Table of Contents Author Guidelines Submit a Manuscript
Wireless Communications and Mobile Computing
Volume 2018, Article ID 5109394, 13 pages
https://doi.org/10.1155/2018/5109394
Research Article

Advanced QoS Provisioning and Mobile Fog Computing for 5G

1Ss. Cyril and Methodius University, Faculty of Electrical Engineering and Information Technologies, Skopje, Macedonia
2Mother Teresa University, Faculty of Informatics, Skopje, Macedonia

Correspondence should be addressed to Toni Janevski; km.ude.miku.tief@jinot

Received 26 December 2017; Revised 20 April 2018; Accepted 23 April 2018; Published 7 June 2018

Academic Editor: Antonella Molinaro

Copyright © 2018 Tomislav Shuminoski et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

This paper presents a novel QoS and mobile cloud and fog computing framework for future fifth generation (5G) of mobile and fixed nodes with radio network aggregation capability. The proposed 5G framework is leading to high QoS provisioning for any given multimedia service, higher bandwidth utilization, traffic load sharing, mobile cloud plus fog computing features, and multi-radio interface capabilities. The framework is user-centric, targeted at always-on connectivity with using radio network aggregation for available mobile broadband connections, and empowered with mobile cloud and fog computing advantages. Moreover, our proposed framework is using Lyapunov drift-plus-penalty theorem that provides a methodology for designing algorithm to maximize the average throughput and stabilize the queuing. Also, we are showing the upper bound of the consumed power and the lower bound of the battery lifetime for the proposed 5G terminal. The advanced performance of our 5G QoS plus MCC framework is evaluated using simulations and analysis with multimedia traffic in heterogeneous mobile and wireless environment. The simulation results are showing that the maximal network utilization, maximal throughput, minimal end-to-end delay, efficient energy consumption, and other performance improvements are achieved.

1. Introduction

Nowadays we are at the point of era when the peak of 5G researchers’ work is reached, the period of implementation of the first pilot 5G networks. The year 2020 is near; it is the year when we should expect the first full-operative 5G deployment. Compared with the previous network implementations and designs, the future 5G systems should require smarter devices able to provide mobile broadband services to the end users, ubiquitous mobility, advanced mobile cloud computing (MCC) features, fog computing features, enormous processing power of the mobile devices, machine-to-machine-based communications, better network utilization, and many other advanced capabilities [18]. Above all, the key goal is provisioning high Quality of Service (QoS) support, as well as faster computing features and longer battery life of mobile nodes (MNs). The exponential growth in the amount of traffic carried through mobile networks and cloud computing is followed by a novel research work towards the advanced computing capabilities of the core part of the networks.

The key trend in the past decade was to push the computing, control, data storage, and processing in the cloud computing [914]. However, in order to meet the intelligent networking and computing demands in 5G network, the cloud alone encounters too many limitations, such as requirements for reduced latency, high mobility, high scalability, and real-time execution. The existing cloud computing mechanisms and data delivery models do not provide the necessary QoS and Quality of Experience (QoE) for the forthcoming large-scale and dynamic services in 5G networks. This is because of the large number of hops of wired networks between the 5G base stations and the cloud which leads to a significant increase in latency. Moreover, when all data generated by the devices is directly forwarded to the cloud may devour the bandwidth and lead to congestion.

A new paradigm called fog computing has emerged to overcome these limitations [15, 16]. Fog distributes computing, data processing, and networking services to the edge of the network, closer to end users. It is an architecture where distributed edge and user devices collaborate with each other and with the clouds to carry out computing, control, networking, and data management tasks [17, 18].

Rather than concentrating data and computation in a small number of large clouds, many fog systems would be deployed at the proximity to end users or where computing and intelligent networking can best meet user needs. The main idea is to take full advantage of local radio signal processing, cooperative radio resource management, and distributed storing capabilities in edge devices, which can decrease the heavy burden on front haul and avoid large-scale radio signal processing in the centralized baseband unit pool.

Undoubtedly, the future generation mobile services and networks move towards the user-centric concept, which is the main reason why the user-centric approach is accepted as a basis for our work.

This paper provides 5G technology framework that could lead to high service performances with excellent QoS provisioning and mobile cloud and fog computing capabilities, using any existing and future RATs (Radio Access Technologies).

The proposed system model framework design and methodology are based on the adaptive queuing Lyapunov optimization techniques [19], which are powerful techniques for optimizing time average queuing networks and are giving joint stability and performance optimization. The solutions and applications are including maximization of the aggregated average throughput subject to average power constraints on the node interfaces (implying longer battery life, as given in Section 3), minimizing average queue backlogs, subject to minimal queue network delay, and leading to achievements of overall system stability. Moreover, this system model supports the fog computing service orchestration mechanisms described in [20] and performs a quality evaluation of these mechanisms in terms of delay, throughput, energy consumption, and energy efficiency.

The remainder of this paper is organized as follows. Section 2 gives an overview of the most relevant research works in this field of 5G nodes, cloud and mobile fog and cloud computing. Section 3 presents the system design of the 5G node with advanced QoS module within and MCC features. Furthermore, Section 4 provides simulation results for the proposed 5G nodes. Finally, Section 5 concludes this paper and provides future work directions.

2. Related Works

The tremendous interest and developments of multimedia services for mobile and wireless broadband networks undoubtedly lead to intensive research works towards advanced mobile cloud computing algorithms and frameworks for high level of QoS provisioning in each telecommunication network. At first place, our proposed 5G nodes implement Lyapunov drift-plus-penalty techniques, which for the first time were applied to wireless networks in [21] by Tassiulas and Ephremides, where stochastic Lyapunov drift is used to develop a joint optimal routing and scheduling algorithm.

Tassiulas-Ephremides in [22] are introducing single-hop RAN (radio access network) with random changing connectivity, stability properties, and a policy that minimizes the queue delay, but they are not studying the recourse allocation problem under the assumption of multivalued connectivities. However, the Lyapunov drift since then has become a powerful technique for the development of stable scheduling strategies for mobile and wireless systems [2225], computer networks and switches [26], ad hoc mobile networks [27], wireless sensor networks [28], and wireless mesh networks [29].

Other methods for joint stability and utility optimization via Lyapunov drift are developed for stochastic networks in [30, 31] for application to flow control and energy minimization. An alternative approach is developed in [32] using stochastic gradient theory and fluid model transformations.

