►
From YouTube: Kubernetes WG IoT Edge 20221214
Description
December 14, 2022 meeting of the CNCF IoT Edge Working Group. General birds of a feather discussion of edge native applications topics.
A
Hi
welcome
to
the
December
14th
meeting
of
the
cncf
iot
edge
working
group
on
the
agenda.
Today,
we've
got
an
item
to
discuss
and
brainstorm
a
cfp
proposal
for
kubecon
Europe,
followed
by
just
general
discussion
and
I,
do
want
to
call
out
that
we
had
a
poll
that
we
started
at
the
last
meeting
just
to
see
what
people's
preferences
were
for
holding
meetings
during
the
holiday
periods
found
around
the
year
beginning
and
end
of
year.
A
B
B
B
So
the
coupon
EU
is
in
Amsterdam.
Let
me
see
all
right
here:
Amsterdam
in
April,
17th,
till
21st,
okay
and
first
thing.
I'm
not
sure,
is
that
I
know
that
the
process
was
changed
changed
the
last
time
right
so
that
we
need.
We
need
to
provide
a
cfp
under
the
the
tag
runtime
right,
so
we
we're
not
guaranteed
the.
A
Yeah
I
think
they've
changed.
It
used
to
be
pretty
much
that
sigs
and
working
groups
that
applied
were
guaranteed
a
slot,
but
now
it's
become
competitive.
So
perhaps,
as
a
program
committee,
reviewer
I
think
in
some
instances
in
the
old
days
people
sort
of
mailed
in
proposals,
but
now
I
think
that
you
perhaps
have
to
invest
a
little
more
in
outlaying
a
plan
and
content
and
kind
of
selling
the
reason
as
to
why
the
world
would
be
a
better
place
with
these
sessions.
So
you.
A
I
think
the
other
thing
that
this
is
just
my
personal
observation
on
these
meetings
be
a
success
in
terms
of
attendance
and
audience.
Participation
and
networking
would
be
that
if
we
can
perhaps
engage
people
from
the
region
that
are
actually
users
as
opposed
to
vendors,
that
there's
an
argument
to
be
made
that
that
would
actually
be
an
ad.
A
I
think
that,
based
on
my
own
experience-
and
you
know,
I've
I've
been
either
at
these
or
spoken
at
them
for
years.
Now
that
going
too
broad
like
if
you
just
announced
this
as
the
general
all-overall
topic
of
Edge
native,
that
it
it
just,
it
doesn't
ring
with
somebody
and
then
that
having
it
focus
on
some
niche
of
edge
native
might
help
build
a
more
compelling
presentation
and
maybe
even
draw
audience,
because
there
there
are
some
people
who
are
there
just
for
the
general
overall
thing.
A
But
I
think
that
at
this
point,
Edge
has
been
out
there
for
a
number
of
years
and
that
there
are
a
fair
number
of
people
looking
for
the
deeper
dive
and
as
long
as
you
call
out
what
your
focus
is
about.
Even
the
newbies
can
appreciate
the
value
of
kind
of
a
deeper
dive
into
one
niche
of
the
edge
native
topic.
As
long
as
you
call
out,
you
know
what
it
is
and
that
it
be
something
that
essentially,
everybody
trying
to
pursue
an
edge
native
use
case
would
be
interested
in.
B
So
first
I
I
completely
agree
about.
If
you
have
a
like
a
local
user
and
and
and
the
user
of
technology,
that
would
be
great.
So
if
anybody
knows
anyone
we
can
join.
That
would
be
awesome
and,
and
the
other
thing
so,
the
only
topic
that
I
have,
in
my
mind
that
we
can
cover
is
that
on
the
last
kipcon
we
had
as
native
principles.
B
A
Okay,
how
about
if
we,
if
we
I,
don't
even
know
the
exact
date
I
know
it's
in
January.
You
know
I
recollect,
maybe
it's
in
mid-January,
but
maybe
we
can
either
drop
this
right
in
our
agenda,
notes
document
or
put
a
link
to
some
secondary
one
where
we'll
go
through
a
phase
say
for
the
next
two
weeks,
where
anybody
who
has
an
idea
for
a
topic
just
drop
it
in
there
and
you
know
flesh
out
whatever
you
can
and
also
maybe
in
that
section
we
can
have
volunteers
who
think
they.
