►
From YouTube: Kubernetes WG IoT Edge 20201104
Description
Kubernetes IoT Edge Working Group November 4 meeting on the North America time cycle. BoF discussion of Intel initiatives in the IoT Edge space
A
So
I
was
just
telling
dion
dion
is
a
co-chair
of
this
group
that
you
weren't
gonna
appear
today
and
introduce
yourself
to
be
honest,
I'm
not
sure
what
kind
of
turnout
we're
going
to
get,
because
you
know
that
here
in
the
u.s
it's
a
elect
day
after
election
day,
and
I
think
a
lot
of
people
are
zoned
out
and
out
of
work
and
we'll
see
what
happens.
But
I.
B
Hear
you
I,
I
want
to
get
engaged
and
get
started
if
any
at
least
right.
So.
A
A
The
right
way
to
do
this,
in
my
opinion,
is
to
go
to
the
kubernetes
community
calendar
and
cut
the
invitation
out
of
there,
because
then
it
auto
adjusts.
But
I
think
a
lot
of
people
just
make
a
personal
entry
and
if
it
doesn't
sync
to
the
time
zones
correctly,
they
get
off.
And
we
have
this
thing
going
on
in
this
group,
where
we
have
alternating
meeting
times
so
that
we
can
accommodate
asia
and
eastern
europe.
So
our
alternating
meetings
are
actually
at
a
different
time.
A
Yeah,
I
don't
know
if
we
had
it
to
do
over
that,
we
would
do
it
that
way.
It's
we
started
out
with
one.
Then
we
announced
that
and
unfortunately,
you
get
these
meeting
times
out
there,
even
in
youtube
videos
that
get
recorded
from
years
ago,
and
somebody
starts
watching
them
and
sees
in
the
video
that
this
is
the
time
and
it
has
become
wrong,
but
there's
no
way
to
update
it
and
I've
learned
that
the
hard
way
that
it's
treacherous
to
shift
any
of
these
things
once
you
get
started.
A
It's
two
minutes
after
we'll
give
it
another
two
minutes
I
see.
Moritz
has
joined
he's.
A
A
Okay,
I
I
pasted
a
link
to
the
agenda
notes
document
in
the
chat.
If
people
would
like
you
can
drop
your
name
in
the
attendees
list,
we'll
get
started
in
one
more.
A
A
Chris,
I
see
you
joined
just
as
I
was
sending
something
out
on
the
chat.
So
let
me
know
if
you
don't
see
it,
but
I
sent
out
a
link
to
the
agenda.
Note
stock,
we're
gonna,
get
started
well,
we'll
get
started
right
now,
but
if
you
would
like,
you
can
add
your
name
there
so
to
get
this
meeting
started.
A
A
So
I'm
about
to
start
recording
and
the
recording
will
be
uploaded
to
youtube.
If
you
don't
like
that,
your
options
are
to
drop
or
to
leave
your
video
off
and
mute
yourself.
I
guess
and
maybe
change
your
name
to
an
alias.
If
even
that,
offends
you
so
with
that
said
I'll
push
record
and
we'll
get
started.
Oh,
I
guess
dion.
You
already
started.
Recording
okay,.
A
Okay,
maybe
it's
set
to
auto
record.
I
didn't
know
that.
Okay,
welcome
to
the
november
4th
meeting
on
the
north
america
time
cycle
for
the
kubernetes
iot
edge
user
group.
A
Today,
we've
got
one
item
on
the
agenda.
Officially,
I
think
we've
got
just
a
reminder
that
diana
will
give
also,
but,
as
always,
members
are
welcome
to
add
agenda
items.
So
if
anybody
has
any
late
ads,
I
think
we
might
have
room
for
it
so
go
ahead
and
just
patch
that
in
there
and
we'll
get
some
time
to
talk
about
whatever
you
want.
I
think
this
is
likely
to
be
one
of
these
relatively
unstructured
things
that
goes
in
a
birds
of
a
feather
discussion
open
for
anything.
A
We
do
have
a
first-time
visitor
here
today,
brian
rodriguez
of
intel-
and
maybe
the
best
thing
to
do
since
we're
small-
is
to
go
around
doing
introductions
just
to
kick
this
off
and
then,
after
that
brian
wanted
to
have
a
discussion
about
what's
going
on
with
kubernetes,
iot
and
edge
at
intel
and
how
those
activities
might
interact
with
this
group
so
for
introductions
dion,
do
you
want
to
kick
it
off.
D
Sure
so
my
name
is
dan.
I
I
work
for
redkit
for
a
long
time
been
doing
a
lot
of
messaging
in
the
past
and
then
for
the
last
couple
of
years,
been
more
involved
into
the
iot
stuff
and
and
particularly
more
on
the
iot
cloud
stuff
and
then
at
some
point
we
we,
we
started
this
iot
edge
working
group
here
to
to
further
discuss,
discuss
how
kubernetes
can
be
used
for
for
iot
and
edge
solutions.
So
that's.
A
That's:
okay,
thanks
all
right
I'll
go
next
I'll!
