►
From YouTube: Kubernetes WG IoT Edge 20230222
Description
Frebruary 22, 2023 meeting of the CNCF IoT Edge Working Group. Discussion of edge related open source projects, recent edge books and articles, The KCP Edge project what it is, and recent milestones
A
Hi
everybody
hey
folks,
Dion.
Maybe
you
can
manage
the
meeting
if
you've
got
the
host
credits
today,
because
I'm
on
my
phone
right
now,
it'll
be
a
few
minutes
before
I
can
get
to
my
laptop,
so
I
think
I'm
operating
without
the
host
setting
right
now.
B
Let
me
just
re-log
in
with
the
working
group
Financial,
so
I
can
record
a
meeting.
I'll
be
back.
You
know
all.
A
Yeah
I
think
record
Auto
triggers,
but
I,
usually
yeah
put
it
off
and
back
on
to
clip
off
the
stuff
before
the
warning
announcement
and
things.
So
maybe
you
want
to
flip
that
game.
A
Okay,
yeah,
okay,
leave
it
we'll
just
have
to
clip
off
the
junk
at
the
beginning,
yeah.
Unfortunately,
it
starts
recording
as
soon
as
the
first
person
arrives.
So
you
end
up
with
a
bunch
of
minutes
of
dead
time
at
the
beginning,
okay,
I
I,
don't
have
the
agenda,
but
when
I
looked
at
it
earlier
this
morning,
there
was
nothing
on
it
which,
in
our
tradition,
doesn't
mean
that
we
cancel
the
meeting,
but
we
just
open
it
up
for
free
form,
discussion
of
whatever
people
want
to
chat
about
or
came
across
recently.
That
was
interesting.
A
And
I
don't
know
if
he's
on
the
call,
but
Brandon
in
the
slack
mentioned
some
things
that
related
to
the
white
paper.
A
I
think
I
saw
Frederick
on
here
and
just
to
let
them
know
we
did
have
a
little
bit
of
it.
Discussion
about
the
list,
we're
compiling
of
the
open
source
software
that
is
related
to
Edge
native
and
I,
want
to
thank
you
for
the
contributions
on
there.
A
I
think.
At
this
point,
the
number
of
things
coming
in
per
week
has
slowed
down
for
a
while.
We
were
getting
a
lot
of
them
per
week
and
now
it's
a
trickle,
but
it's
time
to
classify
those
things
and
I
made
some
attempt
to
come
up
with
a
list
of
possible
categories
on
one
of
the
tabs
of
that
spreadsheet.
I
think
right
now
the
categories
are
too
big.
It's
like
20
I'd,
rather
see
it
down
to
about
10,
but
it
would
require
coming
up
with
better
categories,
to
combine
things
and
I.
A
Think
over
in
the
eclipse
and
Native
meeting
a
week
ago,
Robert
had
suggested-
and
he
gave
me
a
link
to
a
document
put
out
by
the
eclipse
foundation
for
Edge
that
classified
projects
not
on
one
axis,
but
on
two
kind
of
based
on.
You
know
what
it
runs
on
and
where
it
is,
and
maybe
where,
in
the
sense
of
you
know,
far
Edge
at
the
device
level
low,
combined
with
low
resource
to
running
in
mezzanine
and
public
cloud
and
I.
C
Since
we're
doing
more
of
the
open
discussion
today,
I've
noticed
some
new
faces.
Maybe
we
can
just
go
around
and
do
some
quick
intros
and
that
itself
might
kick
off
some
of
the
topics.
C
I'll
go
ahead
and
just
intro
myself,
then
hi
everyone.
My
name
is
Kate
I
currently
work
at
fermion,
which
is
a
company
focused
on
webassembly
in
the
cloud
so
creating
serverless,
microservices
and
webassembly.
C
Previously
I
worked
at
Microsoft,
where
I
focused
on
the
iot
edge
space,
maintaining
a
project
called
awkary,
which
is
all
about
how
kubernetes
interacts
with
small
iot
devices.
So
I
like
any
discussions
around
kind
of
the
tiny
Edge
and
the
webassembly
space.
B
Yeah
yeah,
so
I'm
Dan
I'm
working
for
red
hit,
yeah
I'm,
one
of
the
founders
of
the
group,
so
we've
been
here
for
a
while:
I
worked
in
Eclipse
Society
space
for
for
quite
a
bit
working
on
a
redhead
sponsors,
drug
iot
projects
at
the
moment
and
yeah.
That's
my
whereabouts
write
to
me
is
drugs
or
go
next.
D
Sure,
hey
I'm,
Joel,
Robertson
architect
with
Cisco
Systems,
so
predominantly
in
my
day,
job
and
focused
on
the
infrastructure
or
layer.
Two
three
and
four
of
the
OSI
stack,
probably.
D
Where
the
virtualization
began
sometimes
Beyond
and
have
interest
in
everything,
but
the
the
Wheelhouse
right
now
is
predominantly
in
the
infrastructure.
