►
From YouTube: Kubernetes WG IoT Edge 20230517
Description
May 17, 2023 meeting of the CNCF IoT Edge Working Group. Open community "birds of a feather" discussion: Why have some vendors been end-of-lifing IoT services. Fisher Technik Industrial IoT Test Factory for testing, training, demos.
A
We
are
recording
welcome
to
the
May
17th
edition
of
The
cncf
iot
Edge
working
group
I
just
checked
about
10
minutes
ago
and
didn't
see
anything
on
the
agenda.
There
might
be
some
late
ads,
but
before
this
meeting
formally
got
started,
we
I
just
started
a
little
bit
of
chat,
mentioning
that
I
was
catching
some
of
the
material
that
was
going
on
in
the
eclipse,
Foundation
iot
webinar,
that's
in
its
second
day
today,
it's
second
of
two
days
and
just
I
I
actually
had
to
drop
out.
A
But
there
was
an
interesting
talk
of
an
end
user
going
over
the
an
account
of
the
pain
they
had
when
the
cloud
hosted,
managed
iot
platform,
they
had
been
on
God
received
an
end-of-life
announcement
and
they
had
come
to
the
conclusion
that
they
just
as
soon
moved
to
something
open
source
that
they
could
maybe
host
themselves
and
they
recounted
that
in
recent
time.
Well,
it
actually
said
that
it
in
the
period
of
2022
to
23
that
managed
iot
platforms
have
been
killed
by
Google,
IBM,
sap
and
Bosch.
A
So
you
know
there's
enough
names
in
there
that
perhaps
it's
a
trend,
Philip
or
I-
think
Kate
asked
if
any
reason
was
given
and
I
had
to
say
that,
certainly
not
in
that
presentation
and
I'm
not
aware
of
one.
But
my
guess
is
that
it
was
because
they
weren't
making
money
on
the
platform.
Certainly
if
they
were
making
huge
profits.
I
can't
believe
they'd
shut
it
down
and
then
Philip
was
about
to
say
something.
But
we
decided
to
start
the
meeting
so
go
ahead.
Phil
yeah.
B
I
think
there's
a
plethora
of
things
in
here:
a
lot
of
people
look
at
it
and
go
oh,
what
a
big
Market
we
can
make
loads
of
money
there,
but
iot
is
really
hard
because,
while
there
are
some
underlying
core
principles
like
device
management
over
there
updates,
each
industry
has
very
specific
requirements,
which
are
you
know,
specific
to
agriculture
or
or
manufacturing,
or
something
else
not
just
from
a
functional
perspective,
but
also
from
how
do
you
sell
to
these
people?
How
do
you
discover
them?
How
do
you,
how
do
you
take
stuff
to
them?
B
At
the
same
time,
the
OT
and
where
iot
networks
is,
is
the
antithesis
of
of
IO
of
I.T
and
where
a
lot
of
these
companies
come
from
of
big
it
like
Google
sap,
we're
like
hey.
What's
you
know
hyperscale
when
you
need
to
be
in
an
air-gapped
environment
with
100,
uptime
or
close
to
100
uptime,
it
completely
changes
how
you
think
about
your
architecture
and
your
go
to
market
and
your
productization
and
I
also
like
because
of
all
of
these
things
together.
B
If
you
look
at
the
players
who
are
successful
in
iot,
a
lot
of
them
are
very,
very
focused
on
specific
verticals.
So
you
know
I've
done
some
work
with
a
company,
that's
really
big
in
retail
and
they
you
know
they
they
deploy
into
major
retailers
to
to
manage
fridges
and
energy
and
and
stuff
like
that.
And
while
you
can
start
saying
well,
we
could
take
that
and
now
into
other
locations.
Other
use
cases,
it's
not
paying
Architects
to
any
iot
use
case,
and
therefore
you
have
a
lot
of
verticalization
of
these
Solutions.
B
So
I
think
there
was
a
lot
of
I
want
to
say
10
years
ago,
a
lot
of
people
going.
Oh
we'll
build
an
iot
platform
and
then
a
lot
of
people
found
out.
You
know
it
doesn't
have
the
80
margins
of
of
selling
compute
or
selling
serverless
or
or
all
these
other
functions,
and
therefore
that
that's
why
I
think
they
yeah
they.
They
decided.
It
wasn't
strategic.
C
And
I
think
another
follow-up
of
that
is
I.
Think
big
companies
are
probably
using
iot
as
a
way
to
bring
them
into
their
other
Services.
So
if
they're
not
getting
the
biggest
margins
on
iot,
maybe
they
could
use
it
to
bring
them
into
their
Edge
Solutions,
but
we're
still
working
on
having
good
Enterprise
Edge
Solutions
out
there
and
people
aren't
fully
bought
into
the
edge
yet
or
ready
to
migrate
fully
there.
C
So
I'd
be
curious
if
the
iot
side
of
it
was
too
early,
and
maybe
once
Edge
is
more
mature
than
the
iot
managed
solutions
from
some
of
these
big
companies
will
come
back
as
a
way
to
bring
you
into
their
Edge
managed
Solutions.
B
Good
good
question:
I
partially
Maybe
the
work
that
I
do
around
edge
and
iot
a
lot
of
it
is
environments
where
you
don't
have
internet
or
you
can't
depend
upon
the
internet
or
the
environment
needs
to
be
completely
Sovereign,
because
if
a
tornado
rolls
through
you
need
to
rebuild
on
the
Assumption,
you
don't
have
an
internet
connection
and
that
doesn't
really
fit
to
the
the
business
model
of
those
Mega
providers
who
want
to
to
do
things.
Internet
provided
and
at
the
same
time,
they're.
B
You
know
that
that
if
you
speak
to
some
people
who
work
in
this
space,
for
example,
I'm
doing
some
work
at
the
moment
with
Department
of
engine
to
distribute
energy
resources.
If
you
were
the
company
that
I'm
I'm
working
with
on
it,
they
say
you
know,
we
love
this
stuff
in
the
cncf
space,
but
they're
building
it
for
Mega
data,
centers
and
internet
distributed
and
they're
not
focused
on
how
do
I,
deploy
it
into
a
sovereign
air-gapped
environment
which
has
constrained
runtime
requirements.
