<s>
Edge	B-Device
computing	I-Device
is	O
a	O
distributed	B-Architecture
computing	I-Architecture
paradigm	O
that	O
brings	O
computation	O
and	O
data	B-General_Concept
storage	I-General_Concept
closer	O
to	O
the	O
sources	O
of	O
data	O
.	O
</s>
<s>
Edge	B-Device
computing	I-Device
is	O
an	O
architecture	O
rather	O
than	O
a	O
specific	O
technology	O
,	O
and	O
a	O
topology	B-Architecture
-	O
and	O
location-sensitive	O
form	O
of	O
distributed	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
The	O
origins	O
of	O
edge	B-Device
computing	I-Device
lie	O
in	O
content	O
distributed	O
networks	O
(	O
CDNs	B-Operating_System
)	O
that	O
were	O
created	O
in	O
the	O
late	O
1990s	O
to	O
serve	O
web	O
and	O
video	O
content	O
from	O
edge	O
servers	O
that	O
were	O
deployed	O
close	O
to	O
users	O
.	O
</s>
<s>
In	O
the	O
early	O
2000s	O
,	O
these	O
networks	O
evolved	O
to	O
host	O
applications	O
and	O
application	O
components	O
on	O
edge	O
servers	O
,	O
resulting	O
in	O
the	O
first	O
commercial	O
edge	B-Device
computing	I-Device
services	O
that	O
hosted	O
applications	O
such	O
as	O
dealer	O
locators	O
,	O
shopping	O
carts	O
,	O
real-time	O
data	O
aggregators	O
,	O
and	O
ad	O
insertion	O
engines	O
.	O
</s>
<s>
The	O
history	O
of	O
how	O
CDNs	B-Operating_System
created	O
the	O
first	O
edge	O
networks	O
that	O
later	O
evolved	O
into	O
the	O
first	O
edge	B-Device
computing	I-Device
service	O
in	O
2002	O
,	O
the	O
name	O
itself	O
coined	O
as	O
a	O
trademark	O
for	O
that	O
service	O
,	O
was	O
told	O
in	O
an	O
of	O
ACM	O
SIGCOMM	O
's	O
.	O
</s>
<s>
marking	O
the	O
25th	O
anniversary	O
of	O
the	O
birth	O
of	O
CDNs	B-Operating_System
.	O
</s>
<s>
Internet	B-Operating_System
of	I-Operating_System
things	I-Operating_System
(	O
IoT	B-Operating_System
)	O
is	O
an	O
example	O
of	O
edge	B-Device
computing	I-Device
.	O
</s>
<s>
A	O
common	O
misconception	O
is	O
that	O
edge	O
and	O
IoT	B-Operating_System
are	O
synonymous	O
.	O
</s>
<s>
One	O
definition	O
of	O
edge	B-Device
computing	I-Device
is	O
the	O
use	O
of	O
any	O
type	O
of	O
computer	B-Application
program	I-Application
that	O
delivers	O
low	O
latency	O
nearer	B-General_Concept
to	I-General_Concept
the	I-General_Concept
requests	I-General_Concept
.	O
</s>
<s>
Karim	O
Arabi	O
,	O
in	O
an	O
IEEE	O
DAC	O
2014	O
Keynote	O
and	O
subsequently	O
in	O
an	O
invited	O
talk	O
at	O
MIT	O
's	O
MTL	O
Seminar	O
in	O
2015	O
,	O
defined	O
edge	B-Device
computing	I-Device
broadly	O
as	O
all	O
computing	O
outside	O
the	O
cloud	O
happening	O
at	O
the	O
edge	O
of	O
the	O
network	O
,	O
and	O
more	O
specifically	O
in	O
applications	O
where	O
real-time	O
processing	O
of	O
data	O
is	O
required	O
.	O
</s>
<s>
In	O
his	O
definition	O
,	O
cloud	B-Architecture
computing	I-Architecture
operates	O
on	O
big	B-Application
data	I-Application
while	O
edge	B-Device
computing	I-Device
operates	O
on	O
"	O
instant	O
data	O
"	O
that	O
is	O
real-time	O
data	O
generated	O
by	O
sensors	O
or	O
users	O
.	O
</s>
<s>
The	O
term	O
is	O
often	O
used	O
as	O
synonymous	O
