User mobility and Quality-of-Experience aware placement of Virtual Network Functions in 5G

Abstract

Virtual Network Functions (VNFs) in cloud servers of Fifth Generation (5G) network systems are responsible for executing offloaded codes from mobile users. Placement of VNFs in the cloud is very complicated to get on-time execution service due to many reasons including users’ mobility and resource heterogeneity, which often cause VNF relocations from one data center to another. Minimizing service delay (i.e., maximizing user Quality-of-Experience) for the user applications and the number of VNF relocations are the two main design goals of VNF placement problem; however, they do oppose each other. In this paper, we have formulated the above problem as a Multi-objective Integer Linear Programming (MILP), which is proven to be an NP-hard one. The proposed optimization framework trades-off between the number of VNF relocations and user Quality-of-Experience. We then develop an Artificial Intelligence based meta-heuristic Ant Colony Optimization (ACO) algorithm to achieve sub-optimal placement of VNFs within polynomial time. The performance analysis results, carried out in Cloudsim, depict that the proposed system outperforms the state-of-the-art works significantly in terms of user satisfaction and VNF relocation overhead.

Introduction

The Internet has become an inextricable part of our day-to-day life in recent times. The number of devices connected to the Internet is getting increased rapidly [1]. Almost all types of instruments from communication devices to home appliances like TV, washing machine, toaster, etc. have started to be connected to the Internet [2], [3]. The role of Fifth Generation (5G) cellular network is expected to be very promising for accommodating increasing reliability requirements on Internet-centric mobile applications [4], [5], [6]. 5G heterogeneous network (HetNet) is anticipated to provide more lucrative features such as higher throughput, lower latency, higher mobility range, massive device connectivity, higher network capacity and energy efficiency. It provides up to 20 times accelerated downloading and uploading speeds than 4G, 10 times lesser round trip latency and bandwidth up to 1 Gbps compared to only 20 Mbps in 4G [7].

Software Defined Network (SDN) and Network Function Virtualization (NFV) are the two influential key technologies which contribute significantly for developing the architectural design of 5G heterogeneous network [8], [9]. Virtualization is a middle layer technology between hardware and software layers which creates virtual representation of something such as virtual machines, servers, memory, network functions, etc. The NFV offers the advantage of segregating the network functions from proprietary hardware appliances and executing these functions in software on standardized hardware instead. These decoupled network functions are referred to as Virtual Network Functions (VNFs) [10].

The NFV offered by Cloud Computing has obtained much popularity as constantly growing number of enterprises and individuals are offloading their workloads to cloud service providers and getting served by them [11]. This technology is also being taken into account by 5G mobile operators to deal with the increasing number of data traffics and data intensive applications [12]. In Mobile Cloud Computing (MCC), because of mobile devices having low computational power and battery lifespan, most of the applications are executed on various VNFs operating in different high computational data centers (DCs) in the cloud [13], [14]. Fig. 1 shows the basic concept of user services architecture in 5G. The user equipments (UEs) are connected to respective evolved Node-Bs (eNBs) of their service area in the network. The cloud domain comprises a number of data centers that serve the eNBs in executing their user applications. Users of an eNB are served by VNFs of exactly one data center. In Fig. 1, for example, associated eNBs of the home and university is connected to data center 1 and data center 2, respectively. When any user moves from his/her home to university, the placement of the running VNFs of that user becomes a matter of concern.

Resource management in cloud computing has been well studied in many research problems [15], [16]. However, because of user mobility, allocation of VNFs for running user applications in different data centers is immensely challenging and difficult task in 5G. Because, for optimal placement of VNFs, various parameters are needed to be considered such as total number of VNF relocation, communication delay as well as response time to get service, its cost, etc. If performance of one parameter is attempted to be improved, performances of some other parameters degrade as a consequence. For example, when users move between two eNBs that are taking services from different DCs, keeping the running VNFs of that user at the previous DC eliminates the need for relocations but requires increased response time to get the services. Therefore, an attempt to minimize the number of VNF relocation results in increased communication delay as well as response time and vice-versa. Because of user mobility, static allocation of the VNFs in the DCs will not meet the demand, degrading user Quality of Experience (QoE). The optimal placement of the running VNFs will change with the respective user movements. Therefore, the optimal allocation of resources in different data centers satisfying all the performance parameters becomes a major challenge.

The problem of optimal allocation of the resources from the mobile devices to the DC has been well studied in many papers. However, these existing approaches in the literature encounter several limitations. In [17], the authors have studied the resource allocation problem in the case of static users but the approach cannot be feasibly used when there exists user mobility. For minimizing the load of the Virtual Machines (VMs), the authors in [18] have placed the VNFs efficiently in the DCs but minimizing VNF relocations and service cost have not been considered. However, in A-SGWR [19], minimization of the number of VNF relocations have been taken into account but not the response time for getting the service. In S-PL [19], the authors focused on minimizing the total response time but the total number of relocations and its overhead have been ignored.

The VNF relocation is only necessary in cases when a user moves between two eNBs connected to different data centers. In such case, VNF can be relocated from previous DC to service DC minimizing the communication path between the serving DC and new eNB, which in turn minimize the response time. Alternatively, user can get service from the DC on which VNF was running via the current serving DC. However, this second way causes higher response time. Therefore, minimizing VNF relocations and maximizing QoE are two conflicting objectives. Artificial Intelligence (AI) is a branch of computer science which enables the development of computer programs which possess the ability to make decisions rationally and solve problem by learning from experiences and improving it gradually [20]. Due to the success of AI in solving complex control and decision-making problems, it is anticipated to contribute significantly for developing various aspects of 5G network and solving complicated problems. In this paper, our aim is to allocate the VNFs in the DC for the mobile users maintaining a trade-off between VNFs relocation and response time. The main contributions in the paper are summarized as follows:

• We formulate the problem of allocating the VNFs associated with user mobility as a Multi-objective Integer Linear Programming (MILP) problem.

• We have brought trade off between minimizing the number of VNF relocations and minimizing the total response time (hereafter, we call our system TradeRC) to get the service ensuring user Quality of Service.

• Due to the NP-hardness of our proposed MILP system, we develop an Ant Colony Optimization (ACO) meta-heuristic based VNF placement algorithm. The operational principle of the proposed system is driven by learning from the past experiences.

• We simulate our proposed system TradeRC in CloudSim [21] and compare it with other state-of-the-art works. The result state that the user QoE in TradeRC has been increased by about 25% and relocation overhead decreased by about 15% other state-of-the-art works.

The rest of the paper is organized as follows. Section 2 presents some related research works. Sections 3 System model and assumptions, 4 Design of TradeRC describe system model and problem formulation, solution details respectively. In Section 4.2, we present ACO-based VNF allocation problem in the data center. Subsequently, Section 5 demonstrates the performance of the proposed optimization solution and comparison with other state-of-the-art works. Finally, Section 6 summarizes the paper and outlines of the future work plans.