Chapter 12

Future Trends in Cloud Computing

12.1 Revisiting History of Computing

The journey of computing began with a single user running a single job. We have
come a long way since. In the next phase of evolution, multiple users shared a
computer system. It further evolved into a networked computer system, which was
accessible to remote users. With the emergence of PCs (Personal Computers) in the
1990s, we witnessed PCs being used as a gateway to networked computers, as
shown in Fig. 12.1.

12.2 Current Limitations of Cloud Computing

Thus far, we have studied the benefits of Cloud Computing including Economic,
Elastic Infrastructure, On-Demand Resources, Pay for what you use. However,
there are a few unseen costs before these benefits can be realized, and some are
listed below:
(1) Data Movement: Since servers are located in a remote data-center, any input
data needed for computation needs to be moved there, and results need to be
moved out. Such I/O (input–output) transactions cost additional money in most
Public Clouds and add to the latency as compared to computing on local servers.
An example is the emerging area of self-driven cars, which have a multitude of
sensors including multiple cameras. There may not be sufficient time to run the
image processing algorithms in a remote Cloud due to the dynamic nature of
traffic for real-time decision making while driving. Thus, a self-driven car needs
to have server-like computing on board. By some accounts, a self-driven car in
future may generate up to 5 TB of data per day, all of which needs to be stored
and processed locally, representing a mini data-center on the Wheels.

© Springer International Publishing AG, part of Springer Nature 2018 171

N. K. Sehgal and P. C. P. Bhatt, Cloud Computing,
https://doi.org/10.1007/978-3-319-77839-6_12
172 12 Future Trends in Cloud Computing

Fig. 12.1 An example of Client-Server architecture, with multiple users on left side interacting
with a server

(2) Loss of Control: When a user’s e-mails are hosted in the Cloud, these are often
examined by bots, which then decide on relevant advertisements to display, to
generate revenue for e-mail providers such as Google’s Gmail. However, this
raises a question on who owns the e-mail content and who can access it. For
example, if there is a legal case and court subpoenas the e-mail provider to turn
over the e-mails, it will be hard for the provider to say no. At the end, if a user
wishes to own the content and keep it private, such as pictures or other business
data, then it should be kept on a local computer.
(3) Perception of Cloud Security: While multiple people can access a data-center
in Cloud, it may be no less safe than an enterprise data-center. Due to the loss
of control as mentioned previously, there is a perception of Public Cloud being
less secure. This in author’s opinion is a red herring, and additional steps can be
performed such as to encrypt one’s data in the Cloud and also any virtual
machine when running on a multi-tenanted server, with keys stored separately.
(4) Uncertain Performance: Cloud Computing operators make money by sharing
same hardware infrastructure with many customers. While their virtual
machines (VM) may be isolated in the memory and running on different server
cores, there are other shared resources, such as a memory controller, and net-
working card that each VM’s data must pass through. This creates bottlenecks
similar to traffic jams in a data center at entry and exit points, as well as entry
and exit to the shared servers. This causes the performance drop of a running
VM without any notice. This problem has been described previously as a noisy
neighbor and results in a delay in task completion.
All of the above, and a few similar issues, are causing some customers to rethink
their Cloud Computing approaches, such as Hybrid Computing with critical tasks
being performed using onsite infrastructure.
12.3 Emergence of Internet of Things (IoT) 173