with	O
fog	B-Device
computing	I-Device
.	O
</s>
<s>
However	O
,	O
when	O
the	O
deployment	O
size	O
is	O
large	O
,	O
e.g.	O
,	O
for	O
Smart	B-Device
Cities	I-Device
,	O
fog	B-Device
computing	I-Device
can	O
be	O
a	O
distinct	O
layer	O
between	O
the	O
Edge	O
and	O
the	O
Cloud	O
.	O
</s>
<s>
According	O
to	O
The	O
State	O
of	O
the	O
Edge	O
report	O
,	O
edge	B-Device
computing	I-Device
concentrates	O
on	O
servers	O
"	O
in	O
proximity	O
to	O
the	O
last	O
mile	O
network	O
"	O
.	O
</s>
<s>
Alex	O
Reznik	O
,	O
Chair	O
of	O
the	O
ETSI	O
MEC	O
ISG	O
standards	O
committee	O
,	O
loosely	O
defines	O
the	O
term	O
by	O
essentially	O
suggesting	O
that	O
anything	O
that	O
's	O
not	O
a	O
traditional	O
data	B-Operating_System
center	I-Operating_System
could	O
be	O
the	O
'	O
edge	O
 '	O
for	O
somebody	O
.	O
</s>
<s>
Per	O
Anand	O
and	O
Edwin	O
say	O
"	O
the	O
edge	O
node	O
is	O
mostly	O
one	O
or	O
two	O
hops	O
away	O
from	O
the	O
mobile	O
client	O
to	O
meet	O
the	O
response	O
time	O
constraints	O
for	O
real-time	O
games	O
 '	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
gaming	O
context.	O
"	O
</s>
<s>
Edge	B-Device
computing	I-Device
may	O
employ	O
virtualization	B-General_Concept
technology	O
to	O
make	O
it	O
easier	O
to	O
deploy	O
and	O
run	O
a	O
wide	O
range	O
of	O
applications	O
on	O
edge	O
servers	O
.	O
</s>
<s>
According	O
to	O
research	O
firm	O
Gartner	O
,	O
around	O
10%	O
of	O
enterprise-generated	O
data	O
is	O
created	O
and	O
processed	O
outside	O
a	O
traditional	O
centralized	O
data	B-Operating_System
center	I-Operating_System
or	O
cloud	O
.	O
</s>
<s>
The	O
increase	O
of	O
IoT	B-Operating_System
devices	O
at	O
the	O
edge	O
of	O
the	O
network	O
is	O
producing	O
a	O
massive	O
amount	O
of	O
data	O
-	O
storing	O
and	O
using	O
all	O
that	O
data	O
in	O
cloud	O
data	B-Operating_System
centers	I-Operating_System
pushes	O
network	O
bandwidth	O
requirements	O
to	O
the	O
limit	O
.	O
</s>
<s>
Despite	O
the	O
improvements	O
of	O
network	O
technology	O
,	O
data	B-Operating_System
centers	I-Operating_System
cannot	O
guarantee	O
acceptable	O
transfer	O
rates	O
and	O
response	O
times	O
,	O
which	O
,	O
however	O
,	O
often	O
is	O
a	O
critical	O
requirement	O
for	O
many	O
applications	O
.	O
</s>
<s>
Furthermore	O
,	O
devices	O
at	O
the	O
edge	O
constantly	O
consume	O
data	O
coming	O
from	O
the	O
cloud	O
,	O
forcing	O
companies	O
to	O
decentralize	O
data	B-General_Concept
storage	I-General_Concept
and	O
service	O
provisioning	O
,	O
leveraging	O
physical	O
proximity	O
to	O
the	O
end	O
user	O
.	O
</s>
<s>
In	O
a	O
similar	O
way	O
,	O
the	O
aim	O
of	O
edge	B-Device
computing	I-Device
is	O
to	O
move	O
the	O
computation	O
away	O
from	O
data	B-Operating_System
centers	I-Operating_System
towards	O
the	O
edge	O
of	O
the	O
network	O
,	O
exploiting	O
smart	B-General_Concept
objects	I-General_Concept
,	O
mobile	B-Application
phones	I-Application
,	O
or	O
network	B-Application
gateways	I-Application
to	O
perform	O
tasks	O
and	O
provide	O
services	O
on	O
behalf	O
of	O
the	O
cloud	O
