IFCIoT: Integrated Fog Cloud IoT Architectural Paradigm for Future Internet of Things

0.37 MB·

by  Arslan Munir

1
IFCIoT: Integrated Fog Cloud IoT Architectural
Paradigm for Future Internet of Things
Arslan Munir∗, PrasannaKansakar† andSamee U. Khan‡
Email: ∗[email protected], †[email protected],‡[email protected]
Abstract—WeproposeanovelintegratedfogcloudIoT(IFCIoT)architecturalparadigmthatpromisesincreasedperformance,energy
efficiency,reducedlatency,quickerresponsetime,scalability,andbetterlocalizedaccuracyforfutureIoTapplications.Thefognodes
7 (e.g.,edgeservers,smartrouters,basestations)receivecomputationoffloadingrequestsandsenseddatafromvariousIoTdevices.
1 Toenhanceperformance,energyefficiency,andreal-timeresponsivenessofapplications,weproposeareconfigurableandlayeredfog
0 node(edgeserver)architecturethatanalyzestheapplications’characteristicsandreconfigurethearchitecturalresourcestobettermeet
2 thepeakworkloaddemands.Thelayersoftheproposedfognodearchitectureincludeapplicationlayer,analyticslayer,virtualization
layer,reconfigurationlayer,andhardwarelayer.Thelayeredarchitecturefacilitatesabstractionandimplementationforfogcomputing
n
paradigmthatisdistributedinnatureandwheremultiplevendors(e.g.,applications,services,dataandcontentproviders)areinvolved.
a
WealsoelaboratethepotentialapplicationsofIFCIoTarchitecture,suchassmartcities,intelligenttransportationsystems,localized
J
weathermapsandenvironmentalmonitoring,andreal-timeagriculturaldataanalyticsandcontrol.
0
3
IndexTerms—Fogcomputing,edgecomputing,Internetofthings,reconfigurablearchitecture,radioaccessnetwork

✦
]
C
D
.
s 1 INTRODUCTION AND MOTIVATION to the logical extremes of a network. Fog computing
c
[ significantly decreases the data volume that must be
THE Internet of things (IoT) is a network of physical moved between end devices and cloud. Fog computing
1
things, objects or devices, such as radio-frequency enables data analytics and knowledge generation to
v
4 identification (RFID) tags, sensors, actuators, mobile occur at the data source. Furthermore, the dense
7 phones, and laptops. The IoT enables objects to be geographic distribution of fog helps to attain better
4 sensed and controlled remotely across existing network localized accuracy for many applications as compared
8 infrastructure, including the Internet; thereby creating to the cloud.
0
opportunities for more direct integration of the physical
. Although fog computing alleviates some of the issues
1 worldintothecyberworld.TheIoTbecomesaninstance
facing the realization of future IoT/CPS applications,
0
of cyber-physical systems (CPS) with the incorporation
7 the fog nodes (e.g., edge servers, smart routers, base
of sensors and actuators in IoT devices. Objects in IoT
1 stations) may not be able to meet performance,
v: can possibly be grouped into geographical or logical throughput, energy, and latency constraints of future
clusters. Various IoT clusters generate huge amounts of
i IoT/CPS applications unless fog computing architecture
X data from diverse locations, which advocates the need
is adapted to satisfy these application requirements.
r to process this data more efficiently. Efficient processing This adaptation is needed at both the system-level and
a of this data can involve a combination of different
the node-level for fog computing. This article aims
computation models, such as in situ processing and
to address the architectural challenges associated with
offloading to surrogate devices and cloud data centers.
the realization of scalable IoT and CPS applications
Cloud computing is an Internet-based computing
leveragingfogcomputing.Themaincontributionsofthis
paradigm that provides ubiquitous and on-demand
article are as follows:
access to a shared pool of configurable resources
(e.g., processors, storage, services, and applications) to • A novel integrated fog cloud IoT (IFCIoT)
other computers or devices. Although cloud computing architectural paradigm that harnesses the benefits
paradigm is able to handle huge amounts of data from of IoT, fog, and cloud computing in a unified
IoT clusters, the transfer of enormous data to and from archetype. The IFCIoT architecture promises
cloud computers presents a challenge due to limited increased performance, energy efficiency, quicker
bandwidth. Consequently, there is a need to process response time, scalability, and better localized
data near data source, and fog computing provides a accuracy for future IoT and CPS applications.
