►
From YouTube: CNCF SIG Runtime 2020-06-04
Description
CNCF SIG Runtime 2020-06-04
A
A
A
A
A
C
C
A
D
D
D
E
D
So
it
introduced
the
cloud
and
age
communication
functionality
to
make
work
over
the
limited
unstable
network
between
cloud
and
age
and
also
provide
age
autonomy.
Functionality
to
make
the
age
able
to
work
when
it's
disconnected
to
the
cloud
and
also
cubase
did
some
optimization
for
low
resource
use
cases,
especially
for
iot,
and
also
cuba,
provided
some
extensible
framework
to
able
to
connect
iot
devices
to
make
it
to
simplify
the
communication
between
application
and
devices.
D
D
D
So
basically
it
has
a
lightweight
cubelet
inside
and
we
introduce
the
metadata
consistency
for
node
level
to
make
it
able
to
work
when
it's
disconnected
so
our
currently
it's
it's
able
to
integrate
with
the
standard
container
runtime
and
as
well
as
a
container
network
and
csi,
and
especially
for
iot
cases,
and
we
introduced
a
standard
extensible
framework.
So
on
the
graph
you
can
see
it's
a
mapper
here.
D
D
So
the
device
controller
here
is
to
to
deal
with
the
lifecycle
of
the
device
crt
and
also
do
the
shutter
management
to
synchronize
all
the
relevant
information
to
the
age
and
the
single
controller
is
recently
we
added
to
so.
Basically,
it's
to
to
do
a
periodic
check
to
to
reconcile
just
in
any
case,
the
data
between
cloud
and
edge
are
not
up
to
date.
D
C
Yeah,
the
the
reason
it's
useful
just
because
the
diagram's
up
now
so
I'm
kind
of
curious
how
you
guys
deal
with
inconsistencies
so
so
in
particular
down
at
the
edge
you
have
a
bunch
of
you
have
software
running
and
potentially
you
can.
You
can
modify
the
the
state
of
of
the
nodes
down
there,
but
then
you
also
have
controllers
up
in
the
cloud
and
it's
possible.
C
You
know
not
only
for
them
to
get
out
of
sync,
so
it's
not
just
like
the
edge
behind,
but
but
actually
there
are
mutations
happening
on
both
sides
and
how
do
you
like
reconcile
the
conflicts
there
and
what
is
the
mental
model
that
people
should
have?
Should
they
only
control
things
through
the
cloud
in
which
case
it's
not
controllable
when
the
edge
is
disconnected
or
should
they
control
them
at
both
sides,
in
which
case?
How
do
you
reconcile
the
conflict.
D
D
Every
information
update
is
is
sent
by
by
the
cloud,
so
so
for
the
age.
When
it
it
persistent
any
updates
successfully,
it
will
reply
a
ad,
so
we
will
check
the
if
the
access
successfully
received,
and
we
think
that's
that
one
one
time
synchronization
is
successful.
Otherwise
we
will
retry
and
also
the
sync
controller
here
is
just
the
recording
every
object.
The
version
number
when
every
time
ops
update
success,
so
it
will
compare
to
the
the
original
object
inside
api
server.
D
Welcome
so
for
the
cuba
csi
driver
here
is
so
the
for
the
csi.
A
little
bit
different
is
that
the
upstream
csi
model
integration
model
is
also
designed
by
in
the
data
center
network.
So
here
we
think
for
csi.
In
the
age
cases,
the
whole
csi
backend
would
be
on
the
edge.
So
so
the
cube
age
csi
driver
here
is
trying
to
hook
some
some
requests
and
the
response
to
the
to
make
it
send
it
to
the
backhander
to
the
edge.
F
F
D
So
yeah
so
currently
for
the
iot
devices
management.
We
only
deal
with
the
communication
part.
So
that
means,
when
is
when
a
device
is
ready
connected
to
the
to
a
node.
The
user
can
add
a
custom
resource
to
to
tell
the
device
details
to
the
system.
