►
From YouTube: Kubernetes WG IoT Edge 20221102
Description
November 2, 2022 meeting of the CNCF IoT Edge Working Group. Recapped Kubernetes on the Edge Day event and KubeCon followed by birds of a feather discussion.
A
A
A
If
somebody
is
here
from
the
eclipse
group-
and
they
usually
show
maybe
they'll-
be
a
little
late,
we'll
ask
for
a
recap
of
the
eclipsecon
event
that
conflicted
with
kubecon
but
I
know
some
members
who
regularly
attend
this
group.
Were
there
I'm
going
to
make
a
brief
mention
of
this
scale
conference
coming
up
and
if
members
want
to
add
any
other
agenda,
items
go
ahead
and
edit
the
document
and
edit
the
end
and
we'll
get
to
it
as
time
allows.
But
we
probably
have
a
few
minutes
for
that.
A
If
we
run
out
of
agenda
items,
we'll
try
to
just
switch
over
to
an
ad
hoc
birds
of
a
feather
like
we
usually
do
and
we'll
see
how
it
goes.
So,
the
kubernetes
on
the
edge
Day
event
was
a
pre-event
kubecon
that
lasted
a
full
day.
It
had
topics
all
over
the
map
related
to
kubernetes
and
Edge
will
maybe
I
went
and
in
in
the
engine
notes,
document
I
published
links
to
Dex
and
maybe
about
half
the
sessions
at
speakers
published
decks,
and
then
the
cncf
recorded
these
sessions.
A
So
all,
but
one
session
were
recorded
and
the
one
that
wasn't
recorded
was
unfortunately
one
of
the
better
ones,
but
it
was
ad
hoc.
It
wasn't
even
meant
to
be
on
the
schedule,
but
we
had
a
couple
presentations
that
ran
shorter
than
expected
and
we
actually
drafted
somebody
from
ge
medical
to
come
up
on
stage
and
do
an
impromptu
kind
of
lightning
talk
that
I
thought
turned
out
to
be
one
of
the
better
things,
but
sorry
you
had
to
be
there.
A
The
let
me
just
switch
from
the
agenda
notes
to
this
is
the
actual
conference
session
document.
So
if
you
go
there
and
click
on
these
you
get.
This
is
where
you
go
to
download
the
decks
and
you
can
before
you
bother
reading
the
decks.
I'll
I'll
just
show
you
one.
If
you
click
here,
you
you
can
get
the
abstract
and
speaker
bios
and
if
a
deck
is
there,
this
is
where
you
go,
get
it.
A
The
YouTube
channel
is
up
and,
like
I
said,
I
think
every
session
is
there
now,
so
you
can
go
there
and
watch
these
live
or
use
your
favorite
tool
to
download
these
to
watch
them
offline
and
the
few
I
looked
at
the
video
qualities
really
good,
so
go
ahead
and
the
topics
were
all
over
the
map.
So
I
think
some
people
are
more
likely
to
be
interested
in
some
topics
than
others,
but
I
just
wanted
to.
Let
you
know
that
these
are
all
there.
A
B
Sure,
thanks
so
Amar
and
I
were
able
to
present
on
behalf
of
the
working
group
about
the
white
paper.
We've
been
putting
together
over
the
past
few
months,
and
that
was
all
around
this
idea
of
what
is
Edge
native.
And
can
we
Define
it
in
a
way
where
we
have
a
set
of
principles
that
application
developers
can
look
at
and
see?
Hey
is
my
application
truly
educated
and
how
could
I
possibly
change
it
to
make
it
better
fit
for
the
edge?
B
We
tried
not
to
put
that
too
much
in
the
scope
of
the
paper
but
talk
but
talked
about
how
that
could
be
a
breakout
paper
or
the
more
specific
use
cases
of
5G
on
the
edge.
That
could
be
a
good
breakout
paper
as
well,
and
another
part
of
the
presentation
was
asking
for
feedback,
and
that
was
Natalie
Fisher's
idea,
and
we
had
a
survey
that
we
put
at
the
end
of
the
presentation.
Unfortunately,
we
only
have
three
responses,
and
so
I
think
we
might
want
to
keep
that
as
a
running
survey.
B
So
for
the
announcement
for
this
meeting,
I
put
it
in
the
bottom
of
the
announcements
and
I
think.
Maybe
that's
something
we
continue
to
do
is
just
every
time
we
put
out
anything
for
the
working
group,
whether
it's
an
email
to
the
listserv
or
a
presentation,
or
anything
just
keeping
that
that
survey
there
so
that
we
can
continue
to
gather
that
feedback.
