►
From YouTube: Kubernetes WG IoT Edge 20230405
Description
April 5, 2023 meeting of the Kubernetes IoT Edge Working Group.
Larry Peterson, CTO of the Open Networking Foundation (ONF), well known in software defined networking (both academics and industry), will discuss software defined networking and the Open Networking Foundation (ONF). This talk will be an interesting intersection between networking and the edge. Coverage of the aether project https://opennetworking.org/aether/.
Also a little on the P4 project https://opennetworking.org/p4/
A
Hi
welcome
to
the
April
5th
meeting
of
the
cncf
iot
edge
working
group
on
the
agenda.
Today
we
have
a
guest
Larry
Peterson,
who
is
CTO
of
the
open
networking
foundation
and
a
very
well
recognized
name
in
software-defined
networking,
both
in
academics
and
Industry,
and
one
of
the
one
of
the
people
here
even
indicated
that
he
loved
a
book.
Larry
has
written
on
the
subject
so
without
further
Ado
I'm,
going
to
turn
it
over
to
you
Larry,
and
you
can
talk
about
all
things.
Networking
and
Edge
sure.
B
Sure
happy
thanks
and
thanks
for
the
invitation,
so
I
I
have
a
few
slides.
I
thought
I
would
go
through
I,
don't
want
to
take
up
the
whole
hour,
maybe
half
of
it,
maybe
not
even
that
much
kind
of
give
you
a
sense
of
what
we've
been
doing
and
where
we
see
the
opportunities
and
then
open
it
up
for
discussion
and
questions
and
whatnot,
but
also
just
feel
free
to
jump
in
and
interrupt
me
at
any
point
dupe.
Let
me
go
ahead
and
share
my
screen.
B
Yeah,
so
what
I'm
going
to
talk
about
is
a
project
that
we've
had
at
the
open
networking
foundation
for
the
last
three
plus
years
called
ether,
one,
it's
private,
5G,
two,
it's
an
edge
cloud
and
three.
It
builds
on
a
lot
of
the
sdn
principles
at
o
and
F,
and
before
that,
Owen
lab
have
been
pushing
forward
for
for
the
last
decade.
B
In
case
you
don't
know,
onf,
o,
f
and
Owen
lab
were
jointly
your
two
organizations
originally
created
about
10
years
ago
by
Scott
Shanker
and
Nick
McEwen
as
they
you
know,
the
sdn
initiative
launched,
as
you
probably
know,
nasira
was
acquired.
What
was
a
private
company
and
was
later
acquired
by
VMware,
but
they
wanted
an
open
source
variant
to
to
move
forward,
to
promote
the
ideas
and
that's
what
Owen
owenf
did
the
standards
open
for
originally
and
and
open
networking
lab?
Was
the
software?
B
The
two
merged
I,
don't
know
five
years
ago,
took
the
name
on
F
but
kept
the
open
software
Mission.
Any
case
ether
is
the
latest
project
and
in
some
sense
it's
kind
of
a
Capstone
project
of
the
last
decade
because
it
brings
together.
All
of
these
Technologies
in
one
in
one
platform:
okay,
so
now
why
can't
I?
Okay,
all
right,
so
what
I
wanted
to
do
is
just
5G
has
been
at
the
at
the
core
of
of
what
we've
been.
The
transformation
of
5G
has
been
at
the
core.
B
B
Quick
summary
of
that,
because
it
really
informs
how
we're
approaching
the
edge
and
so
there's
two
big
components:
there's
the
base
stations
in
4G
called
enote
bees
and
there's
the
packet
core,
which
is
basically
a
fancy
router.
It
has
all
the
billing
stuff
that
an
ISP
would
have,
but
it's
basically
a
router.
In
any
case,
it's
closed
at
the
telephony
system
and
certainly
the
the
base
stations
have
been
proprietary,
bundled
systems.
There's
a
software
stack
that
runs
there.
It's
the
standards
open,
but
the
software
has
not
been
for
some
time.
B
So
one
of
the
big
things
happened
in
5G
is
this
is
being
disaggregated
part
of
the
stack
moves
out
to
the
radio
part
of
the
stack
moves
into
a
distributed
unit
and
part
of
it
moves
into
a
central
unit.
The
gray
here
indicates
that
D
to
a
conversion
and
radio
stuff
that
I'm
not
going
to
talk
about
the
blue
is
the
basically
the
data
protocols,
data
layers,
data
plane
layers
and
the
red
is
the
control
plane
layer,
the
core,
which
is
also
bundled,
has
basically
turned
into
a
collection
of
microservices.
B
So
it's
now
called
the
next
generation
core
and
the
big
takeaway
from
all
of
this
there's
a
lot
of
disaggregation
happening
and
a
lot
of
open,
software-defined,
stuff
happening
and
now
pretty
much
everything
on
the
end-to-end
path.
Perhaps
save
the
the
amplifiers
and
the
antennas
in
the
DNA
conversion,
at
the
edge
at
the
very
very
edge,
are
now
open
source
they're,
quite
often
Cloud
native
and
they're.
The
adhered
sdn
principles.
B
B
Oh
one
more
point
before
I
get
there
I
did
call
out
the
sdn
principle.
Let
me
talk
about
that
just
a
little
bit
more,
it's
not
only
disaggregation
into
Cloud
native,
but
we
took
the
control
component
out
of
that
stack
and
turned
it
into
an
sdn
controller,
and
so
now
it's
not
local
in
an
individual
unit.