promising solution to this problem. • We propose an energy-efficient reconfigurable
Fog computing is a novel trend in computing that layeredfognode(edgeserver)architecturethatwill
aims to process data near data source. Fog computing adapt according to fog computing application
pushes applications, services, data, computing power, requirements. The layers of the proposed
and decision making away from the centralized nodes architecture include application layer, analytics
2
layer,virtualizationlayer,reconfigurationlayer,and clients in proximity.” Aazam et al. [4] have defined
hardware layer. The layered architecture facilitates fog computing as: “Fog computing refers to bringing
abstraction and implementation for fog computing networking resources near the underlying networks. It
paradigm that is distributed in nature and where is a network between the underlying network(s) and
different service, application, data and content the cloud(s). Fog computing extends the traditional
providers are involved. cloud computing paradigm to the edge of the network,
• We discuss the potential applications of the enabling the creation of refined and better applications
IFCIoT architecture, such as smart cities, intelligent or services. Fog is an edge computing and micro data
transportation systems (ITS), localized weather center (MDC) paradigm for IoTs and wireless sensor
maps and environmental monitoring, and real-time networks (WSNs).”
agricultural data analytics and control. Distinction Between Fog and Cloud Computing: The
The remainder of this article is organized as follows. word “fog” in fog computing conveys the idea of
Section 2 elucidates the distinction between cloud, fog, bringing the advantages of cloud closer to the data
and edge computing. Section 3 provides a summary of source(cf.meteorology:fogissimplyacloudthatisclose
relatedwork.Section4describeshowfogcomputingcan to the ground). Cloud computing is usually a model
be used for implementation of intelligent transportation for enabling convenient and on-demand network use
systems (ITS). Section 5 presents our proposed IFCIoT of a shared pool of configurable computing resources,
architectural paradigm. The fog node architecture for such as networks, servers, storage, applications, and
the IFCIoT architectural paradigm is presented in services, that may be rapidly provisioned and released
Section 6. Section 7 describes how our proposed fog with minimal management effort or vendor interaction.
node architecture can be applied for implementation of Cloud computing permits options for renting of storage
ITS. Section8 discusses insights into potential consumer and computing infrastructures, business processes, and
electronics applications. Finally, Section 9 concludes our overall applications. Fog computing extends cloud
article. computing and services to the edge of the network.
Fog computing can be distinguished from cloud
2 DISTINCTION BETWEEN CLOUD, FOG, AND computing based on various metrics as discussed in the
following [5]. The proximity of the fog to the end user
EDGE COMPUTING
is one of the main characteristics that differentiates fog
Thedistinctionbetweencloud,fog,andedgecomputing fromcloud,thatis,fogresidesattheedgeofthenetwork
has not been elucidated in many relevant scholarly whereas cloud is located within the Internet. Cloud has
works to the best of our knowledge. To provide readers a centralized geographical distribution whereas fog can
with a clear understanding of fog computing, we havealocalizedordistributedgeographicaldistribution.
elucidate the distinction between cloud, fog, and edge Cloud computing systems typically consists of only a
computing in this section. few resourceful server nodes whereas fog comprises of
Defining Fog Computing: Fog computing has been a large number of relatively less resourceful fog nodes.
defined in a variety of ways in literature by academia Furthermore, the processing at fog nodes frees up the
andindustry.Thetermfogcomputingisoftenassociated core network bandwidth, which helps to improve the
with Cisco, that is, “Cisco Fog Computing” [1], overall network efficiency. The distance between client and
however, fog computing is open to the community server nodes in cloud is typically multiple hops whereas
at large. A coalition of industry and academia has clientscanconnecttofognodesusuallythroughasingle
founded the “OpenFog Consortium” in November hop. Consequently, fog computing reduces the latency
2015 to promote and accelerate adoption of open fog of data transmission from IoT devices to the offloaded
computing [2]. The coalition founders include ARM, server because of the proximity of the fog to the end
Cisco, Dell, Intel, Microsoft and Princeton University. devices as compared to the cloud. Furthermore, cloud
The OpenFog Consortium [2] defines fog computing as: computing platforms typically engender higher delay
“Fogcomputingisasystem-levelhorizontalarchitecture jitter for applications as compared to the applications
that distributes resources and services of computing, running on fog nodes. Hence, fog computing is more
storage, control and networking anywhere along the suitable for real-time IoT and CPS applications as
continuum from Cloud to Things”. compared to cloud computing.
