►
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
You’re probably in a room with at least a dozen objects that are connected to the internet. And if you’re not, it won’t be long before you are. All that connectivity, and all of those devices, require a new way of thinking about cloud infrastructure, and that new way is called Edge Computing.
But creating effective edge computing infrastructure requires much more than just creating a bunch of small-footprint clouds and connecting them to a central core.
Join Mirantis Director of Engineering Satish Salagame for an in-depth technical discussion of:
- What edge computing actually is and why it matters
- How edge enables 5G and why your phone might already be using it
- Why traditional enterprises may find themselves needing edge computing
- Which open source technologies are involved in making edge computing possible
- How Kubernetes fits into the Edge Cloud architecture
- Technological issues you must resolve in order to successfully implement edge computing
A
A
Thank
Johnny
we'd
like
to
welcome
our
presenters,
Satish
salami
and
Nick
chase
from
mer
antes.
The
topic
is
edge
computing
and,
as
you
know,
that's
an
interesting
area
that
looks
at
distributed
cloud
deployments
and
we
are
really
looking
forward
to
hearing
from
Satish
during
the
webinar.
You
will
not
be
able
to
talk
as
an
attendee,
so
please
look
for
the
QA
box
and
drop
your
questions
in
there
and
we'll
get
to
as
many
as
we
can.
A
B
Right
so
good
morning,
everyone
or
whatever
time
it
is
where
you
are
I-
am
Nick
chase
former
antis,
as
you
said,
I'm
here,
to
introduce
Satish
salami
again.
He
is
our
director
of
engineering
and
networking
here
at
moran.
'this
is
background,
is
in
networking
and
design
and
all
kinds
of
things
that
really
set
him
in
good
stead
for
being
responsible
for
edge
cloud
solutions
so
good
morning,
Satish,
it's
good
to
have
you
here.
Thank.
C
You
very
much
Nick
for
the
introduction,
hello,
everyone
and
thank
you
very
much
for
joining
this
webinar
today.
So
let's
go
ahead
and
look
at
the
agenda
for
today,
so
we
are
going
to
get
started
with
each
count.
Computing
talk
about
what
edge
computing
is
and
why
it
matters.
We
look
at
a
few
important
use
cases
that
are
critical
for
edge
computing.
C
We
look
at
key
benefits
that
we
can
get
from
edge
computing
and
based
on
that,
take
a
look
at
what
are
some
of
the
key
infrastructure
requirements
to
make
this
happen
and
and
then
we
get
into
more
interesting
topics
like
how
does
kubernetes
fit
into
the
edge
architecture
and,
more
importantly,
if
we
have
VM
based
vnfs,
this
is
the
NFB
world.
If
we
have
VM
based
vnfs
in
a
kubernetes
edge,
what
are
some
of
the
challenges
and
and
what
kind
of
interesting
solutions
exist
in
that
area?
B
C
That's
a
good
way
to
start
turning,
so
there
are,
there
are
actually
multiple
way
is
to
look
at
edge
computing.
So
let
you
know
from
from
a
pure
infrastructure,
centric
point
of
view.
Edge
computing
is
essentially
two
out
principles
applied
at
the
network:
edge
close
to
the
user.
So
when
I
say
that
it
includes
a
compute,
virtualization
storage,
virtualization,
networking,
virtualization
resources
on
demand,
API
driven
approach,
automated
lifecycle
management
and
use
of
commodity
Hardware.
So
essentially,
these
are
some
of
the
powerful
cloud
principles
that
make
the
network
edge,
highly
flexible
and
and
programmable.
C
Now,
if
we
take,
if
you
look
at
the
edge
computing
from
from
a
network
operating
point
of
view,
it
is
it's
really
the
convergence
of
IT
and
telecom
networking.
As
you
can
see
in
this
picture,
that
that
I
am
showing,
we
can
have
a
variety
of
devices
like
mobile
phones,
IOT,
sensors,
connected
cards,
all
of
them
connecting
to
access
networks
using
different
access
methods.
It
could
be
radio
access,
network,
Wi-Fi,
cable
or
fiber.
So
then
you
have
the
aggregation
network,
the
regional
pop
and
the
core
cloud,
the
on
the
use
case.
C
The
edge
cloud
can
reside
in
any
of
these
networking
layers,
so
this
this
really
allows
network
operators
to
open
up
their
networks
and
open
them
up
for
new
business
opportunities
and
value
chains
that
that
just
did
not
exist
before
now.
Finally,
if
we
look
at
edge
computing
from
an
application,
centric
point
of
view,
it's
it's
really
all
about
new,
innovative
and
immersive
education.