B
In
fact,
egex
Foundry,
who
who's
who's
Big
in
in
the
LF
Edge.
They
they
had
a
I
think
about
two
years
ago.
Their
paper,
which
was
how
do
we
move
to
to
LF
Edge?
Sorry?
How
do
we
move
to
edgex,
Foundry
3.0
and
one
of
the
core
components
was
we
need
to
stop
using
these
cncf
components
which
were
great
to
build
our
first
product.
Instead,
we
need
to
move
to
to
components
which
are
built
for
the
edge
and
which
run
on
you
know,
megabytes
single
megabyte.
B
A
Yeah
to
I
I
think
I
agree
with
you
Philip
and
to
emphasize
that
with
a
tangible
example.
You
know,
you've
got
these
device
management
functions
where
you
might
need
to
monitor
devices
in
a
particular
Factory
or
a
particular
vehicle,
and
you
need
kind
of
the
dashboard
scenario
there
for
the
local
operator
anyway,
and
it
needs
to
still
be
there.
A
So,
what's
the
value
to
if,
if
it's
an
additional
layer
to
also
reproduce
it
up
in
the
cloud
when
you
still
need
it
locally-
and
you
know
I
think
maybe
there's
some
realization
of
the
cloud-hosted
central
not
being
a
complete
solution
that
will
cover
what
you
need
and
if
it
doesn't,
maybe
the
there's
an
argument
that
the
more
important
one
is
the
locally
hosted
rather
than
the
cloud
hosted
and
the
cloud
hosted
is
a
luxury
that
could
be
yeah
added
on
later
or
put
at
a
different
level.
It's
like.
A
If
you
look
at
automobiles
every
one
of
them
needs.
You
know
the
instrumentation
in
the
dashboard.
But
what
is
the
benefit
if
every
Ford
automobile
in
the
world
could
be
centrally
monitored
but
where's
the
value
there
I
mean
somebody
might
like
to
see
that.
But
then
others
are
even
concerned
about
the
Privacy
leakage
of
what's
involved
with
that
and
I
think
in
the
iot
space.
We're
getting
to
a
point
where
some
of
these
issues
of
of
information
leakage
across
borders
starts
to
become
sensitive
too.
D
Yeah
I
think
I
could
agree
to
what
exactly
what
Philip
was
mentioning,
because
I
am
part
of
that
world
and
it's
quite
there
are
two
facets
to
it.
I
think
what
Philip
mentioned
is
the
fact
that
OT
the
operations
technology
part
is
is
in
itself
is
a
very
rigid
world
right
you
let
there's.
No
there's
no
sense
of
oh
I
can
have
a
downtime
over
here,
because
it's
a
very
critical,
it's
a
very
critical
application
right.
D
You
can't
just
tell
the
oil
and
gas
people
guess
what
we'll
have
to
shut
down
the
rig
for
15
minutes,
but
that
would
cost
them
a
lot
of
money,
so
that
part
is
extremely
rigid,
while,
as
on
the
other
hand,
the
I.T
part
is
Extreme
is
designed
in
a
way
to
be
extremely
flexible
right,
adding
a
bit
of
redundancy
baby
scaling
up
scaling
down
this
and
that,
but
when
you
think
about
vertically
scaling,
it's
a
very
difficult
thing
to
achieve
in
in
this
field.
D
So
that's
one
of
the
major
aspects
that
most
of
these
big
platforms
weren't
able
to
achieve
what
they
wanted
and
I.
On
the
other
hand,
I
think
it's
also
the
way
of
monetization
and
Licensing.
It's
not
that
easy.
When
you
have
x
amount
x,
100
times
of
types
of
devices
that
need
to
do
something
very
specific
and
if
they
are
anyways
not
generally
connected
to
the
internet,
the
the
monetization
part
becomes
a
bit
more
of
an
Hassle
and
I.
Think
the
last
part
in
general
is
if
it's
Cloud
related
you.
D
The
companies
need
to
be
very
semantically
sound
right,
like
if
a
certain
x
amount
of
sensors
or
actuators
are
in
this
form,
or
in
this
they
have
to
behave
in
such
a
way.
They
need
to
be
described
in
such
a
way
that
someone
sitting
in
a
completed
Geographic
will
location
would
be
able
to
understand.
E
Yeah
I
I
also
tend
to
agree
that
that
the
the
cloud
provider
iot
platform,
you
know,
promises
the
scale
of
large
number
of
devices,
hundreds,
thousands
and
millions,
but
that's
rarely
the
use
case.
So
so
it's
mostly
a
smaller
number
devices.
That's
much
easier
for
people
to
to
cover
from
that
perspective.
But
the
problem
lies
in
in
this
other
area
like
how
we
vertically
integrate
integrate
all
that
and
and
that's
where
they
need
more
flexibility
than
than
what
what
Cloud
offerings
are
are
are
offering
basically.
A
Okay,
well,
this
wasn't
formally
on
the
agenda,
but
an
interesting
conversation
and
I.
Don't
think
we'll
we'll
solve
this,
but
I
can
tell
you
that.
Having
sat
in
for
most
of
that
presentation
that
talked
about
this,
the
conclusion
of
one
particular
end
user
was
that
they
decided
that
they
were
going
to
go.
A
Take
a
look
at
the
open
source
components
out
there,
and
rather
than
sign
up
with
yet
another
cloud-hosted
iot
management
platform,
they
were,
they
were
choosing
to
go,
stand
up
their
own
open
source
based
infrastructure
to
manage
what
they
had
going
on
for
Edge,
so
found
it
interesting
and
maybe
we'll
see
a
recurring
pattern
evolve.
There
I
think
I
noticed
that
Shan,
you
added
something.
Did
you
want
to
talk
about
these
things
you
listed
in
the
agenda,
notes
on
use
cases
for
Kate's
and
Edge.
D
Yeah,
so
in
a
nutshell,
what
I've
been
talking
to
a
couple
of
my
colleagues?
