A Computation Offloading Incentive Mechanism with Delay and Cost Constraints under 5G Satellite-ground IoV architecture
Minghui LiWang, Shijie Dai, Zhibin Gao, Xiaojiang Du, Mohsen Guizani, Huaiyu Dai
JJOURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 1
A Computation Offloading Incentive Mechanismwith Delay and Cost Constraints under 5GSatellite-ground IoV Architecture
Minghui LiWang, Shijie Dai, Zhibin Gao, Xiaojiang Du, Mohsen Guizani, Huaiyu Dai
Abstract —The 5G Internet of Vehicles has become a newparadigm alongside the growing popularity and variety ofcomputation-intensive applications with high requirements forcomputational resources and analysis capabilities. Existing net-work architectures and resource management mechanisms maynot sufficiently guarantee satisfactory Quality of Experience andnetwork efficiency, mainly suffering from coverage limitationof Road Side Units, insufficient resources, and unsatisfactorycomputational capabilities of onboard equipment, frequentlychanging network topology, and ineffective resource managementschemes. To meet the demands of such applications, in thisarticle, we first propose a novel architecture by integrating thesatellite network with 5G cloud-enabled Internet of Vehiclesto efficiently support seamless coverage and global resourcemanagement. A incentive mechanism based joint optimizationproblem of opportunistic computation offloading under delayand cost constraints is established under the aforementionedframework, in which a vehicular user can either significantlyreduce the application completion time by offloading workloadsto several nearby vehicles through opportunistic vehicle-to-vehicle channels while effectively controlling the cost or protectits own profit by providing compensated computing service. Asthe optimization problem is non-convex and NP-hard, simulatedannealing based on the Markov Chain Monte Carlo as well asthe metropolis algorithm is applied to solve the optimizationproblem, which can efficaciously obtain both high-quality andcost-effective approximations of global optimal solutions. Theeffectiveness of the proposed mechanism is corroborated throughsimulation results.
Index Terms —Satellite-ground networks, Internet of Vehicles,Computation offloading, 5G networks I NTRODUCTION I N past decades, the role that satellite communicationstechnologies can play in the forthcoming 5G Internet ofThings (IoT) has been revisited. Several feasible proprietarysolutions and recent advances in satellite networks such asHigh and Ultra High Throughput Systems (UHTS), whichare built on Extremely High Frequency (EHF) bands andfree space optical links, have ushered in a new era wherethe satellite can be expected to play a fundamental rolein facilitating more demanding broadcast/broadband services,effective resource and mobility management, and achieving alarge population of on-ground mobile users such as cellphones,tablets, and smart cars [1, 2]. With the development of wireless
Minghui LiWang, Shijie Dai and Zhibin Gao (corresponding author) are withXiamen University.Xiaojiang Du is with Temple University.Mohsen Guizani is with University of Idaho.Huaiyu Dai is with North Carolina State University. technologies and applications [3–7], interconnected smart carsare considered as the next frontier in automotive revolution,whereas driverless vehicles have become a future trend withthe number of connected vehicles predicted to reach 250 mil-lion by 2020 [8]. Moreover, many technological advancementssuch as on-board cameras and embedded sensors open upnew application types with advanced, computation-intensivefeatures such as personalized automatic navigation, accidentalerts, and 3D map modeling.Nowadays, the Internet of Vehicles (IoV) formed mainlyby connected vehicles, roadside infrastructures, as well aspedestrians has faced with many challenges principally ow-ing to high vehicular mobility, low transmission rates, localresource limitations and the computational capabilities ofonboard equipment, leaving vehicles struggling to completecomputation-intensive applications locally while ensuring asatisfactory Quality of Experience (QoE). To enhance users’QoE despite increasing demands on applications, a cloud-enabled framework is introduced that allows computation-intensive applications to be executed either partially or fully ona cloud computing server such as a location-fixed cloud com-puting center and nearby Mobile Device Computing (MDC)servers (e.g., neighboring vehicles as vehicular clouds). Thesedevelopments efficiently alleviate resource constraints and easethe heavy execution burden of vehicles by migrating part ofthe workload to resource-rich surrogates. Two commonly usedplatforms in IoV are Dedicated Short-Range Communications(DSRC) based 802.11p networks and LTE cellular networks;however, both have difficulty supporting high mobility, andfrequent handovers associated with different Road Side Units(RSUs) and Base Stations (BSs) become