Karate
is
switching
as
well
as
the
connectivity
to
some
of
these
remote
sites
or
Edge
sites.
F
F
Hello
guys,
my
name
is
Julio
armenta
I
work
for
Dish
Wireless
I
am
part
of
the
technology
team,
I'm
responsible
for
visualization
and
related
to
this
topic,
and
this
Wireless
is
building
a
cloud
native
5G
network
based
on
kubernetes.
B
F
G
I'm
Frederick
with
the
eclipse
foundation,
so
I
managed
there
I
do
iot
and
Nash
Computing
programs,
so
I
had
the
pleasure
to
work
with
Diane
and
our
there
are
a
few
other
names
on
this
call.
I
recognize
anyway.
So
the
eclipse
Foundation
is
a
member
of
the
cncf,
a
member
of
lfedge,
and
so
we're
trying
to
just
well
cross-breed
our
open
source,
ecosystems
and
I'm
really
happy
to
be
here.
I
G
I
Next
so
Andy
Anderson
with
IBM
research
representing
kcp
Edge,
we
are
a
very
small
community,
we're
trying
to
get
a
little
bit
more
momentum
and
I've
been
attending
these
meetings
here
on
and
off
for
a
few
months
now,
so
just
trying
to
stay
in
touch
with
what
the
rest
of
the
community
is
doing,
especially
in
the
care
area.
Iot
Edge.
H
E
I
guess
I
might
as
well
speak
also
I'm
also
with
IBM
research
and
I've.
You
know
looked
into
this
group
a
few
times
before
and
I'm
working
also
on
problems
of
running
kubernetes
clusters
on
the
edge
mainly
focused
on
the
workload
layer
rather
than
the
infrastructure
layer
and
and
currently
pursuing
it
in
the
context
of
kcp
as
I'm
with
the
people
that
Andy
mentioned.
C
So
I
think
that's
pretty
much
everyone
Robin
Wen
I.
Think
if
you
would
like
to
introduce
yourself
there's
still
time
for
that,
but
you're
welcome
to
also
set
it
out.
J
Foreign,
my
name
is
one
I'm,
a
program
manager
in
Azure
I'm,
currently
working
on
building
a
developer
solution
for
in
for
building
the
edge
applications.
I,
don't
have
any
specific
topic
to
discuss
today,
but
I
do
have
some
related
questions.
I
talk
with
Kate
offline,
but
I
will
repost
those
questions
in
the
bigger
Channel.
C
No
worries
so
I
think
there's
obviously
a
lot
of
different
interests
here,
I'm
curious.
Maybe
if,
when
you
mentioned
you
had
some
explicit
questions,
do
you
want
to
share
those
out
with
the
group.
J
Sorry
I
I,
my
name,
my
name
is
I'm
not
at
home
right
now.
I'm
with
my
infant
so
probably
have
I.
My
plan
for
today
is
just
for
listening.
I
probably
will
post
the
questions
in
the
channel.
If
that's
okay,.
A
Okay,
I
guess
I'll
do
an
intro
now
I
finally
got
back
up
on
my
laptop
with
the
camera
functional,
so
I'm,
not
quite
a
founder
of
the
group
like
Diane,
but
I,
think
I've
gone
back
to
pretty
close
to
the
clowning
and
I
work
at
VMware.
As
a
software
engineer,
I'm
got
a
role
at
VMware
that
isn't
a
hundred
percent
agile
aligned,
it's
just
open
source
kubernetes
related.
A
However,
in
my
career,
I
have
worked
in
iot
Edge
industrial
control
for
decades
having
done
a
startup
in
the
field,
so
I
moved
from
there
to
more
corporate
I.T,
but
I
became
convinced
that
kubernetes
is
a
real
opportunity
to
redo
some
of
the
Legacy
implementations
that
are
in
popular
use
for
Edge
native,
particularly
in
industrial
control.
So
I
joined
this
group
and
have
been
frequent
participants
since
then.
B
A
Yeah
you
can
mention
the
title
and
I
did
bring
it
up,
but
maybe
people
missed
it.
My
capsule
summary
was
you
know
we
are
compiling
this
list
of
edge
native
open
source
software.
A
lot
of
them
are
managed
by
the
eclipse
Foundation.
The
book
covers
pretty
much
all
the
ones
that
are
deemed
current
and
excellent
coverage,
and
it's
very
fresh
because
this
thing
just
came
out
it.
A
What
was
it
February,
I
think
early
February,
so
often
the
the
sell
by
date
on
Tech,
Publications
and
books
in
rapidly
moving
Fields
makes
things
you
might
find
out
there
almost
dangerous
because
they
tell
you
things
that
become
wrong,
but
this
one
at
least
at
this
point
in
time,
is
very
good
and
it's
while
it's
still
fresh,
get
it
while
it's
fresh
and
if
there
were
faults,
I'm
sure
they'll
be
fixed
in
version
two,
but
by
Design.
