CLOUD COVER

Fog Computing:
Helping the Internet
of Things Realize
its Potential generated and operational sav-
ings—of $11 trillion per year, which
would represent about 11 percent of
Amir Vahid Dastjerdi and Rajkumar Buyya, University of Melbourne the world economy;2 and that users
will deploy 1 trillion IoT devices.

The Internet of Things (IoT) could enable COPING WITH INTERNET

OF THINGS DATA
innovations that enhance the quality of life, but it IoT environments generate un-
generates unprecedented amounts of data that precedented amounts of data that
can be useful in many ways, par-
are difficult for traditional systems, the cloud, and ticularly if analyzed for insights.
However, the data volume can over-
even edge computing to handle. Fog computing whelm today’s storage systems and
analytics applications.
is designed to overcome these limitations. Cloud computing could help
by offering on-demand and scal-

T
able storage, as well as processing
he Internet of Things (IoT) promises to make services that can scale to IoT requirements. However,
many items—including consumer electronic for health-­monitoring, emergency-response, and other
devices, home appliances, medical devices, cam- latency-­sensitive applications, the delay caused by trans-
eras, and all types of sensors—part of the Inter- ferring data to the cloud and back to the application is
net environment.1 This opens the door to innovations that unacceptable. In addition, it isn’t efficient to send so much
facilitate new interactions among things and humans, data to the cloud for storage and processing, as it would
and enables the realization of smart cities, infrastruc- saturate network bandwidth and not be scalable.
tures, and services that enhance the quality of life. Recent analysis of a healthcare-related IoT application
By 2025, researchers estimate that the IoT could have with 30 million users showed data flows up to 25,000 tuples
an economic impact—including, for example, revenue per second.3 And real-time data flows in smart cities with

40 CO M PUTE R P U B LISH ED BY TH E I EEE COMP UTER SOCI E T Y 0018-9162/16/$33.00 © 2016 IEEE

 EDITOR SAN MURUGESAN
BRITE Professional Services;
[email protected]

many more data sources could easily

reach millions of tuples per second.
To address these issues, edge com-
Provider A Provider B
puting4 was proposed to use comput- Public cloud
ing resources near IoT sensors for local
storage and preliminary data process-
ing. This would decrease network con- Fog computing
gestion, as well as accelerate analysis
and the resulting decision making.
However, edge devices can’t handle Private cloud Private cloud
multiple IoT applications competing
for their limited resources, which re-
sults in resource contention and in-
creases processing latency.
Fog computing—which seamlessly
integrates edge devices and cloud re-
sources—helps overcome these lim-
itations. It avoids resource conten-
tion at the edge by leveraging cloud
IoT sensors ...
resources and coordinating the use
of geographically distributed edge
devices.
Figure 1. Distributed data processing in a fog-computing environment. Based on the
FOG COMPUTING desired functionality of a system, users can deploy Internet of Things sensors in differ-
CHARACTERISTICS ent environments including roads, medical centers, and farms. Once the system collects
Fog computing is a distributed para- information from the sensors, fog devices—including nearby gateways and private
digm that provides cloud-like services clouds—­dynamically conduct data analytics.
to the network edge. It leverages cloud
and edge resources along with its own
infrastructure, as Figure 1 shows. In requirements of widely distributed ap- of available cloud, fog, and network
essence, the technology deals with plications that need low latency. resources and identify the best can-
IoT data locally by utilizing clients or didates to process incoming tasks.
edge devices near users to carry out a FOG-COMPUTING With multitenant applications, the
substantial amount of storage, com- COMPONENTS resource-management components
munication, control, configuration, Figure 2 presents a fog-computing prioritize the tasks of the various par-
and management. The approach bene- reference architecture. Fog systems ticipating users or programs.
fits from edge devices’ close proximity generally use the sense-process-actu- Edge and cloud resources commu-
to sensors, while leveraging the on-­ ate and stream-processing program- nicate using machine-to-machine
demand scalability of cloud resources. ming models. Sensors stream data to (M2M) standards such as MQTT (for-
Fog computing involves the compo- IoT networks, applications running on merly MQ Telemetry Transport) and
nents of data-processing or analytics fog devices subscribe to and process the Constrained Application Protocol
applications running in distributed the information, and the obtained in- (CoAP). Software-defined networking
cloud and edge devices. It also facili- sights are translated into actions sent (SDN) helps with the efficient manage-
tates the management and program- to actuators. ment of heterogeneous fog networks.
ming of computing, networking, and Fog systems dynamically dis-
storage services between datacen- cover and use APIs to build com- FOG-COMPUTING
ters and end devices. In addition, it plex functionalities. Components SOFTWARE SYSTEMS
supports user mobility, resource and at the resource-management layer There are four prominent software
interface heterogeneity, and distrib- use information from the resource-­ systems for building fog computing
uted data analytics to address the monitoring service to track the state environments and applications.