12.3 Emergence of Internet of Things (IoT)

Another emerging trend is a Cloud driven by things, versus current Cloud

Computing mostly driven by people, as cameras and wireless sensors are becoming
pervasive. Their applications include Retail Solutions, Transportation and
Automotive, Industrial and Energy, etc. An example of retail industry is Amazon’s
user-facing portals where customers can visualize things and transact them. An
example of Transportation and Automotive is a Software-Defined Cockpit in a
commercial aircraft, or an autonomous vehicle. An example of manufacturing is a
smart factor with robots, or energy savings in a building. Lastly, additional market
segments such as health, print imaging, gaming, and education are being digitized
at an unprecedented rate. The phrase “Internet of things” was first used by British
technology visionary Kevin Aston in 1999. His perception was to think of “objects
in physical world connected by sensors.” Internet Architecture Board (IAB) RFC
7452 provides the definition of IoT, as follows:
• “Internet of Things” (IoT) denotes a trend where a large number of embedded
devices employ communication services offered by Internet protocols. Many of
these devices, often called “smart objects,’’ are not directly operated by humans,
but exist as components in buildings or vehicles, or are spread out in the
environment.
Four basic communication models for IoT are:
1. Device to device,
2. Device to Cloud,
3. Device to gateway,
4. Backend data sharing model,
We are more interested in #2 and #4, as both involve Cloud services. An
example is shown in Fig. 12.2, of home appliances such as a thermostat-controlled
A/C connected to Cloud for better energy management.

Fig. 12.2 Cloud-based energy management, monitoring, and optimization

174 12 Future Trends in Cloud Computing

Fig. 12.3 Heat map of key IoT opportunities by industries and applications

Bains [1] predicts that by 2020 annual revenues for the IoT vendors could
exceed $470B by selling hardware, software and comprehensive solutions.
Forrester [2] published a heat map in 2016, showing how opportunities vary by the
industry and applications, as shown in Fig. 12.3.
As seen above, the hottest (i.e., most financially attractive) applications are in
transportation, government, and retail. Further discussion of IoT business oppor-
tunities is beyond the scope of this book, but can be found in [2]. These new
opportunities also bring new security challenges. As an example, if these devices
are connected to Internet then a hacker can potentially gain access to read the output
data or alter device configurations to yield unexpected results. We will explain the
security implications and potential solutions in a later section.

12.4 Emergence of Machine Learning

Due to the preponderance of IoT data being generated, it is nearly impossible for a
human to draw any meaningful comparisons. This is reviving expert systems and
artificial intelligence (AI), this time aided by unprecedented compute power and
self-learning systems that improve with more incoming data. Some of the use cases
for IoT-based machine learning are shown in Fig. 12.4, where a smart meter and
building temperature control based on when its occupants are expected to arrive or
leave. Furthermore, different parts of buildings where people are present or absent
can be heated or cooled at different levels, instead of a single setting for the whole
floor.
12.4 Emergence of Machine Learning 175

ADAS, Autonomus Driving, Smart

Parking
Real-!me, Safety systems, parking informa!on, billing,
tra"c mgmt

Consumer
Agriculture Fitness devices, healthcare
op!mize the general monitoring, personal trackers
agriculture produc!on (child/elderly)
including crops and livestock

Smart Home Logis!cs

Ac!on monitoring/access
IoT Tracking of the goods,
control etc Use cases cargo management

Environmental
U!li!es Monitoring
Smart metering, e.g Monitoring of air quality,
electricity, water, water quality , forest fire
gas detec!on, snow condi!ons
etc
Point of Sales
terminals
Payment transac!ons

Fig. 12.4 IoT use case that needs both local and Cloud Computing

At a basic level, we can add intelligence to remote devices if processing ele-

ments and storage are used to locally collect data for rule-based control decisions.
As an example, in a company cafeteria with multiple work-shifts and different
number of employees served on different days, an intelligent refrigerator can check
for the remaining packaged food items, including their expiration dates. If the goal
is to have food ingredients for at least next 2 days in store, cafeteria manager can be
notified to replenish them as needed. Machine learning part comes from not having
a predetermined supply at hand, but making the solution self-learning based on
consumption pattern of employees. If it is a Friday, and company is closed for the
next 2 days, then system will look for supplies needed until the following Tuesday.
Also, on different work days, menu and specific food items needed may vary,
requiring a solution that can intelligently predict what needs to be ordered to
minimize expenses and avoid food wastage, while ensuring that essential ingredi-
ents will never run. In general, such smart devices offer desired functionality and
operate in an energy-efficient manner with minimal compute power and memory,
while connected to a mobile App and Cloud on the backend. These connections are
needed for human intervention and record keeping.
Current machine learning technology has its limitations. A case in point is an
automobile accident [3] in which driver was on a self-driving system, but vehicle’s
camera failed to recognize a white truck reflecting a bright sky, and the car failed to
brake. However, regulatory authorities absolved the automaker and blamed this
176 12 Future Trends in Cloud Computing

