enabling network architectures

FIND A SOLUTION AT Academic Writers Bay

34 0163-6804/19/$25.00 © 2019 IEEE IEEE Communications Magazine • May 2019AbstrActAs the ratification of 5G New Radio technologyis being completed, enabling network architecturesare expected to undertake a matching effort. Conventional cloud and edge computing paradigmsmay thus become insufficient in supporting theincreasingly stringent operating requirements ofintelligent IoT devices that can move unpredictably and at high speeds. Complementing these,the concept of fog emerges to deploy cooperativecloud-like functions in the immediate vicinity ofvarious moving devices, such as connected andautonomous vehicles, on the road and in the air.Envisioning the gradual evolution of these infrastructures toward an increasingly denser geographical distribution of fog functionality, we in this workput forward the vision of dense moving fog forintelligent IoT applications. To this aim, we reviewthe recent powerful enablers, outline the mainchallenges and opportunities, and corroborate theperformance benefits of collaborative dense fogoperation in a characteristic use case featuring aconnected fleet of autonomous vehicles.Advent of IntellIgent Iot ApplIcAtIonsIn December 2017, the 3rd Generation Partnership Project (3GPP) approved the first implementable part of a global fifth-generation (5G)standard in the form of the Non-standalone (NSA)5G New Radio (NR) specification. Being a part oftheir Release 15 delivered in June 2018, the NSA5G NR leverages existing LTE network as a controland signaling ‘anchor’, while the 5G NR air interface acts as a data pipe. Another part of the latestRelease 15 is the Standalone (SA) 5G NR aimingto utilize the next-generation network architecturefor both the control and user planes, which carriesthe actual data. While NSA 5G NR is more suitable for augmenting enhanced mobile broadbandservices as it builds upon the existing fourth-generation Long-Term Evolution (4G LTE) infrastructure,the SA 5G NR becomes a true enabler of ultra-reliable low latency communication [1].As a result, the SA 5G NR technology isexpected to be at the foundation of the emerging 5G networks and facilitate their applicationshaving stringent performance requirements. Theseinclude large fleets of connected and self-drivingcars, swarms of autonomous drones, collaborativeindustrial robotics, and massive augmented, virtual, or mixed reality (AR/VR/MR) services (Fig. 1).Characteristic of all of these advanced use casesis unconstrained mobility of the devices as theyexchange heavy latency-critical and reliability-sensitive data [2] between one another and with theproximate radio access infrastructure. Providingefficient support for these demanding ‘intelligent’IoT applications is very different from operatingmobile broadband or sensor-based services.Despite significant recent developments in5G radio access network (RAN) design, enablingcost-effective and scalable service provisioning forintelligent IoT devices that move unpredictablyand potentially at high speeds requires a matchingeffort from the side of the network architecture[3]. In the past, cloud infrastructures were instrumental to maintain massive IoT deployments byproviding on-demand data processing and storage capabilities. More recently, untethered devicemobility in next-generation IoT deploymentsaggressively pushed storage and computingresources toward dispersion. Along these lines,the edge computing paradigm has emerged toemploy computation resources in the immediatevicinity of the IoT equipment for preliminary information processing and its localized storage.However, relying on the edge nodes alonemay become insufficient when dissimilar IoT services compete for their limited resources [4],thereby imposing longer processing delays andreducing operational reliability. To alleviate theseserious constraints, fog computing was originallyintroduced by Cisco and nicknamed “the cloudclose to the ground.” In essence, fog architecture leverages end devices and near-device edgenetwork infrastructure to achieve improved computation, communication, control, and storagecapabilities as compared to legacy cloud-centricsystem design. To this aim, fog computing deploysvirtualized cloud-like functionality closer to theend-user, by decisively extending the conventional paradigm and thus becoming broader than thetypical notion of edge.Enabling these lightweight cloud-like functionsin close proximity to various moving devices, fogarchitecture permits the shifting of computationally-intensive services from the cloud to the edge[5] and can, therefore, serve them with direct“short-fat” wireless connections as opposed to the“long-thin” mobile cloud connections. In turn, thishelps relieve the burden of fronthaul congestionas well as mitigate the overload of cloud data centers by facilitating more flexible and customizableSergey Andreev, Vitaly Petrov, Kaibin Huang, Maria A. Lema, and Mischa DohlerFOG