C
This
is
really
a
great
opportunity
for
app
developers
to
to
create
applications
that
are
tailored
to
local
conditions,
applications
that
provide
highly
contextualized
services
and,
moreover,
these
applications
can
have
low,
latency
and
high
bandwidth
guarantees
because
of
the
proximity
of
the
edge
cloud
to
the
user.
Not
just
that
also,
the
way
the
edge
infrastructure
is
design.
C
C
So,
let's
talk
about
edge
use
cases
at
a
high
level.
When
you
look
at
edge
computing
use
cases,
they
really
fall
into
two
big
buckets,
so
the
first
one
is
the
third
party
edge
applications.
These
are
actually
the
most
interesting
ones.
So
elipse.
Let's
look
at
some
of
these
categories
here
on
real
time
applications,
so
these
are
applications
that
require
very
low,
latency
and
low
jitter.
This
is
where
augmented
reality
virtual
reality
connected
cards.
Tactile
internet,
smart
cities,
industry
4.0.
All
these
use
cases
fall
into
this
bucket.
C
Then
you
have
the
immersive
applications,
so
these
are
the
ones
that
require
extremely
high
network
throughput.
Think
about
4k
video
360-degree
video,
then
the
cost
reduction.
So
this
is
this
is
interesting.
The
is
where
applications
are
trying
to
save
upstream
bandwidth
and
also
save
processing
in
the
central
cloud.
A
good
example
is
a
video
surveillance.
I
owe
to
Gateway
is
another
good
example
here.
C
If
we
move
to
the
self-contained
category,
this
is
where
you
have
the
stadium
events
or
concerts
or
airports,
and
you
know
we.
We
are
going
to
talk
about
networking
and
network
slicing,
and
that
plays
a
critical
role
here,
because
essentially
each
one
can
be
a
slice
and
and
supporting
the
specific
requirement.
The
private
enterprise
is
about
applications
like
ran
to
you,
know:
radio
access
network
to
enterprise,
breakout,
CPE
type
of
applications.
So
that's
what
that's
what
is
included
here
now
if
we
look
at
the
other
bigger
bucket,
the
other
category,
which
is
operator
applications.
C
This
is
where
we
have
of
applications
that
that
are
deployed
by
the
operators.
You
know
analytics
is
is,
is
is
one
category
here.
This
is
all
about
collecting
large
amounts
of
data
analytics
and
even
running
analytics
on
the
edge
itself,
but
then
sending
small
amounts
of
that
data.
Small
batches
update
that
data
to
the
core
compliance
is
another
interesting
category.
Here
it
includes
copyright,
information
or
geographical
data
placement.
Copyright
enforcement
is
also
interesting
here.
Let's
say
you
are
you're
sitting
in
a
concert
or
in
a
sport
event.
What
edge
application
can
do?
C
C
Security
is
about
moving
the
security
parameter
very
close
to
the
source.
So
this
way
we
can
block
security
threats
at
the
edge
itself
and
protect
the
core
network
and,
finally,
the
NFV
category.
This
is
actually
a
very
important
category
here,
because
edge
is
ideal
for
running
different
types
of
vnfs.
Take,
for
example,
access
via
naps.
You
know,
via
Neffs,
like
virtual,
ran
virtual
cmts
for
cable
virtual
OLTC
ran
if
we
talk
about
security
via
Neffs
virtual
firewall,
ids/ips,
a
unified
trust
management,
the
content
filtering
type
of
via
naps
and
for
service
and
force
enhancement.
C
So
there
are
a
lot
of
things
about
educating.
In
fact,
when
we
talk
about
all
these
things,
there
are
key
benefits
that
that
the
edge
computing
gives
you
and
that
kind
of
na
all
these
I
already
talked
about
some
of
these
in
the
previous
cases.
But
but
let's
go
through
this
ultra-low
latency.
This
is
an
extremely
important
benefit
that
you
get
from
an
edge
cloud
typically
latency
in
milliseconds,
and
this
is
a
significantly
lower
compared
to
what
you
get
from
a
centralized
data
center
again.
C
Text
awareness
is
another
great
benefit
here,
because
the
edge
cloud
has
access
to
the
Radio
Network
and
the
information
that
is
provided
by
the
radio
access
network
can
now
be
efficiently
used
by
the
edge
applications.
We
talked
about
data
filtering
data
reduction
in
in
in
the
previous
slide,
compliance
again
and
and
security.
So
these
are
some
of
the
great
benefits
that
the
edge
computing
gives
us.
B
Okay,
so
those
are
the
benefits
now,
there's
a
lot
of
buzzwords
in
here
and
one
buzzword
that
keeps
coming
up
a
lot
whenever
we're
talking
about
edge
computing.
At
some
point,
somebody
is
going
to
say
something
about
5g
and
somebody's
gonna,
say
something
about
network
slicing.