accident on dead driver. They ruled that driver should have paid attention and not
be depended on the self-driving system. Future liability in accidents will be hotly
contested.
Machine learning systems have proven useful in retail as the vendors can find the
items that customers are buying, or not, and accordingly build next production
order. In addition, they can build customer profiles and suggest additional items to
customers who buy an item, based on what others bought after buying the same
item. This has contributed to enormous success for online retailers such as Amazon.

12.5 Emergence of Edge Computing

With many IoT devices and use-cases, it is imperative to have localized compute
power and data storage. An example is a car, as shown in Fig. 12.5, which can
generate up to 5 TB of data/day. This comes from onboard cameras, IR sensors, and
data collected from the engine, brakes, etc. However, an autonomous car cannot
pause for a server in the Cloud to make a decision to accelerate or brake. Hence, it
needs sufficient compute power in the car to drive safely, by some to dub it as a
“Data center on the wheels.” It can synch up with a remote data center in the Cloud
overnight while parked, but on the road must focus on safe driving with real-time
decision making. Hence, a part of the Cloud is migrated from remote data-center to
field, termed as Edge Computing.
Similar examples can be found in other application domains, such as smart
homes with security cameras, which can decide on the spot if an intruder is a family
member or a stranger, and in the latter case sound an alarm.

Fig. 12.5 A car’s self-driving system with multiple sensors [3]

12.6 Security Issues in Edge Computing 177

12.6 Security Issues in Edge Computing

Security concerns abound with the emergence of Edge Computing. In the car
example, its computers are not behind a firewall but physically accessible to many
people besides the owner. When a car is taken to a mechanic for an oil change or
another repair, there is a risk of someone tempering with the hardware or software
components setting up a future failure of the self-driven car. It is also possible for
someone to access private data stored in the car, e.g., its travel points.
Vulnerabilities in other unprotected devices, such as home appliances (TV, Fridge)
on a network, can be used to launch a cyber attack. A recent DDOS (Distributed
Denial of Service) attack was launched using hijacked home security cameras,
while in another instance private video clips were stolen and posted on Internet.
Even for a simple home automation system, such as an intelligent door lock, it
needs following security features for safety:
(1) A firewall to dissuade remote hackers with login authentication.
(2) Authentication requires identification of phone numbers, password, or bio-
metrics such as face recognition, thumbprint, or retina scan.
Note that any single biometric can be easily defeated, e.g., a pictured mask to
fool a face recognition, or copy of a thumb print image, presented to the door
camera. It is desirable to have a multi-factored authentication system. Furthermore,
a data-logging system is needed to record who opened or locked the door, and
when. This data is immediately backed to a remote Cloud to avoid local tempering.
Machine intelligence can be used to create a regular usage pattern and flag anomaly
if door is opened at unexpected hours or with unusual frequency.
We need to remember that IoT devices are constantly collecting data about an
environment or individuals, which can be potentially shared with third parties
compromising privacy. It can range from personal preferences of Web-browsing
habits, TV channels selection, or images from home security cameras. Some
devices can be programmed to selectively transmit data to a Cloud Service for
processing, e.g., a security camera which has a buffer of 15 s, but records and
transmits a 30 s of clip only if any motion is detected, for 15 s before and 15 s after
the motion is detected. This reduces storage requirements but increases chances of a
mistake. Such devices are designed to render service with minimal intervention, and
yet they need to be directed using voice activation or image recognition. On the flip
side, if there is a continuous recording dashcam, which is a forward-looking
recording device in a car. Purpose of this dashcam is to establish other party’s guilt
in case of an accident in a vehicle. It will also record voice conversations of
passengers potentially violating their privacy rights. It is recommended for the
vehicle driver to inform passengers and seek their consent in advance to make them
aware.
For ensuring trust in Edge Computing, it has to start with a trusted environment,
trusted protocols, and temper proof components. Vendors need to provide
“anti-temper” solutions to start with. Software upgrades in field are needed for any
178 12 Future Trends in Cloud Computing