.	O
</s>
<s>
By	O
moving	O
services	O
to	O
the	O
edge	O
,	O
it	O
is	O
possible	O
to	O
provide	O
content	O
caching	B-General_Concept
,	O
service	O
delivery	O
,	O
persistent	O
data	B-General_Concept
storage	I-General_Concept
,	O
and	O
IoT	B-Operating_System
management	O
resulting	O
in	O
better	O
response	O
times	O
and	O
transfer	O
rates	O
.	O
</s>
<s>
The	O
distributed	O
nature	O
of	O
this	O
paradigm	O
introduces	O
a	O
shift	O
in	O
security	O
schemes	O
used	O
in	O
cloud	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
In	O
edge	B-Device
computing	I-Device
,	O
data	O
may	O
travel	O
between	O
different	O
distributed	O
nodes	O
connected	O
through	O
the	O
Internet	O
and	O
thus	O
requires	O
special	O
encryption	O
mechanisms	O
independent	O
of	O
the	O
cloud	O
.	O
</s>
<s>
Scalability	B-Architecture
in	O
a	O
distributed	O
network	O
must	O
face	O
different	O
issues	O
.	O
</s>
<s>
First	O
,	O
it	O
must	O
take	O
into	O
account	O
the	O
heterogeneity	O
of	O
the	O
devices	O
,	O
having	O
different	O
performance	O
and	O
energy	O
constraints	O
,	O
the	O
highly	O
dynamic	O
condition	O
,	O
and	O
the	O
reliability	O
of	O
the	O
connections	O
compared	O
to	O
more	O
robust	O
infrastructure	O
of	O
cloud	O
data	B-Operating_System
centers	I-Operating_System
.	O
</s>
<s>
Moreover	O
,	O
security	O
requirements	O
may	O
introduce	O
further	O
latency	O
in	O
the	O
communication	O
between	O
nodes	O
,	O
which	O
may	O
slow	O
down	O
the	O
scaling	B-Architecture
process	O
.	O
</s>
<s>
The	O
state-of-the-art	O
scheduling	O
technique	O
can	O
increase	O
the	O
effective	O
utilization	O
of	O
edge	O
resources	O
and	O
scales	O
the	O
edge	O
server	B-Application
by	O
assigning	O
minimum	O
edge	O
resources	O
to	O
each	O
offloaded	O
task	O
.	O
</s>
<s>
Management	O
of	O
failovers	B-General_Concept
is	O
crucial	O
in	O
order	O
to	O
keep	O
a	O
service	O
alive	O
.	O
</s>
<s>
Moreover	O
,	O
edge	B-Device
computing	I-Device
systems	O
must	O
provide	O
actions	O
to	O
recover	O
from	O
a	O
failure	O
and	O
alert	O
the	O
user	O
about	O
the	O
incident	O
.	O
</s>
<s>
To	O
this	O
aim	O
,	O
each	O
device	O
must	O
maintain	O
the	O
network	B-Architecture
topology	I-Architecture
of	O
the	O
entire	O
distributed	B-Architecture
system	I-Architecture
,	O
so	O
that	O
detection	O
of	O
errors	O
and	O
recovery	O
become	O
easily	O
applicable	O
.	O
</s>
<s>
As	O
an	O
example	O
,	O
an	O
edge	B-Device
computing	I-Device
device	O
,	O
such	O
as	O
a	O
voice	B-Protocol
assistant	I-Protocol
,	O
may	O
continue	O
to	O
provide	O
service	O
to	O
local	O
users	O
even	O
during	O
cloud	B-Architecture
service	I-Architecture
or	O
internet	O
outages	O
.	O
</s>
<s>
Edge	B-Device
computing	I-Device
brings	O
analytical	O
computational	O
resources	O
close	O
to	O
the	O
end	O
users	O
and	O
therefore	O
can	O
increase	O
the	O
responsiveness	O
and	O
throughput	O
of	O
applications	O
.	O
</s>
<s>
A	O
well-designed	O
edge	O
platform	O
would	O
significantly	O
outperform	O
a	O
traditional	O
cloud-based	B-Architecture
system	O
.	O
</s>
<s>
Some	O