Just
do
it
in
the
order.
I
see
it
on
the
screen.
I'm
steve
wong,
I'm
a
software
engineer
with
vmware.
I've
been
working
on
kubernetes
since
about
2016,
but
most
of
that
has
not
been
related
to
iot
edge.
I
started
out
working
on
aspects
related
to
storage.
A
My
interest,
though,
in
iot
edge,
is
that
kind
of
a
long
time
ago
in
my
career,
I
was
affiliated
with
a
startup
called
wonderware
that
did
iot
before
it
was
even
called
iot
and
I've
held
on
to
an
interest
in
that
so
started
going
to
the
meetings
of
this
group
and
they
invited
me
to
become
one
of
the
co-tech
leads
of
the
group.
So
that's
how
I
got
here.
Brian.
Do
you
want
to
go
next.
B
Sure
brian
rodriguez,
with
intel,
I'm
in
the
iotg
group,
a
platform
architect-
and
my
background
has
been
you
know:
building
internet
service
providers,
then
cloud
service
providers
to
managing
lot
of
remote
devices
before
docker
containers
was
a
thing
to
where
we
are
today.
Trying
to
you
know,
manage
the
edge
and
orchestrate
the
edge
from
industrial
to
retail,
and
if
you
could
have
multiple
markets,
if
you
had
those
at
two
extremes
to
me,
those
are
two
extremes
but
yeah.
I
include
vehicles
robotics
and
along
that.
B
But
our
objective
is
here:
well,
well
we'll
discuss
that
so
glad
to
meet
you
all.
A
E
Fantastic
yeah,
so
I'm
from
canonical
my
my
background
is
actually
from
electrical
engineering
similar
to
what
you
were
saying.
I
worked
at
an
iot
startup
a
long
time
ago,
called
synapse
wireless
that
did
wireless
and
mesh,
and
so
I've
been
interested
in
iot
and
kind
of
edge
hardware
for
a
while
I'm
at
canonical.
Now
I've
spent
the
last
four
or
five
years
at
canonical
running
kubernetes
and
openstack.
E
So
I've
been
from
kind
of
an
operations
perspective,
but
I've
recently
joined
the
kubernetes
team,
which
is
responsible
for
our
charm,
kubernetes
and
microkites,
and
I'm
really
interested
in
how
those
kind
of
mesh
with
edge
and
iot.
So
I
thought
I'd
start
joining
here
and
see
where
the
discussion's
going
and
where
kubernetes
and
edge
look
like
they're
they're
coming
together,
and
you
know
how
micro
cates
and
all
might
play
into
that
story
as
well.
A
A
So,
if
you'd
ever
like
to
do
that,
we're
up
for
it,
and
I
think
about
two
meetings
ago,
we
had
kind
of
a
nice
birds
of
a
feather
discussion
on
a
lot
of
people
going
to
intro
to
various
radio
protocols
for
a
mesh,
not
in
necessarily
the
istio
context,
but
kind
of
in
the
what
you
might
see
out
at
iot
when
you
didn't
even
have
a
wired
scenario.
Yet
we're
trying
to
run
data
communication
meshes.
So
these
meeting
recordings
are
up
on
youtube
and
you
can
probably
find
that
and
if
you.
A
A
That
one
moritz,
how
would
you
how
about
you
introduce
yourself
next.
F
Yeah
sure
yeah,
my
name
is
moritz,
I'm
currently
working
as
an
phd
student
at
aachen
university
in
germany
for
the
layup
chair
for
laser
technology,
and
I'm
currently,
basically
writing
my
thesis
on
a
on
a
platform
or
on
a
on
a
let's
say.
F
A
platform
is
pretty
good
for
that
on
how
to
control
laser
production
machines
with
kubernetes
so
on
how
to
to
to
basically
build
a
manufacturing
system
that
completely
runs
kubernetes
top
to
bottom
into
the
edge
so
that
you
can
deploy
not
only
like
the
the
software
that
is
running
on
the
machine
itself,
but
you
can
control
it
with,
for
example,
operators
or
stuff
like
that.
A
And
I
think
you
told
us
in
the
past
you're
trying
to
use
image
capture
and
things
for
that
as
a
control
loop
for.
F
Yeah,
so
so
one
of
the
use
cases
we
have
is
that
we
have,
we
have
insane
speeds,
so
our
process
runs
at
like
three
to
four
meters
per
second.
F
So
we
need
image
data
that
is,
like,
I
don't
know,
thousand
to
fifteen
thousand
pictures
per
second
and
we
want
to
basically
run
algorithms
on
that
on
the
fly
to
control
our
process,
and
the
idea
was
to
use
kubernetes
to
to
like
change
the
the
algorithm
that
kind
of
kind
of
yeah
analyzes
these
picture
streams
and
that's
what
I've
been
working
for
or
on
for
basically
one
and
a
half
years
now,
and
it's
still
like
going
hard.
A
Okay,
thanks
yen,
why
don't
you
introduce
yourself.