If
you
are
interested
in
giving
any
feedback
to
the
working
group,
that
is
in
the
the
working
group,
iot
Edge,
slack
and
kubernetes
one
other
thing
from
the
session.
B
We
had
a
lot
of
good
questions.
I
think
one
that
was
interesting
to
me
was,
and
if
other
people
attended
and
thought
that
they
had
any
thoughts
on
the
session,
that'd
be
great
to
hear
too
I
thought
one
question
that
was
interesting
is
we
were
focused
a
lot
on
how
Cloud
native
influenced
Edge
native
and
one
person
had
a
question
of
how
did
Edge
native
influence,
Cloud
native
and
I
thought.
That
was
a
really
interesting
perspective.
B
My
response
at
the
time
was
that
edge
native
has
this
sense
of
more
device
awareness
that
now
the
cloud
is
starting
to
take
on
too
with
more
specialized
Hardware
in
the
cloud
itself,
such
as
gpus
but
I.
Think
that's
something
that
would
be
interesting
to
think
about.
As
a
group
too,
to
kind
of
reverse
the
narrative.
There
did
folks
have
any
other
thoughts
on
the
session
or
anything
they
wanted
to
add.
C
One
there
was
a
question
about
Hardware
security
and
I.
Think
there
that's
another
possibility
for
a
breakout
paper.
I
remember
it
was.
How
do
you
secure
the
data
on
the
edge
device
and
obviously
there's
a
whole
working
group
outside
of
this
one
around
confidential
compute?
And
so
that's
a
potential
area
of
Investigation.
B
Yeah,
that's
a
good
point
mark
and
I
think
Amar
kind
of
answered
that
one
with
the
sense
of
you
kind
of
have
two
paradigms
of
ice
of
gardening
hardening
at
rest
and
hardening
during
transaction
or
in
route,
but
yeah
we
weren't.
We
didn't
really
focus
in
on
that
part
of
security
in
the
paper,
so
I
agree.
That
would
be
an
interesting
way
to
look
at
things
and
looking
at
confidential.
Computing
on
the
edge
is
not
something
we've
done
as
a
working
group.
So
other
that's
a
good
point.
A
I
also
noticed
that
at
kubecon
there
were
a
number
of
other
sessions
that
were
very
similar
to
this.
There
was
one
I
think
by
a
Cisco
speaker
that
was
called
Edge
native
applications,
if
I'm
not
mistaken,
or
something
very
close
to
that
there
were.
There
was
at
least
one
like
on
edge
native
applications
at
the
kubernetes
on
the
edge
day,
and
if
you
just
go
to
the
kubernetes
the
kubecon
schedule
here,
I'll
drop
it
in
the
chat
and
search
for
the
word.
Edge
you'll.
A
You
will
find
probably
about
five
other
sessions
that
somewhat
were
I
I'm,
not
going
to
say
they
were
duplicates,
but
they
were
aligned
with
this.
Certainly
and
I
went
to
most
of
them
and
it
seemed
to
seemed
to
be
that
the
things
people
are
saying
are
indeed
quite
aligned.
But
if
you're
interested
in
the
topic,
I
would
advise
looking
at
those
others
too.
As
far
as
I
know,
they're
not
up
on
YouTube
yet,
but
some
of
the
speakers
may
have
made
the
deck
available
in
the
sched
site.
B
Yeah
going
off
that
the
Cisco
talk
that
you
mentioned
was
really
good.
I
I
posted
a
link
here
in
the
chat
and
I
think
it
was
interesting.
Coming
from
the
perspective
of
the
working
group,
where
we
were
trying
to
figure
out
the
path
towards
defining
principles,
they
started
with
the
definition
of
the
12
Factor,
app
and
zoomed,
into
which
of
those
factors
are
particularly
important
for
the
edge.
B
C
They
talked
about
evolving
the
12
factors
and
12
factors
at
the
edge
and
there's
some
good
slides
in
there
that
were
really
quite
talking
about
Edge
native,
as
opposed
to
Cloud
native
there's.
Also
another
presentation
that
I
attended
by
Harry
Lee
from
milo.ai
talking
about
going
into
the
weeds
of
how
they
actually
have
tackled
building
capes
at
the
edge
for
a
industrial
use
case.
So
I
was
really
really
a
nice
detailed
field
example
of
everything
you
guys
have
been
talking
about
so
far.
A
I
haven't
had
a
chance
to
look
at
the
web
assembly
day
agenda
but
I
know.
In
past
events,
there
have
been
a
couple
of
edge
related
presentations
there,
so
I
mean
the
meaning
to
go
look,
but
I
was
unable
to
attend
that
myself.
If
somebody
was
there,
maybe
they
could
mention
whether
anything
Edge
related
happened.