It's
now
a
global
thing
that
controls
a
set
of
units
and
I'm
not
showing
it
here,
but
there
could
be
many
Central
units.
B
There
could
be
many
distributed
units
as
they
as
they
fan
out
to
cover
some
geographic
area
and
certainly
many
radio
units,
and
just
like
sdn
in
in
the
wired
Network,
there's
a
there's,
a
controller
and
there's
a
set
of
applications
that
run
on
top
of
it.
They
happen
to
be
a
little
bit
different
than
you
might
be
familiar
with
the
you
know.
It's
handle
or
control
and
there's
still
load,
balancing
and
Link
aggregation.
D
B
Still
a
management
plan
on
top
of
that,
if
you
work
for
a
carrier,
that's
their
OSS
BSS
system,
but
for
your.
If
you
view
this
as
now
a
cloud
system,
that's
basically
your
your
management
plane
for
the
for
your
cloud
and
also
just
call
attention
to
there's
lots
of
control.
Well,
there's
three
basic
control:
loops:
there's
a
non-real-time
controller!
B
You
can
change
the
policy
there's
a
near
real-time
control,
Loop,
where
you
can
make
you
know
in
a
sort
of
the
sub
10
millisecond
decision
changes
in
in
configuration
parameters
and
then
there's
the
real-time
control
Loop.
That's
basically
the
scheduler,
the
schedule
switch
little
bits
of
data
gets
transmitted
at
any
given
time
and
again
these
are
available
as
open
source.
There
are
some
restrictions,
as
you
move
out
towards
the
edge
it's
it's
a
little
harder
to
get.
B
B
So
the
first
project
is
called
SD
core
and
that's
basically
just
a
cloud
native
control
plane
that
is
control.
This
is
basically
the
component
that
runs
in
the
core
and
controls.
You
know
your
mobility
and
tracking,
where
you
are
and
authentication
and
so
on,
and
it
don't
need
to
worry.
If
you
don't
know
the
details
of
each
one
of
these
components,
but
it's
literally
a
set
of
microservices
the
the
standardization
body
3gpp
in
my
view,
goes
farther
than
they
they
would
need
to
in
nailing
down
standard
interfaces.
B
These
could
have
just
evolved.
Just
as
you
know,
you
could
have
picked
up
a
directory
service.
You
could
have
picked
up
a
a
a
nosql
database.
You
could
have
picked
up
some
standard
authentication
things
and
built
this,
but
it
wouldn't
have
necessarily
complied
to
all
the
interfaces.
So
that's
that's
an
aspect
of
this.
That's
sort
of
a
legacy
from
the
telephony
days,
there's
actually
also
a
4G
control
plane,
in
addition
to
the
5G
control,
plane
and
SD.
Cork
includes
both
and
right
in
the
middle.
B
Here
is
the
data
plane
or
what
the
3gpp
folks
call
the
user
plane,
but
think
of
that
as
an
IP
router.
It
basically
connects
the
the
ram,
which
is
the
set
of
Base
stations
to
the
rest
of
the
internet
and
it's
under
the
control
either
of
a
4G
or
a
5G
core.
This
collectively
now
is
the
SD
core
project
of
onf
and-
and
it's
essentially
a
a
Helm
chart
defined
in
the
set
of
of
Docker
images
that
you
would.
B
You
would
want
to
deploy
the
control
that
was
mostly
the
control
plane,
the
user
plane.
Piece
of
it
is
interesting
in
the
tone
right
and
we
have
multiple
implementations
of
it,
we're
building
the
switching
fabric
of
our
Edge
Cloud
out
of
programmable
switches
and
smart
next.
So
we
have
a
variety
of
ways
of
of
building
out
the
data
plane
they're.
They
are
all
P4
programmable
that
just
gives
us
a
a
safe
way
of
programming
them.
B
So
there's
a
there's,
a
software
path
that
goes
all
the
way
up
into
a
server,
and
that's
you
could
think
of
that
as
a
microservice.
That's
implementing
the
the
data
data
path
or
there
is
another
path
that
actually
stays
entirely
inside
of
mostly
inside
of
the
programmable
switch,
but
occasionally
has
to
go
up
to
a
Smart
Nick
if
it
needs
to
buffer
some
data.
B
B
B
So
if
you've
followed
onf
at
sdn
for
in
the
last
decade,
you
might
be
familiar
with
onos
the
open
network
operating
system.
That's
been
reconfigured
from
a
set
of
components
into
what
the
telephony
world
calls
the
Rick
the
Rand
intelligent
controller.
That's
that
intermediate
control,
Loop
piece
that
is
near
real
time,
and
so
it's
built
from
a
you
know
a
distributed
store.
B
There's
a
topology
service
that
keeps
track
of
everything
out
in
the
ram,
that's
sort
of
wired
into
the
ram
like
where
the
base
stations
live
and
then
there's
a
a
device
service
that
keeps
track
of
where
everyone
all
the
movable
components
are
at
the
present
time
in
their
parameters,
and
then
you
can
also
configure
it.
So
there
is
an
in
an
organization
called
the
oran
Alliance,
which
is
basically
a
little
history
here.
B
The
3gp
apologies,
if
I'm
preaching
to
the
choir
here-
and
you
know
all
of
this,
but
just
real
quickly
3gpp
is-
is
vendor
dominated
and
the
carriers
wanted
to
have
to
break
vendor
lock-in,
and
so
they
created
the
o-rent
alliance
and
they
is
originally
at
T
and
China
mobile
and
they
set
out
to
Define
some
interfaces
that
would
make
it
possible
to
buy
the
different
components
from
different
vendors.