Yi et al. [3] have defined fog computing as: “Fog The fog’s ability to provide location-based customization
computing is a geographically distributed computing of content, services, and applications to the IoT
architecture with a resource pool consisting of one devices is another distinguishing characteristic of fog.
or more ubiquitously connected heterogeneous devices Cloud, on the contrary, in most cases is unable to
(including edge devices) at the edge of network and deliver specialized content, services, and applications
not exclusively seamlessly backed by cloud services, to devices. The location-based customization of services
to collaboratively provide elastic computation, storage and information is imperative as the information may
and communication (and many other new services and be relevant in a local context (i.e., proximity of specific
tasks) in isolated environments to a large scale of geographic coordinates) and may be irrelevant beyond
3
the physical proximity to that location. Finally, cloud 3 RELATED WORK
provideslimited mobility support to end deviceswhereas
Fog computing has been the subject of many research
mobility of end devices is better supported in fog.
works in recent years. Various fog computing
architectures have been proposed [7] [8] [9], each of
Although fog and cloud computing paradigms have whichaddressesspecific mobility, resourcemanagement
cleardistinctions,theseparadigmsarenotareplacement andoptimizationissues,however,auniversallyaccepted
for each other. In fact, the fog and the cloud are fog computing architecture and standard has yet to be
interdependent and mutually beneficial since certain adopted.
functions are naturally more advantageous to carry Patel et al. [7] discussed the key market drivers,
out in the fog while others are better suited to the benefits, requirements, objectives, and challenges of
cloud. The segmentation of what tasks go to the fog MEC.Thepaperalsopresentedahigh-levelarchitectural
and what tasks go to the backend cloud is application blueprint for MEC. Aazam et al. [8] proposed a layered
specific, and can change dynamically based upon the architecture for fog computing to address the resource
state of the network, such as processor loads, link management challenges, such as resource prediction,
bandwidths,storagecapacities,faultevents,andsecurity allocation, and pricing, in fog servers. The authors used
threats [6]. The cloud provides various services, such a probability-based model that considered the type,
as Infrastructure as a service (IaaS), platform as a traits and characteristics of fog customers to make these
service (PaaS), and software as a service (SaaS), for resource management decisions. Bittencourt et al. [9]
organizations that require elastic scale. Fog computing proposed a layered architecture to facilitate mobility
can provide fog as a service (FaaS) to address various of connected IoT nodes. In their approach, a virtual
business challenges. FaaS may provide services, such as machine (VM) instance was created for each IoT node
network acceleration, network functions virtualization connected tothe fogserver.WhenanIoT device crossed
(NFV), software-defined networking (SDN), content the radio boundary of the fog server, then it was
delivery,devicemanagement,complexeventprocessing, handedofftoanotherfogserverbyexchangingsnapshot
video encoding, protocol bridging, traffic offloading, of the IoT device’s VM instance. Their layered fog
cryptography, and analytics platform, [6] etc. architecture supported the VM migration. Paglierani
[10] discussed the use of hardware accelerators in
network nodes to support fog computing. The paper
Distinction Between Fog and Edge Computing: The
demonstratedthatcombinationofhardwareacceleration
distinction between fog and edge computing is subtle.
and advanced networking concepts, such as SDN and
Most of the prior literature has treated fog and edge
NFV, can significantly improve network performance.
computing as synonymous and has used the word
Although prior works have proposed several fog
fog and edge computing interchangeably. We clarify
architectures to address specific issues, however, a
the similarities and differences between fog and edge
reconfigurable fog node architecture that is able
computing. The term mobile edge computing (MEC) is
to adapt according to application requirements has
also often used in jargon. We point out that MEC is
not been studied. In this article, we propose the
an instance of edge computing where the objective is
IFCIoT architecture that unifies IoT, fog, and cloud
to provide cloud computing capabilities at the edge
computing paradigms to help realization of future
of the cellular network. The edge server in MEC
IoT and CPS applications. We further propose a
is located at the cellular base station. Both fog and
reconfigurable fog node architecture that analyzes
edge computing pushes applications, data, services, and
the applications’ characteristics and reconfigure the
computing power away from the centralized nodes
architecturalresourcestobettermeetthe peakworkload
to the logical extremes of a network. However, fog
demands.
computing paradigm has a more decentralized and
distributed control as compared to edge computing
paradigm that has a relatively more centralized control. 4 FOG COMPUTING FOR INTELLIGENT
Another distinction between edge and fog computing TRANSPORTATION SYSTEMS
is fog’s openness, which is critical for the success of a
ubiquitous fog computing ecosystem for IoT platforms In this section, we discuss fog computing as a key
andapplications.Proprietaryorsingle vendor solutions, driver for future intelligent transportation system (ITS)
as pursued typically in edge computing, can engender implementations. We begin by describing different
limited supplier diversity, which can have a negative agents that constitute an ITS. We then present a
impact on system cost, quality, market adoption, and classification of scenarios for ITS deployment, the
innovation. Furthermore, radio access network in edge shortcomings of modern ITS implementations in these
computing paradigm is typically a cellular network scenarios and how fog computing implementations
whereas in fog computing radio access network can potentially overcome these shortcomings. Finally, we
be WLAN, WiMax, and/or cellular, and is partially list out benefits that different ITS agents get from fog
considered a part of the fog. computing based ITS implementations.