Then
the
application
can
reuse
that
information
to
to
connect
to
the
device
we
not
yet
covered
like
a
firmware
update
in
the
current
implementation.
B
Can
I
add
something?
So
actually
we
are
in
discuss
with
eclipse
heartbeat
project,
so
basically
they
handle
the
firmware
update,
because
when
you
update
the
firmware,
you
need
a
full
backup
of
a
whole
os
on
device.
In
case
the
firmware
update
failed.
They
need
to
roll
back,
that's
provided
by
the
device
os
provider,
so
we
so
our
plan
is
collaborate
with
a
hotbed
to
do
that.
The
do
the
firmware
update
with
on
the
device.
F
D
So
actually
there
are
a
different
applications,
so
so
in
the
cloud
the
the
user
may
just
want
to
have
some
the
applications
like
today
they
have
in
the
data
center
and
also
in
the
cloud
like
so
like.
If
the
multi
there
are
multiple
application
instances
inside
a
age
site,
so
they
want
to
share
some
data
and
also
the
using
csi
on
the
age.
They
can
persistent
some
data
like
for
iot
like
industriality.
D
A
So
I
so
what
would
video
be
one
of
the
the
applications?
I
guess
for
storage
at
the
edge?
Some
people
may
be
trying
to
get
more
video
content
faster
at
the
edge
yes
yeah
yeah,
and
then
how
would
that
communication
happen
between
the
cloud
and
the
edges?
I
mean
there's
this
thing
communication
that
looks
like
that's
happening
through
websocket,
but
did
you
mention
also
that
quick
is
also
supported
between
the.
D
Yeah
so
so,
currently,
a
websocket
is
a
underlying
implementation
to
between
the
cloud
and
age
and
also
we
we
implemented
quick
as
an
alternative,
but
according
to
the
test
result,
there's
no
big
improvement
and
the
websocket
sometimes
is
more
stable.
So
currently,
websocket
is
the
default
protocol.
C
A
C
So
mobile
phones
are
in
a
in
a
way
kind
of
special
cases
of
edge
nodes.
They
have
most
of
the
properties
of
an
edge
node
and
unreliable
communication
with
the
cloud,
and
they
also
have
a
bunch
of
sensors
and
cameras
and
various
other
things
have
you
guys
found
any
use
cases
or
given
any
thought
to
using
this
for
mobile
phone
kind
of
applications,
especially
given.
I
was
interested
in
mobile
phones.
B
We
are
doing
a
inference
of
loading
project,
a
collaborator
with
a
crino
community,
so
in
the
future,
slides
kevin
will
talk
about
our
goal.
So
basically
the
kubernetes
will
focus
on
two
directions:
one
is
iot,
the
other
is
I
mean
you
see
user
case
so
that
mec
use
case
is
mainly
info:
the
technical
provider,
so
that
provided
the
edge
a
mobile
ads
in
the
technical
cases,
so
connect
mobile
phone
to
the
data
center.
Okay,
so
so
emits
nbc
is
mobile,
h
cloud.
B
C
D
D
Basically,
it
will
persistent
part
config
map
and
also
the
the
node
information
and
the
secrets
on
the
age,
so
the
hd.
Currently
we
actually
removed
some
of
the
packages
of
from
the
vanilla
toilet.
So
currently
it's
a
it's.
Basically
a
light
weighted
cooled
in
the
early
days.
We
we
we
did
some
inline
code
change,
but
why
we
are
in
the
middle
way
to
to
reduce
the
inline
code
change
anyway.
D
That's
a
detailed
thing
so
for
the
device
string
that
that
one
is
to
to
to
actually
so
the
the
concept
is
from
iot
scenario,
so
that
one
is
to
deal
with
the
iot
devices.
When,
when
any
information
like
the
sensor,
data
come
from
the
iot
device,
it
will
to
duplicate
the
information
one
persistent
on
the
agent
node
and
another
one,
send
it
to
the
cloud
so
that
one
also
represents
on
the
device
crd.
