Towards intelligent virtual resource allocation in UAVs-assisted 5G networks

Abstract

With the aim of providing novel services and applications having high data rate and low latency, the 5G networks are designed. Thus, it is essential to manage and schedule the physical resources efficiently in 5G era. Virtualization technologies are seen as the most potential enabling approaches towards 5G networks. The known key issue is the virtual resource allocation, called as virtual network embedding (VNE). However, previous researchers simply focus on allocating resources in an inflexible way. As unmanned aerial vehicles (UAVs) communication plays an important role in 5G, we need to add UAVs into the 5G networks in order to expand the coverage and agility of services. In this paper, we investigate the issue of intelligent virtual resource allocation in UAVs-assisted 5G networks. Especially, we propose one intelligent virtual resource allocation algorithm, labeled as Intell-UAV-5G. When receiving one network service, our Intell-UAV-5G implements the network service in an intelligent manner. Though virtualized UAV moves, our Intell-UAV-5G is able to predict all possible connecting access nodes and continue the network service intelligently. Meanwhile, our Intell-UAV-5G can minimize the time of service interruption. Thus guaranteeing the service quality. We also conduct the experiment to highlight the Intell-UAV-5G efficiency. The mostly-similar algorithms are derived for performance comparison. Performance metrics are carefully recorded and discussed.

Introduction

Both academia and industry have reached the consensus that the 5G networks [1], [2], [3], [4] are designed for providing massive network elements connection and fulfilling explosive data growth. Meanwhile, more novel network services and applications [5], having high data rate and low latency, emerge in 5G era. Therefore, it is necessary to manage and schedule the physical network resources efficiently. Considering the heterogeneity nature of dedicated hardware, it is a great burden [6] for telecommunication service providers (TSPs) to invest and upgrade these hardware continuously. Therefore, virtualization technologies (e.g. network virtualization, NV [1], network function virtualization, NFV [4]) are seen as the most potential approach towards the upgrading tendency of 5G. By conducting virtualization, underlying physical resources (e.g. radio, computing, storage, bandwidth) can be abstracted into virtual resources easily. TSPs can manage, schedule and allocate their owned virtualized physical resources easily.

In order to implement 5G network services and load virtualized network elements to their full capacities, TSPs need to be equipped with abundant algorithms, having the function of allocating virtual resources efficiently. In the literature, it attracts significant research attention. In both academia and industry, the resource allocation problem can be called as virtual network embedding (VNE) [7]. There exist abundant technical publications in the literature [7], [8], [9], [10]. These publications enables to help following researchers to start the VNE research quickly. However, existing technical publications [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28] allocate the virtual resources in an inflexible way. For instance, once one network service is allocated and implemented, no intelligent scheme exists in order to adjust the allocation results. In addition, all nodes of virtual network services are assumed to be static throughout their lifespan. None considered the embedding and allocation in case certain node moves. Though a few publications (e.g. [27], [28]) discussed the node mobility, none mentioned which concrete network scenario [27] can be applied. However, ‘the last one mile’ in resource allocation aspect is very important. That is to say, the node mobility cannot be ignored in real networking research. In addition, existing publications (e.g. [28]) did not evaluate their proposed algorithms in a continuous time event. The evaluation scheme further limits their contributions.

Therefore, we research the intelligent virtual resource allocation, on the basis of 5G networks. We also consider the mobile virtual node so as to research the node mobility effect on the virtual resource allocation results. Instead of discussing about node mobility in an abstracted manner, we consider the concrete mobile network scenario. Unmanned aerial vehicles (UAVs) research [29], [30], [31], [32] has emerged in recent years. As UAV will play an important role in 5G era [29], we incorporate mobile UAVs into the 5G networks in order to expanding the coverage and angle of novel network services.

In this paper, we conduct a research on the virtual resource allocation in UAVs-assisted 5G networks. The formal problem model for UAVs-assisted 5G networks is involved. Another novel profit model, quantifying the effect of UAV mobility, is presented, too. With respect to the novel profit model, the penalty of virtual network service interruption is considered. Then, we propose an intelligent virtual resource allocation algorithm, labeled as Intell-UAV-5G. Our Intell-UAV-5G consists of two sub-algorithms: Intell-UAV-5G-Active and Intell-UAV-5G-Reactive. With respect to the Intell-UAV-5G-Active algorithm, it enables to predict all possible connecting access nodes and continue the implement virtual service quickly when the service interruption happens. With respect to the Intell-UAV-5G-Reactive algorithm, it can re-allocate and re-embed the virtual network immediately when the service interruption happens. In order to highlight our Intell-UAV-5G, we conduct the experiment evaluation. Experiment results demonstrate that Intell-UAV-5G outperforms the selected derived algorithms, in terms of income and virtual service acceptance. For example, the virtual service acceptance ratio advantage of Intell-UAV-5G is beyond 14% (10 000 time unit point).

(1) The new formal problem model for UAVs-assisted 5G networks is presented in this paper (Section 3). Previous problem models for solving virtual resource allocation were developed from the fixed core networks and computer networks. No researcher [7], [9], [10] had not considered integrating the UAVs into the virtual resource allocation problem model yet;

(2) Another novel profit model is proposed in Section 3. Besides of quantifying revenue and cost, the novel profit model enables to quantify the penalty. The quantified penalty results from the UAVs mobility. Previous profit models [7], [9], [10] did not consider quantifying the penalty. Not to mention quantifying the UAV mobility and service interruption;

(3) An intelligent virtual resource allocation algorithm is proposed in this paper (Section 4). The intelligent resource allocation algorithm is labeled as Intell-UAV-5G. The Intell-UAV-5G consists of two sub-algorithms. One sub-algorithm is Intell-UAV-5G-Active. The Intell-UAV-5G-Active enables to predict all possible connecting access nodes and continue the implement virtual service quickly when the service interruption happens. The other sub-algorithm is Intell-UAV-5G-Reactive, enabling to conduct the re-allocation and assignments when the service interruption happens;

(4) A comprehensive experiment is conducted in Section 5. The experiment work aims at validating Intell-UAV-5G efficiency. We select the mostly-similar resource allocation algorithms to make up the experiment simulation part. Experiments results are carefully recorded, plotted and discussed.

The rest of this paper is well organized: Related work is presented in Section 2. Problem model and profit model for UAVs-assisted 5G networks are both presented in Section 3. The intelligent algorithm Intell-UAV-5G is detailed in Section 4, including its two sub-algorithms. Experiment work and result discussion are presented in Section 5. In the end, we conclude this paper and outlook the future work.