IEEE Communications Magazine • March 2018 0163-6804/18/$25.00 © 2018 IEEE 157AbstrActOur interactions with the world are increasingly dependent on context-aware services, andthe future of smart cities is coupled with how efficiently and reliably we can deliver these servicesto end users. In this article we present the premiseof personalized IoT systems, by leveraging noveladvancements in user-centric technologies underthe fog computing architecture. This means leveraging the connectivity and processing potentialof the fog to bring IoT control and analytics closer to the user, and improve the coupling of services with local IoT components in user-centriccontexts. The potential gain in access latency andcontext-sensitive service matching will enable amultitude of smart city services. On one hand,data management (collection, pruning, denaturing , and encryption) can take place closer tothe edge, thereby leveraging network load andservice times. On the other hand, service matching in smart city applications will witness higherresponsiveness and resource visibility in areaswith intermittent connectivity or high mobility. Wefirst present the challenges in migrating cloud-IoTarchitectures to the network edge, and detail thehindrances in transitioning the control and management of IoT systems to the user end. As a remedy, we survey recent advancements in the IoT,ubiquitous computing, and user-centric services,which enable us to advance personalized IoTarchitectures. We finally present a framework forIoT in the fog to synergize these advancements,and present a proof-of-concept use case to highlight its utility and impact. We conclude this articlewith prime directions for future work to realize apersonalized IoT architecture, and highlight thepotential gain in prioritizing five high-yield potential research issues.UnderstAnding Fog iotThe case for cloud computing (CC) infrastructuresis widely established. Simply put, the ability to offload computationally intensive tasks on remotedata centers, where you are elastically chargedfor what you use, is growing as a preferable alternative in a large spectrum of applications. Themost prominent everyday use of such services iswitnessed in speech recognition software (e.g.,Samsung S-voice, Google Talk and Apple’s Siri),near-real-time pattern recognition for object identification (e.g., YOLO — You Only Look Once:Unified, Real-Time Object Detection),1 and translation (e.g., Google Translate).However, as we transition into a mobile-drivenworld, today’s users are expecting crisp interaction with their surrounding technologies. The usercan no longer afford to wait for the typically varying response time of a cloud-based computationor service discovery, especially under intermittentconnectivity, high-mobility scenarios, or with stringent demands on tolerated delay. The rising tideof improving quality of experience (QoE) as wellas enabling contextualized user-centric applications is driving forward the migration toward fogcomputing.This user-driven shift in computation and storage to near-edge fog architectures is enablingmany applications that require less interactionwith remote services (or data centers). Fog computing builds on research in edge analytics and leverages recent developments in cloudlets. The interplay between the cloud, cloudlet/edge, fog, mist, and end users is depicted in Fig.1.More importantly, as the Internet of Things(IoT) bridges the physical and virtual worlds ofinteractions, we need solutions that contextualizeour interactions with immediate resources. That is,we now have the technology to probe surrounding resources (sensing, processing, communication, etc.) in real time , but lack the frameworkto deliver a responsive and user-centric IoT experience on the go.In this article we survey the challenges inrealizing IoT systems in the fog, and present anoverview of recent advancements that could besynergized to deliver a personalized IoT ecosystem. We target a framework that will encompassheterogeneous IoT resources in a given region,and the variation in processing and communication offloading that could be leveraged by the fogIoT architecture.Fog compUtingIn its simplest terms, the fog layer resides betweena local resource and the cloud service. In theory,a multiplicity of fog nodes will be geographically distributed to service local resources in theirrespective regions. This multi-tiered architectureis depicted in Fig. 1. Each of these nodes will beable to leverage computing tasks and take part inservice orchestration with underlying resources inthe region .IoT in the Fog: A Roadmap forData-Centric IoT DevelopmentSharief M. A. Oteafy and Hossam S. Hassaneinemerging trends, issUes, And cHALLenges inbig dAtA And its impLementAtion toWArd FUtUre smArt citiesThe authors first presentthe challenges in migrating cloud-IoT architectures