applications	O
rely	O
on	O
short	O
response	O
times	O
,	O
making	O
edge	B-Device
computing	I-Device
a	O
significantly	O
more	O
feasible	O
option	O
than	O
cloud	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
Examples	O
range	O
from	O
IoT	B-Operating_System
to	O
autonomous	O
driving	O
,	O
anything	O
health	O
or	O
human	O
/	O
public	O
safety	O
relevant	O
,	O
or	O
involving	O
human	O
perception	O
such	O
as	O
facial	O
recognition	O
,	O
which	O
typically	O
takes	O
a	O
human	O
between	O
370-620ms	O
to	O
perform	O
.	O
</s>
<s>
Edge	B-Device
computing	I-Device
is	O
more	O
likely	O
to	O
be	O
able	O
to	O
mimic	O
the	O
same	O
perception	O
speed	O
as	O
humans	O
,	O
which	O
is	O
useful	O
in	O
applications	O
such	O
as	O
augmented	O
reality	O
where	O
the	O
headset	O
should	O
preferably	O
recognize	O
who	O
a	O
person	O
is	O
at	O
the	O
same	O
time	O
as	O
the	O
wearer	O
does	O
.	O
</s>
<s>
Additionally	O
,	O
the	O
usage	O
of	O
edge	B-Device
computing	I-Device
as	O
an	O
intermediate	O
stage	O
between	O
client	O
devices	O
and	O
the	O
wider	O
internet	O
results	O
in	O
efficiency	O
savings	O
that	O
can	O
be	O
demonstrated	O
in	O
the	O
following	O
example	O
:	O
A	O
client	O
device	O
requires	O
computationally	O
intensive	O
processing	O
on	O
video	O
files	O
to	O
be	O
performed	O
on	O
external	O
servers	O
.	O
</s>
<s>
Another	O
example	O
is	O
voice	B-Application
recognition	I-Application
.	O
</s>
<s>
Computation	B-Operating_System
offloading	I-Operating_System
for	O
real-time	O
applications	O
,	O
such	O
as	O
facial	O
recognition	O
algorithms	O
,	O
showed	O
considerable	O
improvements	O
in	O
response	O
times	O
,	O
as	O
demonstrated	O
in	O
early	O
research	O
.	O
</s>
<s>
Further	O
research	O
showed	O
that	O
using	O
resource-rich	O
machines	O
called	O
cloudlets	B-Operating_System
or	O
micro	O
data	B-Operating_System
centers	I-Operating_System
near	O
mobile	O
users	O
,	O
which	O
offer	O
services	O
typically	O
found	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
,	O
provided	O
improvements	O
in	O
execution	O
time	O
when	O
some	O
of	O
the	O
tasks	O
are	O
offloaded	O
to	O
the	O
edge	O
node	O
.	O
</s>
<s>
IoT-based	O
power	O
grid	O
system	O
enables	O
communication	O
of	O
electricity	O
and	O
data	O
to	O
monitor	O
and	O
control	O
the	O
power	O
grid	O
,	O
which	O
makes	O
energy	O
management	O
more	O
efficient	O
.	O
</s>
<s>
Another	O
use	O
of	O
the	O
architecture	O
is	O
cloud	B-Application
gaming	I-Application
,	O
where	O
some	O
aspects	O
of	O
a	O
game	O
could	O
run	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
,	O
while	O
the	O
rendered	O
video	O
is	O
transferred	O
to	O
lightweight	O
clients	O
running	O
on	O
devices	O
such	O
as	O
mobile	B-Application
phones	I-Application
,	O
VR	O
glasses	O
,	O
etc	O
.	O
</s>
<s>
Other	O
notable	O
applications	O
include	O
connected	O
cars	O
,	O
autonomous	O
cars	O
,	O
smart	B-Device
cities	I-Device
,	O
Industry	B-General_Concept
4.0	I-General_Concept
(	O
smart	O
industry	O
)	O
,	O
and	O
home	O
automation	O
systems	O
.	O
</s>
