Virtual Resource Allocation in Information-Centric Wireless Virtual ...

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

Virtual Resource Allocation in Information-Centric Wireless Virtual Networks Chengchao Liang and F. Richard Yu Depart. of Systems and Computer Eng., Carleton University, Ottawa, ON, Canada Email: [email protected]; [email protected]

Abstract—Wireless network virtualization and informationcentric networking (ICN) are two promising technologies in next generation wireless networks. Traditionally, these two technologies have been addressed separately. In this paper, we show that jointly considering wireless network virtualization and ICN is necessary. Specifically, we propose an informationcentric wireless network virtualization framework for enabling wireless network virtualization and ICN. Then, we formulate the virtual resource allocation and in-network caching strategy as an optimization problem, considering not only the revenue earned by serving end users but also the cost of leasing infrastructure. In addition, with recent advances in distributed convex optimization, we develop an efficient distributed method based on alternating direction method of multipliers (ADMM)based to solve virtual resource allocation and in-network caching scheme. Simulation results are presented to show the effectiveness of the proposed scheme. Index Terms—Wireless network virtualization, informationcentric networking, alternating direction method of multipliers

I. I NTRODUCTION To accommodate the significant growth in wireless traffic and services, it is beneficial to extend virtualization, which has been successfully used in wired networks (e.g., virtual private networks (VPNs)), to wireless networks [1]. With virtualization technology, wireless network infrastructure can be decoupled from the services that it provides, so that differentiated services can share the same infrastructure, maximizing their utilization [1]. Moreover, wireless network virtualization provides easier migration to newer technologies while supporting legacy technologies by isolating part of the network. Several research projects have been started around the world in the area of wireless network virtualization, such as Environment for Network Innovations (GENI) [2] and Virtualized dIstributed plaTfoRms of smart Objects (VITRO) [3]. The authors of [4] propose a wireless local area network (WLAN) virtualization approach to extend the virtual network embedding from wired networks to wireless networks. Virtualizing eNodeB in 3rd Generation Partnership Project (3GPP) Long term evolution (LTE) is investigated in [5] from the views of node virtualization and software defined networks, respectively. Another new technology called

978-1-4673-6432-4/15/$31.00 ©2015 IEEE

Information-Centric Networking (ICN) has attracted great interests from both academia and industry [6]–[10]. The basic principle behind ICN is to promote content to a first-class citizen in the network. A significant advantage of ICN is to provide native support for scalable and highly efficient content retrieval, and meanwhile with enhanced capability for mobility and security. ICN can realize in-network caching to reduce the duplicate content transmissions in networks. A number of research efforts have been dedicated to ICN, including the EU funded projects Publish-Subscribe Internet Technology (PURSUIT) and the US funded projects Named Data Networking (NDN). Although some excellent works have been done on wireless network virtualization and ICN, these two important areas have traditionally been addressed separately in the literature. However, it is necessary to consider these two advanced technologies together to provide better services in next generation wireless networks. Jointly considering wireless network virtualization and ICN could improve the end-to-end network performance. In this paper, we propose an information-centric wireless network virtualization framework for enabling both wireless network virtualization and ICN in next generation wireless mobile cellular networks. We formulate the virtual resource allocation and in-network caching strategy as an optimization problem, which maximizes the utility of mobile virtual network operators (MVNOs), considering not only the revenue earned by serving end users but also the cost of leasing infrastructure. With recent advances in distributed convex optimization, we develop an efficient distributed virtual resource allocation and in-network caching scheme. Simulation results are presented to show the effectiveness of the proposed scheme. The rest of this paper is organized as follows. Section II introduce wireless network virtualization and informationcentric networking. The proposed information-centric wireless network virtualization framework is presented in Section III. Section IV discusses the virtual resource allocation and in-networking caching issues. Simulation results are discussed in Section V. Finally, we conclude this study in Section VI.