SERVICES AND ENABLING TECHNOLOGIESThe authors offer thevision of dense movingfog for intelligent IoTapplications. They reviewthe recent powerfulenablers, outline the mainchallenges and opportunities, and corroborate theperformance benefits ofcollaborative dense fogoperation in a characteristic use case featuring aconnected fleet of autonomous vehicles.Sergey Andreev is with Tampere University and King’s College London; Vitaly Petrov is with Tampere University;Kaibin Huang is with The University of Hong Kong; Maria A. Lema and Mischa Dohler are with King’s College London.Digital Object Identifier:10.1109/MCOM.2019.1800226Dense Moving Fog for Intelligent IoT:Key Challenges and OpportunitiesIEEE Communications Magazine • May 2019 35performance of large-scale and dynamic layouts.In intelligent IoT, fog architectures are becomingvital not only because they are naturally suitedfor the very large numbers of moving devices, butalso due to their inherent reliance on collectiveand location-based operation for enhanced system-level micro-management.Indeed, future IoT envisions increasingly massive deployments of smarter entities, such as connected and autonomous vehicles, on the road andin the air, as well as moving robots and advancedAR/VR/MR gear. These capable devices are typically equipped with improved means for sensing,processing, storage, and communication, whilethey also have more abundant on-board powersources to substantiate collaborative engagement.This includes but is not limited to collective processing and computing, distributed caching andstorage, and cooperative resource managementand control.However, we argue that cost-eff ective supportof large-scale intelligent IoT services at certain levels of reliability and on the move requires a densegeographical distribution of fog functionality,which we outline in this article and name densemoving fog. We fi rst review the key enablers formoving fog operation. Then we describe thedense moving fog functions by highlighting theadvantages as well as outlining the challengesrelated to its conceptual implementation. Later,we present our case study for dense moving fogassistance in routine operations. Finally, we conclude this work by identifying future prospects ondense moving fog.enAblement offog operAtIon for movIng IotfleXIble softwArIZed 5g networkIngIntelligent IoT systems are envisioned to operatein highly-dynamic and heterogeneous environments. This is aggravated by the increasingly complex mobility patterns, including the unconstrained3D mobility of drones and micro-scale mobility ofwearable devices. Therefore, next-generation IoTdevices need to thoroughly leverage all of theavailable proximate computing, connectivity, andcaching options [6], which goes beyond the conventional fi xed topologies of 4G/4G+ networks.In response to these demands, an emergingcapability of 5G system architecture as definedby 3GPP and the European TelecommunicationsStandards Institute (ETSI) is its modular principleof the network function design.Flexible operation is facilitated by the adoption of software-defi ned networking (SDN), whichdecouples the software-based control plane fromthe hardware-based data plane, thus allowing toadjust the behavior of network nodes in an automated and dynamic manner. Another importantinnovation is in the adoption of network functionvirtualization (NFV), which deploys crucial network functions as software components namedvirtual network functions (VNFs). Intelligent implementation of the SDN and NFV concepts in 5Gnetworks leads to decisive benefits [7], such asmore efficient and close-to-real-time resourceallocation as well as native integration of multipleradio access technologies (RATs), which results inmore fl exible data traffi c management.Softwarized 5G networking additionally introduces the concept of network slicing, which isessentially a virtual separation of heterogeneousdata fl ows from one another by off ering reservation of the network resources to a particular IoTapplication or service. In this case, an SDN-controlled 5G network can provide with a certainlevel of guarantees for the end-to-end on-timedelivery of latency-critical and reliability-sensitivedata traffi c from the network nodes, thus off eringa robust underlying platform for control, intelligence, computation, and data distribution in fl exible fog operation (Fig. 2).multI-hop multI-connectIvIty over mmwAveAs contemporary IoT devices are becoming moreintelligent, “high-end” IoT solutions are alreadyequipped with increasingly advanced sensors,including high-resolution visual and infrared cameras, radars, lidars, and so on. Hence, the aggregate volume of data per hour of their operationcan easily exceed hundreds of gigabytes [8], thuscalling for effi cient mechanisms to exchange suchinformation. Existing microwave technologies cannot provide gigabit-per-second data rates withadequate latency and reliability guarantees. Therefore, utilization of the