The tremendous data traffic demand and the arrival of new mobile services will lead to unwieldy traffic congestion in mobile communications networks over the next decade. This situation, which is similar to the one experienced in the Internet network in the 2000s, has established parallelisms between 5G –the future mobile communications generation- and the Internet network, thus paving the way for similar solutions to similar problems.
In this context, mainly dominated by the surge in video traffic demand, the old network edge caching concept has regained momentum as a technological solution able to alleviate the expected traffic congestion in 5G networks. Specifically, mobile edge network caching is defined as the installation of Cache Units (CU) in the Radio Access Network (RAN) nodes (i.e. eNBs, small cells or gNBs). With this, the mobile network is able to deliver content to end users while reducing latencies and diminishing the traffic across the network.
In view of the above-mentioned points, the mobile network is heterogeneous in nature and has to deal with huge traffic congestion with limited capacity of resources. The network infrastructure is one of the challenges itself, including different network layers: Macro base stations (MBS), Small cells (SC) and users operating in a Device to Device (D2D) mode. Depending on the type of Base Station (BS) or mode of operation, caching can provide different gains. When using caching at the Macro Base Stations (MBS), each BS has large caching space. This allows the users to acquire the requested content directly from the BS, rather than from the internet through backhaul links. Instead, exploiting D2D and caching at the mobile devices improves the spectral efficiency of the system and energy efficiency of the end-user.
In the light of that, the caching control structure is divided into centralized and distributed, both of them to deal with the efficient location of content. In a centralized structure, a Central Control Unit (CCU) decides the content placement and delivery strategies. The CCU collects information, such as the files’ popularity, user requests, and relevant channel information. Using this data, the CCU decides which file should be located in the CU. This kind of structure needs special care due to the high quantity of data that must be processed in a centralized manner. For this reason, algorithms designed under this paradigm are complex in nature. In the distributed structure each access node decides which file should be cached. This structure does not need to collect information from the rest of the network. Therefore, the complexity of these algorithms is lower when compared with the centralized strategies. However, distributed control structure does not ensure global optimality.
In this context, and given the increasing complexity of the network and the overwhelming amount of available data (traffic, popularity, mobility, etc), the range of algorithms offered by the ML framework enables the systematic handling of huge amounts of information. Moreover, the learning capacity of ML allows the exploitation of the available data, which is crucial for the efficient operation and maintenance of complex networks. ML is thus destined to become an ideal complement to network caching, and a pillar of the operation and maintenance of future 5G networks.