If
you
even
look
at
the
title,
it
I
think
it
has
the
word
Eclipse
there.
A
So
some
of
the
projects
that
are
not
under
Eclipse,
like
they're
under
cncf
and
things
are
lightly
covered,
or
maybe
not
even
mentioned,
but
hey.
You
know
I
I,
think
if
you
tried
to
do
everything
now
that
we
have
this
list
underway,
you'd
appreciate
that
that
would
be
maybe
too
much
to
take
on
that
spreadsheet.
A
We
have
now
has
150
plus
projects,
and
but
if
you
try
to
keep
up
with
the
work
like
that,
I
think
by
the
time
you
got
through
a
third
of
it
trying
to
do
the
rest,
the
first
third
would
be
becoming
obsolete
already.
So
I
think,
maybe
that
that
book
is
as
good
as
it
can
be.
G
And
from
from
a
net
Computing
perspective,
I
I
should
say
that
the
edge
Computing
chapter
in
there
is
probably
one
of
the
most
less
eclipsey
one
in
the
sense
that
Eclipse
project
car
mentions,
but
there's
detailed
coverage
of
project
Eve.
We
mentioned
web
assembly
in
a
from
a
generic
perspective,
so
you
know
things
things.
There
are
a
bit
less
project
oriented.
A
H
In
general,
what
what
is
the
networking
Solutions
in
this
area
for
the
edge
Cloud.
B
C
Victor,
if
you're
looking
for
other
people
in
the
working
group
to
kind
of
focus
in
on
5G
on
the
edge
Brandon,
who
recently
posted
in
the
slack
and
Amar,
are
from
Arna
networks
and
they're
very
much
focused
in
on
that
area
and
in
the
past
I've
shared
information
along
those
lines
with
the
working
group.
C
A
Well,
this
is
the
birds
of
a
feather,
so
let's
just
throw
it
out
there
to
things
that
you
may
have
recently
discovered
that
you
were
found
interesting
and
wanted
to
chat
about
I.
Think
I
threw
one
out
recently
or
did
I
in
the
slack
channel
on
an
article
provocatively
entitled
big
data
is
dead
and
I
found
that
kind
of
a
re.
A
It's
a
long
read
but
I
found
it
really
fascinating,
and
it's
apparently
by
someone
who
at
one
time
was
involved
with
Google's
Big
Data
Business,
and
the
point
was
that
there
was
a
period
of
time
when
the
world
was
just
fascinated
by
this
concept
of
Big
Data
gather
everything
you
can
unleash
analytics
to
go,
pull
over
this
to
discover
things
you
might
not
have
been
aware
of,
and
that
if
the
article
is
to
be
believed,
not
too
many
people
actually
form
these,
maybe
somewhat
less
got
value
out
of
this,
and
it
goes
into
the
costs
associating
with
doing
such
a
thing
and
I.
A
Think,
from
my
perspective,
this
article
kind
of
portrayed
it
as
maybe,
if
you
try
to
do
this
big
data
of
gather
everything,
you've
got
put
it
together
in
a
cloud-hosted
reservoir
that
maybe
you've
got
a
hoarding
problem,
because
there
are
issues
with
paying
for
all
of
this
storage.
Obviously,
you've
got
to
pay
for
any
analytics,
there's
privacy
rules
that
mean
that
your
cost
isn't
just
to
Move
It
from
where
it
came
from
and
put
it
in
this
pool,
but
potentially
because
people
have
a
right
to
be
forgotten
under
certain.
A
A
second
video
cameras
and
then
you
could
potentially
say
that
I'm
going
to
gather
everything,
we've
got
from
every
Edge
location,
move
it
up
and
stored
in
the
giant
database
in
the
cloud.
But
you
have
to
ask
yourself
whether
you
really
want
to
do
that
and
how
this
ties
into
Edge
is
that
if
you
start
looking
skeptically
at
the
idea
of
whether
you
really
want
to
pull
this
I
think
it
leads
to
a
scenario
where
you
host
more
intelligence
and
apps
out
at
the
edge
to
filter
this
clean.
A
A
Gather
up
centrally
only
what
you
need.
A
H
E
A
What
is
meant
by
in
the
chat
by
what
is
meant
sorry.
E
It
was
incomplete,
I'm
trying
to
begin
I
just
want
to
ask
Escalon
for
excellent
for
clarification.
He
asked:
can
our
Edge
native
principles
be
applied
to
apps
running
on
medium
and
heavy
Edge
as
well?
I
was
not
sure
whether
he
met
I'm.
Sorry
I,
you
know,
haven't
been
paying
close
attention
here
by
by
just
medium
and
heavy
edge
mean
relatively
high
capacity
nodes
at
the
edge,
or
is
he
referring
to
intermediate
vertices
in
some
hierarchy?.
J
Yeah,
what
I
mean
by
I
can't
give
you
an
example,
for
instance
HCI
on
from
Azure.