There.
B
I
can
give
a
recap
of
the
day.
I,
don't
remember
anything
specific
to
the
edge.
The
day
was
really
focused
on
evolving
the
ecosystem
in
general,
so
it
started.
Three
of
the
eight
main
sessions
were
about
developments
in
Java,
Python
and
c-sharp,
to
make
it
more
fully
featured
for
webassembly,
both
in
the
browser
and
outside
the
browser,
and
so
really
about
developing
what
languages
people
can
use
with
webassembly.
Several
of
the
sessions
were
about
the
upcoming
changes
to
the
webassembly
ecosystem,
to
enable
code
sharing
and
reuse
and
that's
through
the
webassembly
systems.
B
Interface
or
Wazi
is
governed
by
the
bytecode
alliance
and
multiple
companies
of
all
sizes.
Amazon
Microsoft
fastly
are
all
working
together
to
define
something
called
the
component
model,
which
is
helping
with
and
Registries,
which
is
helping
with
with
this
code,
sharing
and
reuse,
so
is
focused
on
on
that
and
then
Docker
had
a
couple
talks
about
how
they're
participating
now
in
the
webassembly
ecosystem
and
then
beyond
that
I.
B
A
A
And
I
went
and
looked
and
I've
I've
done
a
search
on
this
for
the
word
Edge
and
it
looks
like
a
number
of
these
talks
come
up,
and
this
is
just
it.
The
top
note
aligns
it
by
day,
so
it
lasted
three
days
but
I
think
I
randomly
clicked
on
one
and
it
looks
like
it
gives
you
the
abstract
and
you
can
download
the
slides
to
this
deck.
A
So
it
looks
like
this
might
be
worth
taking
a
look
at
and
poking
around
to
see.
If
you
find
anything
interesting
and
I,
don't
know
what
the
plan
is
for
making
these
available
on
video,
but
the
eclipse
Foundation
sponsors
some
number
of
edge,
related
projects
that
are
well
into
the
double
digits.
So
I
think
this
is
likely
to
be
worth
taking
a
look
at
for
anybody
interested
in
Edge
and
maybe
at
a
future
date.
If
you
go
look
at
this
find
something
interesting.
A
Frederick
is
a
frequent
tender
of
our
meetings
he's
just
not
here
today.
It
could
well
be
that
he's
taking
some
time
off
after
a
conference
or
traveling
back,
but
here
it
is
for
what
it's
worth
then
moving
on
I'm
an
organizer
of
this
scale
conference
that
is
a
Los
angeles-based,
open
source
conference
that
now
you'll
see
the
the
title
here
is
scale
20x.
So
next
year
will
be
the
20th
occurrence
of
this
conference.
It
actually
started
literally
in
a
some
college
student
dorm
room
at
USC.
A
They
were
people
who
liked
Linux
in
the
early
days.
So
originally
this
stood
for
Southern
California,
Linux
Expo,
but
it
served
since
expanded
well
beyond
just
Linux
and
covering
all
of
Open
Source.
This
is,
has
become
the
largest
Community
Driven
open
source
conference
in
North
America
before
covid.
This
was
reaching
attendance
of
4
500
people
and
it's
a
somewhat
unusual
conference
that
a
very
non-corporate
Fair,
fairly
large
numbers
of
student
attendees,
as
well
as
children.
A
Also
a
track.
Labeled
developer
embedded
would
arguably
cover
some
sorts
of
iot
Edge,
and
this
track
would
often
get
right
down
into
hardware
and
people
doing
Hands-On
projects
with
drones.
Arduinos
things
like
that.
There's
a
general
catch-all.
There
is
a
topic:
labeled
home
automation,
Pro
server
at
iot.
A
Some
of
these
other
things
like
observability
could
fit
into
Edge.
Open
data
is
the
concept
that
any
data
paid
for
or
collected
by
public
agencies
should
be
required
to
be
open
access.
It,
arguably,
is
more
of
a
political
spin
on
Tech,
but
there
have
often
been
some
really
interesting
sessions
there
and
in
fact,
I
presented
on
that
track
myself.
On
occasion,
we've
got
open
government.
Another
one
that
could
be
considered.
A
A
If
you've
got
any
questions
like
I,
say
I'm
an
organizer,
so
I'd
be
willing
to
help
coach
anyone
who's
thinking
of
presenting
here
or
if
you
wanted
to
co-present
and
looking
for
co-presenters,
maybe
I
could
help
find
some
sort
of
alignment
and
I
don't
know.
If
anybody
here
has
ever
attended
I
give
you
a
moment
to
give
an
honest
review
of
The
Experience,
but
looking
at
the
names
of
who
we
got
today.