You
yourself
have
said
it
in
this
conversation.
What
is
network
slicing?
What?
What
does
all
this
have
to
do
with
that
right?.
C
The
interesting
thing
is
the
networking
characteristics
and
the
transmission
requirements
of
the
applications
are
very
different.
It's
quite
different
for
from
application
to
application.
Let's
take
an
example:
you
know
when
we
talked
about
IOT,
there
will
be
very
large
number
of
devices
these
IOT
devices,
but
each
device
will
probably
require
low
network
throughput.
C
On
the
other
hand,
when
we
talk
about
mobile
devices
running
streaming,
4k,
video
or
360-degree
view
the
network
bandwidth
requirements
are
going
to
be
extremely
high,
but
when
you
think
about
the
number
of
such
devices,
it's
going
to
be
much
lower
compared
to
the
IOT,
sensors
and
other
IOT
devices,
and
with
connected
cars
on
self-driving
cars,
the
critical
requirement
is
low
latency.
So
the
key
goal
of
network
sizing
is
really
to
partition
the
physical
network
at
an
end-to-end
level.
So
here
we
are
talking
about
end-to-end
level.
C
That
means
you
know
all
the
way
from
ran
edge
to
all
the
way
from
ran
slice
to
edge
slice
to
the
cloud
size,
slice
right,
so
so
partition
the
physical
network
at
an
end-to-end
level
and
allow
very
optimum
grouping
of
traffic
and
and
more
importantly,
provide
isolation
at
the
slice
level
itself
right.
So
this
by
doing
this,
it
really
maximizes
the
flexibility
of
5g
networks.
So
because
of
all
this
network,
slicing
is
extremely
critical
for
5g,
and
this
is
again
because
of
so
many
different
type
of
use
cases
5g
promises
to
support.
C
In
fact,
we
can
say
that,
without
without
dynamic
network
slicing,
it's
it's
very
difficult
to
achieve
the
promise
of
5g
you,
you
often
see
network
slicing
being
referred
to
as
the
secret
sauce
behind
5g.
Because
of
this
reason,
so
it's
really
the
combination
of
Sdn
nfe,
because
you
know
network
slicing,
actually
uses
the
Sdn
and
NFV
concepts
heavily.
It's
really
it's.
The
combination
of
STN
nfe
edge
computing
and
network
slicing
that
gives
5g
the
the
ability
to
to
very
efficiently
handle
all
these
diverse
set
of
applications
in
an
extremely
flexible
way.
C
C
C
There
lot
of
edge
infrastructure
will
have
space
constraints,
power,
constraints
and
cooling
constraints.
In
many
cases
there
is
also
need
for
ruggedized
racks
to
be
deployed
and
security.
There
are
a
lot
of
security
concerns,
including
physical
security
of
the
edge
cloud
rack
itself.
It's
very
important
to
have
flexible
infrastructure,
because
we
need
to
iterate
and
evolve
based
on
new
requirements
and
from
a
software
flexibility,
infrastructure,
flexibility,
point
a
few
Sdn
NFV
network
slicing.
All
these
things
play
a
big
role
and
high
performance,
especially
you
know.
Cpu
memory,
storage,
enhanced
platform
awareness
features,
data
plane
acceleration.
C
All
these
become
important
for
hype.
You
know
to
make
sure
that
there
is
good
user
experience
now,
if
we
know
now,
we
are
going
to
have
a
large
number
of
these
geographically
distributed
remote
edge
clouds
and
therefore
distributed.
So
it's
it's
a
highly
distributed
architecture,
which
means
a
powerful
central
common
management.
Plane
also
becomes
extremely
important,
and
once
again,
since
we
are
dealing
with
hundreds
of
these
clouds,
automation
is
key.
We
have
to
have
zero
touch,
provisioning,
zero
touch,
lifecycle
management,
zero
touch
operations,
logging,
monitoring,
alerting.
C
Can
salute
yeah
there
is.
There
is
no
way
you
can
have
anything
manual
happening
in
this
in
this
ecosystem.
Right
compatibility
is
another
important
requirement,
because
edge
edge
clouds
are
not
going
to
function
in
isolation.
They
need
to
work
with
arcus
traitors.
They
need
to
work
with
mech
application,
orchestrators
mono
vnf
orchestrators.
So
how
compatible
are
those
with
industry,
standard
orchestra
tools
and
so
that
that
becomes
an
important
requirement
as
well
workload?
Diversity
is
another
important
requirement
ability
to
support
VMs
ability
for
container
and
in
some
cases
see
when
you
are
close.
C
We
are
going
to
talk
more
about
vnf
workloads.