millimeter-wave (mmWave)band is envisioned as a key enabler for intelligentIoT services.Wireless connectivity over the mmWaveband holds the potential to unlock the muchneeded additional bandwidth and achieveclose-to-noise-limited communication, asenabled by highly-directional transmissions oflarge-scale antenna arrays. Meanwhile, the narrower ‘pencil’ beams may become occludedby most moderate-size objects, including buildings, large billboards, vehicle bodies, pedestrians, and many more. The above challengesthe reliability of data exchange over mmWavefrequencies. As a result, the major standardsdeveloping organizations, including 3GPP andIEEE, are presently investing signifi cant eff ort inimproving the levels of reliability over mmWave(with e.g., 5G NR and IEEE 802.11ay technologies). Here, efficient mechanisms to maintain several simultaneous spatially-diverse 5GNR connections between the communicatingnodes, termed multi-connectivity, are demanded to mitigate link blockage and achieve reliable mmWave networking.As contemporary IoTdevices are becomingmore intelligent, “highend” IoT solutions arealready equipped withincreasingly advancedsensors, includinghigh-resolution visualand infrared cameras,radars, lidars, andso on. Hence, theaggregate volume ofdata per hour of theiroperation can easilyexceed hundreds ofgigabytes, thus callingfor effi cient mechanismsto exchange suchinformation.Figure 1. Envisaged advanced IoT applications andservices.36 IEEE Communications Magazine • May 2019However, in intelligent IoT scenarios wherethe communicating devices are highly mobileand the surrounding environment is dynamic [9], multi-connectivity mechanisms arebecoming vital, as the direct access to a static mmWave RAN deployment or the intended communicating peer may be temporarilyunavailable. Hence, “single-hop” multi-connectivity solutions can be insufficient and mightbe augmented with ‘multi-hop’ functionality,which in turn benefits from the capability of asoftwarized 5G network to support complexdynamic topologies. The establishment, maintenance, and timely update of dynamic “multihop” multi-connectivity links in high-speedmmWave communication is envisioned asone of the emerging enablers in collaborativefog-aided IoT systems (Fig. 2).leArnIng And ArtIfIcIAl IntellIgence In IotThe increased levels of dynamics and complexityin emerging network and access architectures callfor fundamentally diff erent decision-making capabilities to maintain radio connections, process anddistribute data, and collaboratively manage thesystem. Beyond the distributed learning approaches, artifi cial intelligence (AI) solutions emerge tointegrate all of the required fog functionalities[10]. This vision is aggressively pushed forward bythe advent of autonomous and self-driving vehicles, which may produce a disruption in intelligenttransportation systems.Existing solutions for autonomous cars rely onlocalized sensing and control. However, they facefundamental constraints in efficiency, reliability,and safety due to limited perception of local sensors. Recent accidents with autonomous vehiclesconfi rm the importance of coordination betweenthem. Due to non-negligible latencies, the role ofcloud-aided networking has been reduced to providing static or slowly varying support (e.g., traffi cconditions and routes). To unleash the full potential of autonomous fl eets, it is crucial to achievecloud-integrated cooperative sensing and learning. However, this poses major challenges forcomputing and statistical inference with a needto develop hierarchical learning architectures,named federated learning, which will integratedistributed (local) and centralized learning.Owing to high-rate mmWave connectivity andseamless cloud assistance, vehicles can cooperate to substantially enhance the accuracy of theirlocalization and sensing (see Fig. 2). The benefi tof a cloud-based learning solution is in its globalperspective on cooperative sensing over cars. Tothis end, cloud-based cooperative fusion and vehicle-based local sensing can be integrated undera hierarchical learning architecture. The cloudbased solution can provide individual cars withadaptive side-information that refl ects the globaland cooperative view of the system. The vehicle-based solution can leverage the cloud assistance together with local sensor measurements toenhance the accuracy of its perception. Effi cientdata transfer techniques will need to be designedto connect these two inference operation modes.dense movIng fog And ItsmAIn functIonAlItycomputAtIonRecently, fog architectures were envisioned toachieve rapid and aff ordable scaling by enablingcomputation capabilities fl exibly along the entire“cloud-to-things continuum” [11]. Accordingly,they help avoid resource contention at the edgeof the network due to on-device data processingand cooperative radio resource management, further augmented with advanced communication,control, and storage capabilities. As a result, fogcomputing