AUGUST 2016 41
CLOUD COVER

latency, mobility support, and location

Applications and privacy awareness.7

Smart utility services

Fog computing can be used with smart
utility services,8 whose focus is im-
Programming Sense-process-actuate Stream processing
models proving energy generation, delivery,
and billing. In such environments,
edge devices can report more fine-
grained energy-consumption details
API & service Authorization (for example, hourly and daily, rather
API discovery API composition
management & authentication
than monthly, readings) to users’ mo-
bile devices than traditional smart
utility services. These edge devices
Multitenant resource Operator & flow
Raw data Monitoring & can also calculate the cost of power
management placement and resource
management profiling consumption throughout the day and
scheduling
suggest which energy source is most
economical at any given time or when
Edge & cloud Software- Machine-to-
home appliances should be turned on
defined to minimize utility use.
resources machine
networking

Augmented reality, cognitive

systems, and gaming
IoT sensors & Fog computing plays a major role
actuators in augmented-reality applications,
which are latency sensitive. For exam-
Figure 2. Fog-computing architecture. In the bottom layer are end devices—­including ple, the EEG Tractor Beam augmented
sensors and actuators—along with applications that enhance their functionality. These multiplayer, online brain–computer-­
elements use the next layer, the network, for communicating with edge devices, such as interaction game performs continu-
gateways, and then with cloud services. The resource-management layer runs the entire ous real-time brain-state classification
infrastructure and enables quality-of-service enforcement. Finally, applications leverage on fog devices and then tunes classifi-
fog-computing programming models to deliver intelligent services to users. cation models on cloud servers, based
on electroencephalogram readings
that sensors collect.9
Cisco IOx provides device manage- FOG-COMPUTING A wearable cognitive-assistance
ment and enables M2M services in fog APPLICATIONS system that uses Google Glass devices
environments.5 Using device abstrac- Various applications could benefit helps people with reduced mental
tions provided by Cisco IOx APIs, ap- from fog computing. acuity perform various tasks, includ-
plications running on fog devices can ing telling them the names of people
communicate with other IoT devices Healthcare and activity tracking they meet but don’t remember.10 In
via M2M protocols. Fog computing could be useful in this application, devices communicate
Cisco Data in Motion (DMo) enables healthcare, in which real-time process- with the cloud for delay-tolerant jobs
data management and analysis at the ing and event response are critical. One such as error reporting and logging.
network edge and is built into prod- proposed system utilizes fog comput- For time-sensitive tasks, the system
ucts that Cisco Systems and its part- ing to detect, predict, and prevent falls streams video from the Glass cam-
ners provide. by stroke patients.6 The fall-detection era to the fog devices for processing.
LocalGrid’s fog-computing plat- learning algorithms are dynamically The system demonstrates how using
form is software installed on network deployed across edge devices and cloud nearby fog devices greatly decreases
devices in smart grids. It provides re- resources. Experiments concluded that end-to-end latency.
liable M2M communication between this system had a lower response time
devices and data-processing services and consumed less energy than cloud- MODELING AND SIMULATION
without going through the cloud. only approaches. To enable real-time analytics in fog
Cisco ParStream’s fog-­ computing A proposed fog computing–based computing, we must investigate various
platform enables real-time IoT analytics. smart-healthcare system enables low resource-management and scheduling

