Wireless data traffic increases due to the growth in the number of pervasive mobile devices, social networking, and resource-intensive applications of end users [1]. This traffic manifests in diverse domains such as the Internet of Things (IoT) [2], Machine-to-Machine (M2M) communications [3], and cloud computing [4].
Considering the different structures of data traffic stemming from these domains, their organizations are vital. Moreover, enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC) are other challenging services which must be taken into account for controlling data traffic. A brief overview of these services are presented as follows [5]:
This increase in data traffic, mainly driven by mobile video and online social media, motivates mobile operators to look for novel techniques to manage the complex networks and backhaul resources.
The data traffic is one of the key components of the 5G network ecosystem. This ecosystem is diverse and complex due to the aforementioned services and technologies, and so high access data rates, network infrastructure densification, low latency or coverage requirements emerge as new challenges. In order to overcome these challenges, technologies such as massive MIMO (Multiple-input multiple-output), millimeter-wave communications, device-to-device communications, or edge caching are expected to increase both the capacity and the flexibility of the network, but at the expense of additional network management complexity. Most of these technologies deal with the limited backhaul links capacity, exacerbated by the high traffic demand and diversity of quality of service requirements of services. Moreover, the traffic has a dynamic nature which can subsequently impact the backhaul load of the entire wireless ecosystem.
Focusing on the backhaul congestion problem, caching has emerged as a solution to alleviate such hurdle. In that sense, a significant portion of the Web-Caching algorithms has been moved to the edge of the wireless network to manage the backhaul congestion. The aim of caching is to assist the network in controlling the traffic volume and keeping the most requested content over the network edges. The network may collect the popularities over the edges to locate the content with the highest probability of being requested close to the end-user.
Two different caching approaches have been proposed: proactive and reactive caching [6], with differences that can be summarized as follows:
Optimal Content Placement (OCP) [6] aims to design the best strategies to locate popular content over the network edges. The OCP captures the network variations to generate accurate solutions over different network scenarios. The OCP addresses terms such as dynamic traffic of content, the capacity of the network-edges, limited backhaul link capacities, and optimizes all of them. In general, the OCP could be centralized or distributed and use either coded or uncoded caching [7]. Centralized and distributed policies have been built to control the flow and supervise the location of content at the network edges. However, as inferred from their names, such control is conducted in a centralized manner or in a distributed manner. Regarding coded and uncoded caching, their differences can be summarized as:
The OCP techniques that came from web-pages caching systems have been adapted to the 5G scenario to handle backhaul capacity, delay, and so on. Two well-known techniques are the Least Recently Used (LRU) and the Least Frequently Used (LFU). The former removes the least recently used contents from the cache and replaces them with the more popular contents. Conversely, the latter monitors the frequency with which content is requested and updates the caches so that only the most popular contents are stored.
Even though the network complexity can be modeled as an optimization problem, it would not be a practical solution when convergence time is too long for real-time applications. however, the emergence of machine learning (ML) algorithms is envisaged as a possible way to fill such a gap. ML consists of a set of algorithms, computational structures, and statistical scenarios, to efficiently perform a specific learning action to model system behavior. These techniques help in understanding the variation of the demand, thus allowing more efficient management of the caches. The ability of ML to learn and adapt to environment variations has set the basis for its application in 5G. Good examples of the potential application the content popularity dynamics or the mobility of the users, both tightly coupled with OCP problem. This is the reason why SPOTLIGHT is committed to developing efficient ML techniques to enhance the OCP in the network.
References:
[1] Cisco, “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2016–2021 White Paper,” (2016).
[2] “In-network caching of Internet-of-Things data”, S. Vural, P. Navaratnam, N. Wang, C. Wang, L. Dong, and R. Tafazolli, 2014 IEEE Int. Conf. Commun. ICC 2014, pp. 3185–3190, (2014).
[3]“Low cost m2m over LTE.” [Online]. Available: http://www.3gpp.org/news-events/3gpp-news/1714-lc_mtc. [Accessed: 26-Nov-2018].
[4] Cloud based 5G wireless networks, Zhang, Yin. and Chen, Min. Springer-Briefs in Computer Science, (2016).
[5] 5G Wireless Network Slicing for eMBB, URLLC, and mMTC: A Communication-Theoretic View, P Popovski, K. F. Trillingsgaard, O. Simeone, G. Durisi, arXiv:1804.05057.
[6] A Survey of Caching Techniques in Cellular Networks: Research Issues and Challenges in Content Placement and Delivery Strategies, Li et al. IEEE Communications Surveys & Tutorials (2018).
[7] A Survey of Machine Learning Techniques Applied to Self Organizing Cellular Networks, Klaine, P. V., Imran, M. A., Onireti, O., & Souza, R. D. IEEE Communications Surveys & Tutorials (2017).