D
So
the
difference
here
is
currently
we
only
actually
deal
with
where
we
recommend
the
device
train
to
deal
with
like
the
switch
and
some
like
kind
of
static
properties
from
the
device,
because
that
that
one,
the
the
data,
don't
have
much
data.
So
it's
it's.
It's
easy
to
synchronize
for
the
heavy
load.
D
If
there's
a
the
the
iot
device
is
producing
a
large
amount
of
data,
we
would
recommend
end
user
to
deploy
a
tsdb
to
also
and
also
integrate
with
the
other
iot
pipeline
middleware
to
to
kind
of
reduce
the
data
and
pick
the
useful
things
and
the
edge
mesh
here
is
actually
the
service
and
the
node
level
dns
implementation
implementation
here.
So
we
try
to
simplify
the
service
discovery
and
the
service
communication
on
the
age,
especially
in
cases
the
nodes
are
located
in
different
subnets.
D
Okay,
next
so
for
the
community
growth
growth,
I
I
want
to
highlight
that
currently
cubbage
is
becoming
more
and
more
healthy.
So
here
are
some
numbers.
Comparing
to
the
early
days
when
we
being
accepted
as
a
sandbox
project,
you
can
find
out.
The
number
of
contributors
are
now
we
have
more
than
300.
D
D
D
The
early
days
we
have
only
one,
and
now
we
have
more
than
25,
and
also
we
are
currently
working
on
some
of
the
cross-community
collaboration
like
kubernetes
ioth
working
group,
and
also
we
are
closely
working
with
acronym.
We
have.
There
are
two
blueprint
there
using
cubeh,
and
also
the
under
discussion
is
the
eclipse
yeah
on
the
right.
These.
There
are
four
top
new
contributors
in
during
2019.
B
D
D
So
currently
they
are
all
from
china
and
we
are
also
under
interaction
with
the
other
telecos
and
also
we
have
I.t
service
providers
and
also
cloud
providers
contributing
to
the
project,
and
especially,
we
have
some
academics
joining
the
project
as
well,
so
so
for
the
user
adoption.
Currently,
we
have
more
than
20,
but
only
a
few
got
confirmed
to
make
public,
so
I
also
listed
them
here.
D
C
D
So
this
is
one,
so
this
one
is
actually
currently
the
biggest
the
largest
end
user
adoption,
the
end
user
name.
We
are
still
under
confirmation,
but
I'd
like
to
share
the
some
of
the
adoption
information.
So
basically,
this
is
a
is
a
highway,
etc
ecosystem
in
china.
So
the
user
want
to
build
the
the
whole
system
whole
system
based
on
the
cloud
native
technologies,
so
the
architecture.
B
A
B
D
So
the
biggest
challenge
here
is
the
the
actually
the
the
egc
gate
is
really
far
away
from
the
central
cloud
data
center.
The
network
is
very
unstable
and
the
bandwidth
is
very
limited,
so
qh
helped
to
make
it
much
easier
because
it's
able
to
to
synchronize
the
information
be
over
internet
and
make
it
able
to
make,
especially
for
the
age
nodes,
it's
able
to
work
when
it's
disconnected
to
the
cloud.
D
So
currently
they
are
using
coverage
to
manage
more
than
50
000
age
notes
and
also
there
are
500
000
containers
in
total.
Currently
so
the
currency,
the
the
said,
the
containers
are
mainly
the
tow
applications
they
have
planned
to
to
do
to
develop
the
vehicle,
infra
infrastructure,
cooperative
cooperative
system
and
also
that
one
would
use
the
cloud
native
technology
as
well.
D
So
for
the
age
nodes,
the
underlying
architecture
they
have
the
most
of
the
edge
nodes
are
the
arm
can
servers
and
also
they
have
some
x86
ipc.
So
if
ibc
is
industry
or
pc,
it's
kind
of
a
small
pc,
so
for
the
business
perspective,
they
are
currently
managing
dealing
with
300
million
data
records
every
day.
