►
Description
Microsoft recently developed a solution providing 5G to customers, running in on premises Kubernetes managed through Azure. This talk goes over what we did, and answers some of the interesting questions.
What is 5G anyway? Why would anybody want to buy this? Why is Kubernetes such a good fit for 5G? Why manage on premises software through Azure? With Kubernetes on the edge, how do you upgrade and install remotely? How does the networking work where you have edge, cloud, and radio networks, especially when you need fast networking, none of which is natively supported in Kubernetes?
A
Oh,
hey
everybody
hi
good
afternoon,
I
was
just
thinking-
is
it
morning
afternoon,
it's
afternoon
now,
so
welcome
to
track
two
slightly
adjusted
agenda
from
previously
advertised.
Sadly
lewis,
who
is
our
lucas?
Who
was
our
first
speaker
couldn't
make
it
he's
not
well
so
best
wishes
to
him,
but
we
are
ready
to
kick
off
our
track
two,
and
so
our
first
speaker
for
track.
Two,
in
fact,
now
is
peter
white
going
to
be
talking
about
5g
on
the
edge
with
kubernetes
and
azure.
So
welcome
peter,
and
please
take
it
away.
B
B
My
name
is
peter
white,
I'm
a
principal
software
engineer
at
microsoft
and
when
I'm
asked
what
I
do
for
a
living,
I
invariably
mumble
a
bit
and
can't
quite
think
of
what
I
actually
do
for
a
living,
because
I
seem
to
do
all
kinds
of
different
things
in
all
kinds
of
different
projects.
This
particular
project
I'm
going
to
be
talking
about
here-
is
one
where
I
was
involved
in
some
incredibly
tedious
compliance
work.
B
B
So
I'm
going
to
talk
through
the
diff,
the
three
different
areas
and
the
first
thing
is
a
5g
which
obviously
you'll
all
have
heard
of,
but
I
bet
most
of
you
have
no
idea
what
it
is
so
a
little
bit
of
information
about
the
various
telco
standards.
So
this
phone
started
off
with
a
very
primitive
analog.
Things
are
about
the
size
of
a
house,
brick,
they
cost
about
a
million
pounds
each.
They
didn't
work
very
well
at
all,
and
after
a
little
while
people
came
along
and
said,
we
can
do
better
than
this.
B
We've
discovered
that
mobile
phones
are
a
good
thing,
but
we
can
make
them
all
digital
and
they
thought
well.
This
is
the
second
time
we've
done
this.
Let's
call
this
2g
first
second
generation
and
we'll
name
what
we
already
had
one
g
and
then
in
a
rare
piece
of
logical
naming
that
we
can
all
understand
every
time.
They've
renamed
they've
simply
put
a
bigger
number
in
front
of
it.
B
B
A
B
And
you
can
make
phone
calls
and
look
at
videos
and
so
on.
There
are
three
pieces
of
the
infrastructure
behind
that
right.
Without
going
into
an
awful
lot
of
detail,
the
obvious
one
is
the
the
rand,
the
radio
access
network
when
you
pick
up
your
phone
and
you
talk
into
it
and
the
message
goes
somewhere.
Your
phone's,
obviously
talking
to
some
kind
of
cell
phone
nest
somewhere
and
the
those
masts
form
the
radio
access
network
and
there
there's
connectivity
between
them.
B
But
a
lot
of
what
the
where
the
complicated
technology
is
is
in
the
mobile
core,
because
if
you
think
about
making
your
call
a
lot
of
things
have
to
happen
before
you
can
be
sent
out
into
the
wide
world
of
the
internet.
For
example,
you
have
to
look
at
your
phone
and
they
have
to
decide
that.
Well,
yes,
you're
this
sim
card,
so
this
is
your
identity.
B
You
have
this
much
credit
you're
allowed
to
talk
over
the
network
to
these
endpoints
you're,
the
same
person
who
was
talking
to
this
other
base
station
100
milliseconds
ago
before
you
suddenly
dropped
out
of
transmission
range
for
that
one
and
into
the
next
one,
all
of
the
kind
of
logic
for
how
you,
how
you
control
all
those
connections
and
do
all
the
right
things
in
the
mobile
core,
and
if
you
think
about
this
in
terms
of
somebody
like
vodafone's
network
or
something
like
that,
they
ran
well,
that's
obvious.
B
Now,
if
you
look
at
that,
there's
an
awful
lot
of
different
pieces
in
there
there's
a
lot
of
technology
doing
an
awful
lot
of
different
things.