Long
story
short
there's
an
interface
called
E2.
B
That
is
how
this
controller
talks
to
the
various
elements,
and
the
points
I
wanted
to
make
here
is
all
of
this
has
been
running.
Dt
has
been
one
of
Bon
apps
partners
for
a
long
time
and
they
set
up
a
a
Rand
field
trial
in
Berlin
and
all
of
these
components
here
that
I'm
showing
you
will
have
been
running
there,
they're
tearing
it
down
now,
but
it's
it's
been
there
for
quite
a
while,
and
there
are
some
o
ran
service
models
that
have
been
defined
that
is
supported
by
this
software,
so
packets.
B
This
all
up.
This
is
a
second
component.
The
SD
the
SD
ran
and
oh
there's,
just
a
picture
of
the
open
run
trial,
a
little
bit
more
detail
I'm,
showing
the
SD
core
component
with
both
the
5G
and
4G
core
I'm,
showing
the
Rand
component
and
there's
some
apps
from
different
partners
running
above
it.
We've
got
a
disaggregated
split
ran
with
the
CU
and
the
DU,
and
the
ru
separated
Radisson
has
been
a
long
time
partner,
there's
also
a
switching
fabric
which
I'll
get
to
in
a
little
bit.
B
But
remember,
I
told
you
that
the
data
plane
of
the
5G
core
was
implemented
in
P4.
Well,
that's
a
that's
a
basically
a
switching
fabric.
That's
people
are
programmable
back
in
here.
So
Intel
is
certainly
a
partner
and
all
of
this
with
Deutsch
Telecom
wrapping
it
all
up
so
SD
core
SD
fabric,
which
I'll
talk
a
little
bit
more
about
in
a
minute
and
SD
ran,
are
three
software
packages
open
source
projects
of
the
onf?
All
three
are
basically
microservice,
based
sodium
obvious
relevance.
B
So,
in
a
sense,
what
ether
is
is
really
a
prod,
an
integration
project
to
pull
all
those
parts
together?
That's
one
way
of
thinking
about
it.
So,
on
the
one
hand,
it
exists
by
itself
because
there's
a
whole
lot
of
management
stuff
that
goes
on
another
way
of
looking
at.
It,
though,
is
it's
it's
in
some
sense,
a
little
more
than
an
integration
of
all
those
other
pieces.
B
So
this
is
now
getting
closer
to
where
I'm
expect
your
activity
is,
which
is
what
we
are
doing,
is
building
an
edge
cloud
and
I'm
giving
you
a
little.
You
know
a
simple
Edge
Cloud
here
and
by
the
way
Edge
Cloud
could
be
multi-rack
with
a
switching
Fabric
or
it
could
be
as
small
as
a
single
server
and
I'm,
throwing
showing
you
three
servers
here.
It
could
have
programmable
switches
or
it
could
have
dumb
switches.
B
Those
are
all
configuration
parameters,
but
whatever
that
cluster,
however,
whatever
the
bomb
the
bill
of
materials
for
that
cluster
is,
it
runs
the
sd-ran
set
of
microservices,
the
SD
core
user
plane
set
of
services.
That's
the
data
path
and
then
whatever
Edge
apps
you
want
to
run
is
other
microservices
there,
and
this
is
already
an
on-prem
connected
to
a
set
of
small
cells
and
then
back
Central
off-prem,
and
this
could
be
literally
in
Google's
Cloud,
which
is
where
we
run
it
in
our
deployment.
But
it
could
also
be
your
Enterprise's
Central
cloud.
B
B
Going
to
talk
about
next
is
the
basically
the
the
cloud
management
platform
that
turns
those
set
of
components
into
a
managed
cloud
service
in
a
very
familiar
way,
but
the
the
importance
of
of
the
way
this
is
configured
is
that
you
have
something
called
local
breakout,
which
is
to
say,
communication
between
anything.
That's
at
the
edge
your
iot
devices
and
the
edge
apps
that
are
managing
those
devices
in
some
way
never
leaves
the
premises
because
the
ran
and
the
core
data
path.
B
Data
plane
are
all
co-located
on-prem,
and
this
is
something
called
local
Breakout
that
has
obvious
performance
value,
latency
bandwidth,
but
also
security
value,
because
you
never
go
off-prem
okay.
So
how
do
we
build
either
Hardware?
It's
all
commodity,
I'm,
going
to
show
you
smart
switches,
but
these
could
be
dumb
switches
as
well
and
a
set
of
a
set
of
servers.
B
B
It's
all
kubernetes
managed
and
when
I
and
I've
got
kubernetes
here,
you
should
read
all
the
entire
Cloud
native
stack
that
goes
there.
It's
kubernetes,
it's
Helm,
it's
Fleet,
we
use
terraform
so
that
entire
sweet
is
is
actually
running
over
here
and
there's
an
OS
that
runs
in
all
the
switches
and
then
there's
the
network
operating
system
that
manages
the
fabric.
So
these
dotted
edges
means
this
is
the
control
plane
for
all
of
these
switches,
which
is
basically
a
standard
just
like
Google?
Does
it
and
and
Microsoft?
B
Does
it
in
the
in
their
networks,
except
it's
all
open
source,
the
core
user
plane
again?