D
So
for
the
for
the
the
benefits,
the
from
a
driver
perspective,
the
the
time
used
to
passing
through
the
tow
system
tow
station,
you
can
find
out
it's
it's
a
big
improvement
for
the
the
car.
It
now
only
take
two
seconds
because
they
don't
need
to
stop
and
manually
deal
with
the
the
building
stuff
for
a
truck.
You
can
find
it's
it's
it's
a
even
larger
improvement.
D
So
this
one
is
a
a
smart
campus
and
the
user,
so
basically
they
want
to
to
make
the
the
campus
smart
with
some
ai
functionality,
so
the
end
user.
Actually
they
have.
They
already
have
have
some
ip
camera
in
in
the
campus
and
they
don't
have
the
the
wires
for
dealing
with
the
and
also
the
much
power
redundant
power
stuff
for
the
making
it
smart.
D
So
the
keyboard
used
here
is
to
the
way
they
use
qh
here
is
they
have
the
they
have
some
room
to
to
manage
some
pc
and
the
can
servers.
So
they
install
a
keyboard
there
to
manage
some
containerize
the
applications
and
for
the
camera,
because
it
has
the
interface
to
to
to
expose
the
video
stream
through
a
ip
address.
The
applications
can
can
can
get
these
get
their
video
through
the
ip
address.
D
So
in
the
cloud
because
there
are
not
the
the
underlying
server
is
not
that
powerful
on
the
edge.
So
currently,
like
the
the
high
level,
analysis
is
still
doing
in
the
cloud,
so
basically
on
the
edge
there,
the
cuba
age
is
helping
to
manage
the
applications
to
do
like
face
recognition
and
also
flow
analysis
like
how
many
people
are
entering
a
gate
per
hour
like
that.
D
So
for
the
end
to
end
benefits
this.
This
solution
helped
the
campus
saved
around
thirty
percent,
replace
the
work
because
they
they
need
less
security
guards.
They
can
just
rely
on
the
camera
and
the
ai
applications,
and
also
for,
like
the
visitors
entering
the
gate,
it's
much
smarter
with
the
ai
functionality,
so
it
just
needed
seconds
to
for
a
visitor
entering
the
campus.
C
Okay,
next,
that's
a
quick
comment
here.
I
I
would
imagine
that
this
is
a
fairly
big
use
case.
Where,
for
ai
related
systems,
a
lot
of
the
inference
happens
at
the
edge
and
and
you
have
constantly
updated
retrained
models
that
have
to
be
pushed
out
to
all
these
edge
locations
regularly
and
sometimes
constantly.
So,
is
that
a
is
that
a
use
case
that
you've
seen
for
these
kind
of
deployments.
B
Yes,
yes,
yes,
we
are
seeing
that
and
also
that's
why
we
collaborator
with
a
chrono
there's
a
code
machine
learning,
offloading
framework
developer
based
on
kube
edge.
So
that's
the
we
talked
about
in
the
previous
slides
is
a
crano
kubat
offloading
service
in
the
that
blueprint
is
running
on
the
acronym
community,
based
on
cool
badge,
very,
very
cool.
B
And
we
can
set
up
some
demos
if
we
need
so,
we
can
show
how
we
use
a
mobile
phone
and
offload
the
emotion
recognition
to
the
edge.
Then,
if
that's
not
enough,
we
we
have
a
constant
train
model,
push
from
the
cloud
to
the
edge.
So
that's
the
demo,
it's
a
just
early
in
the
early
stage,
but
the
final
release
will
be
in
the
early
q4,
but
we
already
have
the
first
edition
of
them
already.
So
we
are
going
to
set
up
a
environment,
have
a
edge
behind
firewall.
B
We
have
the
cloud
setup
on
aws
to
demo
this
environment.
So
then
you
can
use
cell
phone
to
access
the
edge
server
and
yeah.
Something
like
that.
The
the
cloud
running
on
the
public
cloud.
Your
server
are
running
behind
the
underprivileged
firewall,
but
your
cell
phone
can
connect
to
the
edge
to
offload
your
inference
requirement.