We
consider
it
as
a
heavy
Edge
solution.
So
that's
basically
that's
basically
from
our
lessons
from
customers,
some
of
our
customers.
They
have
like
a
chain
stores
or
like
running
the
chain,
restaurants,
for
each
in
each
of
their
store
or
restaurant.
They
will
have
a
better
mixed
group
of
infrastructure.
J
They
will
use
the
iot
devices
like
sensors
or
on
cameras
to
crack
the
data,
but
due
to
the
data
privacy
concern
or
some
latency
concern,
they
were
also
running
some
heavy
Edge
or
median
Edge
infrastructure
inside
the
store
to
processes.
The
data
collected
from
the
sensors,
but
so
my
question
is
because
we
have
this
white
papers
to
guide
the
users,
the
developers
how
to
write
the
what's
the
best
practice
for
writing
the
edge
applications,
I'm
wondering
for
those
applications
which
are
not
specifically
running
on
the
light
edge
devices
like
iot
devices.
J
That's
can
we
also
apply
the
same
principles
to
those
workloads
as
well.
Does
that
answer
your
question.
A
Yeah
I
think
a
common.
If,
if
I
saw
the
words
light
edge
heavy
it
to
me,
it
would
imply
the
amount
of
compute
resource
and
often
what
goes
along
with
that
is
restrictions
on
the
network
Resource
as
well.
Although
those
are
two
orthogonal
issues,
you
know
it
is
possible
to
have
a
very
heavyweight
compute
at
an
edge
location.
A
You
know
even
something
like
rackmount
servers
and
still
be
challenged
with
regard
to
your
internet
access
capability
or
network
connections,
maybe
just
because
it's
in
a
really
rural
location
or
even
in
a
a
mobile
vehicle,
or
something
like
that.
So
I
think
it's
careful.
It's
important
to
keep
those
things
separate.
The
other
observation
I've
had
being
in
this
group
for
five
plus
years
now
is
that
what
constitutes
being
heavy
versus
light
is
not
a
constant.
You
know.
Compute
has
really
I
think.
A
Maybe
the
old
thing
of
technically
the
Moore's
Law
of
more
transistors
per
device
is
dead
but
still
device.
Even
low
power
devices
have
gotten
far
more
capable
today
versus
five
years
ago,
and
this
applies
not
just
to
traditional
compute,
but
to
that
there
are
certain
classes
of
I'd,
almost
call
them
Edge
native
machine
learning
acceleration
devices
so
that
what
you
have
to
be
very
careful
of
Articles
written
five
years
ago
on
the
best
architecture
for
doing
things,
because
things
have
really
changed
five
years
ago,
something
like
Nvidia
or
sensor.
A
Chip
was
something
you'd
only
be
able
to
run
in
a
large
public
Cloud.
You
didn't
have
the
power
or
dollar
economic
budget
to
put
them
at
Ed
at
Edge,
and
you
prop
for
those
particular
device
five-year-old
devices
you
still
don't,
but
there's
new
ones
out
there
that
are
becoming
pretty
fascinating,
I'll.
Look
it
up
and
post.
A
It
drop
the
link
in
the
chat,
but
as
part
of
that
effort
of
coming
up
with
the
edge
native
software
spreadsheet
list,
I
actually
came
across
a
very
interesting
open
source
project
for
doing
machine
learning
on
camera
Feeds
out
at
Edge
to
the
point
where
people
contend,
that
this
is
runnable
on
Raspberry
Pi's,
but
doing
full
full
ml
based
identification
of
objects
and
I'm.
Reading
this
going.
A
E
General
point
is
true:
right.
I
want
to
make
two
points
here.
Power
increases
over
time
generally
and
as
X1
was
saying,
depending
on
the
scenario
you
may
have,
you
know
very
different.
Unless
power
available
right
in
a
retail
sort
of
setting,
you
may
well
run
it
have
a
small
cluster
of
machines.
You
may
have
a
rack
in
each
store
or
location
whatever
that
that's
perfectly
plausible
for
running
applications
at
the
edge.
A
Yeah,
so
the
project
I
was
talking
about,
is
called
frigate
and
what
it
does
to
pull.
This
off
is
I,
wasn't
even
aware
of
this
device
either
I
just
recently
ordered
one
and
the
backlog
is
months
on
these
things,
but
I'm
expecting
to
get
mine
in
about
a
month
so
device
Google
made
called
Coral.
That
is
not
powerful
enough
to
train
a
model,
but
it's
specifically
been
optimized
to
run
one
and
they
sell
it
in
a
USB
key
form
factor
or
in
these
mini
PCI
Express
cards
and
they're
low
power.
A
Yet
they
are
capable
of
keeping
up
with
these
camera
screens
and
maybe
I'll
I
like
I,
say
I've
got
an
order
in
I
think
they
cost
about
thirty
dollars
so
not
tremendously
expensive
either
when
I
get
mine,
I'll
try
it
out,
and
maybe
report
back
to
this
group
at
some
future
date
in
a
month
or
two
after
I
check
it
out
with
my
own
hands.