G
A
B
B
So
one
of
the
things
that
you
know
and
obviously
intel's
a
hardware
company,
but
we
we
want
to
engage,
and
we
understand
that
orchestration
and
manageability.
B
So
what
we've
been
trying
to
do
is
our
iotg
is
coming
together,
trying
to
create
a
more
of
a
one
voice
story
between
all
of
our
verticals
and
try
to
create
a
common
base
of
sort
of
reference
etc,
and
one
of
the
things
we're
noticing
is
gaps
in
just
scalability
and
deployment
and
the
cost
of
just
to
run
kubernetes
at
the
edge
and
just
assembling.
B
And
so
we've
been
spending
a
lot
of
time
in
autonomy
and
just
even
zero
touch.
You
know
secure
device
onboarding,
you
guys
have
seen
if
you
have
not
that's
on
lf
edge
and
and
beyond
that
and
automating
the
and
self-healing
of
kubernetes
and
cluster
orchestrators
we've
had
a
number
of
customers
come
in
using
nomad
mesos
docker
swarm
still
and
we're.
We've
we've
had
a
lot
of
pushback
for
different
areas
in
the
smaller
iot,
from
a
kubernetes
point
of
just
because
the
sheer
size
of
what
it
takes
to
run.
B
It's
just
not
cost
effective
for
those
retail
folks.
But
when
you
come
from
an
industrial
side,
heck
yeah
they're,
all
so
we're
trying
to
create
this
one
story
and
try
to
go
to
our
customers.
No!
No!
You
should
really
go
to
kubernetes
because
there's
a
lot
out
there
in
the
ecosystem,
but
we
want
to
also
evolve
and
help
give
feedback
to
the
kubernetes
and
working
group
to
say
here's
what
we're
seeing.
How
can
we
make
more
efficient
hybrid
models
as
you
approach
the
far
edge
and
then
another
new
thing?
B
B
How
can
I
have
a
group
of
three
devices
run
on
their
own
with
autonomy
and
there's
a
connection
between
them
that
goes
up
and
down
right,
not
a
single
device
and
let
them
still
orchestrate
orchestrate
of
beyond
amongst
themselves,
and,
let's
say
in
a
robot,
you
know
something
that's
delivering
something
that
needs
to
still
function
while
disconnected
from
the
cloud.
However,
you'd
like
it
to
check
in
because
the
cost
just
to
go
and
check
in
is
there's
an
expense
and
if
you
know,
there's
10
000
of
these
units,
we
we
we're.
B
These
are
the
type
of
use
cases
we're
running
into,
but
people
want
kubernetes,
and
so
what
we
would
like
to
do
is
see
if
this
is
the
right
place
to
engage
with
you
all
start
working
together
start
exposing
some
of
our
technologies
and
our
capabilities
from
a
hardware
aspect.
We
have
a
lot
of
push
to
try
to
increase
our
telemetry
in
a
more
common
bound
way
up
to
kubernetes,
to
make
better
decisions,
help
with
trending
analysis,
there's
a
lot
of
software.
B
That
does
it,
but
you
know
I
would
say,
from
an
intel's
point
of
view,
there's
not
an
organized
way
for
us
to
expose
that
to
kubernetes
in
an
efficient
way,
and
we
want
to
be
engaged
and
hear
from
you
all.
How
can
we
help
and
also
what
our
engineers
can
do
to
help
enable
you
and
you
enable
us
so
it's
you
know
we'd
like
to
see
a
collaborative
approach,
and
you
know,
and
as
an
intel
as
a
hardcore
company,
we'd
like
to
be
more
involved
in
understanding
what
it
means
from
a
community
and
working
group.
B
A
Yeah,
it
does,
is
it
resonating,
it
is,
and
I
think
that
we've
had
historically
a
big
part
of
this
community
showing
up
to
meetings,
maybe
more
so
at
the
at
the
big
tent
things
we've
had
at
physical
conferences
that
recently
have
been
kind
of
moved
to
zooms,
so
they're
they
can't
be
as
interactive,
but
we've
had
events
where
we've
had
in
excess
of
200
people
at
presentations
at
kubecons
and
then
had
hallway
track
things
going
on
afterward
with
major
groups
of
different
use.
A
Cases
for
kubernetes
edge
iot,
they
fell
into
camps
of
telcos
were
a
major
one.
Industrial
iot
is
another
data
capture
and
I'm
throwing
this
in.
But
it's
moritz's
category
of
it's
pretty
broad.
What
that
means
anything
from
security
cameras
to
merits,
trying
to
do
a
control
loop
based
on
video
so
and
I've
even
heard
of
things
like
fast
food
restaurants,
using
cameras
to
monitor
things
like
arrivals
of
pickup
orders
and
there's.
C
A
So
yeah
there's
a
lot
of
that
that
falls
under
the
video
category
and
I'm
not
sure
how
uniform
they
are,
and
then
smart
cities,
although
I
don't
think
we've
had
large
numbers
of
those
people,
show
up
to
meetings,
but
there
were
small
numbers
of
them
and
then,
of
course,
vendors
in
the
space,
but
a
lot
of
times
vendors
can
stand
in
as
a
proxy
for
their
customers.