In [33] the Lyapunov drift is also applied to wireless networks with multireceiver diversity, where an optimal diversity backpressure routing algorithm is developed and shown to improve performance beyond that of related diversity algorithms that do not use backpressure. Moreover, in [34] an optimization problem is presented to minimize the total energy consumed by the mobile users in executing a given service under total execution time constraints in the process of cloud offloading for multi-radio enabled mobile devices.

The given optimization problem does not consider joint stability and performance optimization, since there is no usage of Lyapunov drift-plus-penalty technique and the average throughput is not maximized. Despite all related works, this paper is applying a version of the Lyapunov drift-plus-penalty theorem [35] in the future 5G nodes with adding optimal battery usage algorithm. Such 5G node achieves high performance regarding the aggregated average throughput, minimal service delay, stability, and high level of overall QoS assurance. Also, it has vertical multihoming and vertical multistreaming features [36, 37] and mobile cloud and fog computing capabilities.

The cloud in 5G is distributed on all 5G network architectural levels. It can be centralized in 5G core, geodistributed in 5G RAN (CRAN and FogRAN), and peer-to-peer massively distributed among the 5G mobile smart user devices [3840].

The centralized cloud in the 5G core consists of powerful high-performance computing nodes which provide ubiquitous, pervasive, convenient, and on-demand network access to a shared pool of configurable computing resources such as networks, servers, storage, applications, and services that are rapidly provisioned and released with minimal management effort or service provider interaction. This is an efficient and scalable centralized solution for information management and distribution, for the traditional desktop users that request global information from the remote central server like world news, stock market in different countries, etc.

The smart mobile devices on the other hand massively demand local information around them [39]. For example, a mobile user in a shopping center is interested in the sales, open hour, restaurants, and events inside the attended shopping center, while such information becomes useless once he/she leaves the shopping center. Therefore, the conventional cloud-based Internet is inefficient in serving the local information desired by mobile users. Because the local shops and stores first need to upload their information to a remote cloud server over Internet, only then the mobile users would be able to obtain the desired information from the remote cloud server. Although the physical distance between the mobile user and the shops and stores is short, the actual communication distance, between the mobile user and the cloud server can be far, which may result in increased delay and high congestion.

In addition, because the smart mobile devices have limited data processing and storage capabilities they offload both the data storage and data processing to the cloud computing data centers in 5G core [11], which may require very high throughput and high bandwidth and low delay.

One possible solution to solve these issues is to geodistribute the cloud in 5G RAN. The geodistributed cloud in 5G RAN may appear in the following forms: Cloud-RAN (CRAN), mobile edge computing and fog computing.

CRAN incorporates cloud computing into RANs [41], but the application storing and all radio signal processing functions are centralized at the cloud computing server in 5G core [41, 42]. However, billions of smart user devices need to transmit and exchange their data fast enough with the base band unit BBU pool, which requires high bandwidth and low latency.

Therefore, heterogeneous cloud radio access networks (HCRANs) have been proposed as a solution in which both the user and control planes are decoupled [41, 43]. Here, the centralized control function is shifted from the BBU pool in CRANs to the high-power nodes (HPNs) in HCRANs. HPNs also provide seamless coverage and execute the functions of control plane. The high-speed data packet transmission in the user plane is enabled with the radio heads (RRHs). HPNs are connected to the BBU pool via the backhaul links for interference coordination.

However, the HCRANs still have some disadvantages. For example, traffic data over the fronthaul between RRHs and the centralized BBU pool uses a lot of redundant information, which worsens the fronthaul constraints. In addition, HCRANs do not take fully utilize the processing and storage capabilities in edge devices, such as RRHs and smart user devices, a promising approach to successfully alleviate the burden of the fronthaul and BBU pool. Moreover, the network operators must deploy a huge number of fixed RRHs and HPNs in HCRANs in order to meet the requirements of peak capacity, which makes a serious waste when the traffic volume is not sufficiently large.

Therefore, it is necessary for data processing to be located near the user devices and close to the data source, so that high-speed transmission of 5G can be utilized and data can be processed and filtered out by the time it reaches the cloud. Because of this, mobile edge computing (MEC) has recently started to receive much attention and it is seen as an important technology component in 5G [44].

MEC is a model that enables cloud computing platform to be implemented within the radio access network, nearby the mobile users in order to serve applications that are delay sensitive and context aware. The mobile edge computing entity is positioned next to the radio access network. This entity works with downlink bitstreams from the cloud computing servers to the mobile device and uplink bitstreams from the mobile device to the cloud computing servers. The MEC platform contains standard IT servers and network devices inside or outside of the base station [44]. The external applications are executed in virtual machines (VMs), which are connected with the network devices [45]. Also, the MEC platform can be deployed with standard IT servers, where the network device is implemented as a software entity, such as Open Virtual Switch or Оpen vSwitch (OVS) [46].

However, mobile edge computing devices and entities within the domain are standalone or interconnected ones through proprietary networks with custom security and little interoperability. Although MEC has recently attempted to include some functions of cloud computing like interoperability, local security, etc., it does not extend to the cloud or across domains [47, 48].

In order to provide seamless extension of cloud computing into the edge of the network for secure control and management of domain specific hardware, software, and standard computing, storage, and network functions within the domain and enable secure rich data processing applications across the domain, a new paradigm known as fog computing is introduced [1518]. It represents a completely geodistributed network, multilayered cloud computing architecture for data processing and storage, which contains billions of devices as part of Internet of Things (IoT), multiple local clouds at the edge of the network: fog and main central hyperscalable cloud computing data center. A single application in the fog is distributed to the devices through the cloud components embedded in the nodes in the different network levels, for example, in the RAN, multiservice edge and the core of the network (in the IP/MPLS routers and switches, the gateways of the mobile packet core, etc.). In this way the cloud is closer to the users of mobile devices and is able to offer ultralow latency, higher throughput, lower power consumption, better energy efficiency, quicker response, and high bandwidth, as well as real-time access to the radio information that will be used by the applications in order to offer context related services.

The Fog computing Radio Access Network (FogRAN) architecture uses the advantage of local radio signal processing, cooperative radio resource management, and distributed storing capabilities in edge devices, which can decrease the heavy burden on front haul and avoid large-scale radio signal processing in the centralized baseband unit pool [41]. It contains fog computing nodes that are positioned at the edge of the network, away from the 5G core cloud computing data centers. These fog computing nodes have dense geographical distribution. Therefore, they extend the cloud computing at the edge of the network and provide very low and predictable latency and high support of mobility. If the smart mobile device moves far away from the current servicing fog computing node, then the fog computing node redirects the services and the application to a fog node that is located at the proximity to the smart mobile device. Fog computing nodes also provide support of applications with awareness of device geographical location and device context. They directly communicate with the mobile users through edge gateway and single-hop wireless connections using the off-the-shelf wireless interfaces, like LTE, WiFi, Bluetooth, etc. They also independently provide predefined service applications to mobile users without assistances from cloud computing data centers or Internet. In addition, the fog nodes are connected to the centralized cloud centers in order to leverage the rich functions and application tools of the cloud.

