►
From YouTube: Centralized, distributed or both? - Architectures for Kube... Gergely Csatari, Nokia & Ildikó Váncsa
Description
Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon North America 2021 in Los Angeles, CA from October 12-15. Learn more at https://kubecon.io The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Centralized, distributed or both? - Architectures for Kubernetes on the edge - Gergely Csatari, Nokia & Ildikó Váncsa, Open Infrastructure Foundation
Edge cloud environments are distributed systems that can be implemented in various ways depending on the use case. As edge computing has a broad spectrum, it opens up different architecture choices when Kubernetes is used in the solution. So which parts of Kubernetes should be deployed where? Do you always deploy all components on every site? Does that result in a too large footprint or not? How does it look like with the different cloud platforms that provide infrastructure? In this presentation we will highlight the challenges of different architecture options as well as the problems they solve. You will also learn how Kubernetes fits into the different use cases and architecture options they call for. We will also explore the relevant projects and activities in Kubernetes, such as KubeEdge, K3s, Metal3 and their findings in relation to architecture choices for edge deployments.
A
Hi
everyone
and
welcome
to
our
session
where
we
will
talk
about
edge
architectures.
My
name
is
ildi
kovancha.
I
work
as
senior
manager
of
community
and
ecosystem
at
the
open
infrastructure
foundation
and,
among
other
things,
I'm
a
big
open
source
and
edge
computing
enthusiast,
and
I
have
a
co-presenter
today.
B
B
A
And
now,
let's
jump
into
the
middle
of
all
this
so
computing,
and
there
are
a
lot
of
debates
out
there
about
what
edge
computing
and
edge
are
so
we
will
not
join
and
deepen
the
debate
here.
I
only
wanted
to
just
emphasize
on
that
edge
computing
is
part
of
an
evolution
path,
as
we
are
kind
of
moving
on
from
from
cloud
computing.
Only
and
edge
computing
is
making
it
possible
to
to
get
the
computing
power
out
of
the
large
data
centers
towards
the
edge
and
towards
the
end
users.
A
Let
them
be
humans
or
machines,
and
when
it
comes
to
edge
computing,
a
lot
of
people
and
organizations
are
focusing
on
the
edge
part
of
it,
because
that
is
new.
It
is
exciting,
it
has
a
lot
of
new
opportunities,
and
it
also
has
a
lot
of
new
challenges
to
solve,
with
a
really
small
footprint,
also
with
location
and
and
circumstances
that
you
all
have
to
be
aware
of
and
plan
for,
but
edge
is
also
always
on
the
edge
of
something
on
the
edge
of
the
network
on
the
edge
of
your
system.
A
So
it
is
always
part
of
a
larger
ecosystem
and
architecture,
and
that
is
what
the
opening
for
edge
computing
is
focusing
on.
The
opening
for
edge
computing
group
is
a
top-level
working
group
that
is
supported
by
the
open
infrastructure
foundation
and
we
have
a
broad
industry
outreach
and
a
focus
on
infrastructure
software
and
what
we
are
looking
into
really
is
the
massively
distributed
systems
that
are
behind
edge
computing
use
cases
so
really
how
it
looks
like
from
core
to
the
edge
edge
to
core
or
edge
to
edge.
A
A
The
link
is
on
the
slide
and
also
we
publish
two
white
papers,
so
you
can
see
all
the
exciting
use
cases
that
we've
been
working
with,
that
covers
telecommunication,
retail,
and
we
even
have
a
use
case
on
how
to
modernize
and
automate
shrimp
farms.
So
you
can
also
find
the
links
to
the
white
papers
on
the
side
as
well
go
and
read
them.
A
They
are
not
very
long,
but
they
are
even
more
exciting
and
we
can
move
to
the
next
slide
to
take
a
deeper
look
at
what
we
are
talking
about
when
we
are
talking
about
architecture
models.
So
our
learning
is
that
there
is
no
one-size-fits-all
solution
out
there.
There
is
no
one
particular
tool
or
one
particular
configuration
that
will
work
for
everyone.
I
would
think
that
is
not
coming
as
a
surprise
to
you
either.
A
A
A
Only
and
in
this
case
most
often
if
you
lose
connection,
your
workload
is
still
running
on
the
x
side,
but
you
will
not
be
able
to
launch
a
new
workload
and
you
will
also
not
be
able
to
do
many
other
operations
either
and
in
some
use
cases
this.
This
is
totally
acceptable
and
your
you
may
want
to
rather
focus
on
increasing
the
the
workload's
footprint
on
your
ad
site.
A
A
So
that
is
what
we
have
been
focusing
on
so
far
and
when
it
comes
to
solutions,
we
started
to
build
these
architectures
with
openstack,
also
with
kubernetes
components
and
see
how
it
looks
like
when
you're
putting
this
into
action,
and
we
also
collected
a
couple
of
projects
for
you
that
gergei
will
describe
that
are
matching
one
or
the
other
architecture
models.
To
give
you
some
examples,
so
I'm
giving
the
word
to
gergei.
B
Thank
you.