A
Just
because
you
know
in
many
cases
people
will
use
a
vendor
just
because
they
don't
have
the
time
to
go,
send
their
staff
out
to
open
sports
communities
to
be
active
and
kind
of,
in
effect,
delegate
that
by
going
with
a
commercial
distribution
summarizing
it
all.
I
think,
you're
on
point
with
the
problems
from
big
data
center
versus
out
at
edge.
The
one
that
comes
up
over
and
over
is
lack
of
physical
security,
creating
some.
C
B
That's
actually
our
group.
What
we
spent
we've
been
spending
last
three
to
four
years,
doing
a
lot
of
work,
trying
to
simplify
that
zero
touch,
provisioning
with
your
low-cost
technician,
doing
the
actual
onboarding
and
again
there's
a
big
cost
at
scale
there.
So.
A
A
Other
recurring
thing
with
kubernetes
is
that
the
fact
is
that
there's
tons
of
devices
out
there,
the
probably
the
numbers
dwarf,
you
know,
there's
tons
of
devices
that
don't
have
the
compute
resource
to
run
kubernetes
in
the
sense
of
being
a
kubernetes
work
or
not,
and
in
fact
I'd
say.
The
numbers
dwarf
the
capable
devices
by
maybe
three
orders
of
magnitude,
maybe
even
four
yeah.
A
Yet
kubernetes
is
a.
It
was
designed
from
scratch
to
be
an
extendable
control
plane.
So
there's
a
lot
of
people
who
are
doing
investments
in
writing.
Crds
custom
extensions
to
the
kubernetes
control
plane
so
that,
even
if
it
isn't
a
containerized
workload
or
even
containerize
a
ball
there's
a
chance
that
you
can
use
the
kubernetes
control
plane
to
do
configuration
and
maybe
even
things
like
firmware
updates
of
devices
that
have
don't
even
run
linux.
You
know,
there's
no
reason.
C
A
Edge
you're
talking
decades
pretty
easily
because
of
the
logistical
cost
of
replacing
that
stuff
and
budgets
where
hey.
If
it
works,
if
it
ain't
broke,
don't
fix
it
exactly
that
isn't
going
away
yet
the
pain
points
are
there,
so
that
if
you
could
come
up
with
tech
that
would
allow
kubernetes
to
fit
into
that
story,
and
I
think
managing
configuration
and
updates
is
big.
A
A
B
A
A
Is
the
actual
user
here
and
then
tina
joined
late
just
as
an
intro
tina,
we're
doing
kind
of
a
birds
of
a
feather
and
brian
is
here
from
intel
just
trying
to
learn
the
lay
of
the
land
with
kubernetes
and
iot,
but
tina,
I
think,
has
a
broad
exposure
because
she's
involved
with
the
greno
project,
so
marin
centina.
Maybe
you
want
to
jump
in
and
talk
to
I'll
beat
my
mic
and
shut
up.
F
Yeah
so
yeah.
The
problem
we
run
into
a
lot
is
just
like:
basically
the
diversity
on
the
edge
that,
because
you
have
so
many
different
protocols,
so
many
different
manufacturers,
so
many
different
yeah,
so
many
different
people
doing
stuff
non-standardized
ways
that
it
gets
really
really
complicated,
really
really
fast.
The
funny
thing
about
kubernetes
is
that
you
can
basically
pull
everything
together.
F
It
might
take
a
lot
of
time
to
to
like
implement
something
like
so
we
have
done
modbus,
for
example,
with
kubernetes,
so
you
deploy
basically
in
a
kubernetes
deployment
or
a
daemon
set,
or
something
like
that,
and
this
one
can
talk
modbus
and
then
you
can
control,
for
example,
some
kind
of
robotic
system
or
a
movement
system,
or
something
like
that
and
that
works
wonderful.
F
F
I
would
say
that
there
is
like
not
a
standardized
idea
on
actually
how
to
to
to
to
use
different
protocols,
so
that
might
be
interesting
in
the
future
to
just
think
about
that,
like
how
can
we
leverage,
maybe
cuba
needed
to
jump
from
one?
Let's
say:
communication
system
into
another
yeah,
my
two
cents
on
the
on
the
topic
from
a
user
perspective,
yeah.
B
F
Yeah,
exactly
that's
what
we
are
actually
doing
right
now,
but
the
problem
with
that
is
that
every
vendor
implements
his
own
version
of
that.
If
you
know,
if
you
know
what
I
mean.
F
Like
like
modbus,
for
example,
you
can
communicate
with
your
with
your
manufacturing
machine
with
that,
but
your
your,
the
the
the
position
of
bits
varies
from
manufacturer
to
manufacturer,
and
sometimes
we
have
like,
I
don't
know,
16
bytes.