However, the FogRAN solution does not spread the cloud, i.e., the fog computing capabilities, to the level of the smart user devices, such as such as smartphones, IoT devices, sensors, etc. Our proposed solution framework achieves this. In our solution, 5G smart user devices form locally distributed peer-to-peer (P2P) mobile cloud, where each device shares the resources with other devices in the same local cloud [49]. One of the devices is selected as Local Cloud Resource Scheduler, which performs management on the resource requests and allocates tasks to the devices in the local cloud or Fog Data Center if necessary. The decision about the selection of the Local Cloud Resource Scheduler is done according to the connectivity to the local network, CPU performance, battery lifetime, energy efficiency, etc.

3. System Model

Figure 1 depicts the system model and usage scenario for our proposed 5G mobile cloud computing framework. Also, the main characteristics of our 5G node with incorporated advanced QoS user-centric aggregation module with vertical multihoming and multistreaming features are illustrated in Fig. 1 in [35]. The 5G nodes (5G mobile node in the access part of the network and 5G Cloud-RAN node in the core part of the network) are multi-RAT interface, equipped with several (M) interfaces: each for different RAT. Despite our previous works [35, 49, 50] and using AQUAplus (advanced QoS-based user-centric aggregation plus Lyapunov optimization) algorithm from [35], the novelty in this framework is in

Figure 1: Overview of the system model and 5G nodes.

(i) implementing additional intelligent battery saving module (for optimal use of energy) in the 5G mobile nodes; implementing advanced mobile cloud and fog computing features within the proposed 5G system architecture for rich computational resources (used by AQUAplus for faster and better QoS provisioning),

(ii) placing most of the complex AQUAplus optimization’s calculations and computations within the 5G Cloud-RAN.

Moreover, the 5G Cloud-RAN is placed not in the core network, but near the RANs, for achieving smaller delays and faster responses to the end user’s 5G mobile nodes. In addition, the fog computing is included in each 5G RAN and the smart user devices. The locally distributed P2P mobile cloud has its own advantages, because the workload of the application is managed in a distributed fashion without any point of centralization. This provides scalability, while exploitation of user resources reduces the service cost. The devices possess capacities such as storage space, computational power, online time, and bandwidth. Finally, the P2P mobile cloud has an ability to adapt to network failures and dynamically changing network topology with a transient population of nodes/devices, while ensuring acceptable connectivity and performance. Thus, P2P systems exhibit a high degree of self-organization, self-optimization, and fault tolerance.

Compared to cloud computing, fog computing for mobile users provides enhanced service quality with increased data rate and reduced delay and response time. In addition, by avoiding the duplicated back and forth traffic between cloud and mobile user, not only is the backbone bandwidth significantly saved, but also the energy consumption of core networks can be greatly reduced; i.e., the energy efficiency is greatly improved. This contributes to the reduction of operation costs for the network operators and sustainable development of green networking. Moreover, by reduction of the bandwidth cost of data transmission in the backbone, the service cost for the users is also reduced.

3.1. 5G QoS Algorithm and Mobile Cloud

The Advanced QoS-based algorithm (i.e., AQUAplus) is set within the 5G node (in both nodes: fixed and mobile). But, despite the limited 5G mobile node’s performances (processing resources, limited memory, battery lifetime, etc.), most of the complex and tremendous calculations are transferred in the fixed 5G node, here named 5G Cloud-RAN.

In this way, we have mobile fog computing done in the 5G mobile nodes and MCC done in the 5G fixed node (5G Cloud-RAN).

To emphasize that the used AQUAplus algorithm is explained in detail in [35] (named plus, because it is extension of our AQUA algorithm placed within 5G mobile nodes and proxy servers, presented in related previous papers [49, 50]).

The used system model for the 5G nodes consisted of three main parts: multimedia services (i.e., sources of information for audio, video, and/or data), module for QoS-based intelligent routing (using AQUAplus algorithm [23]), and queues for each different RAT interface. We are considering application sources, which are stationary independent processes with packet arrival rates for every time slot t.

Each source arrival process is entering in the AQUAplus module where, at the first place, the vertical multistreaming and multihoming processes ([36, 37]) of division of one stream (traffic which is originating from one service source) is done, in order to go over different queues. Further, the AQUAplus module is considering queue network with queue vector that evolves in slotted time with update equation:where is arrival rate variable for the m-th queue, and the parameter is the weighting factor for the th traffic flow (i.e., ), which goes over m-th RAT interface; is output serving rate variable of the m-th queue, where . After each queue we are trying to achieve maximal output serving rate on each interface , so the sum of all output serving rates, over the time will be also with maximal value.

Also, we are considering a vector with time average power values for each interface as . Let the be a vector of the time average arrival rates () and be the separable utility function of that vector, which is in the same time the objective function (same as that in [35]). Then, for each 5G node we are applying stochastic utility maximization framework to a simple flow based network model, so the following optimization problem in uplink (in the 5G mobile nodes) and downlink (in the fixed 5G Cloud-RAN node) is considered and solved with the AQUAplus algorithm (see [35]).

Maximize

subject to the following:

(1) Time average flow over the queue (t) is less than or equal to the time average maximal output serving rate on the interface m; i.e.,

(2) All queues are rate stable; i.e.,

(3) The desired time average power constraints are met [31] ((t) is the power incurred in interface of the network on slot t, and is a required time average power expenditure); i.e.,

(4) For the control policy action: .

For t > 0, the used variables as time average over the first slots are defined for , with

For solving the above optimization problem AQUAplus algorithm [35] uses the Lyapunov drift-plus-penalty method (using fixed penalty control parameter V) given in [19]. Here constraints , , and of the optimization problem are enforced with the actual queue (t) (1) and the virtual queue for each m:

where it is easy to show that there is stable mean rate [19] (with finite queue length).

The main outcomes of using AQUAplus algorithm are maximizing the aggregated average throughput and providing minimal average queue delay and optimal power (battery) usage, simultaneously causing network and system stability.

For more details on how AQUAplus solves the above optimization problem and how it gives the optimal flow control decisions, see [35].