G
And
and
Steve
kilton
at
edgeworks,
at
bid
using
coral
boards
in
his
AI
enabled
cameras.
So
we
deployed
that
with
models
in
the
field
and
that
that,
certainly,
from
his
perspective
at
least
a
good,
a
good
product.
A
But
this
is
one
of
these
fields
where
I
expect
that
the
semiconductor
people
are
investing
heavily
in
r
d
in
this,
and
it
is
going
to
change
and
I
think
that
the
net
effect
of
all
of
this
research
going
on
the
world
is
that
it,
the
pendulum,
maybe,
is
moving
for
some
of
these
applications
in
data
analysis,
to
move
from
public
Cloud
out
to
Edge
locations
to
both
offer
better
data
data,
security
and
privacy,
in
other
words,
don't
ship
the
data.
A
Every
time
you
ship
the
data
to
another
location,
it's
another
Vector
for
hackers
to
get
at
it,
tamper
with
it,
steal
it
whatever
keeping
it
only
where
it's
needed
probably
would
be
a
great
idea,
and
if
you
can
get
this
Advanced
capability,
this
might
be
a
good
way
to
do
it.
B
But
I
would
I
would
add
to
to
the
questions
so
I
I
think
the
principles
are
basically
a
good
applied
wealth
to
the
medium
heavy
and
heavy
Edge
right,
it's
their
sketchable,
whether
you
can
apply
them
to
the
light
edge,
but
but
this
is
where
they
would
be
applied
right.
So
just
to
answer
the
question.
C
F
C
Everything
called
but
I
think
when
you're
writing
it
we
said
Tiny
Edge
is
where
we're
not
really
talking
about,
because
we're
talking
about
the
compute,
that's
interacting
with
those
devices
if
we
Define
tiny
Edge
as
sensors
MCU
class
devices,
but
I
agree
with
what
Dahan
said.
I
think
it
can
be
heavy
Edge
I
think
you
have
more
freedom
with
the
more
compute
that
you
have.
So
obviously
you
have
to
be
less
resource
aware.
C
If
you
have
more
compute
resources-
or
you
probably
like
Steve
said-
have
to
be
less
aware
of
variable
connectivity,
because
you
have
probably
a
more
reliable
network
with
larger
devices
and
fewer
devices
assuming
larger
equals
fewer.
B
And
it
may
be
the
good
idea
for
for
some
future
revision
of
this
would
be
to
to
maybe
categorize
the
principles
based
on
the
heavens
on
the
edge.
So
so,
what's
most
important,
more
important
on
the
on
the
heavy
side,
what's
more
important
when
you,
when
you're
going
to
the
medium
and
the
night
edges
or
something
like
that,.
J
Yeah
I
would
say
based
on
our
customer
conversations.
Yes,
the
heavy
Edge
does
provide
more
computer
resource,
but
for
the
network
for
the
networking
issue,
almost
I
would
say.
Almost
90
of
our
customers
have
the
networking
issues
because
they
have
thousands
of
stores
running
in
the
different
locations,
and
sometimes
they
will
have
some
like
a
weather,
very
bad
weather
conditions
or
I.
Don't
know
some
other
issues
make
them
accessible
to
the
networks,
so
they
are
looking
for
and
such
a
solutions
to
help
them
to
solve
the
networking
issues.
C
J
I
think
they
probably
are
looking
for
like
the
end-to-end
solution
like
yeah.
How
do
they
recover
immediately
from
the
downtime
issues
or
maybe
I'm
just
making
up?
Maybe
when
they
designing
the
applications,
should
they
consider
how
to
catch
their
datas,
the
only
so
what
they
want
is
when
this
networking
failures
happens,
this
deal,
can't
the
each
store
or
in
each
location
can
still
operate
normally.
J
E
E
C
C
E
Well,
I
would
say
first
off
one
of
the
questions,
I
think
about
platform
and
applications.
So
there's
a
division
of
responsibilities,
question
so
I'm
trying
to
work
on
platforms
that,
to
the
degree
possible
support
applications
through
interfaces
that
the
interface
does
not
demand
online
connectivity.
The
interface
is
state-based,
and
so
the
implementation
can
either
be
online
or
not,
and
the
application
doesn't
need
to
know,
but
it
in
general
I
suppose
there
are
situations
where
the
application
will
need
to
know.
E
E
So
there's
some
different
issues:
right,
pull
versus
push
is
different
from.
Can
you
expect
to
pull
at
any
time?
Can
you
expect
to
pull
at
full
rate
at
any
time
right,
so
I
I
think
there
are
two
different
issues
that
I
think
the
pull
versus
push
is
more
a
matter
of
you
know
stuff
like
typical
Arrangements
of
firewalls,
who
who
can
succeed
to
open
a
TCP
connection,
and
yes,
it's
often
arranged
such
that
it's
easier
for
the
edge
to
open
a
TCP
connection
to
the
sender
than
vice
versa.