emerges as a unifi ed end-to-end platform for a rapidly growing variety of dissimilar IoTservices, which leverages the on-demand scalability of cloud resources as well as coordinates theinvolvement of geographically distributed edgeand end devices. Hence, it has the potential toprovide a rich set of fog computing functions fora large number of vertical IoT industries.For intelligent IoT equipment with abundantprocessing and computation resources, dense fogseeks to realize their seamless integration withproximate edge equipment and remote cloudfunctions. This vision goes beyond treating thenetwork edges and end devices as isolated computing platforms. Seamless integration of fleetsand swarms of moving IoT entities into a densefog enclave becomes a new distributed computing paradigm that improves scalability, extensibility, and compositionality of cloud-like servicesdeployed closer to the network edge. In conFigure 2. Functionality of dense moving fog for intelligent IoT applications.IEEE Communications Magazine • May 2019 37trast to past fog-like considerations for static andlow-power IoT modules (sensors, meters, actuators, and so on), more advanced capabilities ofintelligent IoT devices (cars, drones, and robots)make the energy costs of collaborative fog operation truly pay off.Challenges in Computation for Dense Fog:Real-Time Task Decomposition and LoadBalancing: Owing to unconstrained mobility andextreme heterogeneity of the underlying computation substrate — a unique feature of thedense moving fog — efficient means are needed for on-the-fly processing, load balancing, anddecomposition of computational tasks. Here, onemay consider applying AI mechanisms, which aretrained on the previous history of dense movingfog operations and thus able to promptly deliver asufficiently appropriate task decomposition for thecurrent system state.Distributed Computing over Unreliable Connectivity: Despite the very dense geographicaldistribution, catering for proximity to end devicesin fog operation may not always be feasible dueto possible lack of powerful computing modules,but also because of limited connectivity in criticalconditions (e.g., intermittent and prone to failurewireless links). One of the possible approacheshere is to determine the minimal sufficient levelsof redundancy in task decomposition, so that thefinal outcome can be reconstructed even in caseswhere a part of the intermediate computationresults is not available on time.communIcAtIonIt is important to continuously provide adequaterates for data exchange in the considered scenarios, given that a connected car produces tens ofmegabytes of data per second, while an autonomous vehicle may generate up to a gigabyte persecond [12]. Here, the dense moving fog can support the accelerated data traffic by heavily exploiting the directional high-rate communications overthe mmWave bands. Dense moving fog can alsoprovide novel ways for intelligent IoT devices tocommunicate with each other as well as with theirproximate network infrastructure in the face ofintermittent connectivity by utilizing multi-hopmulti-connectivity mechanisms, thus combiningthe advantages of centralized and ad-hoc networktopologies into a unified solution.Challenges in Communication for Dense Fog:(Ultra-)Low-Latency Communication: Not limited to capacity demands, the emerging widely-deployed IoT applications may also require (ultra-)low latencies below a few tens of milliseconds:vehicle-to-everything communication, industrialand drone flight control, virtual reality and gamingservices, and so on. These latency-sensitive usecases may challenge the radio communication indense moving fog. Fortunately, recent advancesin embedded AI promise to ‘teach’ the fog whichjob needs to be allocated to which resource todecrease the end-to-end delay and reduce thenetwork loading. Here, delay-tolerant tasks maybe pushed into the cloud, while time-sensitiveoperation can employ nearby fog devices, thusallowing the data to reside close to where it isbeing generated.Reliable Data Exchange over OpportunisticConnectivity: To facilitate dynamic managementof computing, networking, and storage functionalities, dense moving fog architecture needs toexercise real-time control along the continuumbetween the data centers and the end devices[13]. More flexible multi-hop and multi-connectivitysolutions enabled over softwarized 5G radio networks [14] are thus envisioned to address this challenge. However, these mechanisms are currentlyat the early stages of their development and hencecall for further research. This includes determiningthe optimal degree of multi-connectivity (the number of simultaneous links) in particular operatingconditions of the dense moving fog.storAgeAs fog infrastructures bring a plethora of cloudlike services closer to the end devices, efficientstorage is crucial. Correspondingly, elastic memory capacity can be made available to various applications running on top of