We
have
a
so-called
let's
call
it
a
Toy
Factory
we.
This
is
a
very
specific
German
company.
That
does
this
kind
of
things.
It
looks
like
a
Lego
game,
but
it's
it's
a
very
well
designed
product
where
you
can,
for
example,
program
the
whole
Factory,
this
Toy
Factory,
which
is
in
the
link
in
the
docks
through
actual
industrial
controllers,
so
I
have
seen
people
who
have
programmed
this
whole
Factory
using
Siemens,
S7
plcs.
D
We
at
Emerson
have
actually
done
some
progress
in
terms
of
using
our
industrial
plcs
to
control
this
Factory,
and
we
now
also
have
their
infrastructure
to
to,
for
example,
use
some
form
of
tools
that
cncf
provides
or
maybe
using
tools
like
Cube
Edge
or
something
like
that
which
which
we
can,
which
we
are
anyways,
looking
forward
to
giving
it
a
try
with
this
Factory.
It's
it's
a
low
risk
and
a
high
reward
kind
of
a
situation
for
us,
because
it
does
work
like
an
actual
industrial
Factory,
all
the
consequences
being
here.
D
Something
goes
wrong.
We
don't
end
up
getting
sued
or
we
don't
end
up
getting
any
problems.
We
can
always
reset
it
hard
reset
the
whole
system
again
so
yeah.
That's
that's
something.
I
wanted
to
anyways
present
over
here,
I'll
yeah
I.
Think
in
the
coming
meetings.
D
I'll
try
to
prepare
some
couple
of
slides
of
how
it
actually
looks
like
what
are
certain
components
to
it
and
yeah
and
I
think
it
might
be
interesting
for
the
for
the
me
for
the
people
around
here
to
maybe
to
maybe
have
some
ideas
about
if,
if
there
are
certain
things
that
we
would
like
to
do
as
a
so-called
prototype
first
or
maybe
something
with
the
cncf
tools
that
we
have
in
an
actual
so-called
simulated
industry,
so
I
was
I
put
that
thing
in
the
agenda
so
that
maybe
it
might
be
something
that
we
can
actually
work
on
together.
D
I
do
have
some
green
light
from
my
superiors
at
least
a
dimensional,
discrete
automation
that
we
can
actually
use
these
kind
of.
We
can
use
the
infrastructure,
so
I
think
it's
it's
a
good
sign,
and
hopefully
you
can
come
make
something
out
of
it.
A
F
A
D
In
fact,
we
actually
bought
it
as
a
company
because
it
is,
it
is
a
decent.
It
does
take
up
a
decent
space
right,
so
it
is
huge
compared
to
what
what
we
see
in
the
on
the
website,
and
it
does
a
lot
of
things
here
and
there,
so
it
actually
simulates
a
way
of
you
have
like
small
blocks
that
go
through
a
assembly
line.
Some
it
goes
into
a
so-called
simulated
oven
for
five
ten
seconds,
depending
on
how
you
program
your
plcs
to
it.
D
It
then
sorts
it
in
an
high
Bay
storage
area
and
everything.
So
it's
quite
nice.
It
has
a
lot
of
bells
and
whistles
to
it
and
I
think
it
also
provides
certain
Wireless
stuff,
so
I
think
if
I'm
not
wrong,
they
do
have
a
wireless
wi-fi,
router
or
something
like
that,
also
with
some
mqtt
stuff,
and
if
you
use
a
PLC
like
ours
at
least
Emerson
or
S7
or
any
any
of
the
modern
plc's
you
can
actually
have.
D
You
could
control
the
whole
assembly
line,
as
well
as
the
whole
Factory
through
OPC
UA.
So
that
is
a
plus
point
where,
if
we
don't,
we
can
actually
do
some
fun
stuff
with
some
other
tools
that
are
currently
on
the
on
the
cncf
landscape.
A
D
It
I'm
pretty
sure
about
it,
and
the
only
thing
is
the
the
logistics
of
it,
because
it's
so
bulky
that
you
have
to
first
of
all
make
sure
that
it's
it's
not
broken
on
the
way
to
the
conference
or
the
exhibition.
A
D
Thing
is
you
need
to
have
decent
amount
of
like
Industrial
Supply
Alliance,
like
24
volts,
depending
on
how
you
design
it
so
yeah?
It
is
nice
I
have
seen
at
least
I
have
heard
internally
that
at
least
Emerson
uses
it
for
its
exhibitions,
where
we
record
these
kind
of
things
and
present
it
as
a
let's
call
it
a
proof
of
concept
or
an
actual
working
logic
to
to
the
products
that
we
set.
D
So
it
is
quite
common,
but
I
haven't
seen
people
rushing
it
to
different
exhibitions,
because
it's
it's
also
very
sensitive
in
certain
cases,
if
you
don't
handle
it
well,.
A
That'd
be
an
actual
interesting,
like
Bake
Off
contest.
If
the
cncf
was
to
host
some
event,
challenging
projects
and
products
to
go
in
there
and
say:
okay,
here's
the
thing
built
out
with
no
control
system
or
monitoring.
You've
got
one
hour
to
prove
what
you
can
do
it
be
a
people
a
little
too
lacking
confidence
to
go.
Take
that
on
in
a
public
setting,
but
I
would
love
to
watch
it
in
an
audience.
A
D
Yeah
I
think
the
whole.
The
bigger
challenge
would
be
the
operations
technology
part
see.
The
thing
is:
if
you
want
to
program
a
PLC,
it's
not
as
easy
as
using
C
or
something
right.
You
have
to
use
languages
like
ladder,
logic
or
structured
text
which
in
themselves
are
very
like
sometimes
very
hard
to
comprehend.
So
it's
more
or
less
like
designing
a
finite
State
machine
for
how
your
assembly
line
should
work,
and
it's
not
as
easy.
D
Let
me
be
honest:
it
took
us
quite
a
while
just
to
get
things
on
the
PLC
side
up
and
running
beyond
that.