As
far
as
I
know,
maybe
nobody
has
attended
scale
in
the
past.
B
A
B
Sure
so
I
think
we
had
a
big
push
towards
getting
our
draft
of
the
white
paper
out
for
kubecon
and
I.
Honestly
am
very
grateful
that
we
had
a
talk
that
was
kind
of
our
deadline
for
that,
since
it
it
pushed
us
to
have
something
out
there
this
whole.
What
next
I
was
curious
is:
is
there
another
place
to
place
the
draft
that
we
want
to
go
outside
of
GitHub?
Did
people
want
to
put
that
on
some
sort
of
publication
or
blog?
That
was
one
of
the
questions
that
we've
discussed
in
the
past.
B
B
E
Gonna
say
because
I
kind
of
see
like
this
white
paper,
sort
of
as
like
the
I
guess
the
top
level.
And
then,
if
we
continue
to
kind
of
go
down
this
track
of
writing
or
working
on
the
white
paper,
then
we
would
be
a
little
bit
more
specific
and
maybe
a
little
bit
more
detailed
on
the
different
principles
that
we
basically
kind
of
gave
an
overall
view
on.
A
I
think
that
it
was
good
that
we
targeted
the
conference,
but
it
did
get
a
little
rushed
where
I
think
that
this
is
definitely
in
the
draft
stage
and
with
more
time
and
attention
to
detail
put
on
it,
it
could
get
better.
But
it
was
clear
that,
as
we
went
through
this
process,
the
attempt
to
keep
this
concise,
meaning
six
pages
or
less
caused
a
lot
of
material
that
we
were
coming
up
with
to
be
left
on.
A
The
Cutting,
Room
floor,
so
I
think
enabling
spin-off
detailed
papers
was
something
we
talked
about,
and
there's
no
reason
that
we
can't
put
these
additional
reference
documents
out
there
as
drafts
right
away,
so
that
we
have
a
home
for
composing
the
more
deep
dive
drill
Downs
into
specialized
areas
that
the
white
paper
is
an
outline
for.
So
that's
that's
my
opinion
that
we
should
should
have
no
objection
to
people
who
want
to
propose
spin-off
details
that
Branch
off
into
more
detailed
coverage
on
things
outlined
by
the
white
paper
just
go
for
it.
B
So
it
sounds
like
the
call
for
action,
for
that
is
as
a
whole
working
group.
We
settled
on
these
nine
principles
and
that
creates
a
foundation
for
any
working
group
members
on
their
own
in
pairs
whatever
to
create
their
own
draft
and
present
it
to
the
working
group
to
create
their
own
draft
of
any
spin-off
or
focus
focus
of
it
and
present
that
to
the
working
group
at
any
meeting.
Is
that
kind
of
the
idea
that
we're
set
on
working
on
it?
A
Yeah
I
think
that
sounds
like
a
good
procedure
to
just
self-enable
people
to
go
off
and
do
drafts.
Ultimately,
I
think
it
has
to
be
reviewed,
because
this
is
going
to
be
viewed
as
sort
of
an
officially
sanctioned
thing
coming
out
of
a
cncf
working
group.
So
the
idea
that
some
people
would
go
off
unilaterally
compose
something
and
publish
it
without
review
is
probably
not
appropriate,
but
you
know,
as
often
happens,
even
an
open
source
code.
A
People
have
to
go,
do
the
work
and
submit
it,
and
it
should
be
looked
on
as
others,
because
the
whole
principle
was
as
part
of
a
as
as
part
of
the
community
standards
is
that
this
is
looked
on
and
vetted
for
conflicts
of
interest,
and
things
like
that.
We
particularly
don't
want
aspects
of
this
to
be
a
mechanism
for
commercial
advertising
and
things
like
that.
But
you
know
I
I'd
say
what
Kate
said
his
outline
is.
A
plan
is
a
good
way
here.
E
A
E
A
And
Kate,
what
are
the
other
things
you
brought
up
is
whether
we
wanted
to
publicize
this
a
bit
now
that
we've
got
it
out
there.
I
personally
would
say
yes
as
long
as
you
know,
we
are
clear
that
this
is
still
in
the
formative
phases
and
this
isn't
like
a
final,
a
view
of
a
final
polished
document
at
some
point,
I
think
we've
even
potentially
trying
to
get
the
the
steering
people
of
the
cncf
itself
to
recognize
this
more
prominently,
but
maybe
we're
not
ready
for
that.