to the network edge,and detail the hindrancesin transitioning the controland management of IoTsystems to the user-end.As a remedy, they surveyrecent advancements inthe IoT, ubiquitous computing, and user-centricservices, which enablethem to advance personalized IoT architectures.Sharief M. A. Oteafy is with DePaul University; Hossam S. Hassanein is with Queen’s University.Digital Object Identifier:10.1109/MCOM.2018.17002991 https://pjreddie.com/publications/yolo158 IEEE Communications Magazine • March 2018The notion of fog computing is preceded byearlier work on cloudlet access, whereby an intermediate connection/access point is deployed tobridge the computational offloading process frommobile devices to cloud services. Earlier work byM. Satyanarayanan, overviewed in , presents adetailed account on the motivation behind cloudlet design, and highlights its soft-state that is inherently more flexible, modular, and distributed incontrast to cloud platforms.VAriAnts oF cLoUd-bAsed sensingThe notion of leveraging cloud sensing has beeninvestigated heavily in the past decade [2, 4–6],mainly to enable public sensing schemes. Cloudsensing is mainly concerned with distributed datacollection for offline querying , and newermodels attempt to leverage cloud services toenable a real-time association between servicerequesters (i.e., application requiring specificdata in real-time) and currently available resources that are connected to the cloud sensingarchitecture . However, most cloud sensingarchitectures tolerate a significant delay in processing, and are intrinsically designed for offlineoperation, both of which hinder its applicationin newer systems where devices are mobile,intermittently reachable, and more invested inreal-time information services. In recent developments, the case for mobile edge computing(MEC) and novel technologies that bring moreprocessing to the edge of the network areenabling newer forms of cloud sensing in thenearer fog. This means building systems thatmanage, disseminate, and respond to queriesusing in-field technologies rather than foraging resources from distant cloud services. Thisnotion of fog sensing is at the heart of what thisarticle covers.Fog-enAbLed serVicesThe projected proliferation of machine-to-machine (M2M) services, along with an evidenttransition into mobile-driven services and applications, are rendering many cloud-dependentservices inefficient and restricting. In addition,the inherent heterogeneity of all devices that arejoining the mobile resource pool is increasingthe complexity and delay in centralized (cloudbased) management and orchestration of information services over these devices. The adventof big sensed data, in terms of data producedand potential services enabled by the aggregation of all these resources, is further establishedin , and we are in dire need of an architecture that can access, monitor, manage, andrecruit these services in real time and withintheir respective contexts . We next survey themajor challenges facing our development of fogsensing, and then present a roadmap to its development in light of novel technologies.migrAting iot serVices to tHeedge: core cHALLengesThe sheer amount of IoT/M2M traffic projectedin the next five years is mandating novel designconsiderations in both data processing and communication management. Typical traffic generation in IoT devices in 2016 averaged 1614 MB/month,1 mostly from wearable devices. However,with a projection2 of a rise in number of M2Mconnections from 1.1 billion (in 2017) to 3.3 billion (in 2021), there are many scalability challenges to address.Earlier research on cyber foraging by M.Satyanarayanan argued that regardless of hardware advances at the user end, static resources on the Internet (or distributed systems in thegeneral sense) will remain far superior . Thus,cyber foraging was based on a growing disparity between resource capability at the edge incontrast to that in cyberspace. The argument forInternet-based resource foraging grew significantly with the realization of cloud architectures, andmost IoT developments attempted to capitalizeon resource abundance and elastic pricing ofcloud services. However, in revisiting the recentexplosion in data usage and stringency of timelimits, Satyanarayanan and others have arguedfor reducing the dependence on “remote” cyberforaging in the cloud.Migration of data and communication control to the edge of the network has been at theheart of context-aware services for over a decade. Many attempts at personalizing IoT interactions have brought control to the edge, mostly atthe user device or gateway levels. The benefits inresponse latency, hub-free M2M interactions, andpower conservation have been major drivers ofnear-edge operation. This approach opened thedoor for IoT systems that probe nearby resourcesfor real-time service matching  