I
think
the
I.T
side
would
be
interesting,
because
if
you
have
certain
servers
certain
web
services
available,
you
can
you
you
can
leverage
a
lot
of
things
from
from
these
kind
of
infrastructure.
A
Well,
thanks
for
bringing
that
up
that
looks
really
cool
now,
I
want
to
cut
I
want
to
come
across
one
of
those.
You
know
that
I
can
walk
up
to
and
touch.
D
Yeah
so
I
think
on
a
like
in
a
nutshell,
so.
B
D
D
Of
cool
things,
I
think
they're
also
smaller
smaller.
Let's
call
it
quote
and
quote
toys
out
there,
but
this
it
is
still
a
significant
amount
of
money,
sometimes
in
comparison
to
having
just
LEGO
Mindstorms,
but
they
do
decent
stuff
and
they
actually
try
to
replicate
as
much
as
possible
to
an
industry.
So
you
have
components
which
actually
will
work
on
24
volts,
which
is
a
standard
voltage
input
for
a
lot
of
Industries
this
and
that-
and
so
it's
quite
quite
a
nice
thing
to
have.
F
A
D
So
yeah,
so
our
plan
is
anyways.
We
have
got
some
things
up
and
running
with
our
industrial
computers,
the
ones
that
Emerson
actually
takes
I
do
work
on
the
industrial
computer,
part
of
it
where
we
do
support
a
generic
Linux
operating
system
on
on
our
ipcs
or
the
industrial
computers.
So
yeah,
it's
it's
quite
interesting.
However,
we've
never
like
we've
still
not
touched
the
the
cncf
part
of
the
whole
I.T
stuff,
so
I'm
quite
interested
in
naturally
getting
my
hands
dirty
with
some
some
things.
D
That
might
be
interesting,
at
least
for
the
community,
as
well
as
I,
don't
know
for
for
for
DIY
people
out
there,
because
there
are
lots
of
Solutions
out
there.
That
always
insist
on
being.
You
can
have
a
a
single
single
node
cluster
on
a
device
and
it
should
work,
but
when
it
comes
to
actually
doing
it
in
in
practice,
it
might
be
quite
difficult
because
nobody
lets
no
big
company,
would
let
let
you
touch
their
their
OT
stuff
just
like
that
to
have
some
fun
around
with
it
right.
So
right.
A
It's
kind
of
interesting
the
comment
you
made
too
about
The
edgex
Foundry,
trying
to
replace
out
the
cncf
stuff
or
I.
Think
it
was
you
or
was
it
Philip
who
said
that?
But
you
know
I've
I've
heard
indications
that
there
are
sometimes
concerns
with
differing
psych
update
Cycles
on
some
of
these
tools,
where
even
within
the
kubernetes
project,
they
used
to
do
kubernetes
version
releases,
orderly,
they've,
kind
of
cut
it
back
to
three
times
per
year,
but
there
are
some
application
areas
where
that
is
more
often
than
they'd
like
that.
A
They
really
don't
want
to
have
to
deal
with
updates
anymore
often
than
once
a
year,
and
you
can
say
well,
you
could
just
skip
some
of
these,
but
with
kubernetes
they
do
have
a
deprecation
policy.
Where
you
know
beyond
a
certain
number
of
releases,
they
no
longer
continue
to
do
maintenance
releases.
So
you
know
there
are
a
lot
of
elements
here
and
certainly
in
the
manufacturing
sector
of
IO
of
iot
edge
I
think
people
are
used
to
much
slower,
update
Cycles
than
our
typical
in
commercial
I.T.
You
know.
A
Well,
thanks
for
bringing
that
up
and
if
you
Advance
this
at
Emerson
and
want
to
do
a
presentation
or
share
some
pictures
or
reports
of
what
the
experience
was
like.
Please
keep
us
informed.
D
Yeah
absolutely
so
I
think
over
the
course
of
the
coming
meetings.
I've
tried
to
keep
an
update
I'll,
give
you
give
the
team
an
update
on
on
how's
it
going
on
what
are
some
things
that
we
are
actually
doing
so
at
this
point
of
time.
What
I
could
say
is
we
do
have
a
so-called
development
Network
that
is
part
of
this
industrial
Factory
that
we
have
right.
So
we
will
be
able
to
connect
a
lot
of
industrial
PCS
which
are
very
high
performing.
They
are
from
Emerson
themselves
and
we'll
be
I'll.
D
Try
to
see
if
I
can
actually
come
up
with
some
some
decent
scenarios,
where
I
could
have
two
to
three
industrial
computers
with
I,
don't
know
maybe
k3s
or
some
something
similar
and
in
most
cases,
I'll
see
if
it
can
actually
come
up
with
some
decent
way
of
of
using
the
the
I.T
part
of
this
of
the
tools,
as
well
as
the
getting
some
information
out
of
this
Factory.
So
it's
yeah.
A
I'd
love
to
see
that
and
also
hear
honest
opinions.
You
know
I
think
in
my
opinion,
that
a
lot
of
these
you
know
industrial
trade
shows
as
well
as
these
things
operated
by
open
source
foundations.
A
You
have
too
many
things
that
are
essentially
vendors,
giving
pitch
talks
on
whatever
product
they're
trying
to
sell,
even
if
they
couch
it
in
terms
of
Open
Source,
and
they
tend
to
be
mostly
the
happy
talk
stuff
and
the
interesting
thing
was
over
at
this
Eclipse
webinar.
It
ended
up.
You
know
that
they
didn't
give
the
critical
remarks
until
the
actual
vendor
platform
was
end
of
life
didn't
shut
down,
but
you
know
often
there
are
people
who
encounter
problems
trying
to
employ
technology
and
I.
A
Think
I'd
like
to
think
at
least
in
the
open
source
world
that
if
there
are
problems
applying
something
like
kubernetes
to
an
edge
solution,
that
I
would
welcome
talks
recounting
the
pain
points
rather
than
just
giving
the
pitch
Tech
of
oh.
This
was
all
hundred
percent
happy
and
it
ended
up
with
you
know,
balloons
and
butterflies
at
the
beach
in
the
end
and
there's
no
mention
of
anything
ever
having
gone
wrong.