A
A
We
got
after
the
presentation
and
polishing
it
up
just
a
little
bit
before
we
move
on
to
seeking
public
ity
and
we
probably
got
a
fair
amount
of
attention
just
by
virtue
of
going
out
there
with
kubecon,
and
maybe
we
can
wait
until
people
who
weren't
able
to
physically
attend
catch
up
on
the
kubecon
videos
and
things,
and
you
know,
as
the
attention
wave
from
that
starts
to
go
down
a
bit.
We
can
start
a
second
wave
of
publicity.
Perhaps.
E
A
Yeah,
in
fact,
maybe,
if
you're
getting
started
on
forking
off
some
more
Deep
dive
things.
If
we
had
one
or
two
of
those
available
to
talk
to
talk
about
too,
you
know
to
indicate
they're
in
the
formative
stages
and
we're
open
to
other
people
joining
in
on
this
effort
that
that
would
help
I
think
make
it
a
a
more
full-featured
story
for
something
like
the
new
stack
yeah.
E
A
B
Like
I
mentioned
earlier,
we
didn't
get
a
ton
from
the
form
and
so
I
think
understanding
what
it
means
to
have
a
final
version
of
our
principles
are
like
our
original
white
paper
would
be
really
helpful
and
I.
Think
feedback
is
kind
of
a
precursor
to
that.
So
I'd
be
curious
for
folks
on
this
call,
if,
if
any
of
you
all
from
had
immediate
feedback
on
it
and
of
course,
like
that's
putting
you
on
the
spot,
but
finding
a
way
to
get
that
feedback
would
be
helpful
regardless.
C
B
A
A
few
things
I'd
suggest
to
get
feedback.
I
I
confess
that
if
there
was
a
form
or
a
survey,
I
haven't
done
it
myself,
you
know
being
at
kubecon
and
giving
talks.
I
just
was
fully
busy
and
what
wasn't
doing
any
of
the
collateral
stuff
but
make
it
easy
to
find.
So
if
somebody
happens
to
find
the
white
paper
I'm
hoping
there's
a
way
where
the
survey
is
right
there
in
their
face,
like
maybe
added
onto
it.
The
other
thing
is
now
that
the
video
of
that
presentation
at
kubecon
has
been
published.
A
I
believe
that
I
don't
really
know
if
the
cncf
opened
it
up
for
comments
or
not,
but
that's
an
option
when
you
publish
on
YouTube.
But
if
it
does
allow
comments
on
the
video,
maybe
you
can
add
a
link
to
the
form
as
a
comment,
so
that
somebody
who
watches
that
session
in
the
recorded
form
will
discover
the
survey.
A
D
B
A
And
I
don't
know,
but
in
my
experience
maybe
you
already
did
this,
but
whenever
I
put
out
a
survey
I
find
you
get
better
results.
If
you
tell
them
how
long
it
will
take
in
advance,
like
you
know,
will
you
take
this?
Two-Minute
survey
gets
better
response
than
will
you
take
this
survey
and
obviously
that
depends
on
it
being
short.
Hopefully
it
isn't
like
a
40-minute
survey,
because
that
could
be
an
issue.
If
it's
super
long
people,
just
don't
people
just
start
them
and
abandon
them.
A
B
B
I
can
I'm
willing
to
do
that,
and
then
we
have
a
to-do
for
Natalie
to
talk
to
the
new
stack
that
might
be
later
on
when
we're
more
formalized
and
okay,
great
Natalie,
yeah
I'll
reach
out
to
you
about
with
the
newsletter
sending
that
out
and
did
we
have
any
other
to-do's
that
I
didn't
capture
from
that
discussion.
There.
A
C
B
C
C
A
B
A
Okay,
I
think
at
this
phase,
we're
out
of
agenda
items.
Anybody
who
wants
to
add
one
right
now,
verbally
just
go
for
it,
and
otherwise
we'll
switch
to
just
kind
of
a
birds
of
a
feather
anybody's
free
to
shout
out
anything.
They
found
interesting
cool
anything
you
want
to
talk
about
that.
You
saw
if
you
went
to
kubecon
that
was
interesting
in
in
a
booth
after
hours
or
in
a
session
just
go
for
it
and
we'll
chat.
A
C
So
one
of
the
things
as
a
newcomer
to
the
group,
one
of
the
things
I
really
appreciate
about
the
the
group's
work
and
presentation
at
kubecon.
Was
it
reassured
the
the
field?
Practitioners
that
you
know
how
we're
approaching
Edge
you've
you
helped
validate
what
we're
doing
in
the
field
and
how
we're
looking
at
Edge,
because
before
this
before
senior
presentation,
the
question
I
had
coming
into
kubecon
is:
are
we
doing
things
right
right?
What's
the
rest
of
the
world?