You
know
in
a
reg
cloud,
it's
important
to
have
the
ability
to
support
vnf
workloads
that
could
be
either
VM
based
or
even
container
based
and
then
different
types
of
Mecca
plication
'z.
Now
small
footprint,
the
hardware
footprint
is
also
an
important
infrastructure
requirement.
In
some
use
cases
it
could
be
as
small
as
just
one
single
node
or
you
know,
three
to
six
nodes.
Half
rack
full
drag
multiple
racks.
It
all
depends
on
the
use
case.
B
C
A
good
question
Nick,
so
let's
talk
about
kubernetes
in
edge
architecture
right,
so,
yes,
kubernetes
is
actually
well-suited
for
a
small
footprint
edge
cloud.
This
is
because
you
know
we
all
know
that
kubernetes
is
easier
to
manage.
Containers
have
much
lower
overhead
compared
to
VMs,
which
means
we
can
have
higher
density
of
containers
on
a
per
worker
node
basis,
and
because
of
all
this,
more
resources
are
going
to
be
available
for
edge
applications
or
edge
vnf,
even
in
a
small
cloud,
so
using
kubernetes
James.
In
fact,
we
can.
C
You
know
these
features
also
work
with
kubernetes
part,
so
that
shouldn't
be
a
problem
in
designing
the
infrastructure
for
low
latency
and
high
network
throughput
using
kubernetes
right
in
the
in
the
picture
that
we
are
showing
on
this
slide.
We
have
a
very
simplified
small
kubernetes
edge
with
three
worker
nodes
and
running
in
different
edge
applications.
Applications
like
video
surveillance
applications
like
media
streaming,
so
Nick
as
long
as
we
are
running,
containerized
were
containerized
edge,
applications
and
containerized
me
and
vnf
workloads.
Life
is
simple.
Life
is
good.
Ok,.
B
C
So
there
are
a
lot
of
challenges
that
is,
that
is
true.
A
lot
of
challenges
in
running
vnfs
in
kubernetes
containers
I
should
say
that
currently,
there
is
quite
a
bit
of
effort,
quite
a
bit
of
work
in
the
industry
to
make
VN
F's
cloud
native
and
dip
them
as
thinners,
but
as
of
now
mohnish
portion
of
the
VN
is
still
run
as
VMs
and
sadly,
in
many
cases,
these
vnfs
are
really
virtualized
replicas
of
their
hardware
appliance,
which
kind
of
defeats
the
whole
purpose
of
the
NFV.
C
B
C
So,
let's
talk
about
it,
I
mean
we,
we
discussed
how
kubernetes
can
be
used
in
the
edge
cloud.
As
long
as
there
are
containers
we
talked
about.
Some
of
the
problems
in
deploying
vnfs
in
a
component
is
cloud.
So
now,
let's
talk
about
so
the
question
is:
how
do
we
run?
Vn
f's
in
a
coconut
is
edge
cloud
right.
So
if
a
VN
f
has
to
run
as
a
VM,
then
what
we
need
is
a
is
a
notion
of
a
VM
pod
right
in
kubernetes,
which
can
run
a
virtualized
workload.
C
If
we
have
this
it,
it
kind
of
simple
simplifies
things
a
little
bit,
because
darker
pods
can
run
edge,
applications
and
and
also
vnf,
that
are
containerized,
and
we
can
have
VM
parts
which
can
run
edge
applications
and
be
NFS
that
are
virtualized
or
VM
based
right.
Also,
we
need
these
darker
parts
and
VM
parts
to
kind
coexist
and
perhaps
even
be
networked
together
in
a
seamless
way.
So
kubernetes
has
has
the
container
runtime
interface
the
CRI,
and
we
can
have
a
kubernetes
CRI
implementation
for
running
VMware
clothes.
C
So
this
this
this
is
an
interesting
way
of
handling
vnf.
So
one
way
to
do
this,
one
way
to
use
the
kubernetes
ERI
implementation
and
have
and
have
a
VM
running
one
one
way
to
do
this
is
using
the
open-source
word
LED
technology.
So
that's
that's
what
is
shown
in
the
picture
here.
It's
it's
the
same
as
what
we
saw
a
couple
of
slides
ago,
but
but
here
the
kubernetes
edge
cloud
is
running
edge,
applications
that
are
containerized
and
also
a
bunch
of
vnfs
that
are
virtualized
all.
C
C
Now
worklet
can
run
unmodified,
OpenStack
VM
images
using
Q
how
to
format,
and
since
it's
it's
a
CRI
implementation.
The
advantage
is,
you
know
all
higher-level.
Kubernetes
objects
can
be
built
using
vm
parts,
essentially,
kubernetes
sweeps
a
worklet
vm
part
just
like
it's.