3915


Provide Services to End User

InP

Create Virtual Network and Provide to SPs Provide Network and lease to MVNO

SP Information-centric Wireless Virtual Net. Controller

MVNO

SP

MVNO

InP

A. Wireless Network Virtualization and Information-Centric Networking With virtualization, physical cellular network infrastructure resources and physical radio resources can be abstracted and sliced into virtual cellular network resources holding certain corresponding functionalities, and shared by multiple parties through isolating each other. As shown in Fig. 1, three logical roles can be identified after virtualization: infrastructure provider (SP), mobile virtual network operator (MVNO) and service provider (SP). SPs own and operate infrastructures and radio resources of physical substrate wireless networks, including licensed spectrum, RANs, backhaul, transmission networks, and CNs. MVNO leases the network resources from InP, creates virtual resources based on the requests from SPs, operates the virtual resources and assigns them to SPs. The rise of MVNOs breaks the value chain dominated by traditional MNOs [11]. For brevity, we use virtual resources to indicate the virtual mobile network resources. SPs lease, operate and program these virtual resources to offer end-to-end services to mobile users. The principal concern of ICN is to disseminate, find and deliver information rather than the reachability of end hosts and the maintenance of conversations between them. In ICN, the user requests content without knowledge of the host that can provide it, and the communication follows a receiverdriven principle (i.e., the path is set up by the receiver to the provider), and the data follows the reverse path. The network is then in charge of doing the mapping between the requested

Virtual Radio Resource

Virtual Radio Resource

Content ICN BS

BS

Virtual Networks

Radio Resource

Cache BS with Cache

Physical Resources

II. I NFORMATION -C ENTRIC W IRELESS N ETWORK V IRTUALIZATION In this section, we first discuss business models and logical roles in wireless network virtualization. Then, we introduce information-centric networking.

Information-centric Wireless Virtual Network

Content

Physical Wireless Network Controller

Fig. 1. Business models of wireless network virtualization. (a) A twolevel model; (b) A three-level model. SP - service provider; MNO mobile network operator; MVNO - mobile virtual network operator; InP - infrastructure provider.

Service 3

Traditional Virtual Wireless Net. Controller

SP

Service 2 Service 1

Fig. 2. An example framework of information-centric wireless network virtualization. Here, the substrate physical wireless networks are virtualized into two virtual networks. One is running ICN, while another is based on traditional networks.

content and where it can be found, as shown in Fig. 2. The match of requested content rather than the findability of the endpoint that provides it thus dictates the establishment of a communication in ICN. B. Framework of Information-Centric Wireless Network Virtualization In this section, we propose a framework for enabling both wireless network virtualization and ICN, which is called information-centric wireless network virtualization. We present the motivations, radio spectrum resource, mobile network infrastructure, virtual resources, and informationcentric virtualization controller in this novel framework. Traditionally, dedicated physical resources from specific operators are used for for content delivery. As these physical resources cannot be shared by different operators, content delivery increase the complexity of the network, as well as the CapEx and OpEx [12]. Moreover, content delivery is a very volatile market with new protocols, content formats, device types, etc. With dedicated physical resources, operators do not have the flexibility to react on these rapid changes. Fortunately, wireless network virtualization enables the sharing of not only the infrastructure, but also the content, among different service providers. Consequently, the CapEx and OpEx of wireless access networks, content delivery, as well as core networks, can be reduced significantly.

3916


can be easily extended to the multi-cell case. Let I denote the set of SPs. For each SP i, each allocated user is denoted by ki , and Ki is the set of users of SP i. Let xki j denote the association indicator, where x ki j = 1 means that user ki associates to BS j; otherwise xki j = 0. Practically, each user only associates to only one BS; thus xki j = 1. (2)