That's
they
did
plain
piece
runs
at
the
edge
and
it
can
run
as
a
microservice
on
the
servers
or
it
can
run
as
a
P4
program
in
the
switches,
I'm
kind
of
showing
you
both
here
and
the
sdram
runs
at
the
edges
and
again
the
SD
cores
control
plane
runs
off
premises
unless
you
choose
to
configure
it
to
be
on
on
this
cluster.
B
B
B
So
what
is
amp
amp
is
a
an
open
source,
Cloud
management
platform.
It
includes
resource
provisioning,
life,
cycle
management,
runtime
control,
monitoring
and
logging.
It
has
portals
both
for
users
and
think
of
a
user,
maybe
as
the
system
admin
at
an
edge
who
wants
to
set
up
new
iot
devices,
new
5G
devices,
whatever
they
may
be,
give
different
users
different
access
and
an
Ops
portal.
If
this
is
run
as
an
as
a
managed
service,
then
it's
The
Ether
management
team,
that
that
would
use
the
Ops
portal,
but
you
could
it.
B
But
let
me
go
through
the
other
ones.
Life
cycle
management
is
basically
a
continuous
integration
pipeline
and
I'm.
You
know,
and
out
of
that
comes
an
operational
cloud.
It's
built
entirely
from
open
source.
Many
of
these
projects,
from
the
the
cloud
native
from
you
know,
from
from
the
Linux
Foundation
yeah
I'm
not
going
to
go
into
the
details,
but
again
Docker,
Helm,
Rancher,
terraform
and
so
on.
Fleet
are
all
components
in
here.
B
The
monitoring
and
logging
component
should
be
no
surprise.
We
built
that
out
of
a
set
of
Open
Source
components
as
well
and
clearly
you
have
to
instrument
the
5G
components,
so
we've
added
the
instrumentation,
the
dashboards
and
the
panels
and
the
and
the
The
Collection
machinery
and
the
alarms
alerts
that
need
to
go
off
if
something
goes
down.
We've
added
all
that
specialization
for
the
5G
case,
but
it's
otherwise.
It's
it's
a
pretty
familiar
tracking
of
CPU
usage
and
bandwidth,
consumption
and
so
on.
B
So
this
is
all
important,
because
what
it
gives
you
is
now
an
API
that
allows
you
to
do
closed
loop
control.
So
this
runtime
control
is
the
piece
I
was
just
telling
you
about,
and
you
can
obviously
read
a
lot
of
monitoring
and
data
out
of
the
monitoring
side
and
then
and
then
close.
The
loop
I'll
show
you
how
we
use
that
in
a
minute,
but
this
is
kind
of
the
ultimate
thing
that
we
were
after
here.
B
So
that's
I'm,
just
really
quickly
talked
about
the
open
source
components
of
build
the
management
platform
and
and
generating
the
control.
Api
is
maybe
the
the
biggest
intellectual
contribution.
If
you
will
to
all
of
this,
it's
been
deployed
for
some
time.
You
know
not
a
huge
number
of
sites,
but
more
than
two
or
three,
and
you
really
have
to
start
facing
real
operational
concerns
and
as
I
mentioned,
one
of
those
sites
has
been
this.
This
act
first
the
segregated
5G
field
trial,
and
that
was
done
in
collaboration
with
Deutsche
Telecom.
B
That's
my
overview
of
ether
and
I'm.
Just
going
to
take
a
few
minutes
now
to
talk
about
some
of
the
research
that
that
our
collaborators,
Stanford,
Princeton
and
Cornell
have
been
doing.
This
is
a
darker
funded
project
called
Pronto,
so
Pronto
funded
the
development
work,
the
Open
Source
development,
work
of
ether
and
then
the
research
to
use
ether,
and
so
basically
we
take
an
ether
cluster
that
just
like
the
one
that
I've
already
shown
you
and
we
we
instrument
the
switches
which
are
now
to
their
tofina
switches,
programmable
with
with
in-band
network
Telemetry.
B
So
we
can
get
fine-grained
Telemetry
out
of
them
and
we
feed
that
back
into
a
verifier
that,
because
we
program
the
the
data
plane,
we
can
verify
that
in
fact,
what
we're
observing
about
the
behavior,
the
data
plane,
matches
what
we
told
it
to
do
and
if
there
are
any
anomalies
because
we're
under
various
kinds
of
attacks
and
malware,
then
you
can.
You
can
close
group
control
it
and
perhaps
turn
off
a
UE
turn
off
the
base
station
block
this
or
that
or
whatever
whatever
the
right
remedy
happens
to
be.
B
We
have
connected
that
in
a
flight
Lab
at
Stanford,
with
with
drones
that
have
5G
dongles
on
them
and
so
we're
doing
drone
control,
and
so
this
particular
research
group
had
struggled
controlling
the
drones
with
Wi-Fi
needed
the
guarantees
of
5G,
and
so
that's
what
we've
done
and
so
there's
motion
capture
the
cameras
are
wired
in
this
particular
configuration,
the
Drone
control
is
Wireless,
and
then
we
subject
this
thing
to
all
kinds
of
attacks.
This
was
darpa's
sponsored,
and
then
we
show
you
know
you
attack
very
in
various
ways.
B
E
B
Research
papers
there's
some
project
I'll
leave
this
slide
up.
There's
some
project
websites
both
for
ether
as
a
whole
and
for
Pronto
that
used
ether
and
those
demo
videos
are
available.