and enablingcontext-aware IoT .While these drivers are indeed pressing,there are many challenges as we migrateFigure 1. Overview of tiers in a cloud-IoT architecture, highlighting the span offog networks. The major challenges in realizing IoT operation are depicted in bars below each of the architectural components, highlighting thevariation from simple/better (in light blue) to complex/worse (in red) foreach of the operational mandates/design challenges. In some scenarios, asin energy impact on IoT resources, the variation is non-increasing in eitherdirections, but exhibits better results under a subset of the tiers (cloudlet/edge tier in this case) .Service latencyUser MistFog networkCloudlet CloudProcessingcapacityComplexity ofdata zoningEnergy impactComplexity ofdenaturingIoT managementcontrol overheadStatic resourcevisibilityTransient resourcevisibility2 Cisco Visual NetworkingIndex (VNI): Global MobileData Traffic Forecast Update,2016–2021, March 2017update.IEEE Communications Magazine • March 2018 159control and data management to the edgeof the network, far beyond the naive view ofresource limitation. In the remainder of thissection, we overview the major challenges inIoT migration toward the edge, especially incontrast to the dominant approach of centralized and proprietary IoT proliferation that isgoverning most solutions . The major challenges witnessed in cloud IoT architectures aredepicted in Fig. 2, wherein we annotate thespan of fog networks, as well as the challenges under each tier in the hierarchical view ofcloud components.spAtiAL correLAtionDetermining the location of collected data isbecoming an increasing challenge in IoT systems.While advances in GPS as well as indoor localization systems have enabled sub-meter localization,many IoT nodes do not encompass the resourcesto self-localize. More importantly, many IoT systems are building on archaic localization schemesfrom wireless sensor networks (WSNs), whichwere largely static in deployment, or had specificmobility patterns that may not fit most IoT scenarios. As IoT applications are mandating bettercoupling of data generation and coverage accuFigure 2. Contrasting the design factors in delivering IoT operation over Cloud variants. More importantly,the multi-tiered approach to Cloud interaction is highlighted over the different levels of user involvement with personal devices, to neighboring IoT resources (in the Mist), and their combined interactionswith Cloudlets and the Cloud. Cloudlets and Edge nodes are merged under one category, as they areboth handled similarly in Cloud literature [1, 2].ArchitecturalcomponentsPersonal smart devices(phones, tablets,vehicles, wearables, etc.)Home automationdevices (e.g., NESTthermostat), neighboringsmart devices, mostlystandalone IoT devicesHigh-end access pointswith processing,connectivity andstorage resourcesUltra-large-scaleprocessing and storageresources, backboneaccess to Tier-1 InternetData pruningtechniquesLocal homogeneousfusion (temporalsampling, averaging,simple flat fusion)Neighborhood-baseddata alignment andfusionGeo-sensitiveHierarchical datafusion and cleaningLarge-scale dataclassification, mining,and catalogingMobility Highly mobileTypically mobile(often as a group,as in VANET)Typically static, withrecent work onvehicle-mountedcloudletsStrictly staticGoverningarchitectures Personal applications M2M H2M Dedicated IoT hubs VMware Large-scale data centers (e.g., EC2)NetworkingtechnologiesTypically “multi-home”to short- and longrange networksM2M and short-rangecommunicationLTE/5G backboneEthernet (wired) Mostly > 10 GbEGeo-distribution Local to user, directlyaccessibleNon-uniform, oftenagnostic to cloudletlocationsClose to backboneconnections, higherurban densityDemand-centricTime responsiveness(processing)Immediate, zeronetwork delayTypically 1-hop delay,limited queuing delayShort contentiondepending on users,mobility, backendbuffering, and otherfactorsDeployment plan Ad hocMostly ad hoc andmovable.Static exceptions exist(e.g., home automation)Mostly coupled withAPs and high-BWbackbone connectionsStrategically placed byCDN and cloud providers(both surrogate serversand data centers)Current standardsIEEE 802.11/15 familyLTE/4GBLE/BluetoothNFCEC-GSM-IoTDASH7NFCBLZigBeeDSRC (vehicular)RPMAIEEE 802.11acMulteFireLTE / 5GDense wavelengthdivision multiplexing(DWDM)EN 50600-2-4TIA-942-AArchitecturaladvantageZero delayImmune to mobilitychallengesLeast challenge withsecurity andauthenticationHarnessing resourcesfrom immediateneighborhood (sensing,communication,processing, etc)Immune to cloudservice disruptionsInherently geocontextualizedReduces ingress trafficMasks cloudunavailability (due toconnectivity, DDoSattacks, etc.)AnonymizationtechniquesMost economic use ofresourcesEnables large-scale viewof resourcesMost suited for largescale data analyticsIoT layerFactor