Page 90 - Proceedings of the 2018 ITU Kaleidoscope
P. 90

‎ 2018 ITU Kaleidoscope Academic Conference‎




                                  Service requirements
                                                         Virtual Network Operator (VNO)

                                                          QoS requirements   Functions, protocols
                               Application Service
                                  Provider
                                                     Control & management  Analysis
                                                                         Network resource request
                                  Various services  Network slices
                                  Autonomous cars
                                           SNS

                                Sensors & actuators
                                                                 Network resource
                                                                allocation/adjustment

                                                    Resource controller of substrate network



                                IoT devices                  Core network   Cloud
                                             Edge network        Infrastructure Provider

                                      Figure 1.  The concept of network slicing for 5G services.

           discuss the applicability of machine learning to enable the   nodes  (e.g., servers) contain plenty of computation and
           5G  slicing  functions  to be executed autonomously. This   storage resources, the virtualized network slices can  have
           discussion  can  be  useful reference material for the   plenty  of  computation and storage resources in network
           development of standard technologies for the automation of   nodes too. These virtualized network nodes host functions
           5G network slicing functions by using machine learning   such as directory  service, caching, media transcoding,
           techniques.                                        traffic engineering and in-network data processing, besides
                                                              usual data forwarding functions. New nodes can be added
           The remainder of the paper is organized as follows. Section   or their resources can be increased dynamically as demand
           2 gives an overview of 5G network slicing. Section 3 lists   for the scaling of network functions.
           network  automation functions, and Section 4 gives an
           overview  of machine learning techniques for the   Infrastructure providers own physical substrate networks
           automation of network slicing. Section  5  outlines  the  AI   consisting of the edge  network, core network and data
           and machine learning activities being  pursued  in  various   centers, and possess a huge amount of networking (i.e., link
           SDOs and industrial forums, and Section 6 concludes this   bandwidths  and  buffer size) and computing (i.e., CPU,
           paper.                                             memory  and storage) resources. Edge networks are
                      2.   5G NETWORK SLICING                 composed of equipment for collecting, processing, and
                                                              transmitting data to/from various terminals such  as  PCs,
           Network slicing is the key design concept being introduced   mobile/smart  phones,  vehicles, robots, sensors, and other
           in  5G/IMT-2020  networks. The design consideration for   IoT devices. The substrate network is equipped with NFV
           IMT 2020 network architecture specified  in  ITU-T   and SDN capabilities so that it can be sliced and allocated
           Recommendation Y.3102 are network capability exposure,   to virtual network operators.
           separation of control and data planes, common interfaces to
           support access network agnostic common core  network,
           and efficient support of various mobility requirements [2].   An application service provider is the entity  that  offers
                                                              application  services  (e.g., automated driving and smart
                                                              metering) to customers by using  a  virtual  network  slices
           Figure 1 illustrates the structure of virtualized networking   operated  and  managed by a VNO. For this purpose, it
           environment, considering three stakeholders: infrastructure   provides  service requirements (e.g., number and types of
           provider, virtual network operator (VNO), and application   devices to be connected, application type, and reliability) to
           service provider [5]. The virtualized resources are elastic in   the VNO.
           nature so that their amount can be  dynamically  adjusted.
           Unlike the conventional networks in which network nodes   The VNO analyses the application’s QoS requirements and
           (e.g.,  routers)  contain only limited resources to perform   determines about the supporting  network  capabilities,
           predefined  packet  forwarding functions while the end   functions and protocols. It translates the application’s QoS





                                                           – 74 –
   85   86   87   88   89   90   91   92   93   94   95