I'm
not
going
to
talk
you
through
all
the
different
bits.
Here
I
couldn't
anyway.
I
can't
remember
what
half
of
them
are,
but
there
are
a
couple
of
things
you
really
should
take
away
from
this.
The
first
thing
is
that
this
is
so
it's
a
standards
based
architecture,
some
big
standards
committee
sat
down
and
they
wrote
an
architecture
which
they
said.
B
These
are
all
the
components
you
need
to
have.
These
are
what
we're
going
to
call
them,
we'll
give
them
lots
of
three-letter
acronyms
we're
running
out
of
letters
at
the
time
they
got
to
af
and
we're
going
to
define
all
the
protocols
between
them
and
that's
very
much.
A
micro
service,
type
architecture,
they've
ended
up
with
very
deliberately
they've
thought
about
this,
and
they
said,
let's
design
a
microservice
architecture
that
will
run
somewhere
like
kubernetes,
which
is
why
5g
and
kubernetes
are
going
to
go
together
in
the
future.
B
The
5g
architecture
is
designed
to
be
microservices
like
to
run
with
kubernetes
and
in
principle
you
can
take
any
of
these
components,
take
them
out
and
throw
them
away
and
replace
some
of
the
component
written
by
another
vendor,
because
they're
all
standards
based
in
practice.
Of
course,
it's
not
as
easy
as
that,
but
that's
the
idea
behind
it
all.
B
So
the
only
other
thing
I
want
you
to
take
away
from
that
other
than
there
are
lots
of
microservices
components,
each
of
which
are
well-defined
jobs
and
a
clearly
defined
way.
That
works
very
well
with
kubernetes.
The
other
thing
I
want
you
to
take
away
from
this
is
that
piece,
the
upf,
the
user
plane
function
at
the
bottom.
The
user
plane
function
really
is
just
the
thing
that
takes
in
packets
from
some
radio
mast,
which
is
to
talk
to
a
phone
and
sends
them
off
somewhere
else.
B
It's
acting
pretty
much
like
an
internet
router
and
that's
going
to
be
important
and
some
of
the
things
I'll
say
a
little
bit
later.
That
is
one
of
the
components
and
that's
just
running
like
everything
else
in
here
is
running
in
kubernetes
I
should
say
rather
obviously
the
radio
mast
is
not
part
of
the
kubernetes
network.
This
is
the
part
of
the
mobile
core
part.
B
Okay,
so
we've
got
this
wonderful
architecture.
I
said
that
5g
was
going
to
be
better
than
4g,
if
you
hope,
so,
I'm
not
going
to
talk
about
the
technical
ways.
It's
better,
I'm
going
to
talk
a
bit
more
about
the
use
cases.
So
there
are
several
different
use
cases
that
are
important.
The
first
one
is.
We
have
mobile
networks
out
there
already
we
have
vodafone.
We
have
a
moment
of
not
being
able
to
remember
who
any
others
were
vodafone,
o2
3,
all
of
these
networks.
B
All
of
them
are
going
to
be
bigger
and
better
and
faster,
and
that's
frankly,
very
boring.
The
fact
that
you
can
get
better
download
speeds
in
your
phone.
Well,
it's
obviously
nice
to
have,
but
it's
not
something.
That's
really
interesting
from
a
technical
point
of
view,
unless
you're
very
deep
in
the
nerdy
technical
details,
something
much
more
interesting
and
that
5g
will
enable
is
private
mobile
networks.
So
the
idea
here
is,
you
can
have
your
own
network
completely
independently
of
the
big
mobile
phone
providers.
B
B
We
will
set
up
our
own
radio
mast
and
we
will
run
our
own
5g
network
on
it
and
the
standards
are
written
in
a
way
that
allows
you
to
do
that
to
have
your
own
network
and
use
junk
ban
spectrum
just
like
wi-fi
to
set
your
setup,
your
own
mobile
phone
network
within
a
factory,
so
that
so
that
would
allow
you
to
have,
for
example,
a
factory
where
everybody
gets
a
mobile
phone
and
they
can
only
call
other
people
within
the
factory.
B
B
So
the
idea
here
is:
let's
imagine,
that
you
want
to
have
your
own
mobile
phone
network
that
runs
across
the
entire
country
instead
of
building
your
own
radio
networks
everywhere
you
go
and
talk
to
somebody
who
already
has
a
bunch
of
radio
networks
such
as
what
you
say
and
you
pay
them
some
money,
and
then
they
give
you
a
small
slice
of
their
network
which
is
segmented
from
everybody
else's.