F
Sometimes
you
have
32,
sometimes
like,
like
modbus,
for
example,
is
just
crazy
but,
and
the
funny
part
is
that
I
I
really
would
like
to
see
something
like
like
a
convergence
thing
like
going
from,
for
example,
rest
or
grpc
to
something
like
modbus
or
something
like
that,
but
I'm
I'm
not
quite
sure
how
to
actually
deal
with
that
or
if
it's
just
so,
the
kubernetes
we're
actually
just
using
right
now
to
basically
deploy
some
kind
of
gateway.
That
translates
translate.
F
And
of
course
you
can
do
that
for
every
manufacturing
system.
You
own,
that's
not
a
problem.
That's
just
like
you
just
need
to
code
a
lot.
Let's
say
it
like
that,
and
you
have
a
million
services
afterwards
and
but
it
works.
So
we
have
done
that,
for
I
don't
know
seven
to
eight
different
manufacturers
right
now
and
this
works
perfectly,
but
I
think
that
will
not
scale
up
very
good.
A
Look
I
got
one
of
these
running,
but
there's
security
issues
on
it
and
scale
issues
where
really,
when
you
think
about
this,
this,
these
iot
devices
have
simple
protocols
that
maybe
didn't
anticipate
security,
have
very
low
compute
capacity,
and
if
you
just
do
a
translator
that
maps
that
out
on
the
internet,
it's
you're,
exposing
it
to
these
denial
of
service
attacks,
we're
just
reading
asking
for
the
temperature
over
and
over.
You
would
think.
Well,
it's
read
only
what's
the
harm
for
in
exposing
it,
but
some
of
this
stuff
collapses.
A
If
you
ask
for
the
temperature
a
thousand
times
a
second
and
it
falls
apart
and
then
your
actual
use
for
it
goes
out
of
service
and
you
got
a
big
problem.
I
think
a
lot
of
these
seem
simple
until
you
try
to
scale
it
or
really
anal
analyze.
A
The
full
threat
map
of
what
could
happen
with
that
one-to-one
exposure,
so
it
ends
up
being
something
where
you
need
a
little
more
intelligence
and
I
call
it
governance
over
it
to
where,
in
general,
it's
almost
always
going
to
be
a
bad
day
to
get
that
stuff
exposed
up
to
the
higher
cloud
tiers
one
for
one
without
some
intermediary
business
logic.
That
really
is
the
only
one
allowed
to
talk
to
it.
Yeah
the
upgrade.
A
Yeah
and
maybe
even
authentication
of
who
you
are
so
fine,
cache
it
and
let
people
read
it
out
of
the
cache,
but
there
there
isn't
a
direct
pipeline
if
you
really
want
to
expose
that
to
broad
use.
The
other
thing
is
brian,
you
said:
use
kubernetes,
you
know
to
get
the
data
to
these,
but
really,
I
think,
even
in
a
data
center
kubernetes
really
is
just
control
plane
and
if
you
look
at
how
container
images
get
there,
it's
really
sidebar,
where
the
data.
C
A
Do
firmware
patches
into
the
many
computer
devices
inside
an
automobile,
but
there's
a
lot
there
that,
I
think
is
potentially
reusable.
I
think
the
telco
industry
also
has
things
you
know.
Their
own
projects
were
distributing
effectively
binaries
for
various
uses.
B
We
we've
engaged
in
a
lot
of
those
those
different
red.
Exactly
you
know:
telco
plus
vehicles
we've
been
working
very
much
with
even
using
containers
as
a
delivery
mechanism
to
help
do
that,
but
we
spend,
like
I
said
from
a
bottom-up
approach:
we've
been
spending
a
lot
of
time
around
security.
You
know
full
attestation
for
hardware
in
case
it's
lifted.
You
know
this
thing
will
self-destruct
if
it's
not
connected
to
the
network
for
a
matter
of
30
seconds
right,
a
lot
of
those
use
cases,
we've
been
spending
time.
B
I
I
I
think
it'd
be
nice
to
at
least
expose
and
share
with
you
all
and
see
if
we
can
and
what
about
time,
sensitive
networking
does
that
ever
come
up.
A
It
it
does,
you
know,
like
time,
series
data,
certainly
we
haven't
had
like
a
detailed
presentation
on
it,
but
the
stuff
is
clearly
out
there.
It's
actually
been
out
there,
pre-kubernetes
too,
when
you
get
down
to
it
and
in
a
way,
kubernetes
does
use
that
for
some
of
the
metrics
and
data
logging.
But
then
I
think
we
had
a
discussion
a
few
meetings
ago
about
the
really
hardcore
stuff
where
you
need
potentially
single
digit
microsecond
precisions
things
like
doing
root
cause
analysis
for
failures
of
a
power
grid
at
a
large
scale.
A
There
have
been
people
talking
about
the
pain
points,
I'm
not
sure.
We've
ever
had
a
presentation
on
how
you
go
about.
B
A
B
B
We
we've
been
mixing
and
video
inferencing
with
time
series.
You
know
time
since
in
networking.
So
you
know
with
timing
for
life-saving
events.