The
influencing.
C
I
think
there
was
a
fair
amount
of
interest
in
in
using
kubernetes
in
very,
very
different
ways
than
it
was
originally
designed
for
because
this
is
not
what
you
know
the
kubernetes
designs
had
in
mind,
but
it's
clearly
very
useful
for
these
kinds
of
things.
So
I
think
we
should.
We
should
really
kind
of
get
the
message
out.
They're,
not
all
the
details.
B
D
D
So
before
one
zero,
we
are
actually
focusing
on
the
fundamental
applications,
like
so
mainly
it's
to
provide
lightweight
age
components
and
also
to
provide
a
core
api
support
on
the
age.
So
and
then
we
we
are
trying
to
add
some
more
useful
functionalities
like
so
also
like
we,
we
verify
the
container
d
integration
so
for
cube
age
itself.
The
the
component
in
the
age
currently
takes
70
megabytes
and
also
like
recently
in
1.3.
B
Yeah,
I
add
the
one
thing
is
from
this
year,
we're
already
into
the
regular
release
cycle,
so
we
have
a
three
month
release
cycle.
It's
kind
of
a
one
month
behind
the
upstream
kubernetes
release
we're
going
to
incorporate
a
new
release
of
kubernetes
integrate.
Then
we
release
kubernetes,
so
it's
about
a
month
after
a
four
more
kubernetes
release.
So
we
are
also
in
the
quarterly
three
month
release
cycle
now
so
very
regularly.
B
C
One
quick
question:
so
how
do
you
and
this
may
have
been
covered?
I
just
got
interrupted
by
dog,
so
so
you
clearly
there's
there's
a
lot
more
to
cube
edge
than
than
traditional
kubernetes
and
there's
a
lot
of
custom
components
and
things,
but
there's
some
amount
of
it
which
shares
like
apis
and
stuff
with
with
parts
of
kubernetes.
C
D
So,
actually,
in
early
days,
we
are
hosting
some
of
the
inline
changes
for
the
cubelet,
because
we
want
to
make
it
smaller
on
the
edge.
So
during
last
few
months
we
have
moved
the
more
easier
way.
So
we
are,
we
are
currently
rendering
the
the
cubelet
code,
and
so
basically
we
rewrite
the
code
entrance
to
to
only
load
the
packages.
We
think
that's
necessary
on
the
age
and
for
for
the
api
support,
that's
easy.
We
just
rely
on
the
client
code
yeah,
so.
B
C
D
D
B
Yeah,
well,
we
don't
have
we
don't
change
the
upstream
cloud
side
of
kubernetes,
we
just
we.
We
just
take
the
upstream
directly.
However,
we
don't
we
just.
I
said
it
take
a
month
because
we
want
to
see
if
it's
any
new
apis.
If
we
support
or
not
so
basically,
we
only
extend
the
cloud
api.
We
don't
change
any
of
them,
so
we
are
compatible
with
upstream
all
the
time
we
don't
have
any
hard
fork
on
the
cloud
side.
Excellent.
Thank
you.
E
A
They
okay,
so
they
release
every
three
months
now
or
so
four
releases
a
year
and
now
they're
going
down
to.
I
think
three,
but
you
know
one
of
the
challenges.
Is
you
know,
keeping
track
with
that
api
and
and
the
changes
in
the
api
right?
So
so
I
guess
you
have
a
you.
Have
a
matrix
already
that
identifies
this
and
says
this
version
is
it's
works
with
one
grenade
is
115
and
116,
or
one
nod.
Okay,.
B
D
G
D
G
C
E
G
C
G
D
Yeah
and
also
we
are
working
on
to
attract
more
developer
to
joining
the
project
development
and
also
we
we
did
some.
We
did
some
intelligent
evangelist
last
year
and
this
year
we
are
also
joining
like
the
the
google
summer
of
code
and
the
community
bridge
to
help
more
promote
the
project
and
for
the
so
actually
for
for
the
community
governments
we
this
year,
especially
we
want
to
enforce
the
sigs.