constrained IoT devices.Given that dimensioning of fog-aided operationis inherently flexible, the very large numbers ofdensely distributed and potentially mobile intelligent IoT entities may be integrated therein. Abundant storage space becomes accessible by thefog devices collectively with, for example, endnodes coalescing into ad-hoc capacity enclaves.As a result, multiple interconnected fog infrastructures that co-exist in space and time may serve asstorage backup for each other by pooling variousresources of the network edge, access, and enddevices in proximity.Challenges in Storage for Dense Fog:Proactive Storage Selection Procedures: Toprovide with flexible storage capabilities, densemoving fog has to enable informed and timelydecisions on how to dynamically (re-)distributethe data among heterogeneous fog nodes. Here,proactive cell selection procedures, cooperativecaching policies, and radio resource managementstrategies will be instrumental to address this challenge, achieve improved hit ratios at the edge,and thus avoid transferring massive data by reducing bandwidth consumption.Mobile Big Data Analytics: Fog-enabled storage may benefit from mobile big data analysis,especially over densely distributed and increasingly heterogeneous data collection points.However, signaling overheads should be carefully controlled in this context, since the relatively frequent exchange of small-data packets(e.g., for traffic monitoring and logging purposes)may quickly deteriorate the available link budget. Hence, data collection and analysis have tobe offered locally, by utilizing end devices and/or edge-network infrastructure for more efficientmicro-management. Augmenting data analyticswith device-aided content sharing can furtherboost responsiveness and location awareness.securItyNot limited to the above angles, fog infrastructures promise unique security-related opportunities. Massive and dense moving fog with alreadyestablished dynamic chains of trust can act asa trusted authority for external devices and systems. In particular, the moving fog may handleDense moving fogcan also provide novelways for intelligent IoTdevices to communicatewith each other as wellas with their proximatenetwork infrastructure inthe face of intermittentconnectivity by utilizingmulti-hop multi-connectivity mechanisms, thuscombining the advantages of centralizedand ad-hoc networktopologies into aunified solution.38 IEEE Communications Magazine • May 2019the responsibilities of a trusted computation platform, a certifi cation authority, and a secure storage for short-lived sensitive information, amongmany others. Fog can also facilitate localizedthreat monitoring, detection, and protection forits nodes as well as off er powerful proximity-basedauthentication mechanisms by proxying the enddevices for better identity verifi cation.Challenges in Security for Dense Fog:Secure Operations in a Heterogeneous Environment: The central concern here is that of heterogeneity: multiple potentially competing serviceproviders and consumers are utilizing distributedand dissimilar resources across a diverse collection of hardware platforms in multi-tenant environments. Therefore, advanced authorization andauthentication mechanisms need to be coined,which will eff ectively leverage this heterogeneousmedium and mediate between the fog entities[15]. Fortunately, trusted execution environmentssupported by public-key infrastructures maybecome a suitable remedy for the above issues.Still, intelligent integration of hardware-assistedand software-centric security mechanisms remainsan open research challenge for the envisioneddense moving fog.Dynamic Adaptation of Security Measures:In contrast to the state-of-the-art systems primarily operating in known conditions, the prospective dense moving fog will have to handlevolatile environments. Therefore, the employedsecurity mechanisms have to continuously adaptto the current operating conditions. Facing thischallenge, dense fog has to dynamically adjustits overall security level, which calls for designing new security protocols that will be ready torespond adequately to any security compromiseswithout creating disruptions hampering safe anduninterrupted system operation.eXAmple use cAse:fog-enAbled mAssIve fleetscenArIo descrIptIon:urbAn fleet of Autonomous cArsWe address a typical urban deployment characteristic of the City of London, UK and boundedby Charterhouse St. to the North, Queen VictoriaSt. to the South, Moorgate to the East, and A201to the West. Buildings are modeled as parallelepipeds with appropriate heights, while vehicles arerepresented as boxes of 4.8 m long, 1.8 m wide,and 1.4 m tall (Fig. 3). The density of vehicles inthe streets is estimated with the Google StreetView. Each of the autonomous cars is equippedwith mmWave radio transceivers, thus forming aconnected fleet. The stationary mmWave infrastructure is deployed with the inter-site distanceof 200 m. Each of the vehicles is further equippedwith rich sensing capabilities. The rest of the modeling parameters are summarized in Table 1.In our study, we specifically focus on thecapability of a connected fl eet to avoid collisionswith one of the major threats for autonomousvehicles in urban environments, i.e. jaywalking,when a pedestrian is crossing the road illegally.Hence, we mimic these events as a random process inside the area of interest with the intensityof 1 cross per minute. The location of a jaywalkFigure 3. Part of simulation scenario overlaid with modeling considerations.Table 1. Example modeling parameters.Deployment