A
D
D
Sounds
good.
We
ate
the
dejan
shared
in
the
chat
here
that
the
cfps
due
January,
27th
and
I,
put
that
in
the
meeting
notes.
I,
think
another
I
think
adding
some
ideas
would
be
great
and
I.
Think.
Another
idea
would
just
be
to
do
a
panel
itself
and
that
could
be
also
led
by
our
educative
principals,
but
ask
some
experts
from
various
parts
of
edge
within
our
working
group
to
sit
on
a
panel
and
kind
of
just
go
through
and
ask
questions
about
how
they're
managing
and
their
pain
points.
A
Yeah,
the
other
thing
I'll
throw
out
there
is
that
they
they
will,
even
though
the
cfps
aren't
open
yet
I've
been
told
that
they're
planning
another
kubernetes
on
the
edge
Day
event
at
kubecon
Europe
as
well.
So
that
will
be
a
full
day
of
sessions
and
they're.
One
consideration
is
whatever
would
be
put
on
whether
it
would
be
better
for
the
day-long
event
or
for
the
general
thing
associated
with
our
working
group.
B
A
B
A
I'm
kind
of
curious
I
normally
have
found
good
sessions
at
the
Amazon
re
invent
event
around
this
time
of
year,
but
I
wasn't
able
to
go
physically
or
even
watch
any
of
them
live
real
time,
but
wondering
if
anybody
else
caught
anything
interesting
at
that
conference.
I
am
told
that
the
attendance
was
really
high,
that
if
it
was
a
sellout
in
terms
of
what
they
had
allocated
for
attendance
and
like
I,
say,
Amazon
has
been
reaching
out
into
the
edge
space.
So
they've
actually
historically
had
some
pretty
interesting
sessions.
A
Automating
operations
at
a
fish
farming
operation
in
Asia,
and
they
had
this
thing
set
up
intentionally
so
that
it
was
the
default
operating
condition
was
that
it
was
not
connected
that
this.
The
opportunity
here
was
that
they
had
a
fish
farm
out
in
a
rural
area
and
a
feeder
appliance
that
broadcast
fish
food
into
these
ponds.
Almost
like.
A
You
know,
a
sprinkler
system
that
you'd
find
in
agriculture
and
it
was
automated
and
it
didn't
have
internet
connectivity
so
that
the
way
that
they
drove
application
updates
collection
of
data
and
feedback
was
that
someone
would
go
out
there
into
this
rural
area
with
a
cell
phone
and
connect
to
the
equipment
from
cell
phone
to
the
equipment.
Perhaps
deliver
updates
collect
the
data
and
it
would
be
copied
into
the
cell
phone,
not
even
assuming
that
you
had
internet
connectivity
out
at
this
spot.
A
Even
if
you
brought
a
cell
phone
there
and
then
when
the
person
came
back
to
a
more
metro
area
with
connectivity,
it
would
connect
up
to
what's
called
a
cloud
hosted
management.
But
it
was
really
an
interesting
kind
of
extreme
use
case
of
intermittent
connectivity
and
I
think
that
it
provoked
thoughts.
A
Permanent
connectivity
has
great
difficulties
in
transitioning
to
kind
of
that.
More
challenging
Edge
use
case
anyway.
I'll
find
a
link
to
it
and
I'll
drop
it
in
the
chat
because
I
I
thought
it.
It's
a
YouTube
video
presentation
that
lasts
about
30
minutes
and
I've
I
found
that
it
was
both
interesting
and
then
caused
me
to
think
about.
You
know
other
aspects
of
designing
an
edge
native
system.
D
Kind
of
moving
into
a
different
discussion,
I'm
curious
for
folks.
Here
we
moved
from
being
a
kubernetes
working
group
to
being
a
CNC,
F1
and
I'm
curious
of
people
when
they
think
of
edge
in
the
cloud
if
they
think
kubernetes
or
how
much
people
are
starting
to
not
think
kubernetes
for
folks
here
and
how
tied
those
discussions
you
think
are.