So
you
can
only
make
calls
privately.
B
Another
important
thing
is
replacing
existing
networks,
so
5g
is
getting
bigger
and
better
and
faster,
and
it's
now
possible
to
run
private
networks.
One
of
the
things
you
can
do
is
you
could
replace
existing
broadband
instead
of
running
broadband
cables
somewhere,
you
could
just
have
5g
dongles
and
look.
We
don't
have
to
lay
cable.
B
One
of
my
favorite
5g
use
cases
is
inside
an
airliner.
We
have
an
airliner,
we
have
lots
of
video
consoles
in
the
back
of
the
seats
they
need
to
have
video
stream
to
them.
Well,
we
need
a
fast
network.
Well
is
to
run
a
5g
network
inside
the
plate,
and
that
turns
out
to
be
cheaper
and
more
reliable
than
running
cable
under
the
floor
which
gets
damaged.
B
B
I
don't
think
we're
scratching
the
surface
of
the
things,
the
things
that
it
will
end
up
doing
so
we've
got
this
product
called
a
zero
private
mobile
network
that
we're
that
at
the
moment,
is
in
private
preview,
which
means
you
only
get
it
if
you
ask
nicely,
but
in
a
little
while
we'll
become
publicly
available
and
the
kind
of
target
markets
that
we're
looking
at
and
where
we've
got
a
lot
of
interest.
Are
people
like
factories,
so
somebody
running
a
factory?
B
Well,
they've
got
machines,
they've
got
robots,
those
robots
need
to
be
able
to
download
upgraded
software.
They
may
have
3d
printers
that
need
to
get
their
designs
or
whatever
they're
printing
downloaded
from
somewhere.
They
don't
want
to
run
cable
underneath
the
concrete
floor,
and
if
they
did,
it
might
break
and
then
have
to
be
replaced.
So
they'd
much
rather
run
a
5g
network
and
they
really
don't
want
to
run
wi-fi
because
wi-fi
just
isn't
reliable
enough
or
secure
enough.
B
So
there
are
a
whole
bunch
of
industrial
use
cases
and
that's
the
target
that
this
project
was
aiming
at.
So
what
hopefully
I've
done?
There
is
giving
you
a
little
bit
of
background
into
the
first,
the
three
things
I'm
going
to
talk
about,
which
is
5g,
why
it's
important
what
is
going
to
change
and
what's
different
between
5g
and
previous
mobile
standards,
so
the
next
bit
of
the
title
is
on
the
edge
of
the
zero.
B
So
I
should
say
what
I
mean
by
edge,
because
edge
is
one
of
those
terms
that
has
become
cool,
and
so
everybody
means
something
different
edge
to
me.
Just
means
on
premises.
It's
the
thing
which
you
have
in
your
on
in
your
building
as
a
factory
owner
say
rather
than
out
in
the
cloud
or
in
somebody
else's
data
center.
B
The
thing
that
you
control
and
one
of
the
things
that,
if
you're
running
applications
you
have
to
think
about
is,
am
I
going
to
run
it
on
the
edge
or
am
I
going
to
run
it
in
the
cloud
and
the
advantages
of
running
it
in
the
edge
are
reliability
and
availability
factory
owners
get
very
unhappy.
If
the
factory
stop
factories
robots
stopped
working
because
the
internet
connectivity's
failed.
You
can't
have
that
kind
of
thing.
You
need
to
have
much
greater
availability
than
that.
B
B
It
doesn't
help
to
send
it
out
to
the
cloud
and
back
again
that
just
ends
up
being
far
too
slow.
So
there
are
a
whole
bunch
of
case
use
cases
that
require
you
to
use
the
cloud.
On
the
other
hand,
you
sorry
that
require
you
to
use
the
edge.
On
the
other
hand,
there's
a
number
of
use
cases
where
you'd
really
like
to
be
in
the
cloud.
The
cloud's
got
some
great
things.
It's
got
better
apis
and
all
these
wonderful
services
you
can
use.
B
It's
got
a
great,
consistent
authentication
model
where
you
have
one
authentication
database
for
all
your
admins
and
one
and
all
your
users
in
one
place.
Rather
than
having
to
manage
that
yourself.
It's
got
life
cycle
management.