User Mist Cloudlet/edge CloudLongest delay, dependingon congestion towardremote cloud service andaggregated queuing delayWhile advances inGPS as well as indoorlocalization systemshave enabled sub-meterlocalization, many IoTnodes do not encompass the resourcesto self-localize. Moreimportantly, many IoTsystems are buildingon archaic localizationschemes from WirelessSensor Networks, whichwere largely static indeployment, or hadspecific mobility patterns that may not fitmost IoT scenarios.160 IEEE Communications Magazine • March 2018racy, many schemes are challenged by improvingthe latter. This becomes more of a problem whendata pruning and averaging techniques attempt toalign and fuse sensed reports from IoT systems,which is further exacerbated by the heterogeneityof IoT devices.There is promise in establishing localizationin cloudlet zones, especially as they are intrinsically confined to pre-determined regions, anda “local-global” view of available localizationschemes could be fused to improve the spatialcorrelation of data. Moreover, as singular IoT systems may fail to individually localize their data, orestablish coverage in a given region, leveragingcloudlet knowledge of overlapping IoT deployments may increase the spatial knowledge of datafrom a given region.temporAL LimitAtions And serVice LAtencyReal-time access to data sources is pivotal tosense-making systems in the IoT. Many of theproposed solutions for smart cities require highlevels of coordination between IoT systems, andhigh responsiveness is a core mandate. The challenge of leveraging cloud resources is the inevitable queuing delay aggregated over multiplehops toward a cloud service, in addition to servicetime. Many experiments  have been carriedout to demonstrate the challenge with latency insoliciting cloud resources.On the other hand, bringing most IoT management closer to the edge yields significantinteroperability challenges across IoT systems, inaddition to lacking the infrastructure to mediateheterogeneous nodal operations. This is furtherworsened by the mis-coordination of communication between IoT nodes that not only differ intheir duty cycling schemes, but also exhibit varying operation levels as per their power mandatesand accessibility to their tethered devices.energy Footprint oF iot operAtionMost IoT systems are designed to conserve powerin light of their individual operational mandates,so any attempts to interoperate IoT systems yieldssignificant discrepancies in duty cycling schemesand multi-tiered operational levels. More importantly, to conserve power, most IoT nodes aredesigned to switch to low-profile sleep states toconserve power and are rarely open to IP-basedprobing from other Internet devices. While thisis necessary for operational longevity, it affectsinteroperability across IoT systems. More critically,most of these sleep schedules are mandated bygoverning base stations and/or remote controllers, thereby limiting “visibility” of resources toneighboring IoT systems. As we attempt to mergetraffic and data closer to the edge to conservenetworking resources, a pressing challenge in IoTlongevity will prove to be a hindrance.On the other hand, recent research on theenergy footprint of cloud architectures, especially as data centers are ever growing in theirpower demands , is offering new insights intothe potential gain as we migrate IoT operationto the network edge under a broad view of fognetworks. There is evident power gain in reducing overall network traffic, especially as we prunesuperfluous data before burdening cloud servicesup the hierarchy. Moreover, more contextualization of data, due to fog processing, may aid pruning and decision making that does not need toburden cloud systems.The power footprint will likely dominate theoffloading granularity problem. That is, decidingwhat should be processed at the user tier, whatcould be distributed on neighboring resources inthe mist, what can be offloaded to context-awarecloudlets, and what demands high-power processing at the cloud is a major research challenge.These questions build on our collective expertisein elastic processing, network traffic engineering,big data management, and hierarchical fusiontechniques.FUnctionAL mismAtcH UndercLoUd-centrALized operAtionIoT systems have long been developed ashard-coded architectures with pre-determinedoperational mandates. As we witness most oftoday’s things turn into micro-computers withcommunication and identification capabilities,the emphasis on uniform expression of functional capacities of IoT resources is growing inimportance. That is, as we attempt to interoperate between IoT architectures, we need to haveyardstick methods to identify and evaluate thefunctional capabilities of IoT resources across heterogeneous deployments.This is a precursor to enabling IoT cooperation at the network edge, as