It
treats
a
darker
pod,
for
example,
all
familiar
cube,
CTL
part
commands
still
work
with
what
rate
vm
parts
we
can
integrate,
integrate
these
worklet
parts
with
cluster
networking
using
all
the
popular
CNI
plugins
and
and
it's
also
easy
to
deploy.
Now,
let's
look
at
some
of
the
capabilities
right.
C
What
let
capabilities
in
greater
detail?
I
already
talked
about
the
cluster
of
networking
part
IP
is
used
inside
the
VM
and
and
VM
parts
can
be
targeted
by
higher-level
kubernetes
services.
Just
like
any
other
services.
Tcp
and
HTTP
readiness
probes
are
also
supported.
With
what
rate
pods
are
you.
There
is
support
for
cloud
in
it.
Also
that
means
you
can
add
SSH
keys
to
evil
Ravi
MS
running
in
the
part.
C
You
can
add
extra
set
of
scripts
mount
rod,
devices
ephemeral
volumes,
so
it
kind
of
makes
it
a
very
powerful
concept,
and
especially
these
are
you
know
where
clips
are
well
suited
for
running
vnf
workloads,
because
because
of
the
powerful
networking
capabilities,
as
I,
as
I
already
explained,
you
can
have
you
can
run,
you
know:
CNI
geni,
multiple
interfaces
and
various
networking
options.
Well,.
B
C
They
can,
they
can
certainly
talk
to
each
other,
otherwise
it
would
not
be
that
useful
right.
So,
let's,
let's
look
at
this,
as
shown
in
this
picture.
What
do
we
have?
We
have
a
we
have
darker
parts
and
we
have
words
right:
VM
parts
they're
all
networked
together
in
an
edge
cloud,
so
here
we
use
CNI
geni
for
multiple
CNI
support.
If
you
look
at
the
picture
for
the
VM
part,
there
are
two
two
interfaces.
C
One
interface
is
connected
to
a
flannel
network
and
the
other
interface
is
connected
to
an
SSRI
OB
network
for
data
plane
acceleration.
This
can
be.
This
can
be
used
for
a
vnf
workload
running
requiring
high
high
network
throughput.
Now
the
darker
part
also
has
an
interface
connected
to
Faneuil
Network
right,
so
we
can
easily
configure
these
networks
to
achieve
control,
plane,
data,
plane,
kind
of
data,
plane,
traffic
isolation
when
you
are
running
vnf
workloads.
C
These
are
some
of
the
key
requirements,
ability
to
have
multiple
network
interfaces
and
each
interface
connecting
to
a
different
network,
isolating
the
control
plane
or
the
management
plane
and
the
data
plane
traffic
within
the
nfe
environment.
So
this
is
what
really
makes
bartlett
very
well
suited
for
NFV
running
nfe
workloads.
B
C
Management
of
these
distributed
edge
crowds
is
an
is
a
very
important
topic,
and,
and
this
the
this
picture
that
I'm,
showing
here
it's
directly
from
the
cleaner
project
and
and
again
a
Creina
project
for
those
of
you
who
are
not
familiar
with
the
project
is,
is
a
is
a
project
focused
open-source
project
focused
on
the
edge
cloud
and
it's
gaining
lot
of
momentum,
and
here
it
shows
a
hierarchical
deployment
model,
hierarchical
management
model
for
the
edge
clouds.
So,
as
you
can
see,
there
are
two
central
clouds
here:
c1
and
c2.
C
So
these
central
slides,
it's
not
a
central
cloud
central
site,
a
central
site
that
can
deploy
a
collection
of
regional
sites,
r1
r2
r3.
These
are
all
regional
sides
and
the
idea
is,
the
regional
sites
can
actually
managed
the
edge
sites,
the
sites
that
are
here
and
and
for
each
region
there
is
the
concept
of
a
edge
flock,
which
is
a
collection
of
edge
clouds,
and
the
regional
manager
can
actually
manage
the
edge
clouds
within
within
each
within
each
edge
flock.
So
that's
the
idea
here
again.
This
is
one
one
way
to
do
this.
C
C
You
know
we
can
just
have
one
central
region
deploying
one
central
region
deploying
and
managing
various
edge
clouds,
if
that
makes
sense
for
a
given
deployment
architecture
and
the
other
architectures
that
have
been
that
that
have
been
discussed
or
ability
to
have
very
minimal
control,
plane
in
the
edge
clouds
and
and
having
a
more
central,
having
a
centralized
cloud
that
manages
those
those
clouds
and,
lastly,
the
other.
The
other
architecture
that
comes
up
is
having
remote
computes,
having
a
centralized
control
plane
and
a
remote
compute.
There
are
a
lot
of
challenges
there
again.