The complex optimization’s calculations and computations are placed in the 5G Cloud-RAN node, due to the processing potentials of this node, the memory usage, power supplies, and many other superior features in comparison with the mobile terminals. In that way, we have MCC implementation, which helps 5G mobile nodes to relax and offload from the complex AQUAplus calculations. The simple optimization computations are done in the 5G mobile nodes, following the fog computing manner.

The energy efficiency EE represents the amount of data that can be transferred through the power consumed per user, usually on a single cell, and is the ratio between the user throughput R and the power P:

The reciprocal value of the energy efficiency represents the energy consumed per bit per user:

Throughput is the quantity of data that can pass from source to destination in a specific time. The user throughput is calculated as a ratio of the peak data rate Rmax of the particular RAN and the number of smart user devices N and proportional to some weight coefficient μ:

Here μ is a weight coefficient that models the bottleneck problem for the data that carry services from the cloud computing data centers. Due to the increased number of flows for different service requirements, the user throughput given in (10) is decreased for a certain factor.

The weight coefficient μ may receive values between 0.8 and 1, and its value depends on how much the cloud is far away from the radio access network. If the cloud is closer to the base station of the radio access network then the coefficient μ has higher value, and if the cloud is at a greater distance from the base station of the radio access network then the coefficient μ would have lower value. If the mobile device uses a service that is located in the fog computing environment, i.e., in the radio access network, then the weight coefficient μ is equal to 1. For our simulation purposes μ is equal to 1 for a fog computing node, 0.9 for a CRAN, and 0.8 for cloud computing data center.

The peak data rate Rmax for 5G radio access network depends primarily on the Adaptive Modulation Coding Scheme (AMSC) that makes a compensation for the noise, interference, and other factors that have a negative influence on the useful signal in order to deliver higher capacity and better coverage in the presence of noise and other distortions. Depending on the distance between the mobile user device and the radio access network, in the areas where the signal level is good, a modulation with a higher data rate and less robust coding is used. On the other hand, in the areas where the signal level is weak or multipath reflections exist, a modulation with lower data rate and more robust coding is used in order to minimize the errors. For more details about the possible modulation coding schemes for 5G radio access networks, see [51].

In the real world, which modulation coding scheme would be applied at which distance is left to be decided by the network operator itself. In the simulations of this doctoral dissertation the maximum distance between the mobile user device and the base station is taken to be 5000 meters, where at every 500 meters the modulation coding scheme is changed.

The consumed power can be expressed through the user throughput with the following linear equation [52]:

where α is the coefficient that gives the power necessary for data transfer (in downlink or uplink direction), аnd β is a coefficient that represents the idle power [53]. Table 1 gives the values of these coefficients for 5G mobile network, which are empirically (numerically) calculated.

Table 1: Typical values for the power consuming coefficients in 5G network.

Finally, the 5G Cloud-RAN node always can forward part of the AQUAplus computations and calculations in the central cloud computing servers (in Figure 1 cloud servers) for minimizing the delay from solving of the optimization problem, computing load balancing, and reaching the optimal results faster.

3.2. Consumed Energy and Battery Lifetime

Crucially important for the proposed 5G mobile nodes with AQUAplus algorithm is to find the theoretical upper bound of the consumed energy (or power) and the lower bound for the battery lifetime. It is emphasized that the 5G fixed node (the 5G Cloud-RAN node) has theoretically unlimited energy, because it uses power supply units and uses back-up batteries (Uninterruptible Power Supplies (UPS)) when there is no power supply, which is very rare case in the core network.

Let us have upper bound of each value for the energy queue , where the upper bound of the energy queue (t) is defined as

where is maximal length of the real queue (t). Also, we assumed that . From the other side, T is the slot time (or bit period), is maximal energy, which can be consumed on radio interface m, and is a required average value of the energy expenditure on the radio interface m, defined as .

We will prove the following.

Theorem 1. For each positive number T, which is the number of time units for one time slot, the total consumed energy in our proposed 5G mobile fog computing node, for each radio network interface m, for each time slot T, is determined by the upper bound of the expression ; i.e., the following inequality is satisfied:

In order to prove the above upper bound, we will start from (7). So, applying the above defined upper bound for virtual queue (7), i.e., assuming that (t+1)≤, we have the following inequality:

We start from dynamic equation (7), knowing that for each time slot the following is satisfied:

Now, summarizing the above equation over and applying the low of telescoping sums (method of differences, i.e., differentials) we have

Next, replacing = and = +T-1, plus assuming that is a constant value, expression (16) is transformed into

Because the expression from the left side of inequality (16) for sure is equal to or lower than , with multiplying the virtual power queues with T (time slot duration) and with small rearrangements, we proved that (17) is satisfied.

With that, the statement for the total consumed energy for our proposed system model is proved and is smaller than or equal to + . This bound is of significant importance for the 5G mobile fog computing nodes, which have energy scarcity, and it shows that it is proportional to the time for using all included radio interfaces, the average allowed energy (for each interface), and the maximal backlog value of the network queues.

If the values of the above-mentioned parameters are lower, consequently the total consumed energy should be smaller and the power supplies for the 5G node’s battery will be also small. In that direction, the overall lifetime of the battery will be longer (the energy efficiency of the 5G node’s battery will be with large value).

Furthermore, if we observe the radio network interface m, which is wireless transmitter and receiver, let the initially used energy be from the battery with total energy E (where ). Then, the guaranteed battery lifetime () for the proposed 5G mobile node can be found by the derivation from the direct dependency of the total energy and the total battery power, i.e., from the starting point

and from the fact that, for a particular radio interface m, the total energy on the m-th interface is

By summarizing (19) for all used radio network interfaces, the total energy could be found. Consequently, the battery lifetime (duration) for our proposed 5G mobile node is

If on (20) we apply (17) for the total consumed power of the network radio interface m, then the total battery lifetime is determined with the following expression:

As can be noticed, the battery lifetime of the 5G mobile fog computing node is determined by the existence of lower bound, given with the right side of expression (21). Consequently, the battery of the 5G mobile fog computing node will have longer life, if the available energy per RAT interface is higher and is inversely proportional to the number of used interfaces (M), maximal backlog value of the queues, the duration of the time slot T, and the value of the average power (pav) per radio interface. One can notice that the radio network interfaces which are not used (up) at the particular time slot should not be included in the calculations, so we can get longer battery lifetime for the 5G terminal. The derivation above (in expression (21)) is solved by the assumption that all radio interfaces are launched (active) and that all of them are having the same average power expenditures (). But, in the optimal case, there are only several RATs available and the most appropriate ones of them are chosen by the AQUAplus, so the energy consumption is reduced to a minimum. That is how the battery lifetime in our proposed 5G mobile node is longer and in the worst case comparable with the nowadays smartphones (3G or 4G mobile terminals).