There
are
some
background
reading.
We
basically
I
have
collaborated
with
domain
experts
to
to
write
modest
size.
We
call
the
micro
books
are
not
terribly
long,
but
they
focus
on,
for
example,
the
5G
case.
This
one
is
interesting
in
that
there's
an
appendix
that
will
give
you
a
directory
of
all
the
software
components.
B
B
The
runtime
yeah,
so
not
specifically,
the
AIML
is,
is,
is
absolutely
a
lot
of
interest
there.
So
the
Pronto
research
team
was
not
particularly
AI
motivated.
They
were
mostly
networking
people,
but
Intel
Labs
have
has
picked
up
ether
and
has
been
doing
various
ml
control,
demo,
demonstrations
and,
and
so
on.
B
I
can't
tell
you
the
specifics
of
them,
but
I
do
know
that
well,
they've
got
a
what's
it
called
openvino
I
believe
is
a
an
AI
or
an
ml
platform
that
they
have
so
the
other
sort
of
semi-related
is
I'm.
I
didn't
speak
a
lot
about
what
the
edge
apps
were
and
kind
of
leave
that
to
your
imagination.
But
what
what
tends
to
happen?
Is
you
don't
run
an
individual
app?
F
B
That's
that's
exactly
what's
happening
and
and
what's
the
way
I
characterize
it
is
this
used
to
be
the
domain
of
telcos
and
they're.
Very,
very
small,
set
of
incumbent,
vendors
and
I
have
no
idea
who's
on
the
phone
here,
but
you
can,
you
can
imagine
who
those
three
vendors
are
and-
and
it's
been
a
very
tight
relationship
and
it's
been
a
very
closed
system.
B
What
has
happened
as
a
consequence
of
disaggregation
opening
and
so
on
is
now
you've
got
Cloud
companies
interested
in
doing
exactly
that,
and
not
just
Cloud
companies
but
Walmarts
and
and
the
like,
don't
take
them
specifically
but
I'm.
You
know
companies
like
that
Target
and
so
on,
that
are
that
are
building
private
networks.
So
now
the
question
is
why
5G
versus
wi-fi,
because
you
can
build
exactly
that
with
Wi-Fi
and
that's
still
a
little
bit
of
an
open
question,
but
what
what
the
our
experience
seems
to
be
is
there?
B
Are
you
can
get
more
reliable,
Jitter
Behavior,
basically,
that
you
don't
get
the
variation
in
in
in
latencies
that
you
do
with
Wi-Fi
and
if
you're
trying
to
fly
a
drone
that
that
Jitter
can
cause
it
to
crash
now?
Is
it
you
know,
retail?
Is
it
quite
the
same
thing?
It's
it's
hard
to
say
it's
the
outdoor
nature.
Also,
it
just
tends
to
do
better.
I
know.
There's
agriculture,
Pilots,
smart,
AG
yeah,
so
it
all
boils
down
to.
B
Is
our
situation
so
step
one
make
the
mobile
network
look
more
like
a
cloud.
That's
happened
and
is
happening
step
two.
Can
you
find
use
cases
that
really
take
advantage
of
5G
versus
wi-fi,
and
there
are
a
few
of
those
that
I
think
are
out
there
there?
It's
it's
not
about
your
cell
phone,
it's
entirely
about
smart
retail,
smart,
Ag
and
so
on.
Right.
F
Right
yeah
I
mean
I,
see
so
the
a
factory
having
a
private
Cloud,
especially
with
the
ability
to
run
controllers.
A
lot
of
the
the
things
I've
done
with
Cloud
native
at
the
edge
has
to
do
with
controlling
the
tiny
Edge,
where
I
can't
put
kubernetes
but
I
want
to
control
some
robotic
arm
or
something
like
that,
and
so
so
I
I
maybe
run
workloads
in
kubernetes,
but
they
need
to
be
wired
to
to
these
things.
So
it
is
definitely
an
interesting
use
case
or
or
that
and.
B
B
F
F
B
Yeah,
with
the
with
the
exception
of
the
first
workload
that
you
load
on,
it
is
the
5G
workload.
It's
otherwise
looks
just
like
any
other
kubernetes
cloud,
and
you
can
I
mean,
like
I
said
you.
Could
you
quite
often
are
ready
that
you
start
deploying
with
a
single
box,
but
this
will
scale
up
to
a
multi-rack
instantiation.
If
you
wanted
it
to
not.
A
So
to
follow
up
on
Mark's
questioning
it
just
hit
me
I'm
curious
as
to
whether
the
5G
part
of
this
is
sort
of
an
isolatable
plug-in
where
you
could
swap
in
a
different
radio
technology.
You
did
mention
Wi-Fi,
but
you
know
you
could
see
Laura
for
longer
range,
possibly
and
inside
of
a
factory.
Maybe
you
could
use
Bluetooth
low
energy
or
some
other
Wireless
and
as
well
as
whether
this
thing
would
support
something
like
a
service
mesh
to
integrate.
Wireless
Plus,
hardwired
into
one
centrally
controlled,
plane,
managed
connection
fabric.
B
Yeah,
so
hiding
the
specifics
of
the
wire.
The
the
transmission
technology
is
a
it's
a
good
goal.
It's
very
challenging
in
the
sense
of
how
overwhelmingly
prescriptive
the
mobile
networks.
Architecture
is
just
a
little
editorializing.
The
mobile
network,
4G
5G
3G,
it
wasn't
a
fixed.