On the other hand, virtual resource allocation is a significant challenge of wireless network virtualization. Virtual resource allocation schemes need to decide how to embed a virtual wireless network on physical networks (e.g., Which nodes, links and resources should be picked and optimized). As content retrieval (instead of other traditional parameters, such as spectrum efficiency) is put as a high priority in ICN, the processes in wireless network virtualization (e.g., virtual resource abstracting, slicing, sharing and control) will be significantly affected by ICN. Therefore, jointly considering wireless network virtualization and ICN could improve the end-to-end network performance. We propose a framework of information-centric wireless network virtualization, as shown in Fig. 2. In this example, the substrate physical wireless networks are virtualized into two virtual networks. One is running ICN, while another one is based on traditional networks. Different services are provided by these virtual networks. End users logically connect to the virtual network from where they subscribe to the service, while they physically connect to the physical network. A virtual wireless network controller needs to be deployed at the network to realize the virtualization process.

j∈J

yki j ∈ [0, 1] is used to denote the percentage of radio resource allocated by BS j to user k i , yki j ≤ 1. (3) i∈I,ki ∈Ki

The expected instantaneous data rate of user k i is xki j yki j Bj rki j Rki j =

where rki j is the spectrum efficiency of user k i served by BS j and Bj is the total available bandwidth used for data transmission at BS j. In our model, we assume the backhaul bandwidth usage of user k i is the same as instantaneous data rate Rki j [13]. Thus, the total required bandwidth of BS j is Rki j . (5)

III. V IRTUAL R ESOURCE A LLOCATION AND I N - NETWORK C ACHING

i∈I,ki ∈Ki

An important requirement in the proposed framework of information-centric wireless network virtualization is an efficient virtual resource allocation and in-network caching strategy. In this section, we formulate the virtual resource allocation and in-network caching strategy as an optimization problem.

Since the capacity of backhaul is limited, we have the following constraint. Rki j ≤ Cj . (6) i∈I,ki ∈Ki

A. System Model 1) Virtualization Model: With virtualization where slices are bandwidth-based, each SP can schedule next serving users and allocate necessary bandwidth to users based on its own QoS requirements. Assuming the pre-agreed bandwidth of ¯ i , SP can allocate any data rate slice allocated to SP i is R r¯ki to its serving user ki under the constraints ¯i, r¯ki ≤ R (1) ki ∈Ki

where Ki is the set of the scheduled users of SP i. Thus, when MVNO conducts the allocation of substrate resource to user ki , r¯ki requested by SP i should be guaranteed; otherwise the SP will not pay for this user since the agreement is not satisfied. 2) Wireless Mobile Cellular Network Model: We consider a mobile cellular network with J BSs. The set of BSs is J , and j is used to indicate one of the BSs. In this paper, the cellular network is limited in a finite area where there are only one macro BS and several small cell BSs. These BSs belong to different InPs with different leasing prices. Even we only consider a single macrocell system in our paper, it

(4)

j∈J

3) Caching Model: The cache strategy can be controlled by a binary parameter z ∈ 0, 1 [14]. If BS j caches the content that requested by user k, z kj = 1; otherwise zkj = 0. The authors in [15] claimed that the caching gain of mobile networks can be alleviation of bandwidth and reduction of delay. In this paper, we choose the alleviation of backhaul bandwidth as the gain (reward) of caching. Therefore, the expected reward (gain) of a certain caching strategy {z1j , z2j , ..., zKj } is ¯ j xki j zki j ΔCj = qki R (7) i∈I,ki ∈Ki

where qki is the request rate of the content requested by user ¯ j is the average single user data rate of BS j. It ki , and R should be noted that the storage of BS j may be limited. Thus, the cached content cannot be larger than remaining space of cache Z j at BS j. Specifically, we have the following constraint. xki j zki j ski ≤ Zj , (8) i∈I,ki ∈Ki

where ski is the size of the content requested by user k i . In this paper, we assume the size of all the content are the same, and ski = 1.