Right,
you
know,
people's
you
know,
hands
can
get
chopped
off
in
an
industrial
situation
using
video
to
kind
of
watch
these
things
and
but
yeah
okay.
So
it's
it's
there,
but
it'd
be
worth
going
into.
A
A
C
A
B
B
Sure
we
and-
and
you
know
it's
not
just
the
tpm-
it's
the
secure
boot
guard
all
the
way
up
from
this,
this
machine
testing
all
the
way
to
your
your
bootstrapping,
all
the
way
up
using
the
tpm
as
your
facilitator,
but
here's
another
thing.
The
biggest
challenge
has
always
been
getting
the
the
oems,
and
you
know
the
sis:
do
you
want
to
manage
your
keys?
B
H
F
B
Yeah,
it's
it's
a
little
of
both.
It
depends
on
the
situation.
You
know
when
I
say
this
is
it's
more
mainly.
It
starts
being
how
synchronized
the
number
of
nodes
we
have
together
with
the
video
coming
in.
If
we
have
multiple
inferences,
I'm
separated
by
multiple
nodes.
So
how
can
we
make
sure
there's
synchronization
between
them,
interacting
with
devices
on
the
floor?
If
you
will,
in
the
industrial
situation.
F
Because
I've
pulled
my
my
head
around
this
problem
like
how
can
I
introduce
something
like
an
emergency
button
into
a
system
like
that?
So,
for
example,
you
have
like
a
robot
running
around
and
then
the
robot
has
like
a
like
a
service
running,
so
you
can
control
the
robot.
F
When
I
introduce
a
new
button
into
the
system,
I
will
need
to
rewire
everything
and
hook
the
whole
thing
into
the
safety
wiring
again
and
I'm
not
quite
sure
on
how
to
handle
that
elegantly
like
wiring,
something
like
a
security
lifeline
into
something
like
kubernetes
in
real
real
time
like
like,
like
I
don't
know
one
millisecond
or
something
like
that,
so
that
you
can
shut
everything
down
in
like
a
millisecond
but
ethernet
tcp
ip.
You
know
it's
a
good
deal.
A
Yeah,
that
is
a
challenge,
especially
the
way,
if
you
don't
have
guaranteed
reliable
networking
healing
layers
inside
tcp
could
really
burn
up
that
millisecond
on
you.
Historically,
I
know
in
industrial
control,
which
I
used
to
do
a
lot
of
people,
look
at
digitizing
that
and
then
end
up
going
with
even
analog
solutions
or
really
local
loops.
A
You
know
things
like
safety,
for
power
saws
often
are
done
now
with
just
a
circuit
that
detects
a
capacitance
change
if
some
a
finger
touches
the
blade
and
then
detonate
an
explosive
charge
to
do
a
mechanical
break,
and
it's
pretty
tough
with
I
don't
know,
I
wouldn't
want
to
trust
my
life
to
kubernetes
today,
even
though
I'm
pretty
comfortable
with
kubernetes,
I'm
not
sure
that
you
know
I
was
about
to
cut
my
head
off,
though
I'm
that
confident
in
it
well.
B
That's
why
you
know
we
always
say,
like
you,
said,
sidebar
or
something
kubernetes.
Is
that
thing
to
facilitate
in
the
orchestration?
But
it's
not
the
real
time.
It's
not
the
engine,
that's
actually
interacting.
It
just
made
sure
that
thing
got
there
to
manage
that
real-time
sensitive
piece.
So
that's
how
we've
been
treating
it.
B
C
A
To
invite
comment
on
on
your
last
comment,
brian,
as
I
see
kilton
has
joined
us.
I
don't
know,
if
he's
in
a
position
where
he
can
talk,
but
if
he
is,
I
consider
him
sort
of
an
authority
on
the
subject
of
security.
He
contributed
a
lot
to
the
security
white
paper
and
kilton.
If
you
want
to
jump
in
there
and
if
you
can
talk
we'll
give
you
a
chance
on
mute.
B
Yeah
I
watched
some
previous
youtubes
where
he's
from.
B
C
C
There
we
go
sorry.
C
There
was
a
couple
different
mute
layers,
my
car
mute
layer
and
then
the
yeah
so
yeah.
I
was
about
to
to
say
something
before
but
but
steve
you
were
saying
so
many
great
things
I
didn't
want
to
didn't
want
to
interrupt
when
we,
when
you
were
talking
about
how
these
these
lower
layer
protocols
get
get
mapped
up
into.
You
know
the
the
upper
part
of
the
the
stack
you.
C
If
you
weren't
physically
located
right
next
to
the
machine,
then
there
was
no
way
to
communicate
with
it.
So
the
inherent
security
there
made
it
possible
to
just
stick
to
the
you
know
the
bike
transmissions
and
not
worry
about
any
authentication
or
anything.
And,
of
course,
when
you
bring
that
above
to
the
layers
above
you
you,
then
you
know
expose
that
that
implicit
security
is
out
the
window,
and
now
you
have
to
add
something
right.