There
was
too
tight
of
coupling
of
the
architecture
and
the
technology
every
time
the
technology
changed,
which,
of
course,
it
was
going
to
the
architecture
had
to
change.
B
B
There
is
another
control
plane.
You
probably
probably
have
heard
of
called
magma
that
abstracted
the
wireless
control
plane
and
then
controlled
both
Wi-Fi
and
it
was
4G,
didn't
get
to
5G.
That's
a
that's!
Definitely
where
we'll
eventually
end
up,
but
we're
always
from
from
there
right
now
that
being
very
easy,
I
mean,
in
my
view,
the
very
fact
that
the
that
the
5G
control
plane,
core
I,
often
just
say,
control
plane,
because
the
user
plane
comes
out
of
it
really
easy.
A
D
B
So
the
six
G,
the
60,
is
really
interesting
I.
My
view
of
it
is
that
we
won't
care
about
sex
Gene,
because
it'll
just
be
now
that
it's
it's
Cloud
native
it'll,
just
be
a
natural
iteration
of
that,
and
there
was
there
was
well
I,
mean
first
of
all,
5G
60
or
marketing
terms,
a
there's,
a
sequence
of
standards
that
progress
over
time,
and
there
are
boundary
conditions
that
you
need
to
have
the
standards
at.
You
would
like
any
radio
to
talk
to
any
core,
but
internal
to
all
of
that.
B
Those
interfaces
were
over
prescribed.
So
what's
6G
I
don't
know
it's,
it
can
now
be
because
we're
now
on
cloud
feature
velocity
time
scales
instead
of
Telco
vendor
time
scales.
These
things
could
morph
very
fast
and
then
who's
to
say
where
the
line's
drawn.
The
radio
technology
will
probably
change
and
that'll
be
a
perfectly
good
reason
to
call
it
6G,
but
the
rest
of
it
I,
don't
think
it's
going
to
be
as
relevant.
F
I
think
there's
there's
another
challenge
to
this,
which
is
cost
and,
and
that,
like
the
cost
of
pushing
data
through
a
cell
network,
is
exponentially
more
expensive
than
using
the
wireless
or
Wi-Fi.
You
know
the
wired
networks
that
we
have
today,
even
though
they're
ultimately
going
through
the
same
system
eventually
I
feel
like
it
it
just
it's.
F
B
I
think
I
think
that's
the
exactly
point
so
you're
you're,
probably
right
about
the
macro
cell,
that
a
T
and
T
and
Verizon
run
I.
Don't
know,
I
wouldn't
know
how
to
predict
the
future
there,
but
I
think
this.
It's
the
private
version
of
it.
That
is
the
economics
gonna
change,
a
you.
Don't
have
to
have
a
T
and
T
or
whoever
manage
anymore.
A
You
know
in
terms
of
putting
on
the
equivalent
of
your
own
private
cell
network
or
things
and
looking
at
your
demo,
involving
drones,
I
assume
that
drones
maybe
were
endpoints
connecting
to
this,
but
how
much
of
the
radio
control
is
baked
into
the
idea
that
these
cell
towers,
if
you
will
are
fixed
locations
like
they,
would
be
in
the
Legacy
World
versus
an
idea
of,
if
I'm
doing
a
private
Network.
What,
if
I,
deemed
that
trucks
in
my
fleet
are
actually
acting
as
mobile
moving
cell
towers?
A
B
Network
and
count
enough
to
be
dispersed,
yeah
well,
so
yes
and
no
I
I
love
that
idea,
the
challenge
is
going
to
be.
We
still
have
the
problem
of
the
spectrum
and
the
radio
spectrum,
and
so
in
every
this
is
now
a
national
issue,
so
in
the
US
we
have
cbrs
and
it
is
not
as
unlicensed
as
Wi-Fi
is
right
now
and
I,
don't
know
if
you
know
the
history
of
of
this
of
the
spectrum.
It's
actually
quote
unquote
borrowed
from
the
Navy
and
the
Navy
could
call
it
back
at
any
moment.
B
So
one
of
the
challenges
right
now
is
when
you
bring
up
a
private
5G
small
cell,
which
you
could
put
in
a
truck
with
this
with
this
with
the
processor
that
runs
the
software
and
you
can
drive
it
around,
at
least
on
your.
You
could
certainly
drive
it
around
on
your
premises
if
you,
if
you
had
the
rights
to
that
the
Spectrum
there
that
that
that's
it
so
this
is.
This
is
an
important
part
of
this
whole.
How
does
this
play
out?
Because
this
spectrum
is
subdivided
and
I
can't?
B
B
And
so
now,
from
from
a
university
setting,
as
an
example,
did
my
University
go
get
the
rights
to
the
to
that
band
on
their
campus?
Yes
or
no
they're,
not
very
many.
That
did
some
did,
anticipating
being
able
to
use
it,
and
one
way
you
can
use
it
by
the
way,
is
to
put
up
cell
service
and
then
sell
it
back
to
the
the
carriers,
but
that's
a
whole
other
bowl
game.
E
Yeah,
okay,
there
yeah
so
I
know
like
just
I
guess
it's
probably
really
agreed
that
there's
several
areas,
that's
very
exciting.
You
know
cyber
security
at
Computing
and
AI,
so
I,
just
wonder
in
like
going
forward
what
can
be
done
together
between
the
onf
community
and
the
cncf
kubernetes
community.