On the other hand, considering the recently established Koomey law [54], based on analysis of electrical efficiency of computation in the past six decades, the overall outcome is that “the power needed to perform a task requiring a fixed number of computations will fall by half every 1.5 years, enabling mobile devices performing such tasks to become smaller and less power consuming and making many more mobile computing applications feasible. Alternatively, the performance of mobile devices could continue to double every 1.5 years while maintaining the same battery life (assuming battery capacity does not improve).” Roughly speaking, in a period of three years (e.g., from 2017 to 2020) the computational processing in the mobile terminals can increase around 8 times (23=8, due to 2 periods of 1.5 years in the 3-year period) with the same battery life, thus providing possibility of use of several different RAT interfaces at the same time for different multimedia services, including current as well as future ones. So, it is expected that the initial ICT philosophy of keeping the network simple, as much as possible, and giving more functionalities to the end nodes (5G mobile node and 5G Cloud-RAN and fog node), would become reality in the future generation of mobile (5G) networks.

4. Simulation Results and Analysis

In this section, we provide simulation results for the key QoS parameters: average throughput and delays as well as energy consumption and energy efficiency results using the proposed 5G nodes with high level of QoS provisioning with MCC and fog features.

The 5G scenarios include 5G mobile network with 100 to 1000 smart 5G capable user devices that are located at different distances from the serving 5G base stations in the 5G RAN. Therefore different 5G smart devices are using different modulation coding schemes over the 5G radio interface, which results in different peak data rates . The resulting average power for each interface within the 5G nodes is set to be =0.721 W. The control parameter is V=10. All exogenous stochastic inputs (sources) are independent of Poisson processes, with three basic packet arrival rates: =50 packets/s, =24.316 packets/s, and =68.9655 packets/s (corresponding to the packet arrival’s rates of VoIP, VoIP-conference, and data, respectively).

Figure 2 shows simulation results over 108 slots for the average delay values (in units of slots) versus different number of active application sources/services (different values for the total input packet arrival rate), when the complete arrival rate vector for sources is 68.9655; 50; 24.316; 100; 50; 50; 24.316; 50; 68.9655; 68.9655. If the number of sources is , then only the first nine arrival rates will be summarized from the vector , etc. In all seven cases the 250 5G mobile nodes are moving with average velocity of 40 km/h.

Figure 2: Average delay versus number interfaces in 5G nodes.

The maximal output serving rate vectors (for M = 1, 2, …, 7) for all cases are as follows:(i)If M=1, then = (1/250)44643.(ii)If M=2 then = (1/250)44643 12500.(iii)If M=3 then = (1/250)44643 12500 5000.(iv)If M=4, then = (1/250)6694.4 12500 5000 44643.(v)If M=5, then = (1/250)6694.4 12500 5000 44643 6694.4.(vi)If M=6, then = (1/250)6694.4 12500 5000 44643 5000 6694.4.(vii)If M=7, then = (1/250)6694.4 12500 5000 44643 5000 6694.4 12053.57.

Undoubtedly, as the total number of used interfaces in 5G nodes is higher, the average delay is smaller. The case when they are 7 interfaces is achieving superior results for any number of data sources over other six cases.

It is emphasized that in Figures 24 the case when there is only one used interface (M=1) is the case when we are using the basic mobile fog and cloud computing methods [1518] and there is no AQUAplus algorithm within the used mobile nodes. With this specific comparison, the novelty and actual value of the proposed framework is clearly defined. Moreover, the above maximal output rates are carefully chosen to be adequate to the maximal uplink serving rates (by dividing the uplink bit rates with the average number of bits per packet) of LTE/LTE-Advanced, IEEE 802.11n, IEEE 802.16e, IEEE 802.16 m, and IEEE 802.11ac RATs, shared between 250 mobile nodes in one cell.

Figure 3: Average delay versus average velocity of 5G mobile nodes.
Figure 4: Average throughput versus velocity of 5G mobile nodes.

The upper bound for the average time delay is calculated by Little’s theorem [55, 56], i.e., by dividing the average queuing backlog with the average arrival rates, respectively. Consequently, the case with more interfaces within 5G nodes with AQUAplus algorithm will cause minimal average time queue backlogs. It is noticeable that the delays start to rise when the number of sources is equal to the number of RAT interfaces which are launched (M).

Furthermore, Figure 3 presents the average queue delays versus the average velocity of 5G mobile nodes (250 5G mobile nodes in total), when the number of sources is fixed on , with arrival rate vector = []. The maximal output serving rate vectors for all cases are as follows:(i)If M=1, then = (1/250)6694.4.(ii)If M=2 then = (1/250)6694.4 12500.(iii)If M=3 then = (1/250)6694.4 12500 5000.(iv)If M=4, then = (1/250)6694.4 12500 5000 2250.(v)If M=5, then = (1/250)6694.4 12500 5000 2250 12053.57.(vi)If M=6, then = (1/250)6694.4 12500 5000 44643 5000 6694.4.(vii)If M=7, then = (1/250)6694.4 12500 5000 44643 5000 6694.4 12053.57.

It is evident that as the number of RAT interfaces is higher (i.e., for M=5) the value of the average delay is lower for each velocity; even it is with rising trend versus velocity. In the case with only one RAT interface (there is no AQUAplus algorithm within the mobile nodes and when we are using the basic mobile fog and cloud computing methods [1518]) the average queue delay values are maximal (see the curve where M=1 in Figure 3). Also, with smaller number of RAT interfaces the average throughput is smaller and is supporting only those RAT interfaces belonging to the RAT with best mobility support. A key observation about the above algorithm is that it does optimal flow control decisions, simultaneously causing network queuing stability and minimal network queue delays plus optimal battery usage. Moreover, as the number of available RATs is increasing and more RAT interfaces are launched, more superior overall 5G performances are achieved. Also, we must take care of the trade-off between the battery lifetime (which is smaller when there are more active radio interfaces (M)) and the achievable throughputs included in the overall performance of the 5G nodes, as discussed in the previous section.

In the same time, we must to take into consideration the average power constraints for each radio interface, in order to avoid uncontrolled interference between the different RAT interfaces.