C
How
are
they
looking
at
this
and
I
was
really
pleased
and
that's
why
I
wanted
to
join,
and
you
know,
help
the
working
group,
so
my
Impressions
from
kubecon
were
really
about
what
are
others
doing
in
the
space
and,
as
we
discussed
earlier,
there's
other
presentations
like
the
Cisco
presentation.
That
was
really
helpful
and
you
know
being
able
to
you
know
not
only
put
a
lens
on
this,
but
also
corroborate
the
the
working
groups
draft
paper,
and
so
it
was.
A
One
of
the
things
I
was
encouraged
at
at
kubecon.
Kind
of
a
hallway
track
was
even
what
percentage
of
the
generic
kubecon
attendees
were
interested
in
an
edge.
You
know
just-
and
this
comes
from
my
experience
of
just
at
the
lunch
break
sitting
at
some
random
table
with
you
know,
complete
strangers,
who
were
just
there
for
the
event
and
pretty
much
even
these
big
Enterprises
that
started
out
with
Landing
kubernetes
in
the
major
public
clouds
have
use
cases
they're
looking
at
for
also
running
things
out
at
Edge
locations.
A
I
can't
think
of
any
group
that
I
sat
down
with
at
lunch
when
I
say:
hey
I
was
here
starting
the
week
for
the
kubernetes
on
the
edge
day,
and
they
came
up
with
things
that
they
had
going
on
at
Edge.
Maybe
early
formative
stages,
but
I
was
really
con,
encouraged
by
the
percentage
of
kubernetes
users
that
actually
have
viewed
this
as
something
where
hey
kubernetes
is
stable.
We
clearly
have
things
going
on
moving
workloads
to
the
public
clouds
with
kubernetes,
but
we're
also
looking
at
Cloud
native
applications
at
Edge.
C
E
Go
ahead,
sorry
I
was
gonna,
say
that's
super
nice
to
hear,
because
I
was
also
going
to
just
add
some
color
to
that
too.
I've
been
having
a
lot
of
customer
conversations
with
like
customers
like
PNG
and
Chevron
and
Lumen,
and
they
are
all
looking
to
deploy
to
Edge
and
they're
all
very
much
at
the
beginning
stage.
C
Is
one
of
the
most
interesting
conversations
I
had
was
with
the
chief
metallurgist
for
General
Dynamics,
big
gun
board
division
and
his
he
was
came
here
to
learn
more
about
how
they
can
do
metrics
and
observability
at
the
edge?
And
that
really
means
like
getting
you
know:
data
sensors
and
collecting
those
data
sensors
from
you
know
tiny
devices,
Raspberry,
Pi
or
Arduino
level.
C
You
know
work
units,
and,
and
so
when
we
start
to
look
at
how
Edge
is
evolving.
You
know
it's
a
really
interesting
problem,
because
you
know
some
of
my
Automotive
customers
are
also
looking
at
with
their
kubernetes
challenge.
Is
how
much
compute
do
they
have
at
the
edge
to
be
able
to
do
things
and
I
I
really
appreciated
that
geographical
picture
that
you
guys
put
together
to
be
able
to
help
dial
in
and
understand
that.
E
E
E
Yeah,
like
that's
what
I'm
talking
about
too,
like
that's
exactly
some
of
the
orchestration
component
of
it.
It'll
get
definitely
get
a
lot
more
interesting
too
and,
like
I,
said,
I'm
working
more
on
the
ml
side
at
Edge,
but
I'm
also
seeing
a
lot
of
interesting
transitions
with
different
accelerators
at
the
edge
as
well
I
love
that
will
also
be
orchestrated
so
I'm
working
on
a
few
projects
there
around
inferencing
and
model
training,
yeah.
A
Yeah
I
think
people
are
seeing
challenges
not
just
related
to
the
orchestrator,
but
maybe
even
the
container
runtime
too,
like
is
Docker
the
best
container
technology
or
is
maybe
I
think
it's
so
early
days
for
webassembly
that
it
might
be
five
years,
but
there
are
clearly
people
doing
things
with
webassembly
or
wasm
that
look
like
they
have
nice
attributes
for
Edge
applications.
The.
E
Other
one
is
one
of
the
Open
Source
Products,
that's
doing
that
where
they
are
looking
at.
The
reason
was
in
for
neural
networking:
did
you
to
kind
of
look
at
and
how
they're
going
to
do
that
and
kind
of
on
the
walls
and
Forum,
because
that's
another
space
that
I'm
also
looking
into
and
exploring
as
well.