C
B
C
B
C
Right,
so
here
is
a
small
list
of
key
open
source
projects
in
the
edge
flower
area.
There
are
a
few
more
I
hate's,
it's
given
heavily
by
AT&T
Intel,
Wind
River,
a
bunch
of
players
are
involved
and
more
and
this
project
we
had
the
first
summit
last
month
and
there
is
a
network
cloud
group,
a
blueprint,
that's
out
and
also
seed
code
for
the
network
cloud
blueprint
from
AT&T
and,
as
I
said,
this
project
is
certainly
gaining.
C
A
lot
of
momentum
is
another
interesting
cloud:
open-source
project,
it's
it's
an
open
foundation
project
and
the
edge
cloud
software
stack
is
based
on
VIN,
triple-a,
titanium
and
driving.
This
project,
heavily
and
and
starting
X
could
also
have
a
blueprint
that
goes
into
the
fraeno
edge
stack.
So
a
Craner
project
is
more
of
a
relation
open
source
project
and
that
could
have
blueprints
from
other.
B
C
Right
so
then
only
nap
is
another.
There
is
a
edge
automation.
There
is
a
edge
project
within
on
edge
cloud
and
again
own
app
will
be
a
critical
component
for
edge
cloud,
because
because
no,
no
that's
where
the
artesian,
it's
it's,
the
key
orchestration
component
for
orchestrating
vnf
workloads
or
maybe
even
orchestrating
edge
application,
so
own
app
is
likely
to
play
a
play.
A
key
role
there
and
OPN
fe
also
has
a
edge
cloud
project
more
focused
on
on
on
the
BNF
I.
C
You
know
Google
is
also
kubernetes.
Actually,
kubernetes
is
a
edge
related
cig
group
that
has
also
started
up
now.
Let's
the
other
interesting
thing
to
look
at
is
the
POC
that
are
under
the
HC
mech
umbrella.
So
so,
basically
at
sea,
mech
is
G,
which
is
a
standards.
Body
has
has
a
POC
framework
and
there
are
quite
a
few
interesting
pocs
that
have
either
been
completed
or
in
progress.
C
This
is
another
interesting
thing
to
look
at
because
it
gives
us
a
very
clear
picture
of
what
some
of
these
key
players
are
focused
on
in
computing
and
I
have
a
bunch
of
these
mech
POC,
for
example,
videos,
user
experience
optimization
this
was
until
and
china
mobile
and
to
the
mix
POC
umbrella.
So
this
is
basically
King
at
be
used
to
in
and.
C
C
This
is
a
their
interesting
POC,
which
is
based
where
the
application
is
basically
looking
at
the
ran
information,
the
radio
information
and,
depending
on
whether
the
ran
is
congested
or
not.
It's
it's
kind
of
adopting
the
application.
So
it's
kind
of
like
traffic
shaping
on
the
Enterprise
Services
side,
advanced
hakuna.
They
will
involved
in
a
local
breakout.
This
is
users
entering
the
edge
and
the
LAN
network.
C
There
is
a
local
breakout
to
the
enterprise
network
and
similarly,
there
can
be
data
prioritization
and
acceleration,
so
a
mech
application
sitting
in
the
edge
and
looking
at
the
data-
and
you
know
maybe
l4
to
l7
content
and
and
prioritizing
the
data
based
on
the
defined
criteria
and
then
quite
a
few
Stadium
events
and
healthcare
use
cases
that
that
use
the
networking,
a
network
sizing
concept
and
then
the
Nokia
and
Oda
phone.
They
were
involved
in
edge
video
analytics,
and
so
this
is
typically
that
data
reduction.
C
The
clouds,
as
they
are
designed
today
be
able
to
handle
that
much
of
data
bombardment,
so
these
type
of
applications
will
become
really
critical,
mekin
CDN
a
very
good
use
case,
because
now
a
lot
of
the
Syrian
can
move
to
edge
and,
and
and
you
know,
user
experience
little
throughput
latency
all
those
things
can
all
those
things
apply
in
these
use
cases
and
then
make
application
orchestration
with
multiple
met
platforms.
This
is
another.
C
So
it
becomes
important
for
the
edge
applications
and
also
for
the
edge
clouds
to
be
able
to
run
any
application
on
any
cloud,
whether
it's
a
communities
based
edge
cloud
or
in
OpenStack
based
edge
cloud
or
whatever
edge
cloud,
so
that
kind
of
compatibility
needs
to
exist.
So
this
POC
was
along
those
lines
and
then
with
KDDI
and
HP,
there
was
a
IOT
and
v2x
type
of
use
case.
These
are
also
popular
use
cases.
All.