But
you
know
it
had
to
happen
and.
B
F
F
They
had
a
you
know.
They
just
said
that
they
finished
their
rainy
season
right,
so
they
were
all
talking
about.
You
know
how
the
day
before
I
arrived,
that
there
was
rain
and
I.
So
I
gave
a
talk
there
on
that
Argo
CD
scalability
bit
that
we've
been
talking
about
recently
and
it
was
well
received
I
think,
but
the
the
gist
of
it
was.
You
know,
like
you
were
just
mentioning
about.
You
know
vendor
pitches
and
so
forth.
F
They
had
a
bunch
of
vendors
lined
up
out
there
that,
were
you
know
when
you
spoke
to
them.
You
couldn't
really
get
a
good.
You
couldn't
really
pin
him
down
on
whether
they
were
open
source
or
whether
they
were
Enterprise
licensed,
and
so
it
was
a
little
you
know.
Most
of
my
conversations
towards
the
second
day
started
to
become
more
about
hey.
You
know,
could
you
tell
me
more
about
your
product
and
oh
by
the
way,
is
it
open,
source
or
licensed?
First
before
we
get,
you
know
start
did
so
it
was
more.
F
F
Hey
Steve
can
I
share
a
picture.
Real
quick
I
think
it
might
make
everybody.
F
Something
in
Vancouver
that
I
thought
was
interesting
and
I
pointed
it
out.
Actually,
in
my
in
my
talk,
let's
see
Paula
stopped
recording.
How
do
I
do
this.
F
See
it
now
all
right
I'm
just
trying
to
it's
figuring
out
who?
What
what
to
share
now
sorry
about
this.
A
And
you
can
choose
either
a
whole
monitor,
screen
or
just
an
app
yeah.
Let
me
see
if
I
can
get
this.
Do
you
see
the
picture?
No,
it
says
okay,
now
I
do
okay,
that
so
that's
the
convention
center.
So
that's
the
convention
center.
Can
you
see
the
blue.
A
F
F
Super
cool
it
reminded
me,
like
the
weather,
app
on
your
phone
but
yeah
anyway.
The
the
conference
was
good,
but
I
had
to
make
be
careful
about
making
that
distinction
between
open
source
and
licensed
products.
F
Said
OSS
was
just
starting
after.
C
No,
but
my
company
sponsored
OSS
I
was
getting
septum
surgery.
Actually
so
I
was
recovering,
but
I
wanted
to
go.
It
was
only
three
hour
drive,
so
I
should
have,
but
the
timing
was
off
clearly.
A
And
from
where
you
arcade
I
would
really
recommend
my
favorite
way
to
get
Seattle
to
Vancouver
is
actually
that
train.
It
goes
it's
it's
not
quite
bullet
category,
but
it's.
It
is
one
of
the
faster
trains
and
it
goes
right
up
along
the
coast
and
they
have
a
dining
car.
A
So
you
can
eat
breakfast
on
the
trip
and
it's
a
lot
more
laid
back
than
taking
the
airplane,
and
my
experience
is
that
the
Customs
entry
at
the
train
station
is
a
lot
more
laid
back
than
at
the
airport
too,
so
unique
to
Seattle,
where
you
can't
do
that
from
the
rest
of
the
world,
but
that
that
is
how
I
would
choose
to
go
there
from
Seattle
yeah.
C
A
And
in
terms
since
you
brought
up
open
source
Summit,
which
does
have
typically
tracks
on
iot
edge,
I've
been
to
the
North
America,
which
is
often
I'd,
almost
say,
half
the
time
it's
in
Vancouver
I've
been
to
the
Europe
one
and
the
one
in
Japan
as
well.
The
one
in
Japan
tends
to
have
a
stronger
focus
on
iot
edge,
but
it's
very
much
oriented
towards
Automotive
as
well.
A
I'd,
say
so
that
that
is
the
big
one
in
automotive
Linux
per
year
in
terms
of
vehicle,
Electronics
I
think
maybe
a
lot
of
the
companies
that
are
sort
of
the
Leading
Edge
there
are
in
that
part
of
the
world
which
might
explain
part
of
that
focus
and
the
North
America
one
in
my
experience
has
not
been
that
strong
on
iot
edge
presentations,
or
at
least
the
ones
that
I
found
I
enjoyed
a
lot.
Afterward
and
Europe
is
kind
of
mid-range,
but
they
they
all
have
those
tracks.
A
F
Yeah
we
just
speaking
of
of
conferences.
Did
anybody
see
that
kukan
Shanghai
was
announced
for
September.
A
F
A
This
is
pretty
much
online,
but
I
discovered
this
group,
that's
called
tinyml,
there's
a
tiny
ml
foundation
and
the
tiny
there
is
that
this
is
essentially
Edge,
oriented,
Ai
and
machine
learning,
and
it's
I
wasn't
aware
of
it,
but
it
appears
to
have
been
in
existence
a
year
or
more,
and
they
have
been
very
active
just
because
the
world
had
their
eyes
opened
up,
I
think
by
chat
GPT
and
everybody
who
wants
to
get
on
that
AIML
bandwagon
recently,
but
they
have
had
webinars
twice
a
week
and
they
have
been
pretty
interesting.
A
There's
a
little
bit
of
vendor
pitch
going
on
there,
but
I
think
that
this
whole
Space
of
AIML
for
Edge
is
perhaps
going
to
be
even
bigger
than
the
cloud-hosted
one.
That
was
kind
of
the
focus
of
the
talk
I
gave
at
kubecon
Europe
that
you
know
the.
Since
most
of
the
world's
data
is
originated
and
generated
at
Edge.
A
There
is
a
fairly
compelling
argument
that
you'd
want
to
give
the
AI
first
bite
at
that
data
close
to
the
origination
point
if
for
no
other
reason
than
to
reduce
it
down,
but
you've
also
got
the
concerns
of
eliminating
latency
points
of
failure.
A
lot
of
times
that
data
is
massively
bulky
in
size
to
where
the
AI
would
be
really
good.