3917


4) Utility Function: Since the proportion of radio resource allocated to user k is controlled by y ki j , the revenue, radio resource cost and backhaul cost are φ i Bj rki j yki j , αj Bj yki j and βj Bj rki j yki j , respectively, if user k i is served by BS j. Thus, the net revenue of allocating radio resource to user k is defined as aki j yki j , where aki j = φi Bj rki j − αj Bj − βj Bj rki j

optimization problem is expected to be very challenging to find its global optimum. Thus, we have to simplify problem (13). A relaxation of the binary conditions of {x ki j } and {zki j } constitutes the first step of solving the problem. C. Problem Transformation

(9)

According to (7), the expected saved backhaul by caching ¯ j . Thus, the saved cost can be the content of user k is q ki R defined as bki j zki j , where ¯j bki j = βj qki R

(10)

Now, we can formulate the total utility of MVNO by u (aki j yki j xki j + bki j zki j xki j ) Utotal =

(11)

I,K,J

where u(·) is a utility function, which is a nondecreasing and convex function normally. In this paper, we adopt the well-known logarithmic function to our utility function. log x x > 0 u(x) = (12) −∞ otherwise B. Problem Formulation 1) Formulation: The aggregate utility maximization problem is shown as follows: xki j u (aki j yki j + bki j zki j ) max xki j ,yki j ,zki j

I,K,J

s.t. C1 : J xki j = 1, C2 : x I,K ki j yki j ≤ 1, C3 : I,K xki j yki j Rki j ≤ Cj , xki j yki j Rki j ≥ rki , C4 : J C5 : I,K xki j zki j ≤ Zj ,

∀i, j, k ∀j ∀j ∀i, k ∀j

(13)

It is equivalent to take x ki j outside utility function without loss any optimality. If x = 1, we have xu(y, z) = u(x, y, z); if x = 0 that means user is not served by BS so that no resource will be allocated, u(x, y, z) = 0 and xu(y, z) = 0. Note that xki j ≤ 1, yki j ≤ 1, zki j ≤ 1 are eliminated by giving C1, C2 and C5. {x ki j } and {zki j } are Boolean. The first constraint in problem (13) enforces that users can only be associated with one BS at the same time. Constraints C2 and C3 reflect the facts that the sum of allocated resource of all users being served by BS j cannot exceed the total radio resource and backhaul bandwidth. Inequality (13) C4 is due to the minimum virtualization data rate requirements from Subsection III-A1. The constraint of (13) C5 ensures that the caching strategy is limited in the empty space of cache of each BS. Problem (13) is difficult to solve based on the following observations: The feasible set of (13) is nonconvex. The objective function is not convex due to the product relationship of {xki j }. The size of the problem is very large. As is well known, a mixed discrete and nonconvex

Following the approach in [16] and [17], we relax {x ki j } and {zki j } in (13) C6 and C8 to be real value variables that 0 ≤ xki j ≤ 1 and 0 ≤ zki j ≤ 1. Relaxed xki j can be sensible and meaningfully interpreted as the time sharing factor that represents the ratio of time when user k i associates to BS j [17]. Similar to xki j , the relaxed z ki j can be interpreted as the time fraction of sharing one unit cache. However, even relaxing the variables, the problem is still nonconvex due to the nonconvex objective function. Thus, to make the problem (13) tractable and solvable, a second step is necessary. Firstly, we give a proposition of the equivalent problem of (13). Proposition 3.1: If we define y˜ki j = yki j xki j , z˜ki j = zki j xki j and xki j u [(aki j y˜ki j + bki j z˜ki j ) /xki j ] = 0 for xki j = 0, there exists an equivalent formulation of problem (13) as follows: aki j y˜ki j + bki j z˜ki j max xki j u xki j I,K,J s.t. C1 : x = 1, ∀i, j, k J ki j ˜ : (14) C2 ˜ki j ≤ 1, ∀j I,K y ˜ : C3 ˜ki j Rki j − Cj ≤ 0, ∀j I,K y ˜ C4 : rki − J y˜ki j Rki j ≤ 0, ∀i, k ˜ : C5 ˜ki j − Zj ≤ 0, ∀j I,K z This proof of 3.1 is motivated by [18]. The relaxed problem (13) can be recovered by substitution of variable y˜ ki j = yki j xki j and z˜ki j = zki j xki j into problem (14) except xki j = 0. Due to the loss of definition when x ki j = 0, it is not a one-to-one mapping. However, if x ki j = 0, yki j = 0 certainly holds because of the optimality. Obviously, BS j does not allocate any resource to any user if the user does not associate with BS j. Thus, it becomes a one-to-one mapping. Holding 3.1 and well-known perspective function [19], we can have following theory that give the convexity of problem (14). The proof of the convexity is similar to [18]. IV. D ISTRIBUTED V IRTUAL R ESOURCE A LLOCATION AND I N -N ETWORK C ACHING In order to approach problem (14) via distributed method, we introduce local copies of the global association indicators. Each local variable can be considered as the preference of each BS about the association of users. Let us introduce a set of new variables to represent the local copies of our association indicators. To lighten the notation, from now on, we use k to denote all the users instead of k i . If we define the x as the vector of association indicators {x kl , ∀l, k} (note that we change the index of BS from j to l), the local copy ˆ j . Formally, of x at BS j is denoted as x