B
Right,
so
that
gives
us
a
good
idea
of
some
of
the
things
that
are
actually
going
on
right
now
in
terms
of
peer
seasoned
things
move
pretty
quickly,
so
this
could
give
us
a
good
look
at
what
the
real
landscape
is
kind
of
be
in
the
next
year
or
two.
So
that's
that's
interesting.
All
right,
so
we
have
reached
the
end
of
our
content.
We're
ready
to
take
some
questions.
C
That's
a
good
question
Nik,
so
users
approximately
to
the
edge
and
avoiding
multiple
Network
hops
is,
is
certainly
a
key
component
right
in
in
in
achieving
low
latency.
However,
the
edge
infrastructure
design
should
also
ensure
that
unnecessary
delays
are
not
introduced,
especially
in
the
data
path
right.
So
this
is
where
EPA
features
like
no
more
we're
scheduling,
CPU
painting
huge
pages
will
help,
and
you
know
the
data
plane.
Acceleration
techniques
like
SRO
EDP
DK.
They
also
help
in
accelerating
the
data
planes.
C
This
is
why
the
this
is
actually
the
reason
why
the
edge
cloud
infrastructure
has
a
lot
of
commonality
with
nfe
I.
It's
it.
It
resembles
the
nfe
infrastructure
to
a
great
extent.
So
so
the
answer
is
yes.
Proximity
is
an
important
component,
but
the
actual
network
edge
infrastructure,
design,
also
metals.
B
C
Actually,
it
should
not
matter,
I
mean
these,
so
the
way
I'm
looking
at
this
is
again
I'm,
not
an
expert
in
in
different
access
technologies,
but
from
an
edge
cloud
point
of
view.
These
are
two
different
access
methods
right,
so
we
talked
about
radio
access
versus
Wi-Fi
access,
I
mean
wireless
access
or
cable
access
or
fiber
axis.
So
it's
it's
just
another
access
method,
purely
from
an
edge
cloud
point
of
view,
but
there
could
be
other
factors
for
that,
but
I'm
not
very
familiar
with
those
access
technologies.
B
C
This
is
a
loaded
question,
so,
let's
let
this
ad
was
some
of
the
things
so,
first
of
all
vertel,
it
is
certainly
a
critical
building
block
I
think
in
solving
some
of
the
vnf
problems.
We
we
talked
about
right.
So
what
does
it
do?
It
allows
a
VM
based
vnf
to
run
unmodified
in
a
kubernetes
environment
right
and
thanks
to
the
adoption
of
CNI
geni
worklet
supports
using
multiple
interfaces.
We,
including,
as
we
said,
the
SR
iov
interface,
so
SRA.
C
We
support,
is
critical
to
achieve
the
performance
and
latency
requirements
and,
as
I
described
you
can,
you
can
have
network
isolation
and
all
that
right
with
work
types.
So
it's
it's
actually
a
great
option,
but
it
doesn't
solve
all
the
problems.
To
answer
your
question.
So,
for
example,
let's
say
we
have
a
complex
vnf
with
multiple
components,
each
running
as
a
separate
VM
and,
let's
say
a
proprietary
vnf
manager
designed
for
some
other
VM,
let's
say
OpenStack.
It's
going
to
take
some
effort
to
adopt
this
vnf
to
the
kubernetes
worklet
environment.
B
C
C
B
C
Yeah
there
is
a
very
nice
architectural
document
that
shows
that
actually
shows
the
various
components
and
how
how
it
works
and
and
and
again,
as
I
said,
the
nice
thing
here
is
it
is.
It
is
treated
exactly
like
a
docker
pod,
since
it
is
a
CRI
implementation
and
not
just
a
management
add-on.
So
ever
click
part
for
all
practical
purposes
will
be
treated
very
similar
to
a
Tucker
pod
and
anything
that
you
can
do
to
a
darker
part,
pretty
much
you
can
do
to
ever.
Let
me
impart.
C
So
ran
enterprise.
Breakout
is
a
interesting
enterprise,
use
case
where,
where
users
are
joining
a
niche
cloud
from
their
mobile
device
or
any
device,
they
would
be
directly
tunneled
and
connected
to
the
enterprise,
so
there
could
be
instead
of
going
through
the
cloud,
so
the
so,
basically
the
breakout
or
the
tunneling
gets
implemented
all
the
way
from
the
edge
to
the
enterprise
network.
C
So
this
can
this
can
give
you
a
lot
of
interesting
options.
For
example,
there
could
be
security
and
authorization
kind
of
components
running
on
the
edge
itself,
and
so
this
way
and
and
there
could
also
be
you-
know-
intrusion
detection,
intrusion,
prevention,
kind
of
vnf
running
and
and
it
could,
it
could
protect.
