►
From YouTube: CNCF SIG Runtime 2020-05-21
Description
CNCF SIG Runtime 2020-05-21
A
D
D
E
E
D
D
Any
gun,
presentation
matter
I've
left
some
comments
for
people
just
for
information,
so
you
can
also
look
at
the
slides
that
way
so
yeah
my
name's
Renault
has
been
a
software
engineer
at
Nvidia
for
the
past
few
years.
I'm
tech
lead
for
what
we
call
the
cloud
age
today.
I
want
to
talk
about
what
we
call
container
device
interface
and
I'll
go
over
some
background,
just
a
general
idea
of
ways:
the
state
of
third-party
devices
and
run
times
and
orchestrators
just
generally
in
the
cognitive
space.
D
Why
is
it
that
just
typing
darker
or
Putman
or
run
device
/dr
might
have?
No,
it
is
not
sufficient
and
I'll
give
some
background
on
what
we
think.
A
specific
interest
would
be
helpful,
hey
I,
think
I'll
go
or
some
real
word
the
example
that
we've
been
hitting
with
some
other
vendors
in
the
Q&A
space,
but
also
at
Nvidia
and
in
different
I'm,
just
real
world
scenarios,
and
then
I'll
give
an
overview
of
one
of
the
solution
that
we
came
up
with
with
some
of
the
maintainer
x'
and
of
container
D
and
Padma.
D
D
One
of
the
first
tests
that
I
took
on
was
the
odorous
plugin
implementation
in
the
communities
space
with
some
awesome
people
there
of
the
signal
and
resource
management
work
group
I
maintain
the
Nvidia
device
civilian
for
kubernetes
I
maintain
the
container
toolkit,
which
is
basically
our
stack
for
exposing
GPUs
and
different
container
and
time,
whether
it's
stalker,
Padma
and
singularity,
which
extends
to
through
an
ADIZ
and
nomad.
D
We
have
some
kind
of
custom
plugin
for
mezzos
wishes.
Entry
and
I've
been
working
on
some
of
the
OCI
and
runs
the
improvements
to
allow
hooks
for
better
device
supports,
so
I've
been
in
that
space
for
some
time
and
just
wanted
to
call
out
that
this
is
an
ongoing
topic
here.
That's
been
that's
been
brought
up
for
like
a
few
years.
D
If
you
want,
if
you're
looking
at
just
third-party
devices
and
runtimes
and
orchestration,
the
ecosystem
is
kind
of
a
mess
right
now,
because
pretty
much
everyone
has
their
own
concept
of
plugins.
Kubernetes
has
a
concept
of
device
plugin
that
you
can't
really
extend
to
you
other
illustrators
and
really
what
it
does
that
it
reimplemented.
D
Device
support
and
the
orchestration
level
rather
than
implementing
it
at
the
at
the
runtime,
and
that's
why
it's
really
hard
to
have.
It's
basically
extend
to
other
orchestration,
knowing
that
has
its
own
concept
of
the
device
plugin,
that's
not
the
same
interface,
so
I've
cube
entities
at
space,
energy,
RPC,
interface,
docker,
has
an
entry,
plugin
mechanism,
so
there's
an
interface
that
you
can
extend
through
pull,
requests
and
I.
D
Think
right
now,
Nvidia
Singh
when
you
plug
in
which
is
not
ideal
for
us.
We
want
to
make
sure
that
there's
some
kind
of
open
standard,
rather
than
something
that's
locked
down.
Condon
has
a
concept
of
hooks
which
are
like
not
OCI
hooks,
but
it's
mostly
a
followed
you
placed
on
the
host
file
system
and
you
describe
what
OCI
hooks
you
want
to
have
for
containers.
Maybe
maybe
you
I
mean.
Basically
you
write
an
extra,
an
expression
that
says
hey.
D
If
you
see
a
container
that
has
label
X,
then
I
want
you
to
call
this
free
start
hook.
That's
spawn
the
host
at
path,
ABC,
etc,
etc,
etc.
I'm
not
going
like
list
all
the
hooks,
so
just
step
back.
Why
is
it
that
when
you're
a
vendor
you
want
to
have
a
year
you're
not
satisfied
with
just
mounting
device
that
soft
as
my
device
and
well
that
was
true
in
the
past?
D
And
it's
still
true
for
kind
of
simple
devices
when
you
start
looking
at
I
want
to
see
more
advanced
devices
that
vendors
have
been
putting
out
there,
whether
it's
video
GPUs,
because
I'm
from
in
the
video
or
Intel
FPGAs
or
Solar
Flare
Knicks.