The
decentralized
got
its
own
monitoring,
there's
a
whole
bunch
of
things
about
the
cloud
that
just
make
it
a
better
way
of
managing
things,
particularly
when
you
have
multiple
sites
so
you're
in
this
awkward
situation.
B
Where
cloud
has
all
these
advantages,
but
you
have
to
use
the
edge
because
it
has
mandatory
requirements
that
you
just
have
to
get
along
and
use
and
all
of
the
clouds
I'm
going
to
talk
about
a
zero,
because
I
work
for
microsoft.
Other
clouds
exist.
I
will
say
that
azure,
I
think,
is
ahead
in
this
area.
All
of
the
clouds
have
gone
for
a
model
where
you
have.
You
can
manage
things
in
the
cloud,
but
they
have
something
or
other
that
runs
in
the
edge.
B
I'm
not
going
to
try
and
explain
all
the
things
on
this.
The
thing
you
should
take
away
from
this
is
this:
a
nice
glossy
slide,
which
means
I
stole
it
from
some
marketing
slide,
as
opposed
to
putting
it
together
myself,
but
we
have
a
whole
bunch
of
different
think
ways
of
running
software
on
the
edge
that
is
controlled
to
some
greater
or
lesser
extent
from
the
cloud,
and
what
I
want
to
do
is
try
to
talk
here
through
a
little
bit
about
how
this
actually
works
in
practice.
B
So
suppose
I'm
running
my
factory
and
I
want
to
run
something
in
it,
and
I
want
to
manage
it
from
the
cloud.
How
does
that
work?
Well,
there's
two
ways
I
can
get
equipment
running
on
the
edge
that
is
management
zero.
The
first
thing
I
can
do
is
I
can
go
to
something
called
azure
stack
and
I
can
just
buy
physical
equipment
and
that's
a
picture
of
one
of
the
boxes
there.
B
Clearly,
there
are
some
it's
slightly
more
complicated
than
that,
because
you
have
to
agree
to
rent
it
for
some
length
of
time
and
so
forth,
but
effectively
what
you're
doing
is,
instead
of
renting
a
vm
you're,
renting
a
physical
machine
that
you
then
rack
up
and
plug
in
yourself
and
the
other
thing
you
can
do
I'll
talk
about
how
that
works
technically
in
a
moment,
because
it's
the
same
way
as
the
next
point
I'm
going
to
make
as
the
next
example,
which
is
azure
arc,
enabled
servers
and
there
the
idea
is.
B
I've
got
my
vm
and
it's
running
in
vmware
or
it's
maybe
it's
just
a
physical
server.
What
I
want
to
do
is
I
want
to
manage
that
server
through
azure.
So
the
way
I
do
that
is,
I
install
an
agent
on
it,
which
is
exactly
the
same
thing
as
you
get
when
you
buy
an
azure
stack
box
and
that
server
agent
phones,
home
to
azure,
buy
some
secure
connection,
and
it
then
gives
you
various
management
capabilities,
so
the
sort
of
things
it
can
do.
B
If
you
want
to
centrally
manage
your
configuration,
you
can
manage
your
configurations
or
configuration
database
in
azure
and
have
all
your
servers
on
site
download
it
in
such
in
that
case.
Well,
clearly,
if
your
internet
connection
goes
down,
you
no
longer
have
that
metrics
view,
and
you
can't
change
your
config,
but
your
servers
can
continue
to
work
normally
and
there.
There
are
a
whole
bunch
of
different
capabilities
that
you
can
configure
that
these
agents
can
enable
probably
the
most
important
I
see
the
most
important
ones
will
very
well
very,
very
very
enormously.
B
According
to
what
your
use
case
is
the
most
important.
One
generally
is
just
that
secure
connectivity
that
you
can
and
the
fact
that
from
azure
you
can
now
see
all
of
your
deployment.
You
can
see
all
of
your
vms
in
azure,
but
you
can
all
see
all
of
your
vms
and
physical
servers
in
your
factory
site
all
across
the
world,
and
you
can
see
all
of
you
can
get
into
them.
You
can
see
all
their
metrics.
You
can
see
their
state.
B
You
can
enforce
policy,
it's
quite
dull,
but
it's
very
important
to
be
able
to
enforce
policies
such
as
root
access
must
be
disabled
on
this
server.
You
can
look
across
all
your
servers
from
azure
and
see
not
just
all
your
physical
servers,
all
your
vms
everywhere
and
centrally
manage
that
kind
of
policy.