bug fixes during the lifetime of an Edge Computing device. A secure channel must
exist to provide signed binary packets that are transmitted and installed in the field,
e.g., on a car or TV at home. In our door example, vendor needs to provide an
anti-temper solution, to prevent someone locally changing the firmware or settings
in an unauthorized manner. Even remote software upgrades are authenticated, as
unprotected home appliances can be used to launch cyber attacks, e.g., someone
opening doors via remote Internet attacks. Besides security, there are privacy
concerns, as home sensors are collecting data about individuals that can be shared
with third parties for commercial purposes.
Undesirable consequence may emerge if a third party can remotely gain control
of a self-driven car causing an accident on the road, or someone with malice can
access the medicine drip-meters in a hospital with fatal consequences for the
patients. This can be avoided with a balanced approach to interoperability and
access control. This needs to be addressed at different layers of architecture and
within the protocol stacks between the devices. Standardization and adoption of
communication protocols should specify when it is optimal to have standards. Some
vendors like to create a proprietary ecosystem of compatible IoT products. This
creates user lock-in to their particular ecosystem, which from a vendor’s point of
view is desirable because a closed ecosystem approach can offer benefits of security
and reduces costs. However, from a user’s point of view, such practices can create
interoperability problems with solutions from other vendors, thereby limiting user’s
choices in case of upgrades or future system expansion.
Solution-level cost considerations involve technical factors such as limited
internal processing, memory resources, or power consumption demands. Vendors
try to reduce the unit cost of devices by minimizing parts and product design costs.
It may be more expensive to design interoperability features into a product and test
for compliance with a standards specification. A non-interoperable device may lack
in standards and the documented best practices. It may limit the potential use of IoT
device, and the absence of these standards can result in deviant behavior by IoT
devices.

12.7 Security Considerations for Edge Computing

Edge Computing is the most recent inflection point in the history of computing.
With the advent of Edge Computing, evolutionary cycles between a concentration
of powerful centralized computing and an emphasis on distributed powerful com-
puting have changed to a network made of a combination of powerful centralized
powerful computing and distribution of simple computers at the edges of the net-
work. This network has vastly different security requirements. For example, a
central system in the Cloud can send security breaches to the edge, or the edge
computers can send security breaches to a server in the Cloud. A system-wide trust
is difficult to achieve based upon the current start of art strategies, policies,
12.7 Security Considerations for Edge Computing 179