The
Knicks
are
kind
of
a
special
case
here,
because
there's
some
network
involved,
you
kind
of
realize
that
there's
a
lot
more
complexity
that
comes
from
software
and
there's
a
lot
more
constraints
and
how
vendors?
What
can
and
cannot
do
things.
D
So
when
we
would
what
at
least
we've
noticed
and
at
least
for
multiple
vendors
is
that
a
lot
of
the
time
they
need
to
just
expect
exclusive,
multiple
device
nodes?
And
it's
not
practical
to
ask
the
user
to
expose
the
GPU
and
then
some
virtual
control
nodes,
so
I
think
I've
been
video,
zero,
dev,
Nvidia,
CTL,
11,
video,
uvm
or
or
maybe
a
a
graphics
card.
You
might
want
to
expose
the
video
the
display
and
the
audio
devices
there
might
be
AI
pcs
bells
or
even
changing,
prognathous
entries.
D
So
when
I
talk
about
rocket
force
entries,
for
example,
you'll
find
in
Linux,
for
example,
at
proc
driver
video
you'll
find
the
full
list
of
devices
and
if
you
only
wanna,
mount
one
device
and
ordered
by
PCI,
if
you
only
wear
mount
one
device
inside
a
quarry
expose
one
device
inside
the
container,
you
want
to
hide
the
other
device
there
here
on
the
progress,
I'm,
there's
there's
a
lot
of
things
here,
related
to
just
I'm
like
stuff.
I
can
go
over
it.
D
If
we
have
more
time,
I
think
it's
just
probably
more
interesting
to
go
over
some
of
the
other
use
case.
So
maybe
we
maybe
as
a
vendor,
you
want
to
perform
compatibility
checks.
Can
this
container
around
them
this
GPU,
so
an
example
would
be
again
I'm
taking
Nvidia
here
because
I'm
more
familiar
with
it,
but
and
as
we
as
we
as
we
plant
new
GPUs
on
the
markets
at
some
time.
D
At
some
point,
we
have
some
either
backwards
compatibility
or
forwards
compatibility
requirements
where
you've
compelled
your
code
for
a
specific
architecture
and
that
the
architecture
that
you're
running
on
is
going
to
be
slow
and
to
the
point
that
you
might
one
you
might
not
want
to
run
your
GPU
I'm,
so
just
performing
that
compatibility
check.
That
hey
is
this.
The
right
architecture,
for
this
is
the
right,
GPU
architecture
for
this
container
and
filling.
D
If,
if,
if
that
is
not
the
case,
something
that
is
super
useful,
there's
also
other
use
case
where
you
might
want
to
check.
I
can
do
this,
but
if
you
have
questions
like
a
performant
Lorentz
on
specific
operations,
I
think
it's
not
a
big
surprise.
If
I,
if
I
mentioned
that,
if
you're
starting
at
the
VM,
there
is
very
different
operations
that
you
need
to
make
rather
than
just
starting
up
a
good
here.
D
Why
is
there
a
need
for
specification,
as
I
mentioned
I?
Think,
like
the
biggest
one
really
is
just
that
one
times
an
Orchestrator,
don't
really
offer
the
same
experience
to
users
because
of
the
fact
that
the
plug-in
is
not
like
the
plug-in
mechanisms.
Is
such
a
such
a
mess
right
now,
everyone
has
their
own
plug-in
to
give
them
at
least
two
like
things
that
are
not
even
reusable.
D
The
darker
GPU
support
is
not
reusable
in
communities
which
is
kind
of
a
weird
thing,
so
you
can
do
docker
run
GPUs,
but
you're
not
gonna,
get
the
same
experience
or
you
need
to
do
a
lot
of
work
rounds.
I'm
actually
get
that
same
experience.
If
you
want
to
do
that
and
what
that
means
is
well,
the
user
experience
is
not
is
not
super
consistent,
but
worst
of
all,
you
actually
end
up
and
maintainability
Hill
us-made
near
the
container
at
runtime
eight
bears.
D
We
spend
a
lot
of
time
just
either
trying
to
have
some
kind
of
common
interface
and
then
some
kind
of
external
shims
for
other
plugins
to
call
into
or
what
we've
seen
more
commonly
people
either
dropping
runtime
supports.
So
that
means
that
or
first
four
runtimes
that
would
be-
or
that
could
be
easily
supported.
I
think
like
at
the
end
of
the
day,
from
a
plug-in
mechanism,
whether
you're
running
pond
man
or
docker,
really
like
the
differences
at
the
end
of
the
day,
you're
still
using
run,
see
underneath
or
a
new
unique
see
run.