C
I
think
you
can
get
into
a
lot
of
details
with
what
does
the
word
pull
versus
push
mean
for
people,
but
and
what
I
was
trying
to
say
there
with
that
terminology
is
I,
think
having
the
edge
fetch
information
from
the
cloud
is
a
better
model
than
having
the
cloud
have
a
cron
fetch
to
the
edge
where
the
cloud
never
knows.
If
the
edge
is
going
to
be
available
or
not,
but
I
know,
pull
and
push
is
a
like
a
kind
of
a
overloaded
term.
So
there's
also
so
many
new
nuances.
B
I
I
guess
it
also
depends
on
the
semantics
of
the
communication.
Is
it
like
control
from
the
cloud
to
the
edge
or
or
sending
data
from
The
Edge
to
the
cloud
and
and
both
of
those
has
have
different
problems
depending
on
the
you
know,
probability,
connectivity
and
and
the
appropriate
Solutions.
I
We've
got
a
Blog
set
and
a
coding
Milestone,
that's
out
there
now
and
I
just
want
to
I'm.
Just
sharing
that
with
you
now
we're
working
on
a
set
of
a
set
of
initiatives
and
elements
for
a
proof
of
concept
that
will
be
showing
off
at
the
end
of
March
and
I
thought.
It
would
be
useful
for
you
to
see
that
information,
as
well
as
we're
starting
to
build
a
medium
blog
post,
Depot,
so
to
speak
for
curated
information
about
kcp,
Edge
and
kcp,
which
I
just
dropped
in
there
as
well.
E
I
Thought
I'd
update
you
there
so
some
of
the
things
that
we're
working
on
for
this
proof
of
concept
we're
working
on
an
egg
scheduler
and
a
placement
translator.
I
We've
got
some
we're
working
on
the
design
of
a
status
summarizer
and
so
that's
to
be
able
to
summarize
status
from
the
lowest
layers
in
in
an
edge
location
in
a
hierarchy,
a
Spoke
as
we
refer
to
it,
and
so
that
we
don't
overwhelm
The
Hub
site
with
millions
of
different
status
messages
when
they
can
be
semi
as
summarized,
and
we're
also
working
on
a
workload
management
and
an
inventory
management
component
that
can
be
used
to
drive
some
of
the
use
cases
that
we
think
are
viable.
I
We're
tracking
use
cases
in
Telco
agriculture,
space
era.
You
know
and
aerodontics
with
the
we're
working
with
the
European
Space
Agency
we've
started
talking
with
Chick-fil-A,
so
in
the
fast
food
area.
We've
also
done
work
there
too.
So
sure
I
can
recap
that
I
don't
have
the
ability
to
share
right
now.
But
if
you
would
allow
me
to
I
can
show
you
my
screen
as
to
what
I'm,
looking
on
and
Mike
feel
free
to
chime.
In
as
you
feel,
necessary,.
A
I
Yeah,
so
if
you
want
to
follow
along,
if
you
just
click
the
link,
it's
in
the
chat
and
scroll
down
to
the
areas
of
exploration.
I
So
we've
got
in
kubernetes
they've,
got
this
notion
of
being
able
to
to
service
a
deployment
or
send
a
deployment
down
so
that
they
can.
It
can
work
on
any
given
node,
but
that
does
not
include
all
nodes.
So
what
we.
E
Well,
yeah,
I,
I,
think
you
what
you
were
saying
was
specific
to
TMC.
If
you
think
about
sending
kubernetes,
doesn't
sin
right
in
kubernetes?
You've
got
one
cluster
that
holds
the
workload
description
and
runs
the
workload.
E
Okay,
so
yeah
as
I
was
trying
as
I
was
alluding
to
before
I'm,
looking
specifically
at
what
we've
defined
as
the
edge
multi-cluster
problem.
E
So
it's
focused
on
workload,
and
this
is
meant
to
include
things
sometimes
called
policy
running
in
kubernetes
clusters
on
the
edge
and
let's
see
I
have
and
I
have
I
really
actually
closed.
That.
D
E
I
E
It
right
so
I
haven't
had
time
Andy
to
carefully
compare
the
list
of
Investigations
on
that
page
with
the
list
of
problems
in
the
blog
post,
I'm
editing,
so
I
was
going
to
review
the
list
of
problems
in
the
blog
posts
that
I'm
working
on
and
I
was
kind
of
context
setting.
E
So,
as
I
said,
we're
looking
at
specifically
the
issues
of
making
a
platform
that
can
help
with
people
organizations
doing
age
Computing
where
they
want
to
run
applications
on
kubernetes
clusters
at
the
edge,
and
so
the
problems
that
that
you
know
that
arise
are
first,
was
we've
been
alluding
to
so
I
I?
Take
it
as
a
given
that
these
applications
are
going
to
run
on
a
coup
cluster.