4
processing operations are carried out on the cloud.
This places a massive burden on the cloud’s computing
resources.Thisburdenisexacerbatedduringpeaktraffic
hours, when ITS agents have to be rerouted from traffic
congested areas. During such times when the cloud’s
computing resourcesareoverwhelmed, thereis ahigher
possibilityofsignificantlatencyinprovidingresponseto
ITS agents. Late responses during emergency situations
like traffic accidents or natural disasters could leed to
massive casualties and fatalities.
Fog computing offers higher reliability and flexibility
in ITS implementation as compared to using cellular
networks and cloud computing. In urban areas, due
to large number of ITS agents, movement of traffic is
Fig.1. IntelligentTransportationSystem slow. Slow moving traffic opens up the opportunity
of using other modes of communication than cellular
An ITS consists of different agents such as vehicles, networks. For example, ITS agents can communicate
trafficinfrastructureandpedestrians(asshowninFig.1). over a close range using wifi hotspots with multihop
Modern vehicles and traffic infrastructures have a host communication wherein data is communicated from a
of integrated electronic subsystems. Vehicles employ source ITS agent to a destination ITS agent by means of
electronicsubsystemstoincludefeaturessuchas,driving intermediate/relay ITS agents. The distributed network
assistance, safety and security, infotainment, navigation of fog computing nodes further facilitates multihop
etc. Traffic infrastructures have sensor systems that communication. Firstly, the distributed network of fog
sense the number and speed of vehicles through an nodes can be used to provide wifi connectivity to
intersection, road side units for monitoring weather, all agents of ITS. Secondly, fog nodes can also serve
real-time video surveillance cameras, and, signal and as intermediate hops in communication. They can
alert systems. Pedestrians also have electronic systems perform data filtering and analysis operations on the
in the form of smart devices which hold information communicated data to reduce data size. This helps
relevant to ITS such as location, direction of walking, to preserve the bandwidth of the network as well as
walking speed etc. The data generated by each of reduces latency.
these ITS agents is vital to implementing an effective Fog nodes also help to unload the processing burden
ITS. The data has to be processed and communicated ofthecloudinITSimplementation.Insteadofhavingall
to other ITS agents in the system. We discuss data data processing operations of a full ITS implementation
processing and communication mechanisms in modern carried out on a centralized cloud, the processing is
ITS implementations and compare them with fog carried out closer to the edge of the network using a
computing based ITS implementation for different number of distributed fog nodes. Fog nodes process
scenarios in the following text. data in a local context i.e. each node processes data
An ITS implementation can be broadly classified for either a single or a small group of ITS instances.
into two scenarios – urban and rural. In an urban This approach can be leveraged because traffic data
scenario, there are higher number of agents in the ITS. from an ITS instance is usually only pertinent to that
Providingreal-timeresponsetoalargenumberofagents instance or to a few neighboring instances. The fog
requireshighlyreliablecomputationandcommunication nodes only communicate summaries of local traffic data
resources. Modern ITS implementations rely on the to the cloud which uses it for data analytics (traffic
cloud for computing resourcesand on cellular networks patterns, planning construction of new roads etc.). In
forcommunication.Althoughcellularnetworksfacilitate cases of emergency and disaster situations, local traffic
communication in ITS, they are primarily dedicated to data from all ITS instances have global scope [12]. In
mobile telephony. In an urban scenario, wherein mobile these situations, fog nodes communicate data to the
telephonytrafficishigh, cellularnetworks (3GandLTE) cloudmorefrequently.Sincethefognodescommunicate
cannot provide high reliability communication for ITS filteredandlocallyanalyzeddatatothecloud,thecloud
[11].Also,the cloudisnotareliablecomputing resource can swiftly determine an appropriate response.
for ITS. For an effective implementation of a single ITS In a rural scenario, there are fewer number of agents
instance (e.g. at one of the many traffic intersections in in the ITS. Real-time response to a small number of ITS
a city), data from multiple different agents has to be agentscanbedeliveredwithsignificantlylesscomputing
processed to generate information relevant from each and communication resources so, a sparser distribution
agent’s perspective. For a full ITS implementation (e.g. of fog nodes may be used. In such scenarios, fog nodes
city-wide traffic intersections), an enormous number are useful in emergency and disaster situations. During
of unique ITS instances have to be processed. In mass evacuation of urban areas in disaster situations,
cloud computing based ITS implementations, all these traffic may be heavily routed through rural areas. The
5
RADIO NETWORK FEDERATED CLOUD SERVICES
RADIO
WLAN CORE NETWORK
WiMAX
CELLULAR
EDGE SERVER
APPS
EDGE CONTENT
SERVICES
ROUTER APPS
EDGE CONTENT
SERVICES
RAN
RADIO FOG RADIO
$
RADIO
IoT CLUSTERS
Fig.2. IFCIoTarchitecturalparadigm.