3918


V. S IMULATION R ESULTS AND D ISCUSSIONS

xˆjkl = xkl , ∀j, k, l

(15)

ˆj , y ˜ j and z ˜j , let us define By means of the local vectors x a feasible local variable set for each BS j ∈ J ˜ ˜ ˜ ˜ ˆj , y ˜j , ˜ C3, C4, C5 (16) χj = x zj C2, and an associated local utility function, respectively, as

gj =

−

ˆjkj u Kx

∞

(akj y˜kj +bkj z˜kj ) x ˆjkl

ˆj , y ˜j , ˜ x zj ∈ χj otherwise

(17) With this notation, the global consensus problem of problem (14) can be written as ˆj , y ˜j , ˜ min gj x zj J (18) j s.t. xˆkl = xkl , ∀j, k, l In this section, the proposed algorithm for virtual resource allocation and in-network caching via distributed method is described. According to [20], our problem (18) is a global consensus problem, since the constraint is that all the local variables should agree. We will apply ADMM for approach the problem (18). The initial step to apply ADMM to problem (18) is that an augmented Lagrangian with corresponding global consensus constraints should be formed. Let λ jkl , ∀j ∈ J , l ∈ J , k ∈ K be the Lagrange multipliers corresponding to consensus constraints in problem (18). The augmented Lagrangian for problem (18) is

j ˜j , ˜ x , y zj }j∈J , {x}, {λj } = Lρ {ˆ ˆj , y ˜j , ˜ gj x zj + J j j (19) x ˆ + λ − x kl j∈J k∈K,l∈J kl kl 2 ρ ˆjkl − xkl j∈J k∈K,l∈J x 2 where λj is the vector of the Lagrange multipliers and ρ ∈ R++ is a positive constant parameter for adjusting the convergence speed of the ADMM [20]. Based on the iteration of AMDD with consensus constraints [20], the ADMM method applied to problem (18) consists of following sequential optimization steps: [t+1]

˜j , ˜ zj }j∈J {ˆ xj , y

j := ˆ ,y ˜ ˜j arg min{gj x j, z j[t] [t] + k∈K,l∈J λkl xˆjkl − xkl + 2 [t] j ρ x ˆ − x k∈K,l∈J kl kl 2 {x}

[t+1]

(20)

VI. C ONCLUSIONS AND F UTURE W ORK

:=

j[t] j[t+1] − xkl x ˆkl arg min j∈J k∈K,l∈J λkl 2 j[t+1] ρ ˆkl + 2 j∈J k∈K,l∈J x − xkl [t+1] [t] ˆ j[t+1] − x[t+1] {λj }j∈J := λj + ρ x