E
Now
that
can
be
a
single
machine
or
it
can
be
a
you
know,
several
machines,
depending
on
the
actual
scenario
at
hand
right.
There
are
coup
distributions
that
are
engineered
to
run
on
relatively
low
resource
machines,
but
it
doesn't
have
to
be
that
white
now,
Chick-fil-A,
as
you
mentioned,
for
example,
has
the
cluster
of
three
physical
machines
in
each
restaurant.
E
Another
facet
of
the
problem
is
clusterscoped
API
objects,
so,
for
example,
in
the
TMC,
which
is
transparent,
multi-cluster
project
in
kcp
they've
defined
a
different
kind
of
multi-cluster
problem.
They
call
transparent,
multi-cluster
and
in
transparent
multi-cluster.
A
workload
is
something
that
appears
in
one
kubernetes
namespace,
but
we
believe
in
general
workloads
are
not
so
confined.
We
need
to
be
able
to
handle
clusterscoped
API
objects
as
well.
E
Also,
as
I
mentioned
earlier,
there's
a
need
to
support
a
general
hierarchy
right,
as
was
touched
on
earlier.
You
can't
expect
to
transfer
all
the
data
back
to
the
center.
You
don't
always
want
to
it's,
not
always
feasible.
You
have
to
do
some
processing,
it
may
not.
You
know,
and
it
may
not
be
as
simple
as
Center
and
Edge.
You
may
have
intermediate
levels
of
hierarchy.
E
E
Also,
we
touched
on
the
need
for
each
Edge
location
to
be
self-sufficient
and
by
location.
I
really
mean
that
boundary,
this
sort
of
the
lowest
useful
boundary
for
management.
Here
in
the
case,
so
it
may
be
a
restaurant
with
a
cluster
of
three
machines.
It
may
be
what
only
in
one
machine,
whatever
it's
the
lowest
layer
of
the
hierarchy
where
it
it.
You
have
a
that,
it's
a
real
boundary
in
terms
mainly
of
communication.
E
You
know
you
may
have
regulatory
or
data
sovereignty
issues
as
well
as
communication
issues,
but
you
you
want
to
have
each
Edge
location
self-sufficient
and
again
this
is
a
distinction
from
TMC.
Although
again
I
don't
know
how
many
of
you
people
are
familiar
with
TMC
another
one.
Is
that
there's
a
complex
arrangement
of
roles
and
responsibilities
right?
If
you
look
at
a
lot
of
the
kubernetes
world,
it's
focused
on
the
devops
Paradigm
or
a
development
plus
SRE
division,
but
it's
in
Edge.
E
It
gets
much
more
complicated
than
that,
particularly
when
you
have
a
non-trivial
hierarchy
right.
Consider,
for
example,
a
manufacturing
operation.
That's
got
several
plants,
there
may
be
essential
engineering
team.
E
E
There's
a
one-to-many
relationship.
You
can,
unlike
you
over
near
ordinary
kubernetes,
right
between
workload,
description
and
running
copies.
We
want
many
running
copies,
given
one
workload
description,
and
this
has
lots
of
implications
which
I'm
going
to
try
to
summarize
it.
It
does
call
it
a
lot
of
implications
and
then
the
other
is
another.
One
is,
is
again
a
large
one,
not
again
large
number
of
edge
clusters.
E
We
do
want
to
be
able
to
handle
use
cases
with
a
large
number
of
edge
clusters,
and
that
makes
some
simple
and
obvious
solutions
to
the
other
one
other
problems
infeasible.
E
E
So
one
of
the
things
you
need,
when
you
have
multiple
copies,
is
you
generally?
You
need
to
be
able
to
customize
the
workload
for
each
destination,
but
you
wouldn't
want
to
maintain
a
customized
with
k,
a
prescription
for
each
location,
because
that's
that's
just
infusably
tedious.
You
need
pattern-based
descriptions
of
how
to
customize
and
that's
just
one
example.
E
So
anyway,
that's
my
summary.
A
E
Yes,
that's
transparent,
multi-cluster,
that's
something
that
they're
also
doing
in
the
kcp
project.
Okay,.
A
I
Think
kcp
is
attempting
to
be
incorporated
in
the
cncf,
but
I
don't
know
that
they've
made
it
yet
just
I
believe
it
was,
though,
an
activity
that
was
underway
I'll
have
to
check
up
and
get
back
to
you.
F
So
I
have
a
question
on
the
TMC
transparian
multi-cluster
is
the
is
the
main
challenge.
There
is
the
multi-tenancy,
because
you
were
talking
about
namespace
and
not
been
I,
mean
workloads
not
being
confined
to
name
spaces.
So.
E
Well,
I
would
describe
the
the
goal
of
of
TMC
is
really
to
support
a
service
that
gives
you
know
it's
a
multi-tenant
indeed
multi-tenant
service,
where
each
user
has
the
illusion
of
they
have
their
own
curious
cluster.
But
in
fact
the
implementation
is
much
cheaper.