A
C
Exactly
right,
and
so
there's
this
trust
layer
there
that
honestly,
like
I,
don't
know
that
we
would
have
made
it
if
we
had
to
have
everything
internetworks
all
back
in
the
day.
I
don't
know
that
we
would
have
been
able
to
make
performant
closed-loop
systems
right
because
of
the
overhead.
Now
I
think
we
can
handle
the
overhead.
We
just
don't
know
how
to
structure
it,
but
I
wouldn't
want
my
real-time
closed-loop
type
decisions
being
made
above
anyway.
I
would
just
want
to
be
able
to
change
the
directives
right.
C
Orchestration
system
and
then
let
them
get
carried
out
so
yeah,
but
it's
I
was
just
thinking
that
you
know
it
was
like
yeah,
it's
it's.
Actually,
it's
not
that
these
things
never
anticipated
interconnection.
It's
rather
that
they,
they
wouldn't
even
make
sense
right,
and
so
anyway,
I
thought
I'd
jump
jump
in
with.
C
C
And
and
the
earlier
part
which
I
I
came
on
at
about,
you
know
20
minutes
after
the
hour.
The
earlier
part
that
I
heard
was
also
about
just
the
the
nightmare
of
the
the
various
implementations
of
of
you
know.
I
guess
you
could
call
them
different
implementations
of
like
modbus,
etc.
C
It's
more
like
modbus
technique
works,
but
the
registers
change
right
and
the
number
of
bytes
coming
from
the
registers,
changes
and
all
this
nightmare,
and
it's
like
there
are
plenty
of
like
you
know:
modbus
can
bus
to
rest
systems
out
there
that
are
proprietary
and
you
pay
a
license
fee,
and
I
I
hear
that
they
work
really
nicely,
but
the
issue
also
becomes
like
what
do
you
need
where
and
how
do
you
make
sure
that
the
low
latency
real
time
traffic
stays
where
it's
supposed
to
stay?
C
And
then
you
might
need
some
insights
from
it.
And
how
do
you
bubble
that
up
and
you
know-
because
I
definitely
don't
need
my
modbus
data
flow-
to
go
directly
up
to
any
cloud
backend?
I
might
need
it
to
go.
One
hop
higher
on
on
the
edge
environment,
and
maybe
I
make
some
some
changes
in
decision
right
there
and
they're
on
a
very
low
latency,
but
not
what
I
would
call
kind
of
closed-loop.
Real-Time,
that's
possible.
C
C
That's
that's
what
I've
always
figured
is
going
to
emerge
is,
like
you
know,
here's
this
particular
flavor
of
of
modbus
for
this
scenario,
and
how
do
we
detect
that
we
need
it
and
automatically
orchestrate
it
down
for
these
five
edge
nodes?
It's
really
about
putting
the
the
application
workload
close
to
where
it
can
do
that
talking,
but
definitely
better.
If
you
can
have
something
like
I
don't
know,
maybe
a
hundred
or
so
you
know
commonly
used
adapters
from
low
level
protocols
upward
that
have
security
and
stuff
built
in
them.
C
You
know
we
could
probably
cover
a
lot
of
bases
with
about
100
or
so,
but
to
get
100
coverage
is
going
to
be.
You
know
1000
1500,
whatever
there
are,
you
know
number
of
protocol
in
the
in
the
in
the
ot
space
right
so
anyway.
Those
are
those
are
my
thoughts
because
I
didn't
interject
them
before
and
they're
very
similar.
To
what's
been
said,.
A
Yeah,
I
think
really.
A
This
will
be
a
segway
I'll.
Let
deon
talk
about
our
upcoming
talk
at
kubecon,
north
america,
but
one
of
the
ways
I
think
to
generify
this
legacy.
Data
is
to
turn
it
into
events
that
plug
into
can
be
plugged
into
the
existing
open
source
projects
that
deal
with
event-driven
architecture,
and
so
the
keys
there
are
not
one-to-one,
expose
everything
but
cherry-pick
what's
important
data,
wise,
publish
it
in
caches
so
that
it
can
be
used
at
upper
layers
that
maybe
aren't
so
latency
sensitive
and,
at
the
same
time,
potentially
mine.
F
Yeah
what
I
just
thought
about
when
you
meant
like,
like
monitoring
observability
of
system,
what
I
found
really
interesting
or
what
we
started
playing
around
around
with
in
the
last
few
months,
basically
was
combining
our
our
services
with
ontologies,
so
that,
for
example,
you
can
can
do
like
conversion
or
like
like
metric,
not
metrics
like
like.
F
Let's
say
you
have
some
kind
of
of
robotic
system
and
you
send
it
something
like
move
left
two
meters
or
something
like
that.
So
you
send
it
to
two
and
of
course
you
need
to
make
sure
that
actually
what
you're
sending
your
system
is
a
two
and
that
the
client
that
is
actually
connecting
onto
that
iot
gateway
actually
know
that
it
is
working
with
a
with
a
system
that
expects
a
meters
and
not
a
centimeter
value.