D
But
if
you
were
using
run
see
underneath
the
difference
shouldn't
shouldn't
really
matter
from
the
plugin,
of
course,
if
it's
vm
versus
Linux
containers,
then
there's
definitely
a
bit
more
difference
here,
but
I
think
that
the
point
I'm
trying
to
make
here
is
more
than
from
the
user
perspective.
Currently,
the
behaviors
are
not
the
same
across
container
runtimes
and
from
a
plugin
off
of
perspective.
There's
a
lot
of
work
here.
D
You
either
have
to
do
a
lot
of
work
to
have
a
plug-in
spread
across
multiple
runtimes
or
you
have
to
drop,
contain
your
runtime
support,
which
we've
seen
from
the
other
programmers
and
just
generally,
some
plugins,
don't
don't
expose
the
same
capabilities
in
a
consistent
way.
So
we
end
up
responding
or
resorting
to
act,
the
most
common
one
being
what
we
call
the
default
runtime
strategy
where
because,
for
example,
in
communities
plug-in
and
how
it
is
in
the
stack,
you
can't
really
access
OCI
hooks.
We
end
up
writing
a
shim
for
one
C
and
I'm.
D
D
D
So
what
I
gathered,
for
example,
of
the
in
the
intel's
SPG,
I
use
cases
that
the
the
links
they
have
a
fairly
simple
use
case,
where
there's
really
two
things
that
they
need
to
do.
They
need
to
amount
multiple
devices
inside
the
inner
and
so
that's
fairly
straightforward,
and
they
need
to
be
able
to
be
configured
if
pga,
with
the
correct
function.
So
I
mean
in
terms
of
security
from
a
security
standpoint.
D
To
do
that
they
need
to
be
able
to
know
was
the
user
intent
so
aims
have
either
container
runtime
information,
so
the
information
from
the
container
specification
or
they
need
to
have
information
that
is
passed
out,
the
user
so
whether
it
the
information
Frieza?
This
is
the
companies
you
see
where
it
could
be
pence
down
through
pod
labels
or
colonisation,
and
so
they
might
want
to
do
that
in
a
priest
or
work
at
the
u.s.
K
level.
Today,
that's
what
they
do
I
think.
D
Currently
they
use
quite
a
few
inject
hooks
inside
the
container
paired
with
device
plugins
and
they
don't
have
to
contain
the
D
supports.
So
that's
a
use
case
where
the
end
of
exclude
excluding
a
parent,
valid
snicks.
This
one
is
kind
of
I
want
to
say
there
is
a
bit
more
I
think
because
of
the
fact
that
it's
a
networking
interface
we
have
to
take
I
mean
we
have
to
look
at
this
use
case.
D
Maybe
like
it's
a
bit
more
nuanced
here,
but
this
is
an
example
where
what
actually
happens-
and
this
is
a
bit
more
common
than
you
would
think-
is
that
when
you
are
like
device
author,
you
you
traditionally
have
a
or
you
traditionally
have
a
kernel
module,
and
then
you
have
some
user
space
libraries
that
user
can
call
into
to
talk
to
you
to
try
no
model.
What
we've
seen-
and
this
is
something
that
we
see
for
Melnick's
and
we
see
for
also
for
video
I'll
talk
about
it
in
the
next
slide.
D
There's
also
a
few
device
notes
that
you
need
to
mount
here,
so
that
would
be
like
challenges
from
a
software
perspective
and
then
the
other
use
case
that
I'm
a
lot
more
familiar
with
this,
the
Nvidia
use
case.
We
have
these
same
issues
where
you
need
to
mount
device
to.
You
might
need
to
mount
user
LAN
libraries
again
same
problem
as
the
ABI
compatible
wanna
mount
them
at
run
time
rather
than
ships
at
mid
sized
container.
D
You
need
to
not
UNIX
sockets
inside
the
container
filled
a
pocket
century's
perform
a
belly
checks,
I've
linked
to
a
document
that
describes
in
more
details
what
Nvidia
does
if
this
is
something
that
you
want
to
look
more
into
and
we
use
cryo
to
inject
OCI
hooks
or
doctors
default
runtime
with
what
we
call
runtime
shrimp
or
porcupine
I
use
word
runtimes
context.
We'd
have
the
intrepid
individual
thinking
a
step
back.
We
discussed
this
state
a
bit
IQ
comas
on
with
the
day
nerdy
important,
maintainer
z--.
D
The
idea
that
we
came
up
with
this
to
have
this
kind
of
spent
in
space
on
C
and
I'm.
Also,
just
a
Jason
thought
that
you
dropped
on
the
machine
and
then
the
continued
runtime
system
and
it
just
a
shrugs.
What
are
the
devices
that
you
want
to
exploit
that
they're
available
in
the
machine
so
think
of
adjacent
file?