Area of interest
London City, UK
Building models
From the map with real heights (Fig. 3)
mmWave radio
Carrier frequency
28 GHz
500 MHz (allocated for fog with network slicing)
Propagation model
3GPP urban microcell (UMi) – street canyon
Effect of buildings
3GPP LoS → 3GPP nLoS
Effect of vehicles
Non-blocked→ (20 dB degradation)
ISD of stationary BSs
200 m
BS sectorization
3 sectors per site with 102° downtilting
BS transmit power
35 dBm
BS height
10 m
BS computing performance
3 TFLOPS (1012 fl oating-point operations per second)
Car-cell transmit power
20 dBm (windshield transceiver location)
Body models
Parallelepiped 4.8 m  1.8 m  1.4 m
Driving speed
40 km/h (constant)
Mobility pattern
Manhattan grid on all streets
Computing performance
Radar sensing range
50 m
Radar cycle duration
66 ms
Jaywalking speed
10 km/h (constant)
Jaywalking intensity
1 per min for scenario
IEEE Communications Magazine • May 2019 39ing attempt is randomized over the area, whilethe speed of a jaywalking pedestrian is set to 10km/h. We demonstrate that while a single carhas certain chances to miss a running person, thefog-aided fleet of self-driving vehicles is capable ofdetecting this dangerous event and responding toit with appropriate collective reaction even understringent delay constraints.Collective sensing of the area by multipleautonomous vehicles and collaborative data processing of thus sensed data by both the vehiclesand the road infrastructure becomes particularlybeneficial for the considered scenario. We employour in-house system-level simulation framework toquantify the described use case and obtain thefirst-order performance results. The said framework is implemented in Python and follows thelatest 3GPP guidelines for mmWave radio propagation and physical layer modeling.Our tool is a time-driven simulator with thestep of 0.01 s. To improve the accuracy of theoutput results, all of the intermediate output hasbeen averaged over 100 independent simulationrounds, which corresponds to approximately 17hours of real-time operation (each of the roundsmodels 10 min of real time). The correspondinglatencies of data exchange across the connected fleet are carefully taken into account togetherwith other important fog-based enhancementspertaining to moving IoT devices.support of fog enhAncements In movIng IotFlexible Softwarized 5G Networking: In ourscenario, the utilization of softwarized 5G networking allows for constructing flexible temporarily-optimal network configurations withoutsignificant costs. Further, we require completeliquidization of spectrum resources at both theRAN and core network (CN) sides. The amountof spectrum available for a certain link is modeledas a continuous variable, which is infinitely anddynamically (re-)divisible subject to the currentnumber of simultaneous data streams as well astheir rate demands. Here, the virtual network ‘oracle’ observes the status of all radio links to dynamically adjust the connectivity and routing maps formaximizing the target metrics of interest.Multi-Hop Multi-Connectivity over mmWave:Our scenario assumes that all of the mmWavetransceivers support multi-connectivity of degree3 — not more than three simultaneous mmWavelinks — which is a compromise between sessioncontinuity and implementation complexity. Dueto network softwarization, connected vehicles cansupport multi-hop communication, both as initiators and as sink nodes, while assisting the othercars in data forwarding. The maximum numberof hops is not limited at the network layer butis rather determined by the application-specificrequirements (e.g., latency constraints). A nodeis assumed to continuously follow the commandsreceived from the ‘oracle’ as well as update itsactive and backup links according to theseinstructions.Learning and Artificial Intelligence in IoT:A major challenge in urban mmWave connectivity is that of random link quality drops dueto radio signal blockage by an obstacle. Whilethey can be mitigated with softwarized RAN andmulti-connectivity features, these enhancementsremain only a partial solution. Fortunately, mobility in vehicular environments is more predictablethan the directions of human travel [8]. Further,self-driving vehicles are equipped with numeroussensors (radars, lidars, cameras, and so on) toestimate their positions as well as locations andrelative velocity of the nearby