B
Now
you
might
wonder
why
would
I
buy
an
azure
stack
box
rather
than
just
getting
a
box
of
my
own
and
arc,
enabling
it,
and
the
main
reasoning
is
that
the
azure
stack
box
gives
you
a
little
bit
of
azure
capability
locally.
It
allows
you
to
use
azure
apis
to
do
things
like
spin
up
vms
on
premises,
so
you
only
have
one
vm
spin
up
model
if
you,
if
in
use
in
both
cases,
okay,
so
let's
talk
a
little
bit
about
the
private
mobile
network
user
journey.
B
What
it
actually
looks
like
when
somebody
comes
along
and
says
I
want
to
buy
private
mobile
network
from
my
factory.
So
the
first
thing
that
happens
is
they
go
to
azure
and
they
put
in
a
credit
card.
Well,
clearly,
not
a
credit
card
they'll
be
setting
up
their
company's
billing
and
they
buy
some
things
on
azure
and
a
little
bit
later.
Somebody
rings
the
doorbell
and
some
zero
stackage
boxes
and
equipment
arrive.
B
They
then
have
to
find
a
place
for
them
and
plug
them
all
in
they
have
to
set
up
a
radio
access
network
and
plug
that
in
as
well
when
I
say
set
up
a
radio
access
network
there,
you
can
just
buy
these
things,
it's
just
like
buying
a
wi-fi
setup.
You
buy
one,
you
normally
get
a
partner
to
set
it
up
for
you,
you
plug
it
in
and
away
you
go.
B
You
set
up
some
sim
cards
and
all
your
equipment,
and
then
you
have
a
network
running
and
potentially
in
multiple
sites
across
multiple
fights,
multiple
factories
across
your
entire
manufacturing
organization,
where
you
can
see
all
the
endpoints
that
are
connected,
that
is
completely
private,
completely
secure
and
managed
through
azure,
which
is
quite
cool,
and
it
will
then
make
you
lots
and
lots
of
money,
and
you
can
go
away
and
be
rich,
and
I
think
of
the
three
things
I'm
going
to
talk
about.
This
is
the
one
that
I
think
is
probably
most
surprising
to
people.
B
People
are
aware
of
5g
I'm
going
to
talk
about
some
kubernetes
things.
I
think
a
lot
of
people
are
aware
of.
This
is
the
one
that
is
interesting
and
very
is
a
very
strong
capability
and
a
very
strong
sale
for
many
of
our
customers,
just
the
ability
that
you
can
manage
your
software,
wherever
it
is
in
one
place
through
azure,
okay,
so
the
full
title
was
5g
on
the
edge
with
azure
and
kubernetes.
B
So
a
few
kubernetes
specific
things
coming
out
of
all
this
things
we
had
to
do.
Firstly,
it's
kind
of
the
same
point
as
before
about
how
azure
art
gives
you
management
now
azure
art
I've
enabled
servers.
We
also
have
the
concept
of
azure
arc
enabled
kubernetes,
and
the
model
here
is
very
simple:
you
take
a
kubernetes
cluster,
any
kubernetes
cluster.
Only
some
of
them
are
supported,
but
frankly
it's
any
cncf
validated
distribution
of
kubernetes.
There
are
a
very
large
number
that
is
supported
and
most
of
the
others
will
frankly
work.
Just
fine.
B
You
deploy
your
kubernetes
as
usual.
You
then
install
some
agents
on
it
and
then
you
can
manage
that
kubernetes
cluster
from
azure,
with
all
of
the
things
that
gave
you
could
see
into
your
cluster,
you
can
get
inventory
and
you
can
organize
it
just
as
I've
said
before,
you
also
have
all
the
policy
so,
for
example,
if
you're
kubern,
if
you
want
your
kubernetes
cluster
to
be
configured
such
that
root
access
is
never
granted
to
any
containers
all
the
kind
of
security
policy
stuff.
B
B
So
it's
really
just
githubs,
so
you
have
declarative
config,
which
is
stored
in
the
zero
devops
gate
repo.
I
say
an
azure
devops
get
repo!
That's
because
I've
been
working
in
azure
for
long
enough
now
that
I've
forgotten
that
anything
else
exists,
but
it
can
be
in
any
git
repo
anywhere
that
is
accessible
from
the
instance.
You're
running
you
store
your
kubernetes
objects
as
text
files,
you
put
all
your
binaries
in
zero
container
registries.