So,
as
yotico
mentioned,
we
we
collected
some
some
projects
related
to
kubernetes,
which
are
implementing
either
the
sentence,
control,
plane
or
the
distributed
control
plane
in
architecture,
and
we
also
selected
some
projects
to
show
you
in
this
presentation,
which
provides
support
for
for
edge
use
cases
in
a
in
a
different
way,
so
for
decentralized
control,
plane
implementation,
where
we
have
all
the
control
functions
in
a
central
location.
The
two
most
notable
projects
are
kts,
which
is
a
a
very
small.
B
B
So
on
top
of
these
control
plane
functions,
cubed
provides
also
features
for
for
iot,
like
a
message
broker
and
and
advent
bus,
and
also
some
device
management
features.
And
here
also,
we
have
this
capability
that
the
the
control
plane
is
able
to
communicate
over
the
one
via
this
edge
and
cloud
have
components
of
the
of
the
architecture.
B
B
B
Solutions-
and
it
has
a
has-
a
central
management
function
which
controls
the
the
data
and
image
synchronization
of
of
the
different
edge
side.
So
it
has
this
complete
control,
plane,
implementation
for
for
the
different
sites,
and
there
is
one
sensor
function
which
which
is
managing
all
of
these
and
it
can
handle.
B
B
The
other
project
was
selected
is
not
a
complete
solution,
so
in
this
sense
it's
different
from
from
the
others,
but
what
I
described
in
in
the
previous
minutes.
So
it's
it's
cube,
fed
it's
part
of
of
kubernetes.
It's
it's!
It's
an
implementation
to
federate
the
kubernetes
api.
So
basically
what
it
does
is
it's
running
the
federation
agent
in
in
in
one
cuvette's
cluster
and
and
from
there
it
is
capable
to
schedule
workloads
to
different
current
test
clusters.
B
Also,
so
it's
basically
a
scalable
api
for
quebec.
It
is
a
necessary
component
for
building
edge
cloud
infrastructures
purely
based
on
two
benefits
with
a
distributed
control
plane.
If
a
single
entry
point
to
the
infrastructure
is
required,
so
this
this
project
provides
provides
that,
but
to
have
a
complete
solution.
There
are
lots
of
other
components
needed
because
they
are
not
part
of
cube
fed.
B
Layer
of
of
this
so
again,
metacube
is
like
a
necessary
part
of
the
netcloud
infrastructure
solution,
but
this
is
the
very
important
part,
because
the
most
manageability
of
of
the
hardware
is
key
in
case
of
edgecloud
infrastructure,
because
as
we
as
we
learned
from
usdco
in
the
in
the
introduction
side,
h-cad
infrastructures
are
about
massively
distributed
cloud
infrastructures
and
and
to
be
able
to
manage
these
in
a
in
a
scalable
way.
We
need
automation
on
all
layers
of
of
the
second
and
hardware.
Management
is
an
important
part
of
this.
B
Service
discovery
feature
across
these
these
clusters,
so
this
is
a
very
good
baseline
to
implement
a
distributed
edge
application
which
is
able
to
communicate
from
like
edge
to
edge
in
a
in
a
mesh
kind
of
phase.
So
for
this
submariner
is
a
great
networking
solution.
So
these
are
the
the
example
projects
what
you
wanted
to
highlight.
B
We
know
that
it's
not
possible
to
list
everything
we
just
selected
the
let's
see
most
notable
examples
which
are
implementing
the
the
architectures,
what
we
identified
as
the
most
prominent
ones
from
all
the
architectures,
but
also
we
are
now
working
on
on
defining
hybrid
architectures,
when
kubernetes
and
openstack
components
are
part
of
the
edge
cloud
infrastructure
in
different
places
and
in
in
different
roles.
But
let's
hear
more
about
the
future
plans
of
the
group
in
the
next
slide
from
mundico.
A
But
we
would
like
to
understand
it
better
how
the
the
landscape
is
shaping,
how
the
edge
solutions
are
shaping
and
how
the
edge
requirements
are
evolving
over
time.
So
we
are
inviting
you
to
come
and
collaborate
with
us
and
give
us
any
feedback,
for
instance,
about
the
the
two
edge
architecture
models
that
we
have
so
far
the
centralized
and
the
distributed
control
plane
model.
A
A
Please
comment
and
share
the
details
with
us
or
if
you
have
a
use
case,
that
you're
trying
to
identify
your
architecture
for
and
need
some
help
guidance,
or
you
would
like
to
talk
with
someone
about
that.
We
would
really
be
interested
in
learning
about
your
use
case
and
requirements
too.
So
the
the
slide
contains
all
the
information
about
the
the
weekly
meetings
that
we
have.
Also
the
the
maining
list
and
irc
channel
where
you
can
get
in
touch
with
the
group
so
come,
join
us
and
work.
A
Let's
all
work
together
on
finding
solutions
for
the
various
edge
computing
use
cases
that
are
out
there
and
with
that
I
believe
that
we
arrived
to
the
q
a
part.
So
we
are
hanging
out
here
at
the
event
so
come
and
ask
questions
now
or
start
a
discussion
or
find
us
during
kubecon
or
reach
out
to
us
on
the
email
addresses
that
you
saw
at
the
beginning
of
the
presentation
or
well
find
us
anywhere.