F
F
Like
I,
I
wrote
my
my
job
that
I
expect
movement
control
to
be
in
meters
and
not
in
centimeters,
so
you
define
that
beforehand
and
we
have
like
a
little
script
that
checks
the
checks,
if
every
call
would
break
anything
so
that
I
said
I
I
sent
like
a
2
000
instead
of
a
two
and
then
my
my
robot
was,
would
crash
somewhere
or
something
like
that.
That
is
really
interesting
to
run
on
kubernetes,
because
you
have
like
this,
this
definition
or
this.
You
can
basically
embed
that
in
your
service.
F
You
know
like
this
definition
on
what
is
actually
going
on
on
the
edge,
but
in
like
a
structured
way,
and
that
has
been
a
lot
of
fun
to
work
with,
because
it's
like
so
so
far,
it's
basically
open
territory.
You
have
stuff
like
open
api
or
grpc,
but
they
are
not
really
worried
about
this.
F
This
checking
on
on
stuff,
you
know
like
like
they
expect
you
that
you
know
what
you
are
sending,
but
they
don't
basically
check
if
the
value
is
actually
right,
that
you're
sending-
and
that
is,
I
think,
pretty
critical,
especially
when
you
go
into
the
edge
field.
F
C
C
All
the
software
was
done
in
in
the
same
stack,
usually
by
a
small
number
of
vendors
and
implemented
by
the
core
team.
That
probably
were
doing
all
the
checks
and
balances
and
test
runs
here.
You
might
throw
an
edge
application
into
the
mix
for
management
of
a
machine,
and
even
if
it's
made
by
the
very
vendor
that
made
the
machine
it
might
have
been
built
by
an
outsourced
team
that
they
hired
right.
That
doesn't
have
as
much
experience
or
it's
a
new
model.
C
F
G
F
D
A
Oh
one
more
thing
on
the
agenda:
dion
and
I
are
doing
a
event-driven
architecture-
presentation
for
kubecon
north
america
in
a
couple
of
weeks.
Do
you
want
to
tell
more
about
that.
D
Yeah
so
so,
basically,
we
are
covering
event
driven
architectures
for
for
iot
and
edge,
and
it's
been
something
we've
been
doing
for
a
long
time.
D
But
but
specifically,
we've
been
focusing
more
on
k
native
inventing
in
the
session
and
something
I've
been
playing
around
with
in
in
the
last
couple
of
weeks
or
months
or
so,
and
you
know
how
we
can
yeah
use
canadian
preventing
and
it
touches
on
on
top
of
this-
and
you
know
how
maybe
we
can
use
a
cloud
events
and
and
try
to
standardize
on
the
on
the
on
the
basically
payload
so
to
say
metadata
format
for
the
for
telemetry
up
coming
from
different
different
sources.
D
A
Part
of
that
is
we
actually
tried
to
do
a
live
demo,
I'm
not
sure
afterward.
You
know
we
have
to
record
these,
and
some
of
this
was
some
command
line
stuff
that
is
potentially
harder
to
see,
but
I
actually
put
on
the
air
some
temperature
sensors
in
my
house,
and
they
went
out
over
mqtt
to
a
mosquito
gateway
and
then
dion
went
and
rendered
them
graphically
feeding
them
into.
D
A
D
Going
to
yeah,
we
have
a
camel
k
source
that
that
was
reading
from
the
from
the
mqtt
external
entity
broker
getting
that
into
the
k
native
stream
and
then
pushing
things
to
influx
db
and
grafana.
Just
a
little
bit
demo
of
the
flow
of
the
things
that
could
go
but
yeah.
A
Demo
and
there's
a
few
other
things
I
I
don't
know
any
details
about
them,
but
I
went
over
the
kubecon
agenda
and
there's
some
interesting
edge
sessions
there
strange
enough,
there's
one.
This
is
the
story
of
my
life
name
collisions
with
the
name
steve
wong,
so
there
is
actually
a
presentation
by
stephen
wong
on
service
mesh
at
edge.
But
it
isn't
me
it's
somebody
else
who
shares
my
name,
but
I
was
going
to
catch
that
anyway.
A
Then
there's
a
there's
one
on
getting
your
application
bits
to
the
edge
using
the
harbor
image
registry,
and
I
think
there
were
a
couple
more
there's-
definitely
a
cube
edge
session
in
addition
to
ours.
So
I
would
recommend
that
I
don't
know
if
they're
still
out
there,
but
there
were
discount
codes
floating
around
for
kubecon
so
that
I
think
you
could
get
it.
A
A
If
you
can't
make
it,
I
think,
ultimately,
about
a
month
later,
they
end
up
uploading
them
to
youtube
anyway,
but
yeah.
I
I'd,
recommend
catching
that
conference
if
you're
at
all
interested.
The
other
thing
to
check
is,
I
believe
there
are
some
pre-events.
A
A
B
I
just
wanted
to
thank
everyone
and
it
looks
like
we'll
have
a
I'll
be
joining
more
future
and
I'll
add
some
topics
to
the
next
agenda
or,
as
things
go
on,
okay,.