problematic as net-works grow denser. Fortunately, with the booming technologyrevolution accompanying the advent of 5G, high data-ratetransmission capabilities along with soft handover as wellas reduced latency and high reliability can be provided tostrongly support cloud-enabled IoV. These developments areespecially advantageous for edge cloud computing and differ-ent communication modes including vehicle-to-vehicle (V2V)and vehicle-to-infrastructure (V2I), at a high data-rate level.In spite of the effectual part played by 5G cloud-enabledIoV, the demands for different vehicular applications haveexploded given the prospect that the data transmission, storage,and processing capacity of information systems are expectedto grow by 1000 times over the next 10 years. Similarly,information system performance is anticipated to be over 1000times higher for the sake of achieving green development. Fur- a r X i v : . [ c s . N I] A p r OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 2 thermore, the offloading mechanisms of computation-intensiveapplications are vulnerable to signal coverage and viewpointlimitations in addition to being highly susceptible to ineffectiveresource management. As a result, the increasingly complexnetwork environment in IoV requires more powerful central-ized management to overcome the coverage limit, integratemultiple resources, and improve global network effectivenessto the greatest possible.An efficient and future-proof complementary solution to 5Gterrestrial IoV is the 5G satellite-ground cooperative system,which contains open architecture based on Software De-fined Networking (SDN) and Network Function Virtualization(NFV) technologies [9, 10]. To fulfill diverse requirements,the role of the satellite-ground cooperative system can be fun-damental to reaching areas where terrestrial IoV services arelimited as well as managing and optimizing the global systemperformance by taking a macroscopic view. By combining asatellite network with a 5G ground IoV system and adoptingthe efficient fusion of computing, communication, and controltechnology, the satellite-ground cooperative system can pro-vide seamless signal coverage and better support for real-timeperception, dynamic control, and information services to man-age large numbers of vehicular applications. For instance, theIntelsat satellite antenna (i.e., mTenna) can be embedded intothe roof of a vehicle to acquire satellite signals even withoutRSU coverage. Toyota’s Mirai Research Vehicle equipped withmTenna can provide on-the-move services, which has beendemonstrated to achieve a data rate of 50 Mb/s [11]. In regionswith signal coverage, vehicular information such as location,velocity, accident warnings as well as application executionrequests and resource shortage warnings can be gathered byRSUs or BSs and then transmitted to the central controlleracted by satellites via satellite-ground stations, which canmacroscopically rationalize both resource deployment andmobility management. This arrangement provides appropriateresource allocation guidance for every vehicular user to ulti-mately improve the efficiency of the entire network.In this article, which targets computation-intensive applica-tions with high QoE while protecting benefits for all vehicularusers, we establish an integrated satellite-ground cooperativeIoV architecture by considering advanced 5G technologiessuch as V2V communication, cloud computing, and SDN,under which we propose an incentive mechanism based jointoptimization framework of opportunistic computation offload-ing under delay and monetary cost constraints. The satellitenetwork is regarded macroscopically as an integrated controlcenter where most functions, control, and management capaci-ties are supported through SDN-based interfaces. Computationresources are virtualized into pools with resource blocks(RBs) that can be mapped to physical resources. Vehicularusers are classified into two categories: the buyers, who havecomputation-intensive applications waiting to be executed un-der limited resources and computational capability, and thesellers who have idle computational resources that can belent to the buyers for profits. The main duty of the satellitenetwork is to identify a mechanism that can rationally allocateresources of the sellers while guaranteeing their profits so asto motivate them to provide more service in the future from a global perspective. Under the guidance of a central controller,one buyer can appropriately assign its workload to multi-sellers through several opportunistic one-hop V2V channels,significantly reducing application duration and controling costswhile improving wireless spectrum utilization. Sellers are ableto derive appreciable income while increasing the utilizationof idle resources. The separation between the control signalingand the data plane effectively realizes the flexible managementof all network traffic.Based on the constraints of delay, cost, and