we attempt to leverage centralized service matching carried out bycloud services that do not have accurate or realtime feeds of which IoT nodes are currently dutycycling, offering their services, accessible in a givenregion, or reachable via a reliable networkingmedium . The general assumption of cloud-IoTsystems that simplify a global view of what is accessible, and carries out offline matching between service requests and actual IoT resources, is no longerfeasible as our IoT systems grow ever more mobileand independent in operation .priVAcy And secUrityRecent advancements in developing IoT-specific privacy and security mechanisms, such as theO Auth – 2.0 protocol, are tackling one of themost hindering factors in IoT traction. However,many of the challenges with IoT privacy and security result from the remote management of theseimportant operations. For example, in data denaturing (e.g., blurring out the faces of pedestriansin a cloud-camera architecture) is often carriedout at remote cloud services, presenting multipleopportunities for data breaches along the propagation links. Much of the context of the data isalso lost in this cloud offloading approach, whereby important correlations between data from asingle zone might be lost in the mass-scale processing of collected data. In addition, the challenge of data anonymization is magnified as morecentral modules have larger visibility to all datacollected from multiple edge zones. For example, think of a FitBit server that observes all themovement and health patterns of all of its users,and then anonymizes the results prior to company-wide studies.In terms of security, there are significant challenges in centralized “blanket” methods thatare applied to securing data at cloud servers, orAs we witness most oftoday’s things turninginto micro-computerswith communicationand identification capabilities, the emphasison uniform expressionof functional capacitiesof IoT resources isgrowing in importance.That is, as we attemptto interoperate betweenIoT architectures, weneed to have yardstickmethods to identify andevaluate the functionalcapabilities of IoTresources acrossheterogeneousdeployments.IEEE Communications Magazine • March 2018 161attempting to burden low-end IoT devices withencryption and authentication. While advances inthe OAuth – 2.0 protocol are yielding promisingsolutions, much has to be done to enable endusers in securing their own data, and deciding onthe frequency and quality of data that is reportedfrom their end devices to upper-tier cloud components.memoryLess operAtionThere is a rising challenge in maintaining userprofiles in each zone, to establish factors suchas trust in data reporting, weeding out false/malicious reports, and promoting “trusted” users in agiven IoT application, especially in crowd-basedscenarios. However, as users migrate from onezone to another, the exchange of this information between cloudlets opens up many securityand privacy challenges, in addition to challengesin cooperation between heterogeneous cloudletarchitectures. The scope of a designated zone,and what information about it and its ensuingusers could be collected, exchanged, and analyzed, remains a significant challenge as we bringmore processing and decision making power tothe edge of the network.recent AdVAncements: enAbLing Fog iotThe abundance of smart devices in our everyday interactions is mandating novel approachesto viewing what the IoT encompasses, and theaggregated power of these resources. Recentdevelopments in cloud computing are alreadypromising many advancements in reliable servicedelivery via cloudlets , as well as resource provisioning on both the cloud and cloudlet levels.Furthermore, most of today’s smart devicesare able to multi-home, whereby a typical smartphone can communicate over LTE, Bluetooth,WiFi, and ANT+ all in one device. This development is enabling many devices to act as mediatorsbetween multiple IoT systems, and further potentiates IoT interoperation in the mist, wherebyneighboring nodes that have similar multi-homingcapabilities can establish multiple overlay networks for different applications and/or services.The development of nano data centers, building on the growing potential of smart devicesand vehicles, will enable more data storage atthe fog level. This will enable both rapid accessto generated data and near-real-time probing ofresource profiles in a given fog region. That is, aservice can actively explore nearby fog resources(e.g., sensors) and query the data it collects; forexample, to verify if there is a hit to the query,or if other resources should be solicited. Moreimportantly, we can build on existing data aggregation techniques  that require localized datastorage to enable better pruning and data management at the edge. These locations could coincide with cloudlets to bring more processing