At
reducing
huge
frame,
huge
amounts
of
data
I
mean
one
example
is
obviously
video
with
you
know:
Fork
4K
video
coming
in
at
60
frames
a
second.
A
You
don't
want
to
pipe
that
up
to
the
cloud
and
I.
Don't
think
anybody
in
their
right
mind
would
but
there's
plenty
of
other
data
types,
anything
from
Industrial,
Automation
sensors
to
audio
devices,
for
you
know,
preemptive
maintenance,
predictive
maintenance,
applications
that
really
call
for
AI
workloads
running
at
Edge
and
anyway,
this
this
tiny
ml
Foundation
seems
to
have
focused
themselves
on
that
particular
opportunity.
A
Most
of
the
talks
are
being
put
on
by
people
with
chip
vendor
tie-ins
that
I've
caught,
because
essentially
everybody
doing
chips
for
these
Edge
devices
is
adding
on
ML
accelerators
yeah.
You
know
so
this
is
a
completely
different
world
from
things
like
tpus
and
Nvidia
boards
that
consume
three
kilowatts.
This
is
this
is
little
cheap
ml
accelerators.
A
I
think
it's
fairly
typical
that
your
initial
training
set
is
just
your
first
Approach
at
this,
and
you
can
iteratively
tune
these
things
to
keep
making
them
better.
But
if
you
do
that,
and
you've
got
ten
thousand
a
hundred
thousand
Edge
Leaf
nodes
that
you
have
to
distribute
new
training
sets
to
that's
a
that
look.
That
problem
looks
a
lot
like
Distributing,
you
know
Docker
containers
or
webassembly
containers
and
really
maybe
the
ultimate
of
architectural
approach
to
this
is
that
you
take
on
an
edge
native
orchestration
system
that
views
all
of
this
stuff.
A
All
of
these
payloads
is
sort
of
blobs
that
could
be
transported
by
a
management
and
control
plane
to
police
yeah.
What
versions
are
down
at
what
what
Edge
locations?
What's
the
desired
State?
What's
the
current
state
and
just
iteratively
make
it
happen,
and
it's
maybe
a
much
bigger
dream
than
what's
currently
out
there
that,
but
you
know
anybody
embarking
on
a
new
Greenfield
mission
to
take
on
the
needs
of
edge
native.
A
C
I'm
actually
curious
on
that
front
for
down
with
a
drogio
I'm
curious,
because
it's
completely
rust
based
firmware
solutions
for
iot
devices.
Is
there
any
thought
of
going
down
the
path
of
using
webassembly
since
rust?
Already
has
such
a
good
tool
chain
for
webassembly
I
could
see.
C
Joe
gaio
already
has
or
drove
cloud
in
general
already
has
the
infrastructure
for
managing
these
iot
devices
in
a
very
Green
Field
mindset
of
like
the
firmware
level
at
scale
with
the
centralized
control
plane
like
are
y'all
looking
beyond
that
to
maybe
using
webassembly
to
push
updates
to
sandbox
the
firmware
anything
along
those
lines.
E
E
The
thing
that
that
drug-wise,
firmer
is
targeting
is
might
be
too
small
for
for
even
webassembly
and
and
even
the
the
the
native
rust
firmers
are,
are
usually
it's
hard
to
to
to
compress
everything
down
like
if
you
want
https
and
and
things
like
that,
so
it
it
it's
it
targets
at
the
moment,
but
very,
very
small,
small
devices
and-
and
the
argument
as
well
is-
is
that
Ross
is
basically
so
good
in
in
Cross
compiling
that
on
on
that
level.
E
So
so
so
direct
firmware
on
on
this
very,
very
tiny
devices,
It's
A
Hard
Sell
to
to
to
do
anything
then
then
just
compile
to
to
to
to
Native
bits.
We
still
didn't
Target
anything
larger.
So
so
something
where,
where
you
know
replacing
oci,
would
would
make
sense.
So
for
me,
that's
the
that's
the
layer
where,
where
assembly
would
would
make
bigger
sense,
but
that's
just
one
one
point
of
view
right.
C
Now
that's
interesting
saying
how
the
portability
or
the
cross-platform
benefits
of
webassembly
isn't
a
headache
which
I'm
kind
of
surprised
by
that,
like
the
variety
of
tiny
iot
devices
that
y'all
are
using,
it's
so
easy
to
cross,
compile
rust
for
all
of
them
is
a
surprising
thing.
I
wasn't
expecting
that
to
be
a
fact,
but
if
that
is
the,
if
that's
not
a
headache,
I
could
see
why
webassembly
wouldn't
be
really
a
game.
I.
E
E
Yeah,
so
again,
my
knowledge
here
is
limited,
so,
but
most
of
them
are
are
targeting
the
the
arm
architecture
and
in
just
a
couple
of
different
chips
there
and
then
there's
a
there's,
a
loads
of
of
periphery.
Right
that
that's
I,
think
a
bigger
challenge
to
support,
but
but
I
think
most
of
the
people
playing
in
in
that
embedded,
Trust
Field,
are
are
basically
targeting.
A
E
A
Okay,
maybe
I
didn't
make
it
clear,
but
back
in
our
April
26
meeting
when
I
gave
the
summary
of
the
cool
stuff,
I
saw
at
kubecon,
Europe
I
brought
up
that
medikura
Booth,
which
is
an
attempt
to
put
they
described
it
as
a
and
there's
a
picture
of
the
back
wall
of
their
booth
there
in
the
agenda
meeting
notes
on
426,
but
they
self-describe
it
as
an
attempt
to
virtualize
ml,
accelerators
or
Edge
devices,
but
I
just
noticed
when
I
read
that
they
didn't
point
it
out
in
that
description,
but
it
is
webassembly
based,
so
they
took
the
tactic
of
rather
than
writing.
A
You
know
an
API
that
you
write
to
in
technology
of
your
choice
that
this
uses
webassembly
plugins
to
abstract
out
Edge
ml
inference
engine
accelerators,
so
their
intent
is
that
you
would
as
a
vendor,
right
web
assembly
code.