fog nodes in rural areas help in the management of the to IoT devices via intermediary fog. The federated
increased traffic load by providing reliable computing cloud services are provided by a federated cloud that
and communication resources to ITS agents. can comprise of multiple internal and external cloud
Fog computing benefits all ITS agents by improving servers to match business and application needs. As
ITS services. For vehicles, fog computing can provide shown in Fig. 2, the fog comprises of fog nodes (e.g.,
the following improved services: rerouting from heavy edge servers, smart routers, base stations, gateway
traffic areas (during peak hours), repair or towing devices) and partially radio access networks. In a
services, services in case of accidents, emergency fog computing environment, much of the processing
evacuation routes, finding parking space etc. Fog takes place on a fog node. In the IFCIoT architecture,
computing helps in development of traffic infrastucture. the entire fog deployment can be located locally
Traffic flow data collected from fog nodes can be (e.g., in case of building automation, a company
used for the following: changing location of signs that manages a single office complex) or the fog
and signals based on traffic data analysis, surveying deploymentscanbedistributedatlocalorregionallevels
road surfaces for damage, planning construction of that feed information to a centralized parent system
new roads etc. Pedestrians have more safety in and services (e.g., in case of building automation, a
crosswalks on busy streets. They are also informed large commercial property management company). In
about shorter routes for walking to their destinations. the IFCIoT architecture, each operational fog node is
Fog computing also improves transit services with autonomous to ensure uninterrupted operations of the
featureslikeaccuratearrival/departuretimes,delayand facility/service it provides.
cancellation notifications, passenger count, passengers Afognode inthe IFCIoT architecturemanagesallIoT
per stations, vehicle operation duration for repair and devicesthatarewithinitsradionetwork.TheIoTdevices
maintenance, ticket purchase, seat selection, remote typically leverage radio access networks (e.g., WLAN,
check-ins, and information on hotels and restaurants WiMAX, cellular networks) to communicate with the
nearby transit stations, etc. fog whereas the fog is connected to the federated cloud
servers via core network. A fog can be connected to
5 IFCIOT: INTEGRATED FOG CLOUD IOT
other fogs through a radio access network. Specifically,
ARCHITECTURAL PARADIGM
when an IoT device moves from the coverage of one
We propose the IFCIoT architectural paradigm as fog to another, the virtual machines associated with the
depictedinFig.2.Thenovelaspectofthisarchitectureis IoT device are migrated from the original host edge
that the architecture furnishes federated cloud services server to the migrated edge server [9]. The fog nodes
6
in the IFCIoT architecture facilitates the collection and platform services that the edge server can provide to
maintenance of local system statistics and/or locally various applications hosted on the edge server. The
sensed information supplied by various IoT devices application platform services provided by the edge
and/or clusters. These local statistics and information server include services for computation offloading,
caneitherbeusedtoimprovethelocalcontent, services, contentaggregation,databasesandbackup,andnetwork
and applications or to update the federated cloud data information, etc. When an IoT device connected to
center. The federatedcloud data center receivesupdates the edge server requests for a particular application
frommultiplefognodes.Thefederatedclouddatacenter to be executed, a VM environment is created for the
can then perform big data analytics on the received application. This means that each application has its
information to extract information that is representative own instantiation of VM environment running on the
ofabiggergeographicallocationandtodetermineglobal application layer. The application layer for the edge
system statistics. server acts as a PaaS provider, that is, the application
layer abstracts the entire edge server architecture to
6 FOG ARCHITECTURE
provide a standard platform for the IoT application
Thefogcomprisesoffognodesandpartiallyradioaccess developers.
