pdf , 2009。