E
What
they've
done?
One
of
the
things
they've
done
is
they've
introduced
a
higher
level
of
virtualization
in
the
API
server,
so
above
namespaces
they
have
something
you
might
call
Super
namespaces.
They
call
them
logical
clusters,
but
they're,
not
full
clusters,
they're,
really
just
an
extra
level
of
division
of
the
objects
within
the
API
server
and
in
with
the
corresponding
in
the
storage,
but
then
with
the
TMC.
E
So
that's
there's,
there's
layers,
that's
kind
of
one
of
the
lowest
layers
in
in
kcp
above
that
they've
built
something
called
transparent,
multi-cluster,
which
is
about
okay.
You
put
the
workload
description
in
the
kcp
server,
but
the
work,
the
that's
not
actually,
where
the
workload
runs.
E
The
the
workload
runs
in
back-end
compute
clusters
that
are
regular
kubernetes
clusters
and
transparent
multi-cluster
is
about
transporting
desired
State
and
reported
State
back
and
forth
between
the
compute,
the
kcp
The
Logical
clusters
and
the
back-end
compute
clusters,
so
that
each
user
who's,
looking
at
just
one
logical
cluster
in
kcp,
has
the
illusion
that
they're
using
a
regular
kubernetes
cluster
of
Their
Own.
E
No,
it's
not
isolated,
it's
not
a
complete
isolation
and
in
fact
this
is
one
of
the
issues
we'd
like
to
get
Upstream
into
regular
kubernetes,
this
additional
level
of
virtualization,
but
if
we
sell
it
as
multi-tenancy
or
complete
isolation,
that's
just
not
going
to
go
because
it's
not
true
but
I.
Think
for
this
use
case
and
others.
The
idea
of
an
extra
level
of
virtualization
is
useful
and
I
I
would
like
to
see
that
make
it
into
plain
kubernetes.
F
Foreign
yeah
understood
the
reason
I'm
asking
is
I
mean
I
work
for
this
and
we
are
looking
at
multi-tenancy
situation
in
kubernetes
and
the
answer
we
always
get
is
name
spaces,
but
that's
not
completely
separation
between
users
right.
E
Right
indeed-
and
there
is
in
part,
because
in
kubernetes
there's
things
that
are
not,
there
are
objects
that
are
not
namespace
scope,
they're
cluster
scoped,
so
this
extra
level
of
virtualization
solves
that
problem.
So
with
in
with
the
kcp
logical
clusters,
each
logical
cluster
has
its
own
set
of
crds,
for
example
its
own
set
of
API
resources,
okay,
but
it's
still
one
server,
so
you
can
still
have
you
know
performance
interactions
between
users
of
that
server,
and
so
you
know
it
would
be
a
mistake
to
try
to
sell
it
as
complete
isolation.
H
I
Yes,
actually
we're
on
the
kcp
dash
Dev
Channel
over
in
the
kubernetes
slack,
and
the
community
meeting
happens.
Our
next
community
meeting-
we're
bi-weekly,
will
be
on
the
on
tomorrow
morning.
So
if
you'd
like
to,
we
can
rope
you
in
there
there's
a
Google
Google
Calendar
connected
at
the
bottom
of
that
edge
MC.
Let
me
let
me
share
now.
I
guess
now
is
appropriate.
Let
me
see
if
I
can
get
my
Safari
there.
It
is
okay.
I
So
right
so
we've
got
a
a
docs,
kcp,
Edge,
dot,
IO
and
of
course,
then
there's
our
repo
and
at
the
bottom
of
the
for
the
readme
in
the
repo
you'll
see
there
is
a
community
calendar
in
Google
that
you
can
connect
to
and
get
our
contact
info.
If
that's
not
sufficient,
then
feel
free
to
slack
me
in
kcp
Dev,
and
here
we
go
kcp,
Dev
and
I'll
be
happy
to
respond
with.
You
know
a
link
to
an
invite
or
anything
else,
and
that's
where
we
hang
out.
A
Okay,
thanks
does
anyone
else
have
anything
they
want
to
bring
up.
A
Just
jump
in
if
you
do
in
the
meantime,
a
couple
conference
reminders,
the
kubernetes
on
the
edge
days
submits,
came
in
very
strong.
The
program
committee
is
still
going
over
those
to
come
up
with
a
schedule,
but
it
looks
like
that's
going
to
be
a
a
really
nice
conference
going
on
in
Europe
in
April.
A
Another
reminder
that
the
scale
conference
in
the
Los
Angeles
area
is
coming
up
in
mid-march.
A
few
of
us
on
this
call
are
going
to
be
speaking
there,
but
there
is
a
fair
amount
of
edge
related
content
there,
as
well
as
a
lot
of
kubernetes
content.
A
Okay,
well,
if
you
think
of
anything
offline,
put
it
in
the
slack
channel
for
this
group
or
VM,
one
of
us
one
of
the
organizers
either
one
will
work.