networks as depicted in Fig. 2. This section discusses Analytics Layer:
the radio access network and our proposed fog node The analytics layer
architecture for the IFCIoT architectural paradigm. consists of three
modules: platform
6.1 RadioAccessNetwork services use statistics
module, machine Application Platform Services
IoT end devices can leverage a multitude of wireless
access technologies, such as WLAN, WiMAX, and learning module, APPLICATION LAYER
and power manager
cellular access networks (e.g., 4G, 5G), as the radio
module. The platform
access network for accessing the fog. According to the
services use statistics Platform Services Use Statistics
OpenFog Consortium, fog nodes are not completely
module analyzes usage
fixed to the edge, but should be seen as a fluid system
of the application Power Machine Learning
of connectivity. Hence, the radio access network can be services provided in Manager Module
considered partially a part of the fog architecture. Fog
the application layer. ANALYTICS LAYER
computingenablesthedesignofanenergy-andspectral-
The machine learning
efficient radio access network, which can be named as
module takes the service
fogcomputing-basedradioaccessnetwork(F-RAN)[13]. Virtualization
requests’ type and Mananger
The F-RAN can take advantage of local radio signal
volume information as
processing,cooperativeradioresourcemanagement,and VIRTUALIZATION
input and analyzes this
distributed storage capability of fog nodes to decrease LAYER
information to predict
the load on fronthaul (connection between centralized
hardware resource
basebandcontrollersandremoteradioheadsatcellsites requirements, which DVFS Reconfiguration
in a new radio access network architecture) and avoid Manager Manager
can be leveraged by
large-scale radio signal processing in the centralized
the reconfigurable layer. Storage
baseband controllers. The power manager Controllers Reconfigurable
Hardware
module analyzes the Network Modules
6.2 Reconfigurable and Adaptive Fog Node/Edge
service requests’ type
ServerArchitecture HARDWARE RECONFIGURABLE
and volume information
LAYER LAYER
Workload analytics on a server/cloud reveals that to determine the edge
different applications have different peak load server utilization. Based Fig. 3. Layered fog node
hours at different times [14]. We exploit this time- on the edge server architecture.
variance of applications’ peak workloads to propose utilization, the power
a reconfigurable and adaptive multicore architecture manager module dynamically adjusts the operating
for the edge server that can adapt according to the voltage and frequency of the edge server’s hardware
application load being run at a given time to better components.
sustain the projected data velocity, data volume, and Virtualization Layer: The virtualization layer abstracts
real-time requirements of IoT/CPS applications. Our the underlying hardware resources (hardware resources
proposed edge server architecture consists of several can be from different vendors) to provide a common
layers: application layer, analytics layer, virtualization interfaceforapplicationservices.Thevirtualizationlayer
layer, reconfiguration layer, and hardware layer, as acts as an IaaS provider, that is, the virtualization layer
shown in Fig. 3. abstracts the hardware resources of the edge server
ApplicationLayer:Thisisthetop-mostlayeroftheedge from the application services layer, and hence from the
server architecture. This layer consists of application applications running on the edge server.
7
Reconfigurable Layer: The reconfigurable layer processingcapabilitiesoftheedgeserverbyinstantiating
consists of a reconfiguration manager and a set of more hardware modules. With increased processing
reconfigurable modules. The reconfiguration manager capabilities, more applications can be launched in the
in the reconfigurable layer takes input from the application layer.
machine learning module in the analytics layer, and Virtualization Layer: The virtualization layer hides the
reconfigures the architectural resources to better meet underlying hardware from the application layer by
the requirements of the peak workload application at a providing a common interface to all hardware modules.
given time. The ability to adapt edge server hardware For example, a pedestrian can request for platform
to changing workload requirements is a novel concept service from the edge server over a wifi network and
which, to our knowledge, has not been addressed in a vehicle can request the same platform service but,
other works [7] [9] in literature. over an LTE network. The virtualization layer reformats
Hardware Layer: The hardware layer consists of a these requests to remove all hardware and network
dynamicvoltageandfrequencyscaling(DVFS)manager, dependency parameters before forwarding it to the
storage controllers, and network resources. The DVFS application layer. The reformatting process thus, frees
moduleacquiresinputfromthe powermanagermodule up the application layer from all dependencies.
in the analytics layer. The DVFS module adjusts the Reconfigurable Layer: Reconfigurable layer takes
operating voltage and frequency of various hardware input from the analytics layer and reconfigures
components of the edge server (e.g., processor core, the architectural resources of the edge server. This
memory, and peripherals) depending on the workload increases the flexibility of the edge server and makes
demands. The storage controllers and storage units are it capable to adjust to different workloads. For an
used for database services and backup services. The ITS implementation, flexibility is advantageous during
network module manages the connectivity between the traffic congestions, as discussed in the analytics layer.
edge server and the IoT devices, and between the edge Flexibility is also useful in emergency and disaster
server and the cloud. situations. For example, in the event of a disaster, mass
evacuation of several towns and cities has to be carried
7 RECONFIGURABLE AND ADAPTIVE FOG out. During this time, the number of vehicles on the
NODE/EDGE SERVER ARCHITECTURE APPLIED road would be enormous. Reconfigurable edge servers
can adapt to these increased workloads and provide
TO INTELLIGENT TRANSPORTATION SYSTEMS
reliable service.