Furthermore, Figure 4 presents the average uplink throughput versus the average velocity of 250 5G mobile nodes, when the number of sources is fixed on , with arrival rate vector = [50 24.316 68.9655]. The maximal output serving rate vectors for M = 1, 2,…, 7 are as follows:(i)If M=1, then = (1/250)6694.4.(ii)If M=2 then = (1/250)6694.4 12500.(iii)If M=3 then = (1/250)6694.4 12500 5000.(iv)If M=4, then = (1/250)6694.4 12500 5000 2250.(v)If M=5, then = (1/250)6694.4 12500 5000 2250 12053.57.(vi)If M=6, then = (1/250)6694.4 12500 5000 44643 5000 6694.4.(vii)If M=7, then = (1/250)6694.4 12500 5000 44643 5000 6694.4 12053.57.

It is evident that as the number of RAT interfaces is higher (i.e., for M=7) the average aggregated throughput is higher for each velocity value; even it is with diminishing trend versus velocity. Also, with smaller number of interfaces the average throughput is smaller and is supporting only those RAT interfaces belonging to the RAT with best mobility support and omnipresence features.

The energy efficiency and energy consumption results in 5G network for fog, CRAN, and cloud computing environment are provided in Figures 5 and 6, respectively.

Figure 5: Energy efficiency in 5G mobile network.
Figure 6: Energy consumption in 5G mobile network.

Figure 5(a) provides the energy efficiency in 5G network as a function of the number of 5G smart user devices, for a distance range 0 to 500 meters, where the highest order of MCS, i.e., highest peak data rate, is offered. Figure 5(b) provides the energy efficiency in 5G network as a function of the distance for 100 5G mobile nodes.

Figure 6(a) provides the energy consumption in 5G network as a function of the number of 5G mobile nodes, for a distance range 0 to 500 meters, where the highest order of MCS, i.e., highest peak data rate, is offered. Figure 6(b) provides the energy consumption in 5G network as a function of the distance, for 100 5G mobile nodes. In order to compare our results in Figures 5 and 6 the energy efficiency and consumption results are depicted for both local processing in the 5G mobile node and offloading the tasks from the 5G mobile node to the cloud computing data centers, which are obtained according to the scenario given in [53]. It is expected that the offloading to the cloud offers better energy efficiency and lower energy consumption than the local processing. Our energy efficiency and energy consumption results for the case of cloud computing show much more improvement than the energy efficiency and energy consumption results given in [53]. This is due to the application on the adaptive queuing Lyapunov optimization techniques that perform optimization and maximization of the aggregated average user throughput subject to average power constraints on the node interfaces.

Moreover, it can be noticed that our solution fog computing (implemented in the 5G mobile nodes and in 5G FogRAN) provides the lowest energy consumption and the highest energy efficiency, compared to CRAN and cloud computing environment. CRAN environment in terms of energy efficiency and energy consumption is between the fog computing and cloud computing environment, and the cloud computing environment has the lowest energy efficiency and highest energy consumption per bit.

A key observation about the above 5G framework is that it does optimal flow control decisions, maximizes the average throughput, minimizes the end-to-end delay, simultaneously causing network queuing stability, minimizes the energy consumption, and maximizes the energy efficiency in CRAN and fog computing. Overall, it achieves high level of QoS provisioning. Moreover, as the number of available RATs is increasing and more RAT interfaces are up, more superior overall performances are achieved.

5. Conclusion

The paper presents analytical QoS and MCC framework for 5G nodes with multistreaming and mobile cloud features for 5G mobile broadband networks. The 5G nodes are maximizing the aggregated average throughput and stabilize the queuing, simultaneously providing minimal average queue delay. Also, the optimal consumption of energy and the module for longer battery lifetime for the proposed 5G framework with QoS assurance algorithm are presented. The 5G node with Lyapunov drift-plus-penalty technique implemented in the user 5G mobile fog computing nodes and in the 5G Cloud-RAN (fixed) node in the core network handles simultaneously multiple multimedia services via multiple wireless and mobile network interfaces. The analysis shows that the performance gain with AQUAplus module with multi-RAT interfaces in 5G heterogeneous networks with multiple available RANs is higher if we have more available heterogeneous RATs.

The presented framework is leading to the practical policy control algorithms that are provably optimal and stable network (the method guarantees stability of the power queues), with high level of QoS provisioning and mobile cloud computing support.

The cloud in 5G networks would be diffused among the client devices, often with mobility too; i.e., the cloud shall become fog. More and more virtual network functionality will be executed in a fog computing environment, and it will provide “mobiquitous” service to the users. This should enable new services paradigms such as Anything as as Service (AaaS), where devices, terminals, machines, and also smart things and robots would become innovative tools which would produce and use applications, services, and data. This is essential for the success of the future Internet of Everything (IoE), which is a clear evolution of the IoT.

Due to higher penetration of mobile broadband (at the first place smart phones) compared to fixed broadband including developed and developing countries, the proposed 5G framework with mobile cloud and fog computing can be easily generalized in a heterogeneous future 5G scenario, including any existing and future RATs, for different existing and future mobile broadband services. In that manner, efficient MCC and QoS-based usage of available mobile resources is essential for implementation of the 5G desirable performances and goals.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