That
would
be
make
this
generically
useful
as
a
tool
for
ML
on
an
edge
device
and
it
the
the
reasons
they
discussed
for
y
ml
or
why
webassembly
was
cool
is
that
it
even
becomes
platform,
agnostic,
CPU,
agnostic,
language,
agnostic
and
Os
agnostic.
So
you
may
be.
A
You
don't
need
kubernetes,
you
don't
need
Linux,
but
you
can
use
them
if
you
want
to,
and
this
abstraction
layer
should,
in
theory
still
be
good
and
it
sounded
like
a
cool
concept
that
I
hadn't
heard
applied
before
you
know,
but
webassembly
is
new,
so
maybe
people
you
know
it
will
be
interesting
to
see
how
an
approach
like
this
plays
out,
because
it's
a
different
world,
it's
a
different
means
of
abstracting,
something
that
maybe
makes
it
really
versatile
and
useful.
Regardless
of
what
your
other
technology
choices
are.
I.
C
Think
the
model
of
closed
Source
plus
webassembly
plug-in
is
becoming
pretty
big.
Dapper
has
that
now,
which
is
not
closed,
Source
necessarily,
but
there's
also
suborbital,
which
is
a
big
company
in
the
web
assembly.
Space
is
all
around
plugins
and
also
database
Solutions
with
webassembly
plugins
is
becoming
a
big
thing
so
that
you
can
write
your
own
queries
that
run
in
databases
themselves
to
like
specialize
what's
happening
so
the
whole
plug-in
domain
of
webassembly
is
certainly
one.
C
A
This
company
told
me
that
they
plan
to
open
source
this.
It
isn't
out
there
yet,
but
in
a
June
time
frame.
So
maybe
we
can
even
get
them
to
come
on
and
present
here,
but
it
looks
interesting
and
given
your
interest
in
webassembly,
maybe
you'd
be
particularly
interested
in
it,
and
just
this
approach
of
using
webassembly
as
an
abstraction
layer
sounds
really
cool.
C
Yeah
and
there's
actually
a
web
assembly
working
group-
that's
joining
runtime,
so
that
would
be
a
good
opportunity.
They're
just
kicking
off.
C
So
that
would
maybe
be
a
good
opportunity
to
do
a
joint
meeting
kind
of
to
bring
people
into
that
their
space
and
also
bring
you
know,
go
ahead
and
Bridge
from
the
start
of
that
working
group
as
well,
so
because
it
seems
It's,
Edge
and
webassembly.
So
it
might
be
a
good
dual
session.
A
What's
your
perspective
on
if
the
world
were
to
go
with
a
lot
of
these
web
assembly
components,
do
you
think
is
packaging
these
in
oci
containers,
just
a
stop
Gap
or
is
this
you
think
a
long-term
place,
the
world's
likely
to
end
up
in
you
don't
have
to
speak
for
your
employer
either.
You.
G
C
C
Word
container
is
incredibly
confusing
in
the
webassembly
space,
but
I.
Think
oci
is
fine,
I
think
creating
uniformity
around
how
you
define
or
tag
that
image
to
say
this
is
webassembly
and
to
go
further
to
say
this
is
webassembly
that
uses
the
component
model
versus
modules
versus
is
a
specific
app
type.
Like
my
company
spin
apps
versus
a
different
one.
Maybe
a
was
in
Cloud
app.
So
how
do
we
create
uniformity
around
that?
And
does
that
matter?
C
But
the
by
code
Alliance,
which
runs
a
lot
of
the
server-side
webassembly
initiatives,
is
creating
Warg,
which
is
going
to
be
a
registry
for
all
webassembly
components.
So
the
idea
would
be
a
world
where
you
could
import
a
webassembly
component
in
your
webassembly
application
so
that
you
could
not
have
to
worry
about
how
to
create
a
logger
for
your
webassembly
application
or
how
to
handle
in
the
future.
Maybe
an
HTTP
server
and
it
said
import
that
from
one
centralized
award
repository
and
that
can
have
various
backings.
C
So
the
way
that
they've
promoted
work
is
that
org
is
this
one
entry
Central
entity
and
it
could
be
oci
or
it
could
be
bindle,
which
is
a
package
registry
specifically
made
for
webassembly
or
it
could
be
any
of
these
other
ones,
and
the
idea
is
that
it
wouldn't
matter
and
work
would
abstract
it
away.
So
I
think
they're
already
thinking
about
that
question
and
trying
to
build
a
cohesive
solution
around
whatever
way
people
choose
to
package,
it.
A
Yeah,
my
probably
uninformed
attitude
so
far,
is
that
unless
there's
a
lot
of
overhead
in
the
oci
format
that
would
devalue
some
of
the
small
size
of
web
assemblies,
the
world
might
be
better
off
to
stick
with
it.
Just
because
of
all
the
tool
chain,
resources
that
are
there
are
already
there
like
the
ability
to
sign
these
and
to
do
security
scanning.
Potentially,
it
calls
for
metadata
being
wrapped
around
it,
so
that
you
know
what
you're
dealing
with,
and
you
can
maybe
unleash
specialized
tools
that
you
know.
A
A
If,
what's
in
that,
oci
container
is
actually
you
know
a
training
set
for
an
ml
inference,
engine
that
might
look
quite
different
from
that
without
a
layered
structure
like
you'd
have
in
Docker,
so
you
want
enough
Clues
as
to
what
is
in
this
bundle
to
intelligently
deal
with
it
with
the
collateral
tools,
but
the
ability
to
go,
publish
these
at
a
central
location.
Have
them
get
replicated
around
the
world,
maybe
even
stand
them
up
in
dedicated
Registries
for
Edge
locations
that
have
to
deal
with
intermittent
connectivity.
There's
no
point
in
Reinventing
wheels
for
that.
G
I
post
a
question
in
a
chat,
I'm
just
curious:
is
there
any
because
I'm
actually
also
on
YouTube
kubernetes
but
I'm?
Finally,
learning
about
API
Machinery
is
very
helpful.
Just
for
webassembly.
Is
there
any
similar
like
fundamental
or
you
can
say,
internal
information
YouTube
videos
about
how
yeah.