andintelligence to the network edge and reduce IoTtraffic burdening the larger network backbone.Recent advancements in short-range communication protocols, summarized in the table depictedin Fig. 2, are promising low-power and long-rangecommunication between user devices, neighboringdevices in the mist, as well as long-range communication with cloudlets. This is highly utilizable inscenarios where mobility-driven communicationprotocols, such as Dedicated Short-Range Communications (DSRC), are enabling cooperativeoperation in vehicular IoT systems and vehicularclouds. The scale is indeed ever expanding, fromnano-communication with brain-machine interfaces via neuro-dust sensors , to long-range/high-bandwidth communication witnessed in theIEEE 802.11 family. In addition, recent developments in narrowband IoT (NB-IoT) and cloud radioaccess network (C-RAN)-based IoT developments are expanding the scope of which IoT systems we can communicate with on the cellularbackbone, with the added benefits of reliable andhigh-bandwidth channels.Furthermore, we are witnessing the development of many resource discovery protocols thatare enabling real-time probing and utilization ofIoT resources. Whether this is carried out on thecloud, edge, or mist level, there is great potentialin the mechanisms being developed to interrogate and register resources in real time, and scaletheir inclusion in fog-level resource pools, for service matching [7, 12].toWArd A Fog-iot ArcHitectUreAs we advocate for moving from a service-centric (cloud/edge) to a user-centric (fog) approachto IoT systems for smart cities, we focus on thearchitectural components that will enable such aprogressive framework. At its core, a user-centricarchitecture must utilize the context as well asresources of a local fog, and establish real-timemanagement modules that will tap into the potential of neighboring resources in the mist, as wellas cloudlet/edge-level resources when needed.Thus, service matching, mobility monitoring, andoverall offloading granularity are largely servedFigure 3. The interactions between a user-centric fog IoT architecture andcloud variants. The scope of fog IoT lies between user devices and neighboring IoT resources (i.e., the first two tiers). In a simple use case, user-centric health applications can probe local resources around the user (e.g.,smart wristband, chest strap, blood pressure monitor) and correlate withnearby resources (e.g., nearby temperature sensors and weather stations)to establish whether certain readings (e.g., higher heart rates) could beinfluenced by ambient weather conditions or potentially a user-specificcondition. More thorough analysis of history logs and user data could beaccessed via cloudlets and edge computing. Finally, epidemic monitoringand crowd-level analytics could be offloaded to remote (and more powerful) cloud services over their multiplicity of XaaS services.M2MCloudplatform(XaaS)NeighboringIoT resourcesUser devicesFog computing(Mobile) edge computingCloud computingCloudletaccess andresources162 IEEE Communications Magazine • March 2018within the bounds of the fog network rather thanthe coud.On an architectural level, we advocate forestablishing an IoT-in-the-fog controller that is ableto probe local resources and communicate directly with a local fog mediator, which could be thecloudlet/edge access point. The controller operation could be deployed on a dedicated deviceplaced for that purpose (e.g., in a roadside unit)or delegated to high-end resources (e.g., a smartphone or IoT hub).The core operational mandate of this controller would be to respond to policies mandated bythe fog mediator, as passed down from respective cloud services, but matching the currentresources in the fog zone. This includes catering to mobility and resource volatility, especiallyin utilizing mobile/vehicular resources in urbanenvironments. Figure 3 overviews the interactions between cloud variants and what they aredubbed in current literature, highlighting thereach/scale of each cloud variant. A simple scenario for e-health applications is presented in Fig.3, whereby classes of e-health applications running on each tier of the fog-IoT architecture areoverlaid and explained.However, what makes this architecture uniqueis that cloud-based IoT architectures are almostalways service-centric (over the cloud/edge).However, the fog IoT architecture is envisionedto be user-centric, whereby interactions betweendevices, exchange of control messages, and dataflow are governed by user-centric policies. Forexample, a user decides on the granularity of services and data they wish to access, and the associated monetary and energy cost of probing theprovisioned resources.While this entails more processing and power atthe edge, it builds on many advantages in privacypreserving mechanisms, mobility control, and