References

  1. F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Communications Magazine, vol. 52, no. 2, pp. 74–80, 2014. View at Publisher · View at Google Scholar
  2. N. Bhushan, J. Li, D. Malladi et al., “Network densification: the dominant theme for wireless evolution into 5G,” IEEE Communications Magazine, vol. 52, no. 2, pp. 82–89, 2014. View at Publisher · View at Google Scholar · View at Scopus
  3. B. Bangerter, S. Talwar, R. Arefi, and K. Stewart, “Networks and devices for the 5G era,” IEEE Communications Magazine, vol. 52, no. 2, pp. 90–96, 2014. View at Publisher · View at Google Scholar · View at Scopus
  4. C.-X. Wang, F. Haider, X. Gao et al., “Cellular architecture and key technologies for 5G wireless communication networks,” IEEE Communications Magazine, vol. 52, no. 2, pp. 122–130, 2014. View at Publisher · View at Google Scholar · View at Scopus
  5. T. Janevski, “5G Mobile Phone Concept,” in Proceedings of the 2009 6th IEEE Consumer Communications and Networking Conference, CCNC 2009, January 2009. View at Scopus
  6. W. Lu, “An open baseband processing architecture for future mobile terminal design,” IEEE Wireless Communications Magazine, vol. 15, no. 2, pp. 110–119, 2008. View at Publisher · View at Google Scholar · View at Scopus
  7. A. Tudzarov and T. Janevski, “Design for 5G mobile network architecture,” International Journal of Communication Networks and Information Security, vol. 3, no. 2, pp. 112–123, 2011. View at Google Scholar · View at Scopus
  8. D. Soldani, K. Pentikousis, R. Tafazolli, and D. Franceschini, “5G networks: End-to-end architecture and infrastructure,” IEEE Communications Magazine, vol. 52, no. 11, pp. 65–100, 2014. View at Publisher · View at Google Scholar · View at Scopus
  9. M. Armbrust, A. Fox, R. Griffith et al., “A view of cloud computing,” Communications of the ACM, vol. 53, no. 4, pp. 50–58, 2010. View at Publisher · View at Google Scholar · View at Scopus
  10. S. Zhang, S. Zhang, X. Chen, and X. Huo, “Cloud Computing Research and Development Trend,” in Proceedings of the 2010 Second International Conference on Future Networks (ICFN), pp. 93–97, Sanya, Hainan, China, January 2010. View at Publisher · View at Google Scholar
  11. H. T. Dinh, C. Lee, D. Niyato, and P. Wang, “A survey of mobile cloud computing: Architecture, applications, and approaches,” Wireless Communications and Mobile Computing, vol. 13, no. 18, pp. 1587–1611, 2013. View at Publisher · View at Google Scholar · View at Scopus
  12. S. S. Qureshi, T. Ahmad, K. Rafique, and Shuja-Ul-Islam, “Mobile cloud computingas future for mobile applications - Implementation methods and challenging issues,” in Proceedings of the 2011 IEEE International Conference on Cloud Computing and Intelligence Systems, CCIS2011, pp. 467–471, Beijing, China, September 2011. View at Scopus
  13. D. Huang, “Mobile cloud computing,” in Proceedings of the IEEE COMSOC Multimedia Communications Technical Committee (MMTC) E-Letter, vol. 6, pp. 27–31, 2011.
  14. S. Kitanov and T. Janevski, “State of the art: Mobile cloud computing,” in Proceedings of the 6th International Conference on Computational Intelligence, Communication Systems and Networks, CICSyN 2014, pp. 153–158, Tetovo, Macedonia, May 2014. View at Scopus
  15. F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Fog computing and its role in the internet of things,” in Proceedings of the 1st ACM Mobile Cloud Computing Workshop, MCC 2012, pp. 13–16, Helsinki, Finland, August 2012. View at Publisher · View at Google Scholar · View at Scopus
  16. C. Mouradian, D. Naboulsi, S. Yangui, R. H. Glitho, M. J. Morrow, and P. A. Polakos, “A Comprehensive Survey on Fog Computing: State-of-the-art and Research Challenges,” IEEE Communications Surveys & Tutorials, no. 99, pp. 1–51, 2017. View at Google Scholar · View at Scopus
  17. L. M. Vaquero and L. Rodero-Merino, “Finding your way in the fog: Towards a comprehensive definition of fog computing,” Computer Communication Review, vol. 44, no. 5, pp. 27–32, 2014. View at Google Scholar · View at Scopus
  18. Y. Liu, J. E. Fieldsend, and G. Min, “A Framework of Fog Computing: Architecture, Challenges, and Optimization,” IEEE Access, vol. 5, pp. 25445–25454, 2017. View at Publisher · View at Google Scholar
  19. M. J. Neely, Stochastic Network Optimization with Application to Communication and Queueing Systems, Morgan & Claypool, USA, 2010.
  20. S. Kitanov, E. Monteiro, and T. Janevski, “5G and the Fog—survey of related technologies and research directions,” in Proceedings of the 18th Mediterranean Electrotechnical Conference (MELECON '16), pp. 1–6, IEEE, Lemesos, Cyprus, April 2016. View at Publisher · View at Google Scholar · View at Scopus
  21. L. Tassiulas and A. Ephremides, “Stability properties of constrained queueing systems and scheduling policies for maximum throughput in multihop radio networks,” Institute of Electrical and Electronics Engineers Transactions on Automatic Control, vol. 37, no. 12, pp. 1936–1948, 1992. View at Publisher · View at Google Scholar · View at MathSciNet
  22. L. Tassiulas and A. Ephremides, “Dynamic server allocation to parallel queues with randomly varying connectivity,” Institute of Electrical and Electronics Engineers Transactions on Information Theory, vol. 39, no. 2, pp. 466–478, 1993. View at Publisher · View at Google Scholar · View at MathSciNet
  23. M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, P. Whiting, and R. Vijayakumar, “Providing quality of service over a shared wireless link,” IEEE Communications Magazine, vol. 39, no. 2, pp. 150–154, 2001. View at Publisher · View at Google Scholar · View at Scopus
  24. N. Kahale and P. E. Wright, “Dynamic global packet routing in wireless networks,” in Proceedings of the 1997 16th IEEE Annual Conference on Computer Communications, INFOCOM. Part 1 (of 3), pp. 1414–1421, April 1997. View at Scopus
  25. M. J. Neely, E. Modiano, and C. E. Rohrs, “Power allocation and routing in multibeam satellites with time-varying channels,” IEEE/ACM Transactions on Networking, vol. 11, no. 1, pp. 138–152, 2003. View at Publisher · View at Google Scholar · View at Scopus
  26. P. R. Kumar and S. P. Meyn, “Stability of queueing networks and scheduling policies,” Institute of Electrical and Electronics Engineers Transactions on Automatic Control, vol. 40, no. 2, pp. 251–260, 1995. View at Publisher · View at Google Scholar · View at MathSciNet
  27. M. J. Neely, E. Modiano, and C. E. Rohrs, “Dynamic power allocation and routing for time-varying wireless networks,” IEEE Journal on Selected Areas in Communications, vol. 23, no. 1, pp. 89–103, 2005. View at Publisher · View at Google Scholar · View at Scopus