Again,
other
clouds
exist,
you
put
your
beheld
charts
there.
B
You
contain
your
images
and
then
you
have
an
agent
running
in
cluster.
That
just
does
all
the
githubs
looks
at
that
git
repo
and
keeps
everything
up
to
date,
and
this
is
all
just
built
on
standard,
open
source
components.
So
it's
flux
and
the
helm
operator
as
usual.
There's
nothing
very
surprising
about
this.
B
B
So
every
time
a
packet
arrives,
you
get
an
interrupt
when
that
interrupt
arrives,
the
kernel
stops
and
says:
oh,
a
new
packet
has
arrived
I'll,
go
tell
whatever
application,
it
might
be
about
it.
B
Well,
okay,
that's
that's
great,
but
what
that
means
is
the
kernel
has
to
get
an
interrupt
for
every
packet
and
that's
fine
if
you're
running
a
gigabit
in
and
out,
but
it's
not
great
if
you've
got
100
gigabit
piper
at
one
side
and
100
gigabyte
pipe
out
the
other
side
and
you're
trying
to
stuff
traffic
between
them.
So
if
you're
a
rooter,
that's
just
not
fast
enough.
Similarly,
it's
not
great
if
you're
a
telco
app,
because
normally
the
way
you
get
a
lot
of
bandwidth
is
you
have
large
mtu
and
large
packets
being
passed
through.
B
Well,
that's
not
great!
If
you're
trying
to
do
telephony
because
within
a
telephony
call,
you
are
sending
streams
of
fairly
small
packets
with
here's
the
next
10
milliseconds
of
audio.
So
and
you
have
very
long
huge
numbers
of
these
streams,
sending
very
large
numbers
of
packets,
and
you
have
millions.
B
Packets,
a
second
and
again,
the
kernel
just
can't
cope
with
that.
You
end
up
with
a
kernel
just
thrashing
back
and
forth
burning
all
the
cpu
handling
the
network
and
your
your
network
connections
are
running
at
10
capacity
with
your
kernel
running
100
cpu,
so
that's
not
much
use
and
in
5g
well.
Are
we
a
richer?
B
So
the
way
that
you
get
around
this
is
something
called
user
space
networking,
which
is
quite
a
cool
idea,
there's
something
called
dpdk
which
is
written
by
intel
a
little
while
ago.
I
think
it's
data
plane
development
kit,
something
like
that.
It
allows
you
to
write
drivers
in
user
space
and
when
and
when
I
say
drivers
in
user
space.
What
I
mean
by
that
is
that
the
kernel
on
the
box
you're
running
on,
doesn't
even
see
the
network
device,
so
the
network
device
is
only
visible
to
some
application
that
you're
running.
B
B
So
the
the
this
core
runs
along
grabs,
a
packet
off
the
interface
puts
it
where
it
needs
to
be
and
goes
and
checks
checks
again
takes
another
packet
off
of
the
interface
and
puts
it
where
it
needs
to
be
as
fast
as
it
can,
and
it
does
that
in
a
tight
loop.
So
you
have
one
core
running
100
cpu
all
the
time,
but
because
of
that,
it
is
able
to
handle
the
network
traffic
vastly
more
efficiently.
B
So,
instead
of
using
up
several
kernel
threads
at
a
hundred
percent
by
being
interrupted
all
the
time,
you
have
one
core
thread
that
is
running
at
100
grabbing
packets,
so
it's
fairly
specialized
stuff.
But
it's
what
all
reachers
do
it's?
What
most
of
the
high
intel
collapse
do
at
the
moment
and
therefore
it's
what
we
would
tend
to
want
to
do
in
the
upf?
B
So
the
other
problem
with
networking
normally
is
that
the
hypervisor
networking
is
just
too
slow.
So
if
you
have
a
vm
and
it
has
some
network
devices
normally
what
happens
is
there's
some
kind
of
virtual
device
involved
and
packets
are
sent
around
and
it
goes
to
the
kernel
and
the
host
and
the
kernel
and
the
vm,
and
that's
not
going
to
go
anywhere.
B
So
there's
something
called
sri
ovu,
which
effectively
allows
you
to
have
a
physical
network
device
visible
to
the
v
to
a
vm
on
a
host
and
that
that
again
is
something
that
gets
used
by
all
fast
networking
applications.
So
if
you
learn
nothing
else
about
fast
networking,
this
will
kind
of
give
you
a
picture
of
the
space.