B
So
my
recent
experience
is
both
so
still
containers
but,
for
example,
what
what
we
were
experiencing
is
that
we
were
mixing
Boardman
and
and
and
one
node
kubernetes
microshift
on
different
boxes,
depending
on
the
resources
and
things
like
that
and
for
a
lot
of
use
cases
yeah,
like
assistant,
D
and
podman
works.
Just
fine,
but
it's
harder
to
I
I,
don't
see
a
tool
yet
for
managing
managing
all
that
from
from
a
single
single
place.
So
so
so
a
more
larger
deployments
would
still
be
problematicated.
E
We
have
been
working
on
like
a
PLC,
actually
developed
application
using
like
kubernetes,
Cube,
Edge
and
open
art,
so
we
have
not
decided
yet,
but
so
far,
nothing
perfect.
That's
why
we
are
trying
to
make
contribution
back
to
the
main
line,
but
I
think
using
kubernetes
main
line
for
the
edge
iot
devices.
It's
too
much.
We
have
to
pay
many
taxes
right
so.
E
D
E
Yeah,
so
we
actually
indeed
develop
the
application
like
a
live
streaming,
reconfigure
the
data
pipeline
using
open,
Earth,
Cube,
Edge
and
kubernetes
to
compare
the
performance
and
this
kind
of
stuff
and
most
likely
for
the
system,
architecture
that
we
can
use
for
our
use
cases
such
as,
like
entertainment,
music,
this
kind
of
stuff.
So
we
we
haven't
done
decide
this.
You
know
like
we
didn't
make
a
decision
yet,
so
we
are
still
like
open
to
get
more
information,
but
we,
we
are
just
trying
to
find
what's
missing
for
us.
E
E
A
A
So
we
generally
publish
them
on
these,
the
kubecon
site
in
that
sked
platform
and
they
get
published
there
as
PDFs.
But
if
you
want
them
as
a
PowerPoint
or
a
Google
slide,
stick,
let
us
know
and
I
think
we
generally
compose
them.
You
know
in
those
platforms
too
so
who
I
didn't
prepare
the
last
Kate.
A
Maybe
was
a
you
or
I,
don't
know
who
owned
the
deck
that
went
on
there,
but
I've
got
some
of
the
older
ones
like
from
yeah,
kubecon,
Europe,
so
I
know
I've
got
some
of
those
and
they
don't
change
that.
Often,
however,
they
did
change
pretty
dramatically
when
we
transitioned
to
the
cnci.
So
how
about?
If
you
reach
out
to
us,
if
you
want
kind
of
a
summary
deck
in
this
group,.
A
E
A
It's
actually
kind
of
interesting
to
me,
I'm,
looking
at
the
conference
on
the
open
source
projects,
they
call
out-
and
it's
kind
of
interesting,
that
there's
a
there's.
A
group
of
these
that
seem
popular
in
North,
America,
Europe
and
the
ones
coming
out
of
Asia
seem
to
not
get
as
much
traction
and
visibility
over
here
and
that
maybe
some
of
it
is
languages
and
not
people
not
presenting
on
these
at
different
conferences.
But
you
know
a
bunch
of
those
project
names.
A
If
you
come
across
anybody
there
by
the
way
who
is
associated
with
these
projects
and
wants
to
present
at
our
group
meetings,
please
throw
that
offer
out
there,
because
I
think
that
we
tend
to
get
better
attendance
when
we
get
presenters
on
particular
projects,
especially
sometimes
the
attendance
isn't
great.
If
we
can't
announce
it,
you
know
a
month
or
more
in
advance,
but
this
would
help
address
that
situation
where
I
think
sometimes
the
Asian
initiated
projects
aren't
that
well
known
in
North,
America
and
Europe,
and
vice
versa.
Perhaps.
A
F
For
some
of
these
Edge
devices
it
might
not
take
advantage
of
all
the
parts
of
kubernetes
like
you
may
not
want
to
have
a
virtual
overlay
Network
across
all
the
devices.
So
then
the
value
of
kubernetes
comes
down
to
the
cube
API
server,
so
I've
been
mulling
over.
How
can
you
take
advantage
of
the
cube
API
server,
so
you
can
use
it
as
a
common
control
interface
for
these
forage
devices.
A
Yeah
I
think
that
what
we've
been
seeing
is
that
exactly
what
you
said
where
they,
the
the
value,
is
having
it
be
something
that
would
host
kind
of
your
own
custom
apis
that
you
could
plug
in
as
crds
to
manage
things
at
Edge,
and
then
there
are
some
existing
open
source
projects.