A
You
know
one
of
the
talks
I
gave
on
kubernetes
at
Edge
days
nominally
was
on
running
a
home
lab,
but
one
of
the
things
I've
fallen
in
love
with
in
my
home
lab
is
running
an
outbound
only
tunnel
to
ecdn
and,
having
that
be
my
point
of
presence,
rather
than
trying
to
invite
connectivity
right
to
you,
know
some
sort
of
a
Nat
on
a
home
router
that
just
in
a
home
situation,
I'm
a
skeptic
on
the
idea
of
running
kubernetes,
like
networking
and
I.
Think
by
association.
A
Applying
what
like
tunnels
or
kubernetes
like
things
like
load,
balancer
and
Ingress
I,
think
the
answer
is
both
because
I've
seen
people
do
it
not
necessarily
with
uniform
success.
But
it's
you
know
like
I,
say
it's
early
stages
and
people
are
kind
of
trying
everything
and
one
of
the
interesting
aspects
of
groups
like
this.
Is
you
do
get
feedback
from
people
who
went
that
direction
dip
their
toe
in
the
water
and
are
giving
reports
on
how
it
turned
out.
C
To
Rob's
point
I
think:
if
we're
going
to
be
talking
about
the
edge
in
perpetuity
for
a
little
while
I
think
we
do
need
to
explore
constrained
devices,
because
the
categories
of
what
an
edge
device
is
you
know
varies
based
on
the
use
case.
It
could
could
very
well
be
a
Raspberry
Pi
or
it
could
be
a
full
pizza
box.
C
D
The
the
use
case
I'd
be
interested
in
exploring
if
people
are
interested
in
joining
me
would
be
using
a
c
or
rust.
Based
cubelet
running
on
some
of
these
devices
and
to
the
comment
you
made
Steve
stripping
away
things
that
you
don't
need
at
the
edge
like,
maybe
even
orchestration,
and
just
using
the
control
plane
aspects
of
kubernetes
in
order
to
schedule
workloads
onto
some
of
these
small
devices.
A
D
A
Think
some
of
these
versions
of
kubernetes,
like
Cube
Edge,
super
Edge,
open
your
actually
do
involve
or
at
one
time
did
involve
modifications
to
the
cubelet
and
kind
of
what
I
call
a
stretch,
control
plane,
where
at
an
edge,
Leaf
node
they're
attempting
to
declare
that
to
be
a
kubernetes
single
cluster
note,
or
maybe
they
were
even
running
a
handful
of
worker
nodes
there
with
the
control
plane
up
in
a
public
cloud
and
you'll
find
some
projects
went
that
direction,
and
the
reports
I
hear.
That
is
that
it
works
well.
A
If
the
connectivity
to
the
cloud
hosted,
control
plane
is
relatively
good,
but
if
that
becomes
something
that
is
intermittent
by
Design
or
by
by
virtue
of
frequent
failures
that
does
present
some
challenges
and
kind
of
the
original
Orca.
The
original
digital
design
of
kubernetes
was
that
you
would
cool,
centralized
more
or
less
heterogeneous
resources
into
a
pool
and
have
the
schedule
or
move
workloads
to
it.
When
you
invert
that
to
an
edge
thing,
you
have
to
ask
yourself
whether
you've
sort
sort
of
are
causing
the
kubernetes
scheduler
and
design
to
be
morphed
into
something.
A
It
wasn't
optimized
for
right.
In
other
words,
you
care
once
you
care
very
much
that
this
workload
needs
to
go
to
exactly
this
location.
You
know,
say
you
get
to
a
cardinality
situation
where
one
workload
one
one
exact
instance
of
the
leaf
node
that
it's
allowed
to
run
on.
You
can
accomplish
that
with
taints
labeling
Etc.
But
at
the
end
of
the
day
you
have
to
ask
yourself
whether
the
kubernetes
scheduler
is
your
friend
or
something
that
you
know
you're
effectively,
just
more
or
less
disabling.
A
There's
no
right
answer
to
that.
But
I've
heard
some
people
who
came
to
the
conclusion
that
at
the
end
of
the
day,
maybe
they
were
better
off
just
deploying
a
container
runtime
engine
at
those
Edge
locations
and
coming
up
with
a
different
thing
to
model
it.
There
have
been
attempts
that
Define
custom
resource
definitions
and
kubernetes
so
that
you
could
still
use
the
kubernetes
control
plane
to
manage
it,
but
technically
they're,
not
kubernetes
worker
nodes.
D
Yeah
and
I
think
when
you
talk
about
iot
devices
that
operate
in
isolation
that
makes
way
more
sense
because
you're
not
over
engineering,
the
problem.
The
challenge
is
when
you
start
to
have
devices
working
together
and
you
need
that
single
orchestrator
of
workloads
across
devices.