What
kind
of
topics
that
can
be
looked
at
so
going
forward
now,
just
for
any
new
business
and
the
technology
and
opportunities.
B
B
You
know
you
you,
everyone
understands
what
it
means
to
offer
pick
your
favorite
storage
as
a
service.
You
know
how
to
go
to
Amazon
and
acquire
what
is
it
Dynamo
or
whatever,
and
you
you
know
what
you
get
an
API
and
you
now
can
control
storage
and
you
can
build
apps
on
that.
The
counterpart
for
the
community
is
is
basically
connectivity
as
a
service.
We
don't
have
connectivity
as
a
service.
We
we
go
and
acquire
connectivity
from
a
carrier
historically,
but
we
get
EX.
We
get
virtually
no
control
over
it
now
we're
giving.
B
B
So
I
mean
beyond
that.
There
are
a
few
specific
things
about
managing
small
cells
that
are
a
unique
kind
of
device
a
little
bit.
How
do
you
I
mean?
Let
me
see
if
I
can
take
an
example.
I
know
how
to
deploy
kubernetes
on
I
know
how
to
I
know
how
to
set
up
the
infrastructure
for
kubernetes
on
commodity
servers.
I
know
how
to
deploy
workloads
on
the
kubernetes
I've
I've
set
up
as
infrastructure.
B
A
B
B
G
B
B
G
B
Well,
so
the
devices
that
connect
to
this
so
the
small
cell
is
the
rating
is
the
radio
with
just
a
little
bit
of
software
on
it.
That
is
just
bundled
with
the
radio
okay,
so
the
devices
are.
Certainly,
you
know,
that's
an
entire
range.
You
you
know
you
break
into
my
Android.
You
can
do
whatever
you
want,
it
could
be,
it
could
be
a
Raspberry
Pi
running
Linux
there
you
can
do
whatever
you
want.
B
And
you
know
once
that
thing
blows
up
into
a
macro
cell,
that
you
know
you
find
out
in
your
along
the
highway,
the
amplifiers
in
the
and
there
are
highly
specialized
and
that's
hard
to
break
that
vendor
lock
in,
but
on
the
on
the
on
the
femtose
femtocells.
Sometimes
they're
called
the
you
know:
they're
they're,
basically
like
Wi-Fi
access
points.
B
Those
can
be
built
out
of
commodity
hardware
and
and
and
be
programmable
there
there's
a
I,
don't
know
the
specifics
of
it,
but
there's
a
project
at
Intel
called
flexrand
and
it's
basically
open
source
with
the
stipulation
that
the
code
has
to
run
on
an
Intel
processor,
but
it
it's
basically
that
bottom
layer
of
of
the
radio
this
what
could
be
a
small
cell,
but
when
I
use
the
word
small
cell,
essentially
it
what
I?
What
I'm
saying
is
it's
it's.
B
No,
no,
no,
those
it
could.
If,
if
that
were
in
fact
programmable,
then
that
would
be
an
aspect
of
it,
but
in
most
of
what
I
was
talking
about,
the
control
of
all
of
that
sits
back
on
a
commodity
server,
running
Linux
as
a
canoe
as
a
kubernetes
worker.
So
we
run
the
control
of
it.
You
connect
the
radio
to
the
control.
We
do
still
have
to
provide
a
data
path
to
get
to
it,
which
is
like
the
functionality
of
our
IP,
router
plus,
plus,
because
it
has
to
track
mobile.
C
And
kind
of
going
off
of
that
so
you're
mentioning
how
these
small
cell
small
cells
have
a
shorter
range,
and
so
you
need
a
server.
That's
running
kubernetes!
Have
that
control
plane
near
it?
What
does
it
look
like
in
the
when
you're
setting
all
this
up
to
say?
Hey
this
node,
that's
near
this
small
cell!
This
is
your
small
cell.
What
does
it
look
like
to
connect
that
small
cell
to
the
management
node
near
it?
Oh.
B
So
the
management
node
in
the
small
seller
are
over
a
fixed.
It's
an
overlay
network
over
fixed
fixed
infrastructure,
so
you
might
just
literally
run
you
could
run
it
over
your
Wi-Fi
or
maybe
you
wanted
more
predictability,
so
you
run
it
over
something
else,
yeah
that
it's
it
it.
Basically,
the
the
small
cells
connect
to
each
other
in
effect
and
back
to
the
control,
centralized
control
point
over
as
an
overlay
over
some
wired
Network.
B
B
B
Well,
no,
so
so
the
femto
cells
AKA
small
cells,
are
while
they
can
be
software-defined
radio
and
there
are
open
versions
of
them.
They
are
currently
bundled
devices
that
have
their
own
internal
protocol.
Stack
that
stack
as
I'm
trying
to
explain
is,
is
shrinking
and
more
of
that
is
becoming
open
and,
and
actually
more
of
it
is
centralized.
So
what
actually
runs
in
each
individual
radio
is
is
less
and
less,
but
there
is
still
software
there,
okay
and
and
it
you
know
it
it
just
there's
the
lowest
level
of
control.
B
E
See
thanks,
Ellen
Singh
I've
been
learning
about
the
overall
landscape
for
the
cncf,
kubernetes
and
networking
one
thing
I
find
is:
whenever
we
talk
about
networking
in
the
in
the
cncf
kubernetes
world,
mostly
it
stopped
at
the
the
service
mesh
layer
and
basically
it
will.
The
underlay
network
is
not
discussed.