That
kind
of
do
that,
for
particular
things
like
doing
an
API,
related
device,
Twins
and
things
a
broader
thing
that
has
come
up
over
and
over
again
in
the
past
year.
A
I
think
is
even
aspects
of
things
that
are
below
the
the
water
line:
surface
for
kubernetes,
like
managing
the
OS
and
the
container
runtime
that
has
to
exist
kind
of
as
a
base
platform
for
kubernetes
itself,
managing
those
at
scale
at
Edge
and
keeping
the
things
updated.
Dan
I
think
some
of
the
aspects
that
have
been
going
on
at
Red
Hat
are
kind
of
related
to
that
right,
specialized
things
going
on
in
container
runtime
and
operating
system
level.
C
A
A
Yeah
I
think
it.
Those
things
are
probably
perhaps
not
even
running
a
Linux
based
OS
and
I.
Think
in
some
aspects
some
some
of
them
don't
even
support
imp
network
connectivity
so
that
you
get
into
the
additional
challenge
of
having
to
do
Communications
gateways
or
transitions
right
to
in
order
to
update
applications
and
the
base
platform
at
those.
F
B
A
A
A
C
C
D
C
Things
like
that,
along
with
AI
pipelines,
down
to
edge
devices
and
primarily
focusing
around
video
streaming.
So
that's
actually
a
question.
I
was
going
to
pose
to
the
group
if
you
guys
had
any
familiarity
with
the
different
Technologies
to
stream
video
amongst
Edge
applications
and
also
from
Edge
to
Cloud.
You
know
usually
with
Edge
you're
talking,
like
you
said
about
mqtt
or
rabbitmq,
or
some
of
these
other
messaging
protocols
to
get
the
information
up
to
a
central
Cloud
location,
but
with
video.
D
I
guess
I'm
curious
in
that
scenario,
for
the
ml
processing.
Is
that
what
are
you
pushing
up
to
the
cloud
versus?
What
are
you
processing
on
the
edge
and
in
part
because,
for
example,
with
rtsp?
Is
that
stream
just
going
to
the
local
Edge
node,
which
has
the
ml
application?
Or
is
that?
Are
you
doing
something
where
this
is
like
a
smart,
IP
camera
that
can
actually
hold
a
model
itself.
C
But
that
said,
there
are
a
lot
of
situations
where
customers
may
want
to
see
that
live
rtsp
stream,
either
for
debugging
or
to
get
some
training
data
or
to
even
just
position
the
camera
and
so
being
able
to
stream
that
video
back
at
least
sometimes
can
be
really
useful,
and
so
we've
been
looking
at
basically
repackaging
that
video,
either
into
SRT
or
webrtc,
to
actually
do
the
streaming
back
to
the
cloud
in
those
situations.
But
there's
still
a
bit
of
an
open
question.
I
can't
say
I'm.
C
D
C
C
We
want
to
be
able
to
stream
it
both
to
an
AI
model
through
inference
and
then
also
to
the
cloud
if
somebody
wants
to
see
it
for
debugging
or
some
other
reason,
and
maybe
even
you
know
when
you
have
that
functionality
built
in
you
could
start
looking
at
doing
blue
green
deployments,
where
you
could
run
two
models
side
by
side
on
the
same
stream
and
see
if
the
inference
Compares
or
differs
in
some
way
for
testing.
So
there's
a
lot
of
really
interesting
things
you
do
with
kubernetes
on
the
edge.
A
But
then,
after
training
put
your
model
back
down
on
the
edge
site
and
have
the
model
spot
anomalies
or
things
of
interest,
and
then
you
only
stream
deductions
based
on
the
frame
by
frame
video
and
if
they
find
something
interesting.
Maybe
you'll
selectively
take
some
frames
up
to
the
public
Cloud,
but
try
to
reduce
the
bandwidth,
so
that
is
making
that
transition.
All
the
way
up
to
the
public.
A
A
Consumption
I
produced
an
interest,
a
link
that
I
found
interesting
on
a
video
about
the
challenges
of
edge,
AI
related
to
hardware
and
I.
Think.