objects. Smart carsmay also utilize microwave vehicular communication systems, IEEE 802.11p, dedicated short rangecommunications (DSRC), and similar, for assisted node discovery or mmWave medium accesscontrol tasks. If supplied with appropriate intelligence, an autonomous vehicle can thus predict ablockage event of its mmWave link(s) in advance.We model such proactive connectivity (re)selection, where a car may anticipate a link failure andreconnect to another more reliable alternative.understAndIng the benefIts of dense movIng fogHere, we summarize our first-order analysis ofthe benefits brought by dense moving fog to aconnected fleet. We begin with Fig. 4, which illustrates the number of misses in jaywalking detection as a function of the density of vehicles in thestreets (four lanes assumed). As can be observed,higher densities of cars impact jaywalking detection negatively in the baseline scenario, since vehicle bodies block the sensory capability to detectthe event on time: pedestrians suddenly appearfrom behind the other vehicle. Once the densemoving fog becomes operational, this negativeeffect of vehicle densification is complemented bythe positive trend of the increased sensing capability in a collaborating fleet as well as by collective data processing.To this aim, Fig. 4 confirms that even a smallfraction of vehicles in the fog (e.g., 20 percent)allows for a significant — over 50 percent —reduction in the miss rate (24.7 vs. 38.6 for 50vehicles per 100 m). For higher penetration of themoving fog functionality, this positive effect startsdominating, since the connected fleet can assessthe same spot from different angles, predict thejaywalk trajectory, and warn proximate cars aboutthe threat. Here, two important observations areFigure 4. Impact of dense moving fog on detection of jaywalking pedestrians.40 IEEE Communications Magazine • May 2019made as a result: (i) the moving fog is more beneficial at higher densities of the involved entities;and (ii) it enables assistance in real-world applications and services far beyond the vertical information and communication technologies (ICT) usecases, e.g., improved chances of on-time jaywalking detection.Continuing with Fig. 5, we generalize theabove point use case to investigate the mechanicsof dense moving fog in enabling the aforementioned gains. We first compare the aggregate benefit of the vehicular fog vs. the performance of asingle autonomous car (set as 100 percent for reference). One can observe that a random vehiclein the fog may receive a boost of up to 6 timesin its on-time data processing rate by sharing thecomputational load with other vehicles in proximity. In particular, we consider only those vehiclesthat respond to a computation offloading requestwithin 50 ms (a car moves only 0.5 m during thattime, which ensures its relevant response). Similartrends are observed for other threshold values.The connected fleet may further benefit fromoffloading its tasks onto the edge RAN infrastructure. Hence, we assume that each mmWave basestation (BS) shares its computational resources equally between all served vehicles. Higherinvolvement factors yield additional improvementssince a larger and denser fog is more likely tointerconnect across multiple underloaded BSs.Better performance naturally leads to lower processing times. More specifically, a fixed-size jobof 1012 floating-point operations (1 TFLOP) willbe computed by a standalone vehicle within 200ms, while the dense fog (assuming efficient parallelization) is able to decrease this value down to69 ms (by three times). Then, ≈15 percent more isgained if the network infrastructure also becomesinvolved in collaborative data processing.future prospects ofdense fog operAtIonIn conclusion, our numerical results confirm thebenefits of the moving fog infrastructure not onlyfor a particular application (e.g., a connected carfleet) but also in terms of collaborative data processing across many practical use cases along thelines of vehicular and airborne fog computing.The benefits will grow even further if the datasensed by a dense moving fog are complemented with those produced by the networking and/or road infrastructure. The latter requires resolving additional challenges related to matchingthe formats of heterogeneous data streams. Thisbecomes a promising research direction to extendour present work.The advantages of dense fog stretch above andbeyond rapid and affordable scaling: it will unlockemerging IoT services and disruptive business models as well as help accelerate the roll-outs of newproducts by creating a more open marketplace.As cloud and fog architectures converge within anintegrated end-to-end platform, numerous anglesof this rapidly materializing multi-faceted vision willneed to be aligned in future studies on the subject,which includes performance, reliability, safety, business, and regulatory issues.AcknowledgmentThis work was supported by the Academy of Finland (Projects PRISMA and WiFiUS). The work ofV. Petrov was supported in part by the Nokia Foundation and in part by the HPY Research Foundationfunded by Elisa.references[1] M. R. Palattella et al., “Internet of Things in the 5G Era:Enablers, Architecture, and Business Models,” IEEE JSAC, vol.34, Mar. 2016, pp. 510–27.