In the simulations, we consider two RAN InPs, two backhaul InPs, one MVNO and three SPs. RAN InP 1 owns a twotier HetNet with one macro BS with price of 10 units/MHz and 6 small BSs with price of 8 units/MHz. RAN InP 2 owns only 6 small BSs with 9 units/MHz. The price of backhaul InPs 1 for macro BS and small BSs of RAN InP 1 is 1 units/Mbps, while the price of backhaul InP 2 for small BSs of RAN InP 2 is 2 units/Mbps. The average numbers of the scheduled users of 3 SPs are assumed to be equal, and the prices for requesting virtual resources are 15units/Mbps, 18 units/Mbps and 22 units/Mbps, respectively. Transmit power of 49dBm for macro BS and of 20dBm for small BSs are considered in our simulations. The bandwidth is 20MHz. In our simulations, the location of the macro BS is fixed in the center and the locations of 12 small BSs are uniformly distributed. To compare our proposed algorithm, two benchmarks are also considered. The first baseline is a centralized algorithm based on solving problem (14) by interior methods directly and round up. The second baseline is a traditional maxSINR association algorithm, all users associate to the BSs who provide the maximum received SINR, and each BS performs proportional fairness resource allocation. The gain of deploying in-network caching is evaluated as another performance measure. Fig. 3 shows the average aggregated backhaul used by all users. From Fig. 3, we can observe that the proposed ADMM-based scheme with in-network caching significantly reduces the total backhaul usage, compared to the max-SINR scheme and the proposed ADMMbased scheme without in-network caching. This is because the proposed information-centric wireless virtual network framework enables in-network caching, which reduces the duplicate content transmission in networks. Next, we compare the number of users who are satisfied with the minimum data rate requirements requested by SPs, as shown in Fig. 4. We can observe that, by deploying our proposed ADMM-based algorithm, virtualization can be realized without violating data rate requirements, because we put virtualization as constraints in our optimization problem. By contrast, compared to our algorithm, the max-SINR scheme cannot guarantee the isolation of virtualization, since some users are affected by other users, and may not get the requested data rate.

(21)

(22)

In this paper, we jointly studied wireless network virtualization and information-centric networking in next generation cellular networks. We proposed an information-centric wireless network virtualization framework for enabling both wireless network virtualization and ICN. Then, we formulated the virtual resource allocation and in-network caching strategy as an optimization problem, which maximizes the utility of mobile virtual network operators. In addition, we developed

3919


R EFERENCES 1100 1000

Total Backhaul Usage (Mbps)

900 800 700 600 500 400 300 Max−SINR w.o. caching Proposed ADMM−based scheme w.o. caching Proposed ADMM−based scheme w. caching

200 100 15

20

25

30 Number of Users

35

40

45

Fig. 3. Backhaul alleviation by deploying in-network caching.

45

Number of Total Satisfied Users of SPs

40

Max−SINR Proposed ADMM−based Scheme

35 30 25 20 15 10 5 0

15

18

21

24

27 30 33 36 Number of Users

39

42

45

Fig. 4. The number of total satisfied users of SPs.

an efficient ADMM-based distributed virtual resource allocation and in-network caching scheme. Simulation results were presented to show that the performance of backhaul alleviation can be substantially improved in the proposed scheme with in-network caching. The InPs, SPs and MVNOs can benefit from the proposed information-centric wireless network virtualization framework. Future work is in progress to consider energy efficiency [21] and full-duplex relaying [22] in the proposed framework. ACKNOWLEDGMENT This work was supported in part by the Natural Sciences and Engineering Research Council of Canada and in part by Huawei Technologies Canada CO., LTD.