You
ideally
want
the
same
control
system
for
your
compute
Edge
devices,
as
well
as
your
far
Edge
devices
that
are
very
resource
constrained,
so
I'd
be
very
interested
in
finding
out
what
other
people
ran
into.
But
the
problem
I'm
trying
to
look
at
is
having
that
single
control
plane.
C
Yeah,
that
is
a
challenge
in
design,
and
it's
always
going
if
we
look
at
the
current,
like
Stephen
had
pointed
out,
is
that
the
original
kubernetes
control
plane
design
it
does
not
was
never
designed
in
with
the
edge
in
mind.
It
was
always
designed
with
a
pool
of
resources,
and
so
we
started
looking
at
the
SCD
communication
requirements
and
response
times.
You
can
fine
tune.
Those
to
you
know
really,
you
know
small
latency,
but
any
hiccup
in
that
and
you
you.
A
Yeah,
a
few
other
challenges:
I'll
throw
out
there.
You
know
public
Cloud
versus
Edge
are
even
the
scale.
You
know,
kubernetes
really
was
made
to
scale
up
to
5
000
cluster
nodes,
I
think.
As
far
as
I
know,
I
haven't
looked
at
the
spec,
but
I
think
it's
still
five
thousand
and
there
are
age
deployments
that
are
way
over
5000
Edge
notes.
You
know
hundreds
of
thousands
now
granted.
A
You
could
go
with
multiple
kubernetes
clusters,
but
even
if
you
do
that
and
break
them
into
five
thousand
node
increments
kubernetes
at
5000
nodes
is
actually
really
at
an
upper
limit.
You
know
I,
believe
people
find
that
they're
Etsy
d
doesn't
necessarily
scale
linearly
in
terms
of
the
demands
for
just
keeping
your
NCD
healthy
and
that
you
might
have
to
go
up
to
five
nodes,
seven
node,
if
NCD
clusters-
and
if
you
then
needed
multiple
kubernetes
clusters,
the
thing
is
going
to
have
a
pretty
big
appetite.
A
The
other
concern
is:
what
do
you
do
with
these
Edge
locations
when
they
perhaps
undergo
in
they're
in
different
failure
domains
for
power?
So
they
power
cycle,
and
maybe
they
power
cycle
and
everything
doesn't
come
up
quickly
or
maybe
at
all.
In
other
words,
the
Internet
contract
activity
is
down
at
the
moment
they
recover
or
access
to
things.
They're
using
like
image
repositories
is
down.
A
You
walk
yourself
into
something
where
you're
saying:
okay,
if
I
have
to
anticipate
being
able
to
tolerate
loss
of
internet
connectivity
intermittently
I
have
to
run
air
gap,
so
that
means
I
have
to
stand
up
a
local
container
image
registry.
By
the
time
you
get
there
more
and
more
resources
are
being
duplicated
up
at
these
Edge
locations
and
gobbling
up
resources,
and
this
this
can
be
a
slippery
slope
that
isn't
real,
attractive
right,
I,
don't
know
on
a
universal
solution,
but
I'm
just
telling
you.
C
So
that's
the
third
rail
or
the
the
third
angle
on
this.
We
have
Network
constraints,
we
have
resource
constraints,
and
now
we
have
failure,
domain
constraints
and
those
three
things
are
going
to
dictate
how
you
design
your
Edge
and
and
the
polar
opposite.
How
you
design
your
clusters
that
are
going
to
happen,
support
the
edge
you
know.
Are
you
gonna
do
like
you
said
Steve?
If
you're
gonna
do
seven
node,
you
know
SCD
Central
cluster
or
you
go.
C
You
know,
go
crazy
and
do
Submariner,
which
is
cluster
of
clusters,
to
support
your
you
know,
20
000
or
30
000
Edge
devices,
you
know
again.
Sometimes
kubernetes
doesn't
solve
the
edge
problems.
It's
now
you're
solving.
It's
you're
pigeonholing,
a
a
design
Choice
into
the
solution,
so
yeah
I
think
those
three
things
are
going
to
potentially
dictate.
You
know
like
a
spin-off
article
or
document
for
that.
Rob
was
talking
about.
A
Okay,
at
this
point,
we're
three
minutes
after
the
hour,
so
I
think
we're
gonna
have
to
call
it
quits
here,
but,
as
usual,
the
birds
of
a
feather
discussion
got
kind
of
interesting
and
we'll
resume
again
in
a
couple
of
weeks.
Thank
you,
everybody
for
attending.
If
anything
else
didn't
get
covered,
just
bring
it
up
on
the
slack
channel.
So
thanks
everybody.