opportunisticcontact, we establish a novel mathematical model in whichthe simulated annealing algorithm is utilized to identify near-optimal solutions for the offloading data rate and the knock-down price per RB. The practical effectiveness and efficiencyof the proposed mechanism are corroborated through simula-tions and experiments.The rest of this article is organized as follows: after describ-ing related work in both computation offloading and integratedsatellite-ground networks, the architecture of the integrated 5Gsatellite-ground cloud-enabled IoV is introduced. The problemdefinition and system models of the proposed mechanism arepresented in the following section. Then, the joint optimizationframework of opportunistic offloading under delay and costconstraints is presented in detail, after which a solution basedon simulated annealing is designed. Finally, we analyze theperformance of the mechanism and present numerical resultsbefore concluding the article.R
ELATED WORK
The integration of satellite and 5G IoV has attracted con-siderable scholarly attention in recent years as a novel andjudicious idea for supporting diverse services in a seamless,efficient, and global manner. Several studies have focusedon different issues including framework design and coverageextension, but comparatively few are attentive in the issue ofglobal resource management. Computational resources war-rant particular attention for reasons that both computationaland analytical capabilities have come to occupy increasinglyimportant positions in different applications due to swiftly de-veloping advanced technologies such as 3D modeling, artificialintelligence (AI), and augmented reality (AR). In this section,we conduct a comprehensive investigation on both computa-tion offloading and integrated satellite-ground architecture inIoV from the perspectives of motivation and feasibility.
Motivations and Feasibility of Computation Offloading in IoV
With the rapid development and widespread popularity ofIoV, computation-resource-hungry applications have becomeincreasingly necessary among vehicular users, especially interms of historical traffic analysis, personalized navigation,dangerous driving warnings, etc. All the above mentionedapplications require significant computational resources andstrong computing abilities in order to support big data analysis.However, the resource limitations and unsatisfactory computa-tional capabilities of onboard equipment result in undesirableQuality of Service (QoS) and QoE. An innovative solution toaddress this issue is cloud-enabled IoV, established over the
OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 3 past few years to allow computation-intensive applications tobe executed either in part or in full on reliable cloud comput-ing servers. The first widely used paradigm was the remotecloud [12, 13], but it resulted in huge transmission delays,serious signal degradation, and low reliability [14] due tovariability in the network topology, wireless network capacitylimitations, and delay fluctuations in transmission on the back-haul and backbone networks. Then, Mobile Edge Computing(MEC) technology at the edge of pervasive Radio AccessNetworks (RAN) in close proximity to vehicles became pop-ular [20] but continued to be plagued by resource constraintsas well as RSU radio coverage limits. To overcome communi-cation range constraints and make full use of the opportunisticcontacts between moving vehicles, vehicular clouds open upnew schemes that allow resource exchange between vehiclesthrough one-hop V2V channels by capitalizing on economicefficiency. In a vehicular cloud scenario, a vehicle can flexiblyplay the part of either a seller providing computing serviceswhile charging a certain fee or a buyer who has a computationrequest to be executed. Literature [15] investigated a cloud-assisted vehicular network architecture in which each cloudhad its own features, and a corresponding optimal scheme wasobtained by solving a Semi-Markov Decision Process aimed atmaximizing the system’s expected average reward. To improvenetwork capacity and system computing capabilities, authorsin [16] extended the original Cloud Radio Access Network(C-RAN) to integrate local cloud services and provide a low-cost, scalable, self-organizing, and effective solution calledenhanced C-RAN with essential technologies of device-to-device (D2D) and heterogeneous networks based on a matrixgame theoretical approach. Although several studies havesolved the problem to some extent, limitations can still remainin the strict requirements regarding contact and inter-contactduration between vehicles and effective resource managementsolutions from macroscopic and longer-term perspectives forsustainable development (e.g., incentive mechanisms to protectsellers). Therefore, attempts to logically improve both resourcemanagement and the system framework while insuring therights of all users are foreseen to be urgent.