Maybe
the
trend
is
to
move
some
of
that
analysis
out
towards
the
edge
to
do
massive
amounts
of
data
reduction.
You
know
by
massive
amounts
you
know
thousand
to
one
or
better,
are
only
sending
what
you
need.
C
I,
remember,
I,
think
the
first
week
I
was
here,
I
heard
Kate
was
a
contributor
to
acri.
Is
that
right,
and
that
was
one
of
the
projects
we
were
really
interested
in,
because
it
made
discovery
of
cameras
so
easy.
So
that
was
really
awesome
to
hear
and
completely
agree
with
you
and
that's
exactly
what
we're
looking
at
is.
This
idea
of
we
could
do
inference
at
the
edge.
C
Send
the
inference
results
out
via
amqp
is
what
we're
looking
at
to
a
cloud
to
then
do
reporting
on
and
show,
and
do
you
know
a
single
pane
of
glass
to
give
you
information
and
it's
a
really
powerful
concept,
but
it
tends
to
be
a
bit
tricky
and
I
completely
agree
with
you
too.
That
Nvidia
seems
to
have
held
a
bit
of
a
monopoly
in
this
space
for
a
long
time,
but
that's
been
really
changing.
C
I
assume,
that's
what
you
meant,
because
we've
been
talking
to
Intel
a
bit
and
with
openvino.
It
seems
like
that's
a
really
powerful
project
to
help
kind
of
use,
different
alternative
silicone,
basically
from
a
standard,
Nvidia
GPU,
so
I
don't
know.
If
anybody
here
has
a
lot
of
experience
with
openvino
I
was
digging
into
it,
but
I'm
not
too
far.
Yeah.
E
E
C
E
E
Yeah,
so
there's
so,
for
we
don't
know
the
same
number
of
the
senders
so
for
our
use
case.
So
once
the
sand,
the
device
comes
online,
we
are
taking
advantage
of
the
demon
set
to
to
you
know
just
let
the
kubernetes
to
start
to
deploy
the
application,
send
those
application
for
each
senders
and
that
their
service
will
be
discovered
by
Qwest
orchestration
and
create
the
receiver
on
the
cloud
to
get
the
you
know,
like
stream
from
it
to
Cloud
and
distribute
live
streaming
for
everyone.
That's
the
that's
kind
of
like
a
PLC.
C
E
D
Did
you
just
set
up
essentially
a
grpc
endpoint
for
the
pre-processing
app
so
essentially
the
pre-processing
app
connects
to
the
rtsb
stream.
It
opens
up
some
sort
of
grpc
service
at
the
next.
Maybe
that's
the
inferencing
app.
Does
the
next
step
on
or
are
you
looking
for,
maybe
something
more
lightweight
and
Native
than
that.
C
I
was
looking
for
something
a
bit
more
centralized
I
guess
where
the
there
would
be
some
sort
of
service,
Discovery
and
agreement
between
the
two
apps
based
off
of
you
know
the
encoding
and
how
to
fetch
the
video
with
grpc.
One
app
would
then
connect
to
the
next
in
the
Stream,
which
is
probably
fine
if
you
had
a
central
Discovery
service,
but
ideally
it
would
go
through
a
broker,
whether
that's
just
for
proxying
or
like
an
actual
message-
queue
I'm,
not
sure,
but
it
seems
like
if
you
were
doing
a
message.
C
But
then
you
lose
a
lot
of
the
benefits
of
you
know
in
a
coding
like
h.264
or
reducing,
reducing
the
size
of
the
Stream,
so
I
I
think
you're,
right
and
I
was
digging
into
something
like
grpc,
but
it
would
be
more
complicated
to
have
to
tell
one
app
where
the
other
app
is.
Basically,
when
they're
both
deployed.
D
The
kubernetes
service
of
the
grpc
endpoint
would
be
what
I'm
expecting,
but
I'm
I'm
sure
I'm
missing
something
there,
but
we've
done
that
at
least
with
awkward
with
demos
where
you
can
connect
to
kind
of
a
broker
of
sorts.
That
is
an
app
that
is,
you
know
it
is
representing
the
device
in
a
way
it's
Gathering
all
the
stream
from
the
device,
and
it
just
has
a
kubernetes
service
in
front
of
it.