elasticoffloading when the need arises. The granularity ofdata handled by users could further be controlledfrom both the user (to reduce access latency) andfrom the cloud (to enforce access rights).HigH-yieLd reseArcH directionsmemory-preserVing operAtion in tHe FogAs we attempt to enable smart city applications,there is a major opportunity loss in our memoryless view of contributing IoT systems. Withina given region, on a fog network scale, thereis much that can be inferred and stored aboutregion and user profiles per zone. That is, wecan establish history-based logging of data andcontributing users, adding to the developmentof trust-ranking schemes for each user known tocommonly access IoT systems in a given zone.Furthermore, there is great promise in establishing time-series-based inference of potentialresource needs in a given region, based onmaintaining memory of what is being produced(in terms of data) and accessed (in terms of IoTresources) in a given region.in sensing ArcHitectUresWe advocate for a migration from sensing as anintrinsically event/sampling-based paradigm to aservice paradigm. That is, data is only collectedwhen there is a demand for a given service, andthe decision of which nodes in an IoT ecosystemare to take part in sensing (i.e., load balancing)should be made on a fog level rather than an individual IoT system level. While many applicationswill mandate that their own sensors report data(e.g., for reliability and calibration constraints),there is significant data redundancy across IoT systems, which is causing significant big sensed datachallenges in IoT scalability and management.bUiLding on icnsThere is great promise in the recent developmentof information-centric networks (ICNs) that handle data at an intrinsic network primitive. In ICNs,data is automatically encoded and distributedover the network architecture, masking many ofthe challenges of data naming and cloud-basedquerying over dedicated IoT systems. However,much investigation is needed in terms of enablingremote “subscriptions” to data from given IoTsystems, to enable IoT nodes to act as data/content providers in an interactive environment thatresponds in real time to demand and popularitymetrics, rather than collect data in the hope offuture interest.incentiVe scHemes And interpLAybetWeen cLoUd VAriAntsA major challenge in crowd-based IoT systemsis soliciting data and resources from users. Manyrecent research endeavors have investigatedincentive schemes that address this challengeand the promise of these systems  in yieldinghigher user contributions. However, much is tobe discovered in incentivization across IoT platforms in smart city environments. This includeshow to establish incentives across IoT systems tosolicit the best-fit resources for a given task in amarket-driven architecture that reacts to resourceabundance, and responds to urgency in servicematching and timely delivery of IoT services.interpLAy oF iot serVices AndserVice orcHestrAtion pLAtFormsOne of the great promises of IoT is the potentialto build larger services (e.g., weather predictionand route planning) based on atomic/simpler services (e.g., temperature sensors and road monitoring cameras). The premise of service orchestrationhinges on the accessibility of reliable services thatare closer to the edge, with capped access latencies, and contextually enforced data collectionand pruning mechanisms. That is, we need todevelop more robust and reliable atomic servicesto feed larger service orchestration platforms. Thefog IoT architecture can synergize heterogeneousservices and architectures at the edge level, anddevelopments in policy management, data pruning, and information dissemination at fog serviceswill potentiate service orchestration.concLUding remArksThe potential of IoT proliferation in the fog is evident in the development of many technologiesthat bring more resources to the network edge.For over 20 years, sensing systems and IoT wereenvisioned as technologies that require lightoperation at the edge, with emphasis on cyberforaging and cloud offloading to enable reliableservices. The rise of mobile edge computing,Much is to be discovered in incentivizationacross IoT platforms insmart city environments.This includes how toestablish incentivesacross IoT systemsto solicit the best-fitresources for a giventask in a market-drivenarchitecture that reactsto resource abundance,and responds to urgency in service matchingand timely delivery ofIoT services.IEEE Communications Magazine • March 2018 163cloudlet access, and M2M communication modesare all providing ample resources for migratingmore processing and resource management atthe network edge. In this article we survey manyof the challenges in attempting to remotely operate sensing systems, and the ensuing big senseddata challenges that warn us of data generationbeyond