standards, or implementations. Edge Computing expands the potential threat plane

and introduces the possibility of attacks from many directions.
Following classifications describe the types of security issues as related to the
Edge Computing, with a few examples:
1. Identity authentication: By definition, the number of players in Edge
Computing is large and these may not belong to the same organization. It is
infeasible to verify their identity in a foolproof manner. Trust needs to be
extended, as new customers buy their devices, such as security cameras, and
bring these online with a remote registration. Central authority then must depend
on the ability of these remote customers to protect their own devices.
2. Unauthorized access: Depending on the nature of sensors at the Edge, their
access into data-center may be bidirectional in nature. If someone hacks into a
remote device to impersonate a previously trusted remote device, it is nearly
impossible to differentiate between genuine or fake users. Similarly, someone
pretending to act as a central computer can access the remote devices and get
critical user-data, such as on remote medical devices.
3. Denial of service attacks: An attack launched by hijacking multiple remote
devices and simultaneously contacting the central server. This will cause the
server to be overloaded, denying access to a genuine device in a timely manner.
4. Data theft: Depending on where data is stored and for how long opens the
possibility of it being stolen. An example is a security camera at home with local
storage. In event of a theft, it may be possible for an intruder to simply remove
the local storage, thus circumventing the purpose of a security camera.
However, if camera immediately uploads an image to Cloud upon detecting a
motion, then any physical tempering will still protect the image of intruders.
5. Data integrity and falsification: A key difference between confidentiality and
integrity is that in the latter case, an attacker does not need to read the protected
data, but merely modify it, e.g., with a buffer overflow, rendering it useless. This
system level attack can happen if multiple devices from different sources are
writing back to a central server memory or database. This can be protected with
assigning a virtual partition or container to the data coming from each distinct
source, and checking the address range of each access to prevent data integrity
of other users on the same server.
6. Invasion of privacy: Since multiple players may combine their data inputs from
different sources to arrive at a desired conclusion, e.g., for real-time traffic
updates, their identities need to be protected. This may include an individual’s
location, movements, and any other aspects of personal nature.
7. Activity monitoring: A simple example of cell phone which constantly pings
the signal tower is sufficient for someone to monitor the location of phone’s
owners, their movements, etc. Furthermore, if a remote App can turn on the
microphone or camera in a phone, then additional information and activities can
be monitored in an illegal manner. Similar effects can be achieved with fixed
cameras at commercial or public locations, e.g., in a shopping center.
180 12 Future Trends in Cloud Computing

It is recognized that traditional Trusted Compute Boundary (TCB) expands with

Edge Computing to include domains that are physically outside the control of
remote device or central data-center owners. The best they can do is to monitor/
track a threat, identify an attacker, launch a recovery, and prevent false positives.
These steps are outlined below:
1. Monitor/track a threat: This is possible by establishing a normal usage pattern
and then looking for anomalies. Any deviation represents a potential threat.
2. Identifying attackers: Once a threat is detected, then attackers need to be
identified. These could take the form of an IP address that is repeatedly pinging
the central server, to launch a denial-of-service attack.
3. Attack recovery: This can take the form of blocking the offending IP address.
However, situation is not always so simple as an attacker can corrupt the critical
data before the attacker’s presence is detected. In such a case, frequent check-
points must be taken to do a rollback to the known good state.
4. Accidental and unintentional failures confused with security attacks: Any
detection method suffers from the risks of false positives, e.g., mistaken flagging
of a genuine access as a potential threat. An example of this is a stock market
trading computer that detects unusual activity, which is genuine yet may flag a
false alarm. Similar situation can happen with security alarms due to false sensor
activity data, etc. This calls for a learning system that becomes smarter over
time.

12.8 Future Work Needed

Internet Engineering Task Force (IETF) has identified the problem of interoper-
ability, as many suppliers build “walled gardens” that limit users to interoperate
with a curated subset of component providers, applications, and services.
Interoperability solutions between IoT devices and backend systems can exist at
different layers of the architecture and at different levels within protocol stack
between the devices. Key is the standardization and adoption of protocols, which
should specify when and where it is optimal to use standards. More work is needed
to ensure interoperability within the cost constraints for Edge Computing to become
pervasive.
There are other regulatory and policy issues at play, such as device data being
collected and stored in a Cloud may cross-jurisdictional boundaries, raising liability
issues if the data leaks. This is especially important if data is of personal nature,
e.g., related to shopping patterns or patient health records.
12.9 Example of an IoT-Based Cloud Service 181