Motivations and Feasibility of Integrated Satellite-ground 5GIoV
Generally, a satellite network is composed of several satel-lites, ground stations (GSs), and network operations controlcenters (NOCCs), and usually provides services for naviga-tion, emergency rescue, communication/relaying, and globalresource and geographical information management. On thebasis of altitude, satellites can be categorized into eithergeostationary orbit (GSO), medium Earth orbit (MEO), or lowEarth orbit (LEO) satellites [11]. Owing to many advantages ofsatellite networks such as wide-area coverage, reliable accessproviding mechanism, global information coordination andbroadcasting/multicasting capability on supporting massiveusers, the motivations for integrating satellite networks withIoV can be summarized as follows: • To provide a kind of seamless coverage service to over-come the coverage and distribution limitations of RSUs in sparsely populated rural areas (e.g., mountainous areasand the desert). • Different applications in IoV cannot be served efficientlyby a single technology; hence, the convergence of variousnetworks is likely to become a major trend in the future. • Potential network congestion may occur even in suburbanareas with high spatial-temporal dynamics in traffic loadsdue to vehicular mobility. • A lack of macroscopic management to improve overallnetwork efficiency and resource utilization without satel-lite networks.Several proprietary solutions and open standards have beendeveloped to enable data broadcasting via satellite to mobileusers over the years. A plan called "Free Space OpticalExperimental Network Experiment Program (FOENEX)" wasimplemented by the Defense Advanced Research ProjectsAgency (DARPA), which brought the data transmission rateof Mobile Backbone Communication Networks (MBCNs) to10Gb/s and even 100Gb/s when using Wavelength DivisionMultiplex (WDM) technology between 2010 and 2012. Forsatellite networks, Free Space Optical Communication (FSO)is the primary choice of MBCNs but fails to cut throughclouds. Attempting to design and execute an airborne com-munication link with equivalent capacity and optical com-munication distance, DARPA prepared a plan entitled "100GRF Backbone Program" in early 2013 [17]. Moreover, someexisting researchful studies have focused on different issuesbased on the satellite-ground vehicular network framework.Authors in [10] examined a use case for the realizationof end-to-end traffic engineering in a combined terrestrial-satellite network used for mobile backhauling. Literature [18]proposed an analytical assessment of the cooperation limitsin the presence of both a satellite and a terrestrial repeater(gap filler) and derived exact expressions and closed-formlower bounds on coverage in a setup of practical interestby using the max-flow min-cut theorem. Furthermore, theystudied a practical implementation of the Random LinearNetwork Coding (RLNC) cooperative approach for the DigitalVideo Broadcasting-Satellite services to Handheld (DVB-SH)standard. All previous research have proved that advancedfeatures from different segments can be exploited to supportmultifarious vehicular applications and scenarios in an efficientmanner through interworking, which spurred opportunities tointegrate satellite networks with IoV successfully.I
NTEGRATED ARCHITECTURE FOR THE SATELLITENETWORK AND G CLOUD - ENABLED I O VSDN is regarded as an emerging network architecture in5G that separates the control plane from the data plane,introduces logically centralized control with a global andmacroscopic view of the network, and facilitates networkprogrammability/reconfiguration through open interfaces [11].In this section, we provide an introduction to the proposedsatellite-ground 5G IoV. As shown in Fig. 1, the 5G satellite-ground IoV mainly incorporates a satellite network segmentand the terrestrial IoV segment. SDN controllers exist eitheron powerful servers or cloud computing centers of both
OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 4