what we can communicate and process.As more researchers are advocating for migrating IoT data management to the network edge,utilizing variants of cloud computing paradigms,this article surveys the core challenges in thismigration and proposes a roadmap for IoT interactions on the fog/edge/cloud tiers, based onthe aforementioned developments in edge technologies. Finally, we present a number of highyield directions that will further propagate IoTdevelopment in the fog. It is important to notethat many developments are taking place in parallel research domains, and it is at the heart of thisarticle to highlight the potential benefits in synergizing some of these mainstream efforts. This hasbeen surveyed in Fig. 2 as a building block to instigate further discussions on cross-domain synergytoward more potent fog IoT architectures.reFerences M. Satyanarayanan, “The Emergence of Edge Computing,”IEEE Computer, vol. 50, no. 1, Jan. 2017, pp. 30–39. S. Sarkar, S. Chatterjee, and S. Misra, “Assessment of theSuitability of Fog Computing in the Context of Internetof Things,” IEEE Trans. Cloud Computing, Oct. 2015, pp.1–14. A. Al-Fuqaha et al., “Toward Better Horizontal Integrationamong IoT Services,” IEEE Commun. Mag., vol. 53, no. 9,Sept. 2015, pp. 72–79. F. Bonomi et al., “Fog Computing and Its Role in the Internet of Things,” ACM Wksp. Mobile Cloud Computing, Aug.2012, pp. 13–16. S. Oteafy and H. Hassanein, Dynamic Wireless Sensor Networks, Wiley-ISTE, June 2014. ISBN: 978-1-84821-531-3. J. Preden et al., “The Benefits of Self-Awareness and Attention in Fog and Mist Computing,” Computer, vol. 48, no. 7,July 2015, pp. 37–45. S. Oteafy, “A Framework for Heterogeneous Sensing in BigSensed Data,” IEEE GLOBECOM, Dec. 2016, pp. 1–6. M. Satyanarayanan, “Pervasive Computing: Vision and Challenges,” IEEE Personal Commun., vol. 8, no. 4, Aug. 2001,pp. 10–17. S. Oteafy and H. Hassanein, “Resilient IoT Architectures overDynamic Sensor Networks with Adaptive Components,”IEEE Internet of Things J., vol. 4, no. 2, Apr. 2017, pp. 474–83. A. Al-Fuqaha et al., “Internet of Things: A Survey onEnabling Technologies, Protocols, and Applications,” IEEECommun. Surveys & Tutorials, vol. 17, no. 4, 2015, pp.2347–76. F. Jalali et al., “Fog Computing May Help to Save Energyin Cloud Computing,” IEEE JSAC, vol. 34, no. 5, May 2016,pp. 1728–39. J. Sahoo, S. Cherkaoui, and A. Hafid, “Optimal Selection ofAggregation Locations for Urban Sensing,” Proc. 2014 IEEEIC), Sydney, Australia, Aug. 2014, pp. 239–44. M. Maharbiz et al., “Reliable Next-Generation CorticalInterfaces for Chronic Brain-Machine Interfaces and Neuroscience,” Proc. IEEE, vol. 105, no. 1, Jan. 2017, pp. 73–82. A. Radwan et al., “Low-Cost On-Demand C-RAN BasedMobile Small-Cells,” IEEE Access, vol. 4, May 2016, pp.2331–39. D. He, S. Chan, and M. Guizani, “Privacy and IncentiveMechanisms in People-Centric Sensing Networks,” IEEECommun. Mag., vol. 53, no. 10, Oct. 2015, pp. 200–06.biogrApHiesSharief M. a. Oteafy [M] is an assistant professor at the Schoolof Computing, DePaul University. He is actively engaged in theIEEE Communications Society, serving as the ComSoc AHSNStandards Liaison and on the ComSoc Tactile Internet Standards WG. He co-authored a book, Dynamic Wireless SensorNetworks (Wiley), and has presented over 50 peer-refereedpublications and delivered multiple ComSoc tutorials on sensingsystems and IoT. He has co-chaired a number of IEEE workshops, in conjunction with IEEE ICC and IEEE LCN. He is anAssociate Editor of IEEE Access and on the Editorial Board ofWiley’s Internet Technology Letters.hOSSaM S. haSSanein [F] is a professor and director of theSchool of Computing at Queen’s University, Kingston, Ontario,Canada. His record spans more than 500 publications, in addition to numerous keynotes and plenary talks at flagship venues.He is also the founder and director of the TelecommunicationsResearch Lab at Queen’s University. He is an IEEE Communications Society Distinguished Speaker (Distinguished Lecturer2008–2010). He is the past Chair of the IEEE CommunicationSociety Technical Committee on Ad Hoc and Sensor Networks.He has received several recognitions and best paper awards attop international conferences, and led a number of symposia atflagship ComSoc conferences.For over 20 years,sensing systems andIoT were envisionedas technologies thatrequire light operation at the edge, withemphasis on cyberforaging and Cloudoffloading to enablereliable services. Therise of Mobile EdgeComputing, Cloudletaccess, and M2Mcommunication modesare all providing ampleresources for migratingmore processing andresource managementat the network edge.
- Assignment status: Already Solved By Our Experts
- (USA, AUS, UK & CA PhD. Writers)
- CLICK HERE TO GET A PROFESSIONAL WRITER TO WORK ON THIS PAPER AND OTHER SIMILAR PAPERS, GET A NON PLAGIARIZED PAPER FROM OUR EXPERTS