So
that's,
that's,
that's
reasonable
because
you
know
that's
the
way
it
is,
and
you
usually
don't
need
to
worry
about
that.
B
Yeah
I
think
that
I
mean
that's
really
putting
your
thumb
on
on
the
key
issue
here
and
that
if
you
remember
that
early
slide,
where
I
said
the
access
network
Edge
has
moved
that
is.
That
is
the
important
it
is
one
takeaway.
That's
that's
what
it
ought
to
be,
because
now
everything
to
the
right
of
that
looks
just
like
any
other
I'm.
B
It's
it's
microservice,
it's
open
source,
it's
Cloud,
native
and
and
so
on,
and
the
only
thing
that
is
you
know
now
outside
of
that
and
sort
of
beyond
our
reach
or
concern.
Is
that
literally
the
radio
piece,
the
radio
unit
and
as
soon
as
that
happens,
and
that
I
mean
that
we're
in
the
midst
of
that
happening,
everything
about
operational
practices,
everything
about
feature
velocity
comes
into
play
and
and
as
a
consequence,
because
there's
turmoil
there
there's
opportunity,
because
it's
not
the
old
ways
of
doing
it.
Just
are
not
going
to
apply
anymore.
B
E
B
D
B
To
that
URL,
you
will
find
a
description
of
5G
written
by
someone
who
intelli
started
to
try
to
understand
it
had
no
clue
it
was
just
a
pile
of
acronyms
and
some
obscure
information,
Theory
math,
but
basically
we've.
You
know
this
describes
from
my
perspective,
how
it's
being
transformed
into
something
looks
very,
very
familiar
and.
B
The
reason.
So
this
has
been
an
interesting
journey
in
the
sense
that
who
writes
network
software.
It
is
not
the
consumers
of
the
software,
it's
not
the
telcos.
It
has
historically
been
their
vendors
funded
by
the
telcos
onf
set
out
to
build
an
open
source,
software-defined
substrate
that
would
break
the
vendor
lock-in
on
behalf
of
the
telcos.
B
That,
in
a
slightly
Twisted
way,
is
exactly
what's
happened
over
the
last
10
years.
But
what
ended
up
happening
is
that
we
also
wrote
software
that
was
now
more
cloud-friendly
than
it
was
teleco
friendly
and
therefore
anyone
that
understands
the
cloud
can
now
do
that
so
who's.
Our
participants,
we've
been
in
the
business
of
disrupting
first
the
vendors,
so
no,
they
did
not
help
us.
Do
it
we're
now
in
the
business
of
disrupting
the
carrier.
So
no,
they
originally
funded
us,
but
they
really
didn't.
Have
the
engineering
staff
the
funding
was.
B
The
work
was
all
done
by
onf
Engineers
to
a
first
approximation.
There
were
that's
not
to
say
there
weren't
some
many
contributors,
but
it
it
wouldn't
have
happened
without
the
onf
engineers.
One
year
ago
today
those
onf
Engineers
got
acquired
by
Intel.
So
today,
most
of
the
software
coming
back
into
these
projects
is
coming
from
Intel.
A
You
know
that's
a
that.
That's
real
interesting
to
hear
of
that,
because
I've
read
some
fascinating
articles
on
this
too,
and
it
could
be
that
some
of
the
disruptions
that
people
here
posed
as
questions
like
taking
this
in
New
Directions
of
you
using
even
non-5g
radios,
are
really
interesting.
But
it
sounds
like
maybe
people
would
have
to
contribute
labor
to
to
to
use
these
abstractions
to
make
that
happen,
but
maybe
there's
a
way
to
put
together
an
organization
to
bring
together
the
people
interested
to
work
together
on
that
and
the
the
typical
open
source
fashion.
A
B
And,
and
over
the
past
year,
onf
has
transitioned
from
being
an
engineering
organization
to
being
a
community
organization.
Just
like
LF
right.
There
are
companies
that
are
taking
the
software
and
they're
building
product
out
of
it.
They
don't
generally
contribute
back
in
quite
the
same
way
as
you
would
like
Intel,
certainly
still
contributing
in
a
big
way,
we're
finding
a
lot
of
exploratory
organizations
that
really
aren't
in
the
space
using
it
to
come
up
to
speed
and
they're
contributing
some
back.
Some
of
those
are
universities
or
like-minded
affiliation.
B
B
A
F
G
B
F
B
E
B
A
Thank
you
well,
thank
you,
Larry.
That
was
a
fantastic
presentation.
We'd
invite
you
to
come
back
again
for
a
refresh
if
you're
interested
in
terms
of
logistics.
The
next
meeting
is
going
to
be
postponed
and
slip
by
a
week
because
it
conflicts
with
kubecon,
and
we
speculate
that
a
number
of
people
who
normally
come
to
these
are
going
to
be
physically
at
kubecon
in
a
different
time
zone
where
this
meeting
time
just
it
doesn't
work
out.
So
let
me
look
up
it's.
A
The
next
meeting
is
going
to
be
on
the
26th
of
April.
We
have
one
item
on
the
agenda
now,
which
is
the
edge
native,
a
white
paper,
supplemental
assets
with
five
names
listed
there,
and
that's
only
going
to
take
a
portion
of
the
meeting.
So
if
anybody
wants
to
add
something
else,
as
usual
practice
for
this
group,
members
are
welcome
to
add
their
own
agenda
items
just
list
it
with
an
estimated
time
and
go
for
it.