Fig. 1. Integrated architecture of the satellite network and 5G cloud-enabled IoV. the satellite network and terrestrial networks and can havedifferent functions given the diverse characteristics of eachsegment. We develop hierarchical SDN controllers to coor-dinate various features and operations of different segmentswhile generating gradational information from microscopic tomacroscopic perspectives. Cases such as vehicular velocityand direction information can be collected by RSUs. Localtraffic information (e.g., traffic density and accident warning)can be transmitted from RSUs to the satellite-ground stations,and macroscopic perceptions will be gathered in the satellitenetwork to facilitate decision making at different tiers ofthe SDN controllers of each segment. In remote regions likemountainous areas and deserts without terrestrial signal cover-age, vehicles can get contact with the satellite network eitherdirectly or through the satellite-ground station by installing aspecific transceiver antenna so as to enjoy seamless coverageservice.It is noteworthy that vehicles will not encounter interferencefrom the various levels of service based on the premise thatnetwork slicing technology is performed in each segment topartition resources of the entire network into various slices fordifferent services, wherein operations are executed in isolationso as not to interfere with each other.P
ROBLEM DEFINITION AND SYSTEM MODELS
In this section, we provide a detailed overview of theproblem definition and system models. For one snapshot attime t , the whole terrestrial IoV can be divided into severalcooperative groups, each containing one buyer and several sell-ers who can interact with the buyer through an opportunisticone-hop V2V channel. A buyer can offload partial applicationdata to sellers and obtain resultant feedback after the sub-applications are completed. Notably, one seller can provide service for different buyers simultaneously, and these servicesare independent without affecting each other. The main dutiesof a central controller for each group can be defined as follows.Firstly, a scheduling of a computation-intensive applicationwhich can be explained as the appropriate data size allocatedto one seller in order to reduce the application durationunder cost constraints; then, the unit knockdown price of thecomputational RB for each seller is decided while controllingthe buyer’s monetary cost and guaranteeing the seller’s profitsas well as willingness to provide idle resources in the future.The schematic diagram of computation offloading in individualgroups and interactions in local areas is shown in Fig. 2. System Models
The proposed terrestrial scenario consists of B buyers b i ∈ { b , b , . . . , b B } and S sellers s j ∈ { s , s , . . . , s S } in which buyer b i and seller s j can be denoted as tetrad b i = { D i , C Bi , T max i , P Bi } and s j = { C Sj , CC Sj , c j , p Satis f y j } ,respectively. D i , T max i and P Bi are defined as the data size,the tolerant completion time of the computation-intensiveapplication and the cost limitation of vehicle b i respectively; C Bi and C Sj are the computational capabilities (RB per second)of the buyer and seller; CC Sj represents the idle resources atthat time, and c j , p Satis f y j are the unit cost and satisfied unitprice of seller s j . It is worth noting that if the price paid by abuyer is less than c j , then s j will no longer provide service forthe said buyer due to its loss. As the offloading mechanism andpricing strategy of each group must follow the same guidanceof the central controller, our discussion below is focused ononly one group and the members in it. Assuming there are m sellers in b i ’s group, the application of buyer b i can be denoted OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 5
Fig. 2. Schematic diagram of local interaction and computation offloading. as X i = (cid:205) mk = x k + x B , where x k and x B represent the workloadassigned to seller s k and local execution respectively. Vehicular Mobility Model: it is assumed that B + S vehiclesare moving in a network Ω N = [ , (cid:112) ( B + S )/ µ ] , where µ is the density of vehicles per kilometer on both the east-west and south-north bound roads. Vehicles move accordingto mobility process Q . Assume that L i ( t ) and L j ( t ) denote thelocations of vehicles i and j , and the mobility process of avehicle is stationary and ergodic such that location L i ( t ) hasa uniform stationary distribution in the network scenario [19].Moreover, the mobility processes of vehicles are i.i.d (inde-pendent and identically distributed). We call one contact event Υ C between two vehicles, which occurs during t ∈ [ t , t ) ifthe following conditions are satisfied: (cid:107) L i ( t − ) − L j ( t − )(cid:107) > R , (cid:107) L i ( t ) − L j ( t ) (cid:107) ≤ R and (cid:107) L i ( t ) − L j ( t )(cid:107) > R , where R represents the transmission radius. Assuming that vehiclesmaintain a uniform linear motion during a small offloadingperiod, the contact duration ∆ t C can be calculated easily. Communication Model:
A pair of vehicles can commu-nicate with each other at time t when their locations satisfy (cid:107) L i ( t ) − L j ( t ) (cid:107)≤ R . All vehicles will report information suchas velocity, direction, as well as other parameters gatheredby RSUs to the central controller periodically, which is fur-ther routed through satellite-ground stations to the satellitenetwork. Due to the mobility of vehicles, different channelconditions lead to direct differences in the data transmission rates of m links, represented as r k ∈ { r , . . . , r m } , where r k is the data transmission rate between buyer b i and seller s k , and can be regarded as a fixed average value relatedto several factors including channel condition, packet lossretransmission, transmission power and the outage probability.The delivery duration of the corresponding sub-applicationcontent x k can be calculated as t Tr ans k = x k / r k . Computation Model:
Considering a situation where anapplication is computation-intensive, and the V2V channelsare unavailable, the local processing duration is typicallysmaller than or equal to T max i , which can be denoted as K X i / C Bi ≤ T max i , with K , a constant that serves as a mapingbetween the data size and computational RBs. When the V2Vchannels are available, the sub-application execution durationat seller s k can be described as t execk = K x k / C Sk while t B = K x B / C B is the local execution time.Overall, the sub-application duration for the seller s k is t k = t Transk + t execk ; correspondingly, the total application com-pletion time can be obtained as T i = max { t , t , . . . , t m , t B } .T HE JOINT OPTIMIZATION FRAMEWORK OF INCENTIVEOPPORTUNISTIC OFFLOADING UNDER DELAY AND COSTCONSTRAINTS
In this section, a joint optimization problem is modeledunder delay and cost constraints. Due to different user
OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 6 preferences, we introduce weight factors denoted as ω , ω , and ω to emphasize either the application completiontime, monetary cost, or the incentive mechanism. Fornotational simplicity, we create four diagonal matrices: H = dia g ( / r + K / C S , . . . , / r m + K / C Sm , K / C B ) ; P Satisf y = dia g ( p Satis f y , . . . , p Satis f y m , ) ; X = dia g ( x , . . . , x m , x B ) ; and P = dia g ( p , . . . , p m , ) is definedas the knockdown price per RB for each seller. Constraintsare defined as follows: a) available idle resource constraint K x k − CC Sk ≤ ; b) contact duration constraint t k − ∆ t C ≤ ;c) monetary cost constraint (cid:205) mk = p k x k − P Bi ≤ ; d) datasize constraint x B + (cid:205) mk = x k = D i ; e) non-negative constraint ∀ k ∈ { , , . . . , m } , x k ≥ and p k ≥ ; and f) incentiveconstraint CC Sk = p k ≤ c k . Thus, the objective functionfor each cooperative group that includes one buyer and m sellers can be specified by using infinite-norm, 1-norm, andF-norm as in the following optimization problem: Φ = arg min X , P ω (cid:107) H X (cid:107) ∞ + ω (cid:107) PX (cid:107) + ω (cid:13)(cid:13) P − P Satisf y (cid:13)(cid:13) F s . t . a ) , b ) , c ) , d ) , e ) , f ) (1)The first part of the objective function has the physical signif-icance of reducing application completion time. The monetarycost constraint is shown as the second part, and the incentivemechanism is reflected in the third part which means the closerthe knockdown price is to the seller’s satisfied price, the morewilling a seller will be to provide service in the future. Solution:
Owing to the fact that the objective function(1) with constraints is non-convex and NP-hard, a simulatedannealing algorithm is utilized to obtain feasible solutions.Simulated annealing is a heuristic algorithm based on theMarkov Chain Monte Carlo (MCMC) as well as the metropoliscriterion, which can lead to high-quality and cost-effectiveapproximations of global optimal solutions.P
ERFORMANCE EVALUATION
100 300 500 700 900 1100
Number of Sellers A v e r a g e A pp li ca ti on D u r a ti on ( s ) The Proposed MechanismLocal ComputingAverage Offloading
Fig. 3. Average application completion duration with the number of sellers.