C
C
So
we
actually
had
a
discussion
about
this
at
kubecon
Europe,
where,
like
it
was
an
impromptu
discussion
with
a
lot
of
the
people
in
the
server-side
space
and
webassembly.
There
isn't
great
learning
tools
out
there,
whether
it's
coming
from
the
perspective
of
trying
to
build
up
the
underlying
technology
or
just
use
the
technology
itself.
C
The
fundamentals
of
it
I
can
send
you
what's
out
there,
but
there
isn't
one
good
source.
Quite
yet
I
think
people
are
trying
because
it's
moving
so
fast
at
least
the
Wazi
or
the
systems
interface
they're
running
it
on
the
server,
isn't
fully
stabilized
yet
to
a
point
where
there
there's
one
book
on
it,
but
we're
getting
to
a
point,
probably
like
preview,
two
which
launches
the
component
model
is
supposed
to
come.
A
I'll
give
you
my
perspective.
Having
tried
to
learn
it
is
that
just
the
sell
by
date
on
articles
written
about
webassembly
should
be
six
months.
So
when
you
go
out
there
and
just
do
a
Google
search
looking
for
things,
you'll
find
plenty
of
blog
posts
and
anything
that's
two
years
old,
I'd
be
really
skeptical
of
because
when
I've
tried
things
like
that,
they
blow
up
on
me
because
things
change
to
the
point
where
they're
no
longer
valid
and
in
terms
of
YouTube
videos.
A
A
G
Web
assembly
have
a
huge
amount
of
information
for
me:
I'm
a
I'm,
a
database
consultant,
so
I,
don't
really
need
to.
You
know,
learn
to
develop
on
it,
so
I'm
interested
in
the
in
the
underneath
just
like
API
Machinery
I,
don't
need
to
actually
implement
the
kubernetes.
You.
G
Lot
but
I
it'd
be
interesting,
like
just
knowing
the
you
know,
the
the
components
of
API
machinery
and
the
difference
between
the
examination,
API
yeah.
That's
very
interesting
to
me.
A
Yes,
I
can
tell
you
there
I
can't
give
you
the
name
and
the
link
offhand,
but
if
you're
a
database
person
at
rejects
in
Los
Angeles
about
two
years
ago,
there
was
a
presentation
specifically
on
webassembly,
combined
with
database,
where
they
were
injecting
web
assemblies
into
a
a
transactional
database
to
use
as
stored
procedures.
You
know
so
the
storage
procedures
could
operate
server-side
client
side
and
they
would
just
extend
the
database
to
go
shovel
webassembly
in
there
as
an
object
within
the
database.
A
That
could
be
done
during
a
transaction
as
it
something
equivalent
to
the
old
stored
procedures,
and
it
was
an
interesting
talk.
G
C
A
And
Andy
in
the
slack
channel,
even
though
he
said
he
couldn't
be
here
to
Moya-
gave
a
reference
to
that
doc
on
edge
native
application.
Design,
behaviors
and
I
had
hoped
to
get
into
that
here.
But
it
looks
like
we're
almost
out
of
time,
but
I
went
and
took
a
look
at
that
Dock
and
it
looks
like
that.
There's
been
a
lot
of
good
material
put
in
there
and
I.
A
Don't
know
what
your
assessment
is
on
the
state
of
the
dock,
whether
you
know
percent,
complete
or
whatever,
but
that
looked
like
some
really
valuable
information
in
there.
It's
been.
F
Going
on
I
think
the
team
is
making
good
strides
there.
We
intend
a
meeting
again
very
soon
and
the
the
document
for
the
most
part
you
know
was
put
together
in
two
sections.
The
top
section
was
mostly
the
initial
ramblings
and
then
we've
been
able
to
fortify
those
that
information
and
spread
the
word
and
get
other
people's
input
on
the
bottom.
F
A
So
we're
getting
to
the
point
where
for
kubecon
North
America
in
Chicago,
that
would
have
be
an
event
taking
place
in
November.
We
probably
would
have
an
opportunity
to
do
a
maintain
or
track
session
to
talk
about,
what's
been
going
on
in
the
group
I'm
thinking.
If
we
don't
have
a
generic
thing
we
put
on
the
you
know,
maybe
we
could
do
something
around
AIML,
just
because
it's
so
hot
and
a
lot
of
people
that
was
the
king
of
the
hallway
track
at
kubecon.
A
Could
probably
do
a
summary
of
that
as
as
well
I
think
that
isn't
quite
ready
that
the
next
step
for
that
is
to
classify
those
into
categories,
because
right
now
dealing
with
150
of
these?
It's
just
like
too
much.
It
needs
to
be
divided,
and
but
we
could
probably
have
that
ready
by
the
November
time
frame
too.
That
would
be
cool.
A
Okay,
it's
two
minutes.
After
the
hour,
we
started
a
little
like
905,
but
any
last
remarks.
Otherwise
I
think
we'll
close
this
meeting
and.
F
I
did
have
one
we
just
went
through
a
Rebrand
kcp
Edge
is
now
known
as
coopstellar
dot,
IO
and
We've
made
a
bunch
of
noise
about
it
and
we're
releasing
our
quick
start
guide
tomorrow,
and
this
afternoon
we
have
we're.
We
invited
Victor
farchak
from
get
Ops
Fame
to
come
visit
our
Channel
and
we're
going
to
do
a
fireside
chat
with
him.
So
look
for
that
posting
at
so.
If
you're
looking
on
YouTube,
the
channel
is
at
kubesteller.
F
It's
not
gonna,
be
we're
not
doing
it
live
we're
gonna
record
so,
but
I
think
in
the
future.
That
would
be
wise
to
do
but
yeah
but
check
it
out.
It'll
be
recorded
and
dropped
in
that
channel
fairly.
Shortly
this
afternoon,
okay,.
A
And
feel
free
to
drop
those
links
either
in
the
agenda
notes
or
the
slack
Channel
or
both
because
yeah
I'll
I'll,
see
if
I
can
catch
that
when
it
comes
out.
That
sounds
cool
great.