[1] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, some research issues and challenges,” IEEE Commun. Surveys Tutorials, 2014, dOI: 10.1109/COMST.2014.2352118. [2] L. Peterson, T. Anderson, D. Blumenthal, D. Casey, D. Clark, D. Estrin, J. Evans, D. Raychaudhuri, M. Reiter, J. Rexford et al., “Geni design principles,” IEEE Computer, vol. 39, no. 9, pp. 102–105, Sep. 2006. [3] L. Sarakis, T. Zahariadis, H.-C. Leligou, and M. Dohler, “A framework for service provisioning in virtual sensor networks,” EURASIP J. on Wireless Commun. & Netw., vol. 2012, no. 1, pp. 1–19, 2012. [4] L. Xia, S. Kumar, X. Yang, P. Gopalakrishnan, Y. Liu, S. Schoenberg, and X. Guo, “Virtual wifi: bring virtualization from wired to wireless,” ACM SIGPLAN Not., vol. 46, no. 7, pp. 181–192, Mar. 2011. [5] K. Pentikousis, W. Yan, and H. Weihua, “Mobileflow: Toward software-defined mobile networks,” IEEE Commun. Mag., vol. 51, no. 7, pp. 44 – 53, Jul. 2013. [6] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman, “A survey of information-centric networking,” IEEE Commun. Mag., vol. 50, no. 7, pp. 26–36, Jul. 2012. [7] J. Choi, J. Han, E. Cho, T. T. Kwon, and Y. Choi, “A survey on contentoriented networking for efficient content delivery,” IEEE Commun. Mag., vol. 49, no. 3, pp. 121–27, Mar. 2011. [8] S. Paul, J. Pan, and R. Jain, “Architectures for the future networks and the next generation internet: a survey,” Comput. Commun., vol. 34, no. 1, pp. 2–42, Jan. 2011. [9] C. Fang, F. R. Yu, T. Huang, J. Liu, and Y. Liu, “A survey of energyefficient caching in information-centric networking,” IEEE Comm. Mag., vol. 52, no. 11, pp. 122–129, Nov. 2014. [10] C. Fang, F. R. Yu, T. Huang, J. Liu, and J. Liu, “A survey of green information-centric networking: Research issues and challenges,” IEEE Commun. Surveys Tutorials, 2015, dOI: 10.1109/COMST.2015.2394307. [11] T. K. Forde, I. Macaluso, and L. E. Doyle, “Exclusive sharing & virtualization of the cellular network,” in Proc. IEEE Symp. on New Frontiers in Dynamic Spectrum Access Net. (DySPAN), Aachen, May 2011. [12] ETSI, “Network functions virtualisation (NFV): Use cases,” Tech. Rep. 2013001, Mar. 2013. [13] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource allocation in multi-cell ofdma systems with limited backhaul capacity,” IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 3618–3631, Sep. 2012. [14] P. Blasco and D. Gunduz, “Learning-based optimization of cache content in a small cell base station,” arXiv preprint arXiv:1402.3247, 2014. [15] X. Wang, M. Chen, T. Taleb, A. Ksentini, and V. Leung, “Cache in the air: exploiting content caching and delivery techniques for 5g systems,” IEEE Commun. Mag., vol. 52, no. 2, pp. 131–139, Feb. 2014. [16] Q. Ye, B. Rong, Y. Chen, M. Al-Shalash, C. Caramanis, and J. G. Andrews, “User association for load balancing in heterogeneous cellular networks,” IEEE Trans. Wireless Commun., vol. 12, no. 6, pp. 2706–2716, Jun. 2013. [17] D. Fooladivanda and C. Rosenberg, “Joint resource allocation and user association for heterogeneous wireless cellular networks,” IEEE Trans. Wireless Commun., vol. 12, no. 1, pp. 248–257, 2013. [18] S. Gortzen and A. Schmeink, “Optimality of dual methods for discrete multiuser multicarrier resource allocation problems,” IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 3810–3817, 2012. [19] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge university press, 2009. [20] S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimization and statistical learning via the alternating direction method of multipliers,” Foundations and Trends in Machine Learning, vol. 3, no. 1, pp. 1–122, 2011. [21] S. Bu, F. R. Yu, Y. Cai, and P. Liu, “When the smart grid meets energyefficient communications: Green wireless cellular networks powered by the smart grid,” IEEE Trans. Wireless Commun., vol. 11, pp. 3014– 3024, Aug. 2012. [22] G. Liu, F. R. Yu, H. Ji, V. Leung, and X. Li, “In-band full-duplex relaying: A survey, research issues and challenges,” IEEE Commun. Surveys Tutorials, 2015, dOI: 10.1109/COMST.2015.2394324.

3920