Simulation results are presented in this section containing100 buyers and multiple sellers (from 100 to 1100) in a widecoverage area under the satellite-ground 5G IoV framework.The communication range of a vehicle is set to be R = 250m; the application data size D ∈ [5Mb, 6Mb] and datatransmission rate r ∈ [3Mb/s, 6Mb/s]. The weight parametersare set to be ω = ω = ω = / . Figure 3 showsthe average application completion duration with the numberof sellers by comparing the proposed mechanism with twoother schemes: the Local Computing algorithm where oneapplication is totally executed locally without offloading, andthe Average Offloading algorithm, where one application isdistributed evenly to every seller in the cooperative groupunder contact duration constraints. As can be seen, the LocalComputing algorithm has the highest application completionduration, whereas that of Average Offloading decreases asthe number of sellers increases but still remains higher thanthat of the proposed mechanism without considering differentcomputing capacities among sellers. The proposed mechanismdemonstrates the best performance by jointly accounting forcomputational capacities, contact duration, and different chan-nel conditions among the sellers.
100 300 500 700 900 1100
Number of Sellers A v e r a g e K no c kdo w n U n it P r i ce -7 The Proposed MechanismAverage Satisfied PriceAverage Offloading
Fig. 4. Average knockdown unit price with the number of sellers.
Figure 4 shows the average knockdown unit price with thenumber of sellers. Due to the competitions among the sellers,the knockdown unit price trends downward in the proposedmechanism but still is much higher than that of AverageOffloading, where the knockdown price is only slightly higherthan the seller’s cost without an incentive mechanism. In otherwords, central controllers in the Average Offloading scenarioare not concerned with how much sellers earn as long as theydo not experience a loss. In contrast, sellers in the proposedframework receive better payments and are more willing toprovide idle resources for sustainable and green developmentof the satellite-ground IoV.An example of convergence in the simulated annealingalgorithm is shown in Fig. 5, where 100 buyers and 600sellers are tested for the rate of convergence, which is regardedas a critical factor in a rapidly changing environment likeIoV, especially when vehicles communicate with each otherthrough opportunistic V2V channels. As illustrated in Fig. 5,the algorithm begins to converge when the number of iterationsexceeds 100, which corresponds to a negligible period withthe current computing technology; therefore, the offloading
OURNAL OF L A TEX CLASS FILES, VOL. XX, NO. XX, MARCH 2018 7
Number of Iterations A v e r a g e V a l u e o f O b j ec ti v e F un c ti on Fig. 5. Convergence of the simulated annealing algorithm. decision-making at the central controller can handle the highmobility of vehicles effectively.C
ONCLUSION
In this article, we devise an integrated architecture of satel-lite networks and 5G Internet of Vehicles that effectively pro-vides both seamless coverage and resource management froma macroscopic point of view. Then, an incentive mechanismbased joint optimization problem for computation offloadingamong vehicles is modeled under delay and cost constraints,where a service buyer can significantly reduce the applicationcompletion duration and control monetary costs while servicesellers are motivated to promote sustainable green network de-velopment. Using a simulated annealing algorithm, simulationresults are presented to substantiate the practical effectivenessand efficiency of the proposed mechanism.R