12.9 Example of an IoT-Based Cloud Service

A Cloud Service where intelligence extends beyond a data-center to the edge based
sensors is also known as Fog Computing [4], which is a clever name for gathering
and processing data at the local computing devices. In this model, sensors and other
connected devices such as cameras send data to a nearby edge-computing device,
which has processing power to analyze this data, make some local decisions, and
then send the results to the Cloud. BI Intelligence forecasts that 5.8 Billion IoT
devices owned by enterprises and governments will use Fog Computing in 2020, up
from 570 million devices in 2015 (Fig. 12.6).
An example comes from mining industry [5], where drilling equipment is
working in harsh conditions, with autonomous trucks and trains, tunneling and
boring machines, moving at high speeds. In order to ensure worker safety and
increase productivity, decisions need to be made locally. Even though mining
equipment can generate terabytes of data/hour during normal operation, there may
not be a reliable connection to backend Cloud given 100 s of feet of underground
operation, say in a coal mine. This is where Fog Computing can help by processing
the data locally, makes appropriate decisions, and, as shown in Fig. 12.7, sends
only small uploads to the backend Cloud every few hours or at the end of each day.
However, this also increases risk of accountability and security as different legal
entities may own the local sensors, edge gateway, and backend Cloud. If something
goes wrong, e.g., in the event of a mining accident, then finger pointing will begin
with hard to assign liabilities. This is where companies offering end-to-end services

Fig. 12.6 Growth of data being generated by IoT and Cloud together [4]
182 12 Future Trends in Cloud Computing

Fig. 12.7 An example of Fog Computing to support real-time decisions [5]

will have an advantage in Edge Computing deployment resulting in a business

success. In the long run, as industry evolves and standards emerge, there will be a
room for horizontal service and equipment providers to excel at competitive price
points, but initially Fog Computing is likely to be a vertical play.

12.10 Summary

Combination of locally intelligent devices with backend Cloud-based processing is

giving rise to a new class of Edge or Fog Cloud Computing, which offers new usage
models, but also raises potential of new vulnerabilities with possibility of wide-
spread cyber attacks. There are additional concerns of user lock-ins if vendors do
not follow interoperability standards in their edge-based devices in proprietary
Cloud solutions. Additional issues of user-data privacy and legal jurisdiction cur-
rently lag the fast evolution of Edge Computing domain with IoT-based solutions.
This requires policy framework to be discussed by vendors and Cloud Service
Providers with the users for avoiding any legal pitfalls.
As shown in the historic computing spiral of Fig. 12.1, industry has oscillated
between large central computers, to localized computing and hybrid models leading
12.10 Summary 183

back to the spiral growth. This now requires large central computers to handle the
distributed Edge Computing demand.
This trend is likely to continue as networks will become faster and machines will
become more intelligent to recognize patterns of data to make decisions. In this
evolution, it is important to develop standards for interoperability of computing
devices on the edge and servers on the backend, to ensure a level-playing field for
all players.

12.11 Points to Ponder

1. There is potential to have more devices and machines in an increasingly

automated world, and next wave of Cloud Computing growth is coming
from IoT. Can you list additional areas to drive the growth of Cloud
Computing?
2. How could one improve the Cloud’s performance and support for IoT?
3. Why is Edge Computing needed for self-driven cars in future?
4. Can you think of another example of Edge Computing devices on a road?
5. What is the trust and security model for edge devices?
6. What kinds of attacks are possible using IoT and Edge devices?
7. What is the impact of vendor lock-in on Edge Computing devices?

References

1. https://www.forbes.com/sites/louiscolumbus/2016/11/27/roundup-of-Internet-of-things-forecasts-
and-market-estimates-2016/#634d80ab292d
2. https://www.cloudera.com/content/dam/www/static/documents/analyst-reports/forrester-the-iot-
heat-map.pdf
3. https://www.nytimes.com/2017/01/19/business/tesla-model-s-autopilot-fatal-crash.html
4. http://www.businessinsider.com/Internet-of-things-cloud-computing-2016-10
5. http://www.nanalyze.com/2016/08/fog-computing-examples/