  28. A. Sridharan, S. Moeller, and B. Krishnamachari, “Making distributed rate control using lyapunov drifts a reality in wireless sensor networks,” in Proceedings of the 6th International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (Wiopt '08), pp. 452–461, Berlin, Germany, April 2008. View at Publisher · View at Google Scholar · View at Scopus
  29. M. J. Neely and R. Urgaonkar, “Cross-layer adaptive control for wireless mesh networks,” Ad Hoc Networks, vol. 5, no. 6, pp. 719–743, 2007. View at Publisher · View at Google Scholar · View at Scopus
  30. L. Georgiadis, M. J. Neely, and L. Tassiulas, “Resource allocation and cross-layer control in wireless networks,” Foundations and Trends in Networking, vol. 1, no. 1, pp. 1–149, 2006. View at Publisher · View at Google Scholar · View at Scopus
  31. M. J. Neely, “Energy optimal control for time-varying wireless networks,” Institute of Electrical and Electronics Engineers Transactions on Information Theory, vol. 52, no. 7, pp. 2915–2934, 2006. View at Publisher · View at Google Scholar · View at MathSciNet
  32. J.-W. Lee, R. R. Mazumdar, and N. B. Shroff, “Opportunistic power scheduling for dynamic multi-server wireless systems,” IEEE Transactions on Wireless Communications, vol. 5, no. 6, pp. 1506–1515, 2006. View at Publisher · View at Google Scholar · View at Scopus
  33. M. J. Neely, “Optimal Backpressure Routing for Wireless Networks with Multi-Receiver Diversity,” in Proceedings of the 2006 40th Annual Conference on Information Sciences and Systems, pp. 18–25, March 2006. View at Publisher · View at Google Scholar
  34. S. E. Mahmoodi, K. P. Subbalakshmi, and V. Sagar, “Cloud offloading for multi-radio enabled mobile devices,” in Proceedings of the IEEE International Conference on Communications, ICC 2015, pp. 5473–5478, London, UK, June 2015. View at Scopus
  35. T. Shuminoski and T. Janevski, “Lyapunov Optimization Framework for 5G Mobile Nodes with Multi-Homing,” IEEE Communications Letters, vol. 20, no. 5, pp. 1026–1029, 2016. View at Publisher · View at Google Scholar · View at Scopus
  36. Rec. ITU-T Y.2052 (02/2008): Framework of multi-homing in IPv6-based NGN.
  37. Rec. ITU-T Y.2056 (08/2011): Framework of vertical multihoming in IPv6-based Next Generation Networks.
  38. L. Gao, T. H. Luan, W. Z. B. Liu, and S. Yu, “Fog Computing and Its Applications in 5G,” in 5G Mobile Communications, pp. 571–593, Springer International Publishing Switzerland, 2017. View at Google Scholar
  39. S. Barbarossa, S. Sardellitti, and P. Di Lorenzo, “Communicating while computing: Distributed mobile cloud computing over 5G heterogeneous networks,” IEEE Signal Processing Magazine, vol. 31, no. 6, pp. 45–55, 2014. View at Publisher · View at Google Scholar · View at Scopus
  40. M. Peng, S. Yan, K. Zhang, and C. Wang, “Fog-computing-based radio access networks: Issues and challenges,” IEEE Network, vol. 30, no. 4, pp. 46–53, 2016. View at Publisher · View at Google Scholar · View at Scopus
  41. M. Peng, Y. Li, Z. Zhao, and C. Wang, “System architecture and key technologies for 5G heterogeneous cloud radio access networks,” IEEE Network, vol. 29, no. 2, pp. 6–14, 2015. View at Publisher · View at Google Scholar
  42. M. Peng, Y. Li, J. Jiang, and J. Li, “Heterogeneous cloud radio access networks: a new perspective for enhancing spectral and energy efficiencies,” IEEE Wireless Communications Magazine, vol. 21, no. 6, pp. 126–135, 2014. View at Publisher · View at Google Scholar
  43. Y. Yu, “Mobile edge computing towards 5G: Vision, recent progress, and open challenges,” China Communications, vol. 13, pp. 89–99, 2016. View at Publisher · View at Google Scholar · View at Scopus
  44. R. Buyya, J. Broberg, and A. Goscinski, “Cloud Computing: Principles and Paradigms,” Cloud Computing: Principles and Paradigms, 2011. View at Google Scholar · View at Scopus
  45. B. Pfaff, J. Pettit, T. Koponen et al., “The design and implementation of open vSwitch,” Networked Systems Design and Implementation, 2015. View at Google Scholar
  46. Nebbiolo Technologies, “Fog vs Edge Computing v1.1,” Nebbiolo Technologies Inc. Report, Milpitas, California, USA, 2015.
  47. R. Roman, J. Lopez, and M. Mambo, “Mobile edge computing, Fog et al.: A survey and analysis of security threats and challenges,” Future Generation Computer Systems, 2016. View at Publisher · View at Google Scholar · View at Scopus
  48. H. Kavalionak and A. Montresor, “P2P and Cloud: A Marriage of Convenience for Replica Management,” in Proceedings of the 6th IFIP TC 6 International Conference on Self-Organizing Systems, pp. 60–71, Delft, Netherlands, 2012.
  49. T. Shuminoski and T. Janevski, “5G mobile terminals with advanced QoS-based user-centric aggregation (AQUA) for heterogeneous wireless and mobile networks,” Wireless Networks (Springer journal), vol. 21, no. 6, pp. 1–18, 2015. View at Google Scholar
  50. T. Shuminoski and T. Janevski, “Radio network aggregation for 5G mobile terminals in heterogeneous wireless and mobile networks,” Wireless Personal Communications, vol. 78, no. 2, pp. 1211–1229, 2014. View at Publisher · View at Google Scholar · View at Scopus
  51. ITU, “Draft New Report ITU-R M.[Imt-2020.Tech Perf Req] - Minimum requirements related to technical performance for IMT-2020 radio interface(s),” ITU-R SG05 Contribution 40, 2017.
  52. J. Huang, F. Qian, A. Gerber, Z. M. Mao, S. Sen, and O. Spatscheck, “A close examination of performance and power characteristics of 4G LTE networks,” in Proceedings of the 10th International Conference on Mobile Systems, Applications, and Services (MobiSys '12), pp. 225–238, ACM, Low Wood Bay, Lake District, UK, June 2012. View at Publisher · View at Google Scholar · View at Scopus
  53. C. Ragona, F. Granelli, C. Fiandrino, D. Kliazovich, and P. Bouvry, “Energy-efficient computation offloading for wearable devices and smartphones in mobile cloud computing,” in Proceedings of the 58th IEEE Global Communications Conference, GLOBECOM 2015, December 2015. View at Scopus
  54. J. G. Koomey, S. Berard, M. Sanchez, and H. Wong, “Implications of historical trends in the electrical efficiency of computing,” IEEE Annals of the History of Computing, vol. 33, no. 3, pp. 46–54, 2011. View at Publisher · View at Google Scholar · View at MathSciNet · View at Scopus
  55. D. P. Bertsekas and R. Gallager, Data Networks, Prentice-Hall, Inc., New Jersey, NJ, USA, 1992.
  56. ITU-T Study Group 2 (Working Party 3/2) and ITU-D SG 2/16 & ITC (Draft 2001-06-20), "TELETRAFFIC ENGINEERING", Handbook, editor: Villy B. Iversen., Geneva, December 2003. Link: https://www.itu.int/dms_pub/itu-d/opb/stg/D-STG-SG02.16.1-2001-PDF-E.pdf.