B
Okay,
so
I've
said
this
stuff
about
fast
networking,
but
this
is
the
kubernetes
section.
Where
does
kubernetes
come
in
most
likely?
My
slide
would
like
to
go
forward
that
would
help
there
we
go
so
this
is
the
standard
kubernetes
networking
model.
I
don't
want
to
spend
ages
talking
about
it.
I'm
not
sure
I've
got
a
huge.
A
B
Of
time
anyway,
but
the
point
I
would
make
about
this
is
that
there
are
lots
of
different
lots
of
different
pieces
in
this
picture,
so
typically
every
pod.
Well,
every
node
will
have
a
network
device
kind
of
more
than
that.
It'll
have
a
linux
bridge
that
is
routing
all
the
traffic
from
that
network
device
to
all
the
pods.
Every
pod
lives
in
its
own
network,
namespace
with
one
physical,
sorry,
one
virtual
interface,
which
are
marked
as
v,
vest,
zero
and
vet1.
They
all
have
their
own
ip
addresses.
B
And
if
you
look
at
this,
there's
an
awful
lot
of
things
going
on
inside
that
node
there's
the
network
interface,
which
has
got
kernel
routing
on
it,
that
there's
the
traffic
to
the
linux
bridge,
which
again
is
going
through
kernel,
routing
and
then
there's
a
vehicle
which
guess
what
more
kernel
routing
that
doesn't
perform
very
well.
It
performs,
surprisingly
well
actually
considering
what
it's
all
the
things
it's
doing,
but
it's
just
not
going
to
work
when
you
need
very
high
traffic.
B
So
the
first
one,
just
just
to
recap
what
I
said:
one
ip
for
pod,
shared
overall
containers,
the
ips
known
used
by
kubernetes
and
then
those
ips
are
set
up
by
what's
called
cni
plugins.
This
is
quite
an
important
piece.
B
The
cni
plug-in
has
the
job
that
every
time
a
new
pod
is
created
or
destroyed,
the
cni
plugin
is
called
until
got
a
new
new
pod.
Do
something
with
it,
and
the
cni
plugin
will
then
go
and
create
these
virtual
interfaces
and
so
forth,
and
then
report
back.
This
is
the
ipi
set
and
there
are
lots
of
different
cni
plugins.
B
B
A
B
Time
you
call
it
to
configure
to
configure
a
pods
networking.
It
calls
multiple
cni
plug-ins.
So
the
first
thing
it
looks
at
some
annotations
in
the
pod.
That
say
I'm
going
to
be
on
the
fast
network
and
the
pod
network,
or
maybe
just
says,
I'm
just
on
the
pod
network
and
then
from
that
it
decides
which
cni
plug-ins
you
need
to
call
in
which
order
to
set
up
all
the
various
networks.
So
you
have
a
support
network
and
you
have
additional
networks
and
there's
a
standard.
B
That's
written
by
the
kubernetes
network,
plumbing
working
group,
which
describes
how
this
all
works.
Multis
is
the
standard
reference
implementation
and
the
effect
of
all
this
is
that
you
can
now
have
multiple
networks.
Kubernetes
only
sees
one
of
those
networks,
but
the
other
network
is
set
up
under
the
covers
and
you
can
do
what
you
like
with
it,
to
get
fast
networking
and
in
the
practical
way
that
you
tend
to
do.
B
B
The
customer
just
has
to
enable
the
service
in
the
azure
portal,
and
this
is
going
back
to
the
whole,
managing
it
all
centrally.
In
a
zero
point,
you
just
turn
it
on
the
a0.
Once
you've
got
private
mobile
network,
you
push
the
button
and
everything
just
happens,
and
then
you
have
everything
beautifully
set
up
for
you,
so
the
customer
never
gets
to
see
any
of
this
complexity.
B
Okay.
So
on
that
time,
I'm
going
to
wrap
up,
I
think
the
three
key
things
I've
said
here.
The
first
thing
is
5g
is
here
at
last,
and
it's
probably
not
what
you
think
it
is
if
you're
thinking
of
it
as
a
mobile
phone
system,
it's
really
a
little
bit
bigger
than
that.
It's
about
private
networks,
particularly
and
about
new
use
cases
that
are
surprising
and
a
little
bit
more
decentralized
than
you
might
have
thought.
I
don't
think
we
even
know
what
those
use
cases
are
yet.
B
A
That
was
that
was
super
peter.
Thank
you.