D
So
because
the
this
it's
kind
of
different
with
the
iot
so
for
the
rme
c6,
currently
are
mainly
the
telecos
joining
to
discuss
how
to
use
cubase
to
provide
a
the
common
underlying
first
infrastructure
to
enforce
the
the
whole
mec
system
so
yeah.
So
the
the
current
status
of
these
two
things
are
are
discussing
the
the
secret
charts
and
probably
will
be
announced
in
the
following
month
and
for
the
technical
thing
way,
especially
so
for
the
two
things
we.
D
We
definitely
will
do
some
more
development
and
add
some
more
advanced
features
to
better
serve
the
device
iot
and
mec.
But
in
general
we
want
to
integrate
more
to
do
more
integration
with
csf
projects
as
as
well
as
rfh
projects,
and
also
we
we
will
keep
working
on
to
improve
the
reliability
and
the
security
to
better
serve
the
end
user
adoption
yeah.
D
D
C
I
haven't-
I
haven't
spoken
to
the
rest
of
the
city
about
this,
but
but
just
based
on
purely
on
what
I've
seen
here.
I
know,
as
you
know,
quite
a
lot
about
the
project
from
from
years
gone
by
when
I
was
working
at
huawei.
I
mean
this
looks
like
a
perfect
candidate
for
incubation
to
me.
It's
it's
clearly
got
some
momentum.
It's
clearly
not
you
know
mature
enough
for
for
finalization
for
graduation.
Yet,
but
it's
it's
a
great
candidate
for
incubation.
C
It
has
enough
momentum,
it
has
a
bunch
of
companies
involved
and-
and
it's
in
use
in
in
you
know,
production
and
at
least
a
few
use
cases.
So
sounds
like
a
very,
very
good
incubation
project
to
me,
and
it's
also
the
first
that
I'm
aware
of
anyway,
the
first
major
kubernetes
project,
which
is
sort
of
a
very
specific
specialization
of
kubernetes
that
that
was
not
really
intended,
for
which
I
think
is,
is
interesting
in
itself.
A
C
Yeah
sure
do
we
have
any
volunteers?
Does
anyone
have
the
bandwidth
and
the
knowledge
available
to
be
able
to
do
a
good
job
of
having
a
look
at
this?
So
to
be
clear,
we
need
you
to
go
and
look
at
the
code.
Maybe
you
know
install
it.
Try
it
out
talk
to
a
couple
of
the
customers,
if
possible,
and
and
just
make
sure
that
I
mean
I
our
job
is
to
make
sure
that
what
kevin's
told
us
is
all
true,
that's
one
of
our
jobs.
C
C
C
Yeah
something
like
six
weeks,
so
we
would
need
to
give
the
you
know.
I
I
don't
they're
busy
working
on
what
the
process
is,
and
I
haven't
been
following
that
discussion
too
closely,
but
I
would
imagine
that
the
toc
need
at
least
a
couple
of
weeks
just
to
kind
of
open
this
up
for
public
scrutiny,
etc.
C
Ideally
in
the
next
two
weeks,
I'd
say
the
in
the
worst
case
by
the
end
of
june,
just
to
give
about
two
weeks
for
a
public
review,
and
then
the
toc
can
hopefully
vote
after
that,
so
ricardo,
maybe
you
can
just
get
in
contact
or
with
diane's
left
already.
Is
that
why
she's
still
here?
Yes,
she's
still
here.
A
C
C
So
I
can,
I
can
do
my
best
to
try
and
crank
out
a
brief
sort
of
due
diligence
document
in
the
next,
let's
say
three
weeks,
but
we
we
do
then
need
to
make
sure
that
we
have
the
dates
lined
up
for
public
review
and
toc
vote
before
kubecon
china
starts
so
ricardo.
Maybe
I
can
ask
you
to
to
line
all
of
that
stuff
up
and
I'll
focus
on
the
due
diligence.
Unless
there
is
anybody
else
who
would
like
to
help
with
that.