Evolution of mobile network infrastructure

Driven by the prosperity of mobile applications which demanding vast network and com- putation resources, leveraging the idle computation power and storage space distributed at the network edge becomes an invertible trend for enabling ubiquitous and person- alised mobile service. Compare to the cloud computing which still suffers from long propagation delay, the MEC server deployed at various RAN nodes (e.g. LTE eNB or Wi-Fi access point) can have a close proximity to mobile user, thus is able to provision “local” content delivery or customised computation with the consideration of real-time user and network context information. As a comprehensive service enhancement en- tity in the network, the functionalities that can be integrated on a MEC server include not only the traditional service optimisation techniques (e.g. content caching/pushing or HTTP/TCP proxy) but also the newly proposed radio resource based techniques

(e.g. mobility-aware content prefetching, D2D-based offloading). In terms of the tra- ditional optimisation approaches, HTTP/TCP proxy has been already widely deployed to perform a group of optimisation methods [62]. For instance, in order to absorb the network uncertainty on backhaul, the end-to-end TCP connection between the content server and the client is split to two connections at the proxy [63], achieving increased throughput [64]. Meanwhile, holding the packets at network proxy side can seamlessly leverage the error recovery techniques at the base station, thus the connection failure or packet loss at radio interface could be fast recovered. Therefore this enhanced feature can achieve a reduced number of connection re-establishment or packet retransmission from the original server which may stay far away from mobile network [65]. On the other hand, to extend the protocol compatibility at both the content server and client sides, the proxy can also perform as a protocol translation gateway. For instance, as proposed in [66], Information Centric Network (ICN) data plane is an alternative option for mobile backhaul and its hierarchical caching can significantly reduce the content de- livery time as well as the traffic volume at backhaul network [4]. To practically support the coexistence of different protocols, Li et al. proposed a new architecture to augments the mobile HTTP/TCP users with the ability to utilise the ICN hierarchical caching without any modification at client side [4] (see in Fig. 2.5). Similarly, the backhaul can also seamlessly access the external HTTP-based web content, and with the help of more name mapping algorithms like [67], popular content service (e.g. HTTP live streaming [68]) can be beneficial from such ICN-enabled caching network.

Once these traditional optimisation approaches are moved to the MEC, with the help of computation ability and storage space at the network edge, the QoE of the end user can be further improved with the consideration of context-aware information like the availability of radio resources and the trackable user information. Figure2.6 depicts an example of MEC framework with an intelligent video acceleration module enabled [5]. Given the fluctuating capacity caused by the varying radio channel or device mobility, the lagged reaction of upper layer protocol like TCP may fail to adaptively tune its congestion window accordingly. As a consequence, the under-utilisation of available radio resource will result in a sub-optimal user experience. To overcome this issue, the radio analytic module deployed at MEC server can collect and analyse the radio variation, and thus the MEC can periodically hint the original video server the estimated achievable throughput measured at radio interface. With the assistance of this information, the video server

Figure 2.5: HTTP ICN Gateway [4]

can perform its own intelligences like dynamical congestion window (cwnd) tuning and video coding selection, ensuring the application throughput as well as the user QoE.

Figure 2.6: MEC Assisted Video Acceleration [5]

Meanwhile, to overcome the performance deterioration caused by long backhaul de- lay and traffic exposition, many cache-enabled MEC frameworks which utilise the dis- tributed storage at the mobile edge have emerged. For instance, Wang et al. proposed a novel edge caching scheme based on the concept of content/information centric network- ing [38]. The evaluation results validate that caching at edge nodes (e.g. EPC and RAN)

can significantly reduce user-perceived latency as well as redundant traffic volume over the network. Shanmugam et al. further investigated the file downloading latency under different scales of radio cache helpers and backhaul, with a greedy algorithm proposed to achieve an optimal allocation between each cache point [69]. User mobility is also considered as a key metric to further enhance the cache efficiency (e.g. driving cache relocation and collaboration between different edge nodes [70] or determining offloading policy by greedy or random algorithm [71]). Furthermore, leveraging the downloaded data based on numerous mobile devices and performing D2D communication managed by a MEC server can be another effective way to improve the cache availability and the reduce the content access latency [72–74]). However, as mentioned in 2.1.3, since web content has limited content cacheability and high security consideration, MEC frame- work will inherit the same constraint when applying such caching or prefetching based approaches.