42 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R

techniques including the placement, environments because their architec- further development of fog comput-
migration, and consolidation of ture is based on static configurations. ing could thus help the IoT reach its
stream-processing operators, applica- Fog environments require the ability vast potential.
tion modules, and tasks. This signifi- to add and remove resources dynam-
cantly impacts processing latency and ically because processing nodes are REFERENCES
decision-making times. generally mobile devices that fre- 1. R. Buyya and A. Dastjerdi, Internet
However, constructing a real IoT en- quently join and leave networks. of Things: Principles and Paradigms,
vironment as a testbed for evaluating Morgan Kaufmann, 2016.
such techniques is costly and doesn’t Security, reliability, 2. J. Manyika et al., Unlocking the Poten-
provide a controllable environment for and fault tolerance tial of the Internet of Things, McKinsey
conducting repeatable experiments. Enforcing security in fog & Company, 2015.
To overcome this limitation, we devel- environments—­ which have multiple 3. R. Cortés et al., “Stream Processing
oped an open source simulator called service providers and users, as well as of Healthcare Sensor Data: Studying
iFogSim.11 iFogSim enables the model- distributed resources—is a key chal- User Traces to Identify Challenges
ing and simulation of fog-­ computing lenge. Designing and implementing from a Big Data Perspective,” Proce-
environments for the evaluation of re- authentication and authorization tech- dia Computer Science, vol. 52, 2015,
source-management and scheduling niques that can work with multiple fog pp. 1004–1009.
policies across edge and cloud resources nodes that have different computing 4. W. Shi and S. Dustdar, “The Promise
under multiple scenarios, based on their capacities is difficult. Public-key infra- of Edge Computing,” Computer, vol.
impact on latency, energy consumption, structures and trusted execution envi- 49, no. 5, 2016, pp. 78–81.
network congestion, and operational ronments are potential solutions.12 5. F. Bonomi et al., “Fog Computing: A
costs. It measures performance met- Users of fog deployments also must Platform for Internet of Things and
rics and simulates edge devices, cloud plan for the failure of individual sen- Analytics,” Big Data and Internet of
datacenters, sensors, network links, sors, networks, service platforms, and Things: A Roadmap for Smart Environ-
data streams, and stream-processing applications. To help with this, they ments, N. Bessis and C. Dobre, eds.,
applications. could apply standards, such as the Springer, 2014, pp. 169–186.
Stream Control Transmission Proto- 6. Y. Cao et al., “FAST: A Fog Comput-
CHALLENGES col, that deal with packet and event re- ing Assisted Distributed Analytics
Realizing fog computing’s full poten- liability in wireless sensor networks.13 System to Monitor Fall for Stroke
tial presents several challenges in- Mitigation,” Proc. 10th IEEE Int’l Conf.
cluding balancing load distribution Power consumption Networking, Architecture and Storage
between edge and cloud resources, API Fog environments consist of many (NAS 15), 2015, pp. 2–11.
and service management and sharing, nodes. Thus, the computation is dis- 7. V. Stantchev et al., “Smart Items, Fog
and SDN communications. There are tributed and can be less energy effi- and Cloud Computing as Enablers
several other important examples. cient than in centralized cloud sys- of Servitization in Healthcare,” J.
tems. Using efficient communications Sensors & Transducers, vol. 185, no. 2,
Enabling real-time analytics protocols such as CoAP, effective fil- 2015, pp. 121–128.
In fog environments, resource man- tering and sampling techniques, and 8. J. Li et al., “EHOPES: Data-Centered
agement systems should be able to dy- joint computing and network resource Fog Platform for Smart Living,” Proc.
namically determine which analytics optimization can minimize energy 2015 Int’l Telecommunication Networks
tasks are being pushed to which cloud- consumption in fog environments. and Applications Conf. (ITNAC 15),
or edge-based resource to minimize 2015, pp. 308–313.
latency and maximize throughput. 9. J. Zao et al., “Augmented Brain

F
These systems also must consider other og computing enables the seam- Computer Interaction Based on Fog
criteria such as various countries’ data less integration of edge and Computing and Linked Data,” Proc.
privacy laws involving, for example, cloud resources. It supports the 10th IEEE Int’l Conf. Intelligent Envi-
medical and financial information. decentralized and intelligent process- ronments (IE 14), 2014, pp. 374–377.
ing of unprecedented data volumes 10. K. Ha et al., “Towards Wearable
Programming models generated by IoT sensors deployed for Cognitive Assistance,” Proc. 12th Int’l
and architectures smooth integration of physical and cy- Conf. Mobile Systems, Applications, and
Most stream- and data-processing ber environments. Services (MobiSys 14), 2014, pp. 68–81.
frameworks, including Apache Storm This could generate many benefits 11. H. Gupta et al., iFogSim: A Tool-
and S4, don’t provide enough scal- to society by, for example, enabling kit for Modeling and Simulation of
ability and flexibility for fog and IoT smart healthcare applications. The Resource Management Techniques

AUGUST 2016 43
CLOUD COVER

in Internet of Things, Edge and Fog

Computing Environments, tech. report
CLOUDS-TR-2016-2, Cloud Comput-
ing and Distributed Systems Labora-
tory, Univ. of Melbourne, 2016; http://
cloudbus.org/tech_reports.html.
12. I. Stojmenovic and S. Wen, “The Fog
Computing Paradigm: Scenarios and
Security Issues,” Proc. 2014 Federated
Conf. Comp. Sci. and Info. Sys. (FedCSIS
14), 2014, pp. 1–8.
13. H. Madsen et al., “Reliability in the
Utility Computing Era: Towards Reli-
able Fog Computing,” Proc. 20th Int’l
Conf. Systems, Signals, and Image Pro-
cessing (IWSSIP 13), 2013, pp. 43–46.

AMIR VAHID DASTJERDI is a

Research Fellow with the University
of Melbourne’s Cloud Computing
and Distributed Systems (CLOUDS)
Laboratory. Contact him at amir
[email protected].

It’s already at
RAJKUMAR BUYYA is director of
the CLOUDS Laboratory and CEO
of Manjrasoft Pty Ltd., a university
spin-off company that commercial-

your fingertips
izes cloud-computing innovations.
Contact him at rbuyya@unimelb
.edu.au.

If you do computational work,

you are going to love Computing
in Science & Engineering (CiSE).
Because CiSE appears in the
IEEE Xplore and AIP library
packages, representing more
than 50 scientific and engineering
societies, your institution is bound
to have it.

Selected CS articles and

columns are also available for
free at http://ComputingNow
.computer.org.

44 COMPUTER  W W W.CO M P U T E R .O R G /CO M P U T E R