In this section, we describe how our proposed Hardware Layer: The hardware layer consists of
reconfigurable and adaptive fog architecture maps to computation and communication hardwarecomponents
the intelligent transportation system use case that we which runs the edge server applications and,
presented in Section 4. reconfigures hardware modules in the reconfigurable
Application Layer: The application layer provides layer. In an ITS implementation, the hardware layer
platform services to various applications hosted on the components aggregate data from sensors (induction
edge server. For example, in an ITS implementation, loop sensors, weather sensors, speed-radar sensors etc.)
consider real-time video surveillance as a platform and manage signal and alert displays (traffic signals,
service. This service can be used by vehicles waiting warnings for closed roads, bad weather conditions,
to turn at an intersection to look for vehicles coming ongoing construction etc.)
from behind or from the sides; by traffic management
8 INSIGHTS INTO OTHER POTENTIAL
department for routing traffic; by public transport
management department to track its buses; or by CONSUMER ELECTRONICS APPLICATIONS
emergency response department to access severity of This section discusses the potential applications of
vehicular accidents to dispatch personnel accordingly. the IFCIoT architecture in various sectors, such as
Allthese applicationsusethe sametrafficinfrastructure, smart cities, localized weather maps and environmental
but, process the collected video data differently. All monitoringand,real-timeagriculturaldataanalyticsand
of these applications are hosted on edge servers. Each control.
application is run on a VM environment which hides its Smart Cities: The IFCIoT architectural paradigm can
processing operations from all other applications. provideabasisarchitectureforvarioussubsystems(e.g.,
Analytics Layer: The analytics layer analyzes the smart grid, smart buildings, industrial plants, hospitals,
volumeofplatformservicerequestsandforwardstuning schools, and law enforcement) in smart cities. A major
parameters to hardware and reconfigurable layer. For challenge in establishing smart cities is the requirement
example, consider the case of traffic congestion at an of ubiquitous broadband bandwidth and connectivity
instance of ITS during peak hours. During this time, availability. While most modern cities have multiple
therearelargenumberofITSagentsrequestingplatform cellular networks that provide adequate coverage, these
servicesfromtheedgeserver.Theanalyticslayerdetects networksoftenhavecapacityandpeekbandwidthlimits
the increased volume of request and forwards tuning that just meet the needs of their existing subscribers.
parameters to the reconfigurable layer to increase the This limited bandwidth of cellular networks makes the
8
realizationofadvancedmunicipalservicesenvisionedin processing with reduced latency and delay jitter, real-
a smart city (e.g., real-time surveillance, public safety, time responsiveness, mobility support, and location-
on-time advisories, smart buildings) a challenge. The based customization. However, fog computing is not a
IFCIoTarchitecturehelpsinreducingtheloadoncellular replacement for cloud computing as cloud computing
networksbyleveraginglocalradioaccessnetworks,local will still be desirable for high end batch processing jobs
radio signal processing, and cooperative radio resource that are very frequent in the business and scientific
managementinfognodes.Theconservedbandwidthcan worlds. The synergy of fog and cloud computing will
then be used for providing smart city services. help in realization of future IoT and CPS applications.
Localized Weather Maps and Environmental In this article, we have proposed a fog-centric IFCIoT
Monitoring: Localized weather maps can be an architecture that promises increased performance,
interesting application of the IFCIoT architectural energy efficiency, reduced latency, scalability, and better
paradigm. Various IoT devices measure temperature, localized accuracy for IoT and CPS applications. To
humidity, and atmospheric pressure, and send this better meet the performance, energy, and real-time
information to nearby edge servers. The edge servers requirements of applications, we have also proposed
process the received information from IoT devices to a reconfigurable fog node architecture that can adapt
obtainamorerefinedandlocalizedweatherinformation according to the workload being run at a given time.
for customers as opposed to the weather information We also elaborate the potential applications of the
available from news outlets for the whole city. The proposed IFCIoT architecture, such as smart cities, ITS,
edge servers further update the back-end cloud localized weather maps and environmental monitoring,
servers for refined weather information and better and real-time agricultural data analytics and control.
weather forecasting. Environmental monitoring is a
similar application that can be realized in the IFCIoT REFERENCES
architectural paragon. The environmental monitoring
[1] Cisco, “Cisco fog computing,” accessed on, June 2016. [Online].
system that leverages the IFCIoT architecture can Available:https://goo.gl/ZUPQLD
provide more localized (e.g., geographical precision [2] OFC, “Openfog consortium,” accessed on, June 2016. [Online].
Available:http://www.openfogconsortium.org/
rangingfromazipcodetolessthanamile)andaccurate
[3] S. Yi, Z. Hao, Z. Qin, and Q. Li, “Fog computing: Platform and
information regarding air quality, allergens, pollution, applications,”inIEEEWorkshoponHotTopicsinWebSystemsand
and noise in an area. Technologies(HotWeb),Washington,DC,USA,November2015,pp.
73–78.
Real-Time Agricultural Data Analytics and Control:
[4] M. Aazam and E.-N. Huh, “Fog computing: The cloud-iot/ioe
The IFCIoT architecture can improve agricultural health middleware paradigm,” IEEE Potentials, vol. 35, no. 3, pp. 40 –
toensurepeople’saccesstosafe,plentiful,andnutritious 44,May2016.