The
enterprise
networks,
make
the
enterprise
networks
more
secure.
C
Right
so
this
is
another
interesting
question.
So
so,
basically
we
talked
about
worklets
and
we
said
that
this
is
one
way
of
doing
this.
One
way
of
making
VMs
run
in
a
cooperative
environment
so
Qbert
is,
is
a
is
a
very
different
way
of
doing
it.
Cooper
it
is
a
virtual
machine
management
add-on
for
kubernetes
and
it
controls
VMs
as
a
kubernetes
custom
resource.
C
So
that's
the
key
difference,
so
over
Atlit
is,
is
a
CRI
implementation,
which
means
kubernetes
will
see
the
VMS
the
world
ATMs
exactly
the
way
it
sees
docker
containers,
whereas
with
with
cubeworld
it's
it's,
it's
a
management
add-on.
So,
having
said
that,
you
know
both
approaches
have
their
pool
is
one
is
more
suitable
than
the
other
which
it's
not
like.
C
So
in
some
ways
this
is
an
advantage
right,
because
now
kubera
developers
have
the
freedom
to
implement
whatever
they
want,
as
opposed
to
being
limited
by
the
pod
definition
right
and
at
the
same
time
it
also
has
the
disadvantages,
because
because
many
parts
of
the
kubernetes
need
to
be
reimplemented
you,
you
can
end
up
with
a
bigger
code
base
right
so
with
worklet.
On
the
other
hand,
so
all
VMs
are
defined
as
very
much
similar
to
cover
Nettie's
parts
and
they
are
treated
as
first-class
citizens.
C
So
the
advantage
is
that
anything
you
can
do
with
parts
in
worklets.
You
can
do
with
anything
you
can
do
with
the
darker
parts.
You
can
pretty
much
do
those
with
word
rate
VMs
right
out
of
the
box,
and
the
disadvantage
here
is,
that
is
that
you
are
kind
of
limited
to
the
to
the
functionality
that
comes
with
the
parts.
If
you
want
to
do
anything
fancy
that
needs
to
be
implemented
separately
now.
Having
said
that,
the
biggest
advantage
of
worklet
parts
is
really
in
networking,
and
especially
nfe,
workloads
and-
and
the
question
is.
C
Why
is
this?
Because?
Because
you
know
we
talked
about
CNI
geni,
our
clips
actually
supports
multiple
interfaces.
4Sr
iov
is
there
and
I'm
support
for
multiple
interfaces
network
isolations.
So
these
are
all
the
key
factors
that
make
the
word
parts
more
powerful
and
Cooper
doesn't
have
this
much
of
flexibility
when
it
comes
to
vnfs
and
networking.
Okay,.
B
C
Yeah
I'm,
familiar
with
the
service
mesh
concepts
and
I,
have
to
look
into
how
that
fits
into
the
edge
cloud
and
and
the
nfe
framework
that
we
have
been
talking
about
so
essentially
those
are
offloading.
Some
of
the
general
networking
functions
to
two
side.
Cars
right.
That's
that's
kind
of
the
idea
there
and.
C
B
C
Mek
application
is,
you
know,
MEK
is
multi
axis
edge
cloud
right,
so
Mac
application
is
a
generic
term
used
for
any
edge
application.
I
mean
it
became
popular
because
most
of
the
Mac
biosis
came
under
the
EXCI
Mecca
SG
umbrella.
Now
who
develops
them?
Who
manages
them?
That's
a
very
interesting
question
and
this
is
what
is
going
to
evolve
right
as
5g
and
edge
cloud
takes
over.
The
dream
here
is
that
the
Mac
applications
would
become
very
much
like
mobile
apps,
that
we
see
on
iPhones
and
Android
phones,
so
think
of
application
developers.
C
So
that's
that's
very
much
possible
and
then
just
like
the
Android
Market
or
the
iPhone
market,
you
can
have
these
Mac
applications
have
their
profiles.
You
know
an
application
can
say
hey.
These
are
the
things
I
need
to
run.
These
are
the
type
of
clouds
I
need
to
run.
I
need
a
VM.
I
need
a
container
I
need.
You
know
these
are
the
things
I
need
and
then
potentially
edge
clouds
when
they
register
with
the
orchestrator
can
also
publish
those
kind
of
profile,
information
and
an
intelligent
application.
C
B
B
C
A
Thank
You,
Nick
and
sudhish
I
think
we're
right
at
the
top
of
the
are
so
we'll
close
right
away
here,
thanks
Nick
for
handling
the
questions,
and
you
made
my
job
easier
today
and
punished.
That
was
a
great
presentation
and
I'm
sure
the
audience
will
be
looking
forward
to
getting
for
the
information
from
the
links
you
provided.
Thank.