[2] R. Inam et al., “5G Network Programmability for Mission-Critical Applications,” Ericsson Tech. Review, Jan. 2018,pp. 2–11.[3] V. Petrov et al., “Achieving End-to-End Reliability of Mission-Critical Traffic in Softwarized 5G Networks,” IEEE JSAC,vol. 36, Mar. 2018, pp. 485–501.[4] A. V. Dastjerdi and R. Buyya, “Fog Computing: Helping theInternet of Things Realize Its Potential,” Computer, vol. 49,Aug. 2016, pp. 112–16.[5] Y. Mao et al., “A Survey on Mobile Edge Computing: TheCommunication Perspective,” IEEE Commun. Surveys &Tutorials, vol. 19, Aug. 2017, pp. 2322–58.[6] S. Andreev et al., “Exploring Synergy between Communications, Caching, and Computing in 5G-Grade Deployments,”IEEE Commun. Mag., vol. 54, Aug. 2016, pp. 60–69.[7] P. Yang et al., “Catalyzing Cloud-Fog Interoperation in 5GWireless Networks: An SDN Approach,” IEEE Network, vol.31, Sept. 2017, pp. 14–20.[8] J. Choi et al., “Millimeter-Wave Vehicular Communication toSupport Massive Automotive Sensing,” IEEE Commun. Mag.,vol. 54, Dec. 2016, pp. 160–67.[9] J. Xu et al., “Cooperative Distributed Optimization for theHyper-Dense Small Cell Deployment,” IEEE Commun. Mag.,vol. 52, pp. 61–67, May 2014.[10] W. Zhang et al., “Deep Learning over Multi-Field Categorical Data: A Case Study on User Response Prediction,” Proc.European Conf. Information Retrieval, Jan. 2016, pp. 45–57.[11] M. Chiang et al., “Clarifying Fog Computing and Networking: 10 Questions and Answers,” IEEE Commun. Mag., vol.55, Apr. 2017, pp. 18–20.[12] M. Chiang and T. Zhang, “Fog and IoT: An Overview ofResearch Opportunities,” IEEE Internet of Things J., vol. 3,Dec. 2016, pp. 854–64.[13] Y. Yang, “FA2ST: Fog as a Service Technology,” Proc. IEEECOMPSAC, vol. 1, July 2017, pp. 708–708.[14] I. Stojmenovic, “Fog Computing: A Cloud to the GroundSupport for Smart Things and Machine-to-Machine Networks,” Proc. ATNAC Conf., Nov. 2014, pp. 117–22.[15] W. Fang et al., “A Source Anonymity-Based LightweightSecure AODV Protocol for Fog-Based MANET,” Sensors,vol. 17, June 2017.bIogrAphIesSergey Andreev [SM’17] ([email protected]) is an assistantprofessor of electrical engineering at Tampere University, Finland. Since 2018, he has also been a Visiting Senior ResearchFellow with the Centre for Telecommunications Research, King’sFigure 5. Impact of dense moving fog on collaborative data processing.IEEE Communications Magazine • May 2019 41College London, UK. He received his Ph.D. (2012) from TUT aswell as his Specialist (2006) and Cand.Sc. (2009) degrees fromSUAI. He has (co-)authored more than 150 published researchworks on intelligent IoT, mobile communications, and heterogeneous networking.vitAly Petrov ([email protected]) is a Ph.D. candidate at theFaculty of Information Technology and Communication Sciences at Tampere University, Finland. He received the Specialistdegree (2011) from SUAI University, St. Petersburg, Russia andthe M.Sc. degree (2014) from TUT, Tampere, Finland. He wasa visiting scholar with Georgia Institute of Technology, Atlanta,USA, in 2014. He has (co-)authored more than 30 publishedresearch works on millimeter-wave/terahertz band communications, Internet-of-Things, nanonetworks, cryptology, and networksecurity.KAibin HuAng ([email protected]) is an assistant professorin the Department of EEE at the University of Hong Kong. He isan editor for IEEE Transactions on Wireless Communications andIEEE Transactions on Green Communications and Networking.He received a Best Paper Award from IEEE GLOBECOM 2006and an IEEE Communications Society Asia Pacific OutstandingPaper Award in 2015.MAriA leMA ([email protected]) holds a Ph.D. in wireless communications and an MSc. in telecommunication engineering & management from UPC. She is currently responsiblefor technical operations in the 5G Testbed and Project Management for multiple activities related to Technology Transformation of industry verticals with 5G. She is involved in thedefinition of applications for 5G, working together with variousindustries to identify the main requirements and challenges tosuccessfully bring 5G to market.MiScHA doHler ([email protected]) is a full professorin wireless communications at King’s College London, drivingcross-disciplinary research and innovation in technology, sciences, and the arts. He is a Fellow of the IEEE, the Royal Academyof Engineering, the Royal Society of Arts (RSA), the Institution ofEngineering and Technology (IET), and a Distinguished Memberof Harvard Square Leaders Excellence. He is a serial entrepreneur, a composer and pianist with five albums on Spotify/iTunesand he is fluent in six languages.

Order from Academic Writers Bay
Best Custom Essay Writing Services