EFERENCES [1] C. Sacchi et al., “Toward the “space 2.0” Era [GuestEditorial],”
IEEE Communications Magazine , vol. 53,no.3, 2015, pp. 16-17.[2] M. Corici et al., “Assessing satellite-terrestrial integra-tion opportunities in the 5G environment”, Sept. 2016;https://artes.esa.int/sites/default/files/Whitepaper.[3] H. Zhang, S. Chen, X. Li, H. Ji, and X. Du, “InterferenceManagement for Heterogeneous Network with SpectralEfficiency Improvement,” IEEE Wireless Communica-tions Magazine, Issue 2, Vol. 22, pp. 101-107, April2015.[4] X. Du, M. Zhang, K. Nygard, S. Guizani, and H. H.Chen, “Self-Healing Sensor Networks with DistributedDecision Making,” International Journal of Sensor Net-works, Vol. 2, Nos. 5/6, pp. 289-298, 2007.[5] L. Wu, and X. Du, “MobiFish: A Lightweight Anti-Phishing Scheme for Mobile Phones,” in Proc. of the23rd International Conference on Computer Communica-tions and Networks (ICCCN), Shanghai, China, August2014.[6] X. Du, Y. Xiao, M. Guizani, and H. H. Chen, “AnEffective Key Management Scheme for Heterogeneous Sensor Networks,” Ad Hoc Networks, Elsevier, Vol. 5,Issue 1, pp 24-34, Jan. 2007.[7] Y. Xiao, X. Du, J. Zhang, and S. Guizani, “InternetProtocol Television (IPTV): the Killer Application forthe Next Generation Internet,” IEEE CommunicationsMagazine, Vol. 45, No. 11, pp. 126-134, Nov. 2007.[8] N. Lu et al., “Connected Vehicles: Solutions and Chal-lenges,”
IEEE Internet of Things Journal , vol. 1, no. 4,2014, pp. 289-299.[9] M. Qian et al., “A super base station based centralizednetwork architecture for 5G mobile communication,”
Digital Communications and Networks , vol. 1, no. 2,2015, pp. 152-159.[10] R. Ferrus et al., “Towards SDN/NFV-enabled satelliteground segment systems: End-to-End Traffic Engineeringuse case,”
IEEE ICC Workshops , May. 2017, pp. 888-893.[11] N. Zhang et al., “Software Defined Space-Air-Ground In-tegrated Vehicular Networks: Challenges and Solutions,”
IEEE Communications Magazine , vol. 55, no.7, 2017,pp. 101-109.[12] T. Taleb and A. Ksentini, “Follow me cloud: interworkingfederated clouds and distributed mobile networks,”
IEEENetwork , vol. 27, no. 5, pp. 12-19, 2013.[13] H. Zhang, Q. Zhang, and X. Du, “Toward Vehicle-Assisted Cloud Computing for Smartphones,” IEEETransactions on Vehicular Technology, Issue 12, Vol.64,pp. 5610-5618, Dec. 2015.[14] E. Ahmed, and M. H. Rehmani, “Mobile edge com-puting: opportunities, solutions, and challenges,”
FutureGeneration Computer Systems , vol. 70, 2017, pp. 59-63.[15] H. Meng et al., “A utility-based resource allocationscheme in cloud-assisted vehicular network architecture,”
IEEE ICC Workshop , Jun.2015, pp. 1833-1838.[16] R. Yu et al., “Optimal resource sharing in 5g-enabledvehicular networks: A matrix game approach,”
IEEETransactions on Vehicular Technology , vol. 65, no. 10,2016, pp. 7844-7856.[17] DAPAR Strategic Technology Office Broad Agency An-nouncement. DARPA-BAA-13-15, “100Gb/s RF back-bone (100G),” Jan. 2013.[18] C. Giuseppe, N. Alagha, and C. Ibars, ”Cooperativecoverage extension in vehicular land mobile satellitenetworks,”
IEEE Transactions on Vehicular Technology ,vol. 65, no. 8, 2016, pp. 5995-6009.[19] Y. Li, L. Sun, and W. Wang, “Exploring device-to-devicecommunication for mobile cloud computing,”
IEEE ICC ,Jun. 2014, pp. 2239-2244.[20] Li. Tang and S. He, “Multi-User Computation Offloadingin Mobile Edge Computing: A Behavioral Perspective,”