[5] M. Abdelshkour, “Iot, from cloud to fog computing,” March
food by enabling real-time agricultural data analytics
2015.[Online].Available:https://goo.gl/7zNxEd
and automated control where possible. The IoT devices [6] OpenFogConsortium,“Openfogarchitectureoverview,”inWhite
(mainlysensors)inanagriculturalareaprovidelocalized Paper,February2016.
[7] M. Patel et al., “Mobile-edge computing,” in ETSI White Paper,
information regarding soil moisture, precipitation, rain
September2014.[Online].Available:https://goo.gl/DS5x2L
water,waterestimatefrommeltingsnow,pollutionlevel, [8] M. Aazam and E.-N. Huh, “Fog computing micro datacenter
pestlevel,andtypesofpeststothenearbyedgeservers. based dynamic resource estimation and pricing model for IoT,”
in IEEE 29th International Conference on Advanced Information
The edge servers process and analyze the received
Networking andApplications, Gwangju,SouthKorea, March 2015,
informationfromtheIoTdevicesintheagriculturalfield pp.687–694.
and then determine an accurate scheduling of water [9] L. F. Bittencourt, M. M. Lopes, I. Petri, and O. F. Rana,
“Towardsvirtualmachinemigration infog computing,”in Proc.
sprinkler systems, fertilizers and pesticides supply in
ofInternationalConferenceonP2P,Parallel, Grid,CloudandInternet
the area to preserve the crop quality. Our proposed Computing(3PGCIC),Krakow,Poland,November2015,pp.1–8.
edge server with real-time analytics engine can provide [10] P. Paglierani, “High performance computing and network
function virtualization: A major challenge towards network
real-time control of the IoT actuator devices in each
programmability,” in IEEE International Black Sea Conference
agricultural area as opposed to sending and processing on Communications and Networking (BlackSeaCom), Constanta,
all the sensed information in the cloud. The edge Romania,May2015,pp.137–141.
[11] X.Hou,Y.Li,M.Chen,D.Wu,D.Jin,andS.Chen,“Vehicularfog
servers also raise triggers and alarms for the respective
computing:Aviewpointofvehiclesastheinfrastructures,”IEEE
agriculturalauthorities. The edgeserversfurtherupdate TransactionsonVehicular Technology, vol.65,no.6,pp.3860–3873,
the cloud data centers with the information periodically Jun2016.
[12] O.Khalid,M.U.S.Khan,Y.Huang,S.U.Khan,andA.Zomaya,
so that the cloud data center can perform detailed
“Evacsys:Acloud-basedserviceforemergencyevacuation,”IEEE
analytics and make agricultural decisions for larger CloudComputing,vol.3,no.1,pp.60–68,2016.
geographical areas. [13] M. Peng, S. Yan, K. Zhang, and C. Wang, “Fog computing
based radio access networks: Issues and challenges,” June 2015.
9 CONCLUSIONS [Online].Available:https://arxiv.org/pdf/1506.04233.pdf
[14] R.Singh,U.Sharma,E.Cecchet,andP.Shenoy,“Autonomicmix-
Fog computing provides various advantages over aware provisioning for non-stationary data center workloads,”
in Proceedings of the 7th International Conference on Autonomic
cloud computing for applications that require faster
Computing,Washington,DC,USA,2010,pp.21–30.
9
ArslanMunirisanassistantprofessorintheDepartmentofComputer University,Institute ofEngineering,PulchowkCampus,Nepal.Contact
Science and Engineering at the University of Nevada, Reno. His [email protected].
research interests include embedded and cyber-physical systems,
computer architecture, multicore, parallel computing, fault tolerance,
and computer security. Munir has a PhD in electrical and computer
engineeringfromtheUniversityofFlorida,Gainesville.He’samember
[email protected]. Samee U. Khan is an associate professor in the Department
of Electrical and Computer Engineering at the North Dakota
State University, Fargo. His research interests include optimization,
robustness,andsecurityofcloud,grid,clusterandbigdatacomputing,
social networks, wired and wireless networks, power systems, smart
PrasannaKansakarisaPhDstudentintheDepartmentofComputer grids, and optical networks. Khan has a PhD from the University of
Science and Engineering at the University of Nevada, Reno. His Texas, Arlington. He is a senior member of IEEE, a fellow of the
research interests include embedded and cyber-physical systems, Institution of Engineering and Technology, and a fellow of the British
computerarchitecture,multicore,andcomputersecurity.Kansakarhas [email protected].
aBEinelectronicsandcommunicationengineeringfromtheTribhuvan