I'm
sitting
here
feeling
a
little
like
my
mind,
has
been
blown.
We've
had
a
couple
of
comments
actually
in
the
in
the
slack,
mostly
about,
and
so
tim
banister
said,
the
summary
of
going
from
1g
to
5g
seems
like
every
time
they
increment
the
number.
The
definition
of
improvement
becomes
slightly
unclear.
A
B
B
B
And
more
powerful
every
time,
as
well
as
everything
else
yeah,
but
it
the,
I
think,
what
the
the
each
generation
they
look
at
the
previous
generation
and
decide
what
it
didn't
do
very
well
and
sometimes
they've
added
enormous
new
features.
So
the
the
big
jump
to
me
was
from
2
to
3g
and
2g
was
phone
calls
worked
pretty
well,
you
could
have
a
mobile
phone
and
actually
make
phone
calls
on
it
and
it
was
okay,
but
data
was
just
you
unusably
awful
and
in
3d
it's
like.
B
Oh
we've
got
data's
connection
now,
we've
added
the
ability
to
do
the
internet
on
your
phone,
and
that
was
the
thing
that
made
things
like
the
iphone
possible,
but
nobody
would
have
thought
of
that
unless
they'd
gone
through
having
mobile
phones
as
phones
and
gone
through
the
connection
before
so
I
don't
actually
have
the
faintest
idea
what's
in
60
or
what
they
think
is
wrong
with
the
5d
sounds
of
what
they
can
do
better,
but
each
each
time
they
look
at
the
look
at
the
standards
they
think
of
something
clever.
They
can
add
on
top.
A
Yeah,
it
is
interesting
to
wonder,
and
one
of
the
one
of
the
comments
you
made
just
at
the
end,
there
was
obviously
5g
is
here
and
it's
sort
of
in
in
the
wild.
Are
there
any
case
studies?
B
That's
a
good
question,
I
don't
know
to
be
honest,
I
know
that
there
are
so
one
thing
about
5g
and
the
big
mobile
networks
is
that
there's
two
parts
to
5g:
there's
the
ransidens
and
the
core
side,
so
those
two
can
be
upgraded
independently.
You
can
have
a
5g
core
and
a
4g
network
and
inevitably.
B
So
all
the
4g
networks
are
still
going
to
be
there,
but
you
want
to
upgrade
your
core
to
the
latest
and
greatest
core,
so
you
can
get
onto
the
5g
backbone
from
so
you
can.
So
you
can
upgrade
the
ram
and
the
core
independently
of
one
another
in
a
complicated
way,
and
I
think
some
of
the
big
providers
are
upgrading
their
cores
now,
but
they
haven't
got
very
far
in
terms
of
upgrading
their
rams
because
that's
actually
a
bigger
deal
in
many
ways,
because.
B
Similar
thing,
a
lot
of
the
improvements
are
going
on
inside
some
carriers-
data
centers,
so
you
just
can't
see
them
if
you're
using
a
phone,
you
don't
notice
that
the
carrier's,
you
know
saved
three
percent
and
is
now
being
able
to
oper
on
its
operating
costs
and
is
now
being
able
to
offer
different
services
to
different
people.
The
the
five
key
use
cases
in
kind
of
factories
and
things
I
think,
that's
relatively
new.
I
know
that
we're
in
in
terms
of
what
is
public
from
the
azure
work.
B
B
But
I
there
are
other
vendors
out
there,
and
so
I
think
we're
at
the
stage
where
a
small
number
of
factories
and
installations
are
trying
these
things
out
at
the
moment-
and
I
don't
know
whether
they're
very
many
of
them
are
actually
using
them
for
real
live
workloads
that
are
actually
production
critical
yet,
but
I
think
it'll
be
very
very
soon,
because
at
least
I
know
of
a
number
that
are
trialing
and
setting
things
up
and
a
lot.
A
Yeah
interesting
well
we're
actually
over
time
slightly.
Thank
you
so
much
for
that
talk.
It
was
really
interesting.
If,
if
anyone
has
any
questions,
then
please
feel
free
to
post
them
in
the
slack.
Hopefully
peter
will
have
a
chance
to
look
at
them,
maybe
later
on.
That's
it
for
now
it's
lunch
break
time.
So
thanks
again
peter
and
thank
you
for
watching
we'll
take
a
break
now
and
we'll
be
back
at
two
o'clock.
So
thanks
very
much
everybody
and
see
you
later.