D
So
an
example
here
is
I,
don't
know
if
this
is
big
enough,
for
everyone
would
be
that
you
have
a
vendor
or
JSON
file
here
in
this
standard
pass
at
CCD
I
just
described
the
CD
I
wear
them.
It
describes
who
you
are
so
vendor
comms
device,
and
then
it
describes
the
devices.
So
here
there's
a
single
device
that
you
call
my
device
and
it
describes
the
operations
and
would
be
here
at
inside
or
when
you,
the
runtime,
uncie
runtime
run
device
vendor
come
slash,
my
device.
D
What
you
do
to
the
spec
is
that
you
would
merge
these
two
device
nodes
and
add
them
to
the
container
space,
and
so
here
I've
give
an
example.
What
run
time
you
might
want
or
what
CLI
example
you
might
want.
So
there's
a
verbose
example
here
where
you
would
do
run
time
run,
and
this
would
be
the
fully
qualified
name
vendor
dot-com
slash
device
across
white
device.
But
you
might
want
to
try
and
infer
this
through
that
stretch,
device
equals
my
device
there's
only
one
vendor
or
one
device.
D
D
Container
spectrum
specifies
the
operation
to
be
performed
and
there's
another
field
here.
If
you've
noticed
your
container,
spec
is
device
specific,
so
here
and
that's
to
my
device
and
there's
another
here
scene
I
may
feel
that
would
be
named
container
spec.
That
would
be
a
level
higher
so
not
at
the
level
of
the
CDI
device
fields,
but
at
the
level
of
the
just
general
container,
and
if
any
of
these
ter
devices
mounted
inside
the
container
you
would.
You
would
also
merge
this
field,
so
this
would
be
the
case.
D
For
example,
she
had
a
control
device
instead
of
saying
for
each
device
adding
a
container
spec.
You
know
that
says
dev
my
device
control.
You
might
want
to
my
only
this
device
control
once
inside
the
inner
explosives
device
once
inside.
So
that
would
be
this
example
here,
where
it
just.
It
is
a
an
operation
to
be
performed
if
any
of
the
devices
are
requested
for
rickety.
D
D
Is
you
either
need
to
update
the
environment
variable
on
to
these,
the
LD
preload
environment
variable
to
come
to
these
SL
files,
which
can
be
breaking
in
some
cases,
because
you
want
to
make
sure
that
you're
not
overriding
the
users
LD
preload
or
you
just
run
an
OD
config
before
the
container
starts
and
the
last
one
is
you
might
want
to
have
like
runtime
specific
operations,
but
I
think
these
are.
These
are
more
details,
or
at
least
that
one
is
more
of
a
detail.
D
That's
the
general
gist
of
these
specs
is
we
to
just
say:
I
have
devices
that
or
just
tell
the
runtime
I
have
devices
that
you
can
use,
here's
how
you
could
use
it
and
then
actually
probably
had
probably
forgot
to
yeah.
That's
probably
a
site:
I
won't
go
generally.
D
We've
talked
about
how
a
runtime
might
use
this,
and
the
idea
is
that,
at
the
end
of
the
day,
what
we
want
the
runtime
to
be
doing
is
instead
of
our
sorry.
The
orchestrator
to
be
doing
is,
instead
of
reimplemented
their
own
device
plugin.
We
want
the
orchestrator
to
call
in
runtime
run
device.
My
device,
rather
than
do
runtime
run,
should
I
step
in
video
zero
then
should
be
label,
slash,
etc,
etcetera.
D
Generally,
the
the
did
I
think
the
on
principle.
People
agreed
that
there
needed
to
be
some
kind
of
way
or
their
a
device
supports
that
the
date
of
the
I
think
all
men
was
kind
of
at
least
and
planning
dinners
were
and
I
interested
in
the
idea,
and
they
said
just
sent.
Yours
I've
been
focusing
right
now
or
just
I'm.
Getting
the
OCI
hugs
first
and
I
wanted
to
make
sure
that
there's
like
a
bit
a
bit
broader
consensus
than
just
having
like
the
101
specification
or
plugin.
D
D
The
goal
here
is
not
reach:
have
your
runtime
exposed
to
the
orchestrator?
What
device
are
on
the
host?
The
goal
would
be
to
have
your
Orchestrator
still
have
a
device
plugin
mechanism
here
is
right.
Now
the
device
plugin
in
humanities,
the
model
that
it
is
has
really
two
responsible.
The
first
one
is
that
it
talks
to
the
orchestrator
so
to
nakula
and
says
I
have
3G
and
video.com
/
GPUs
on
the
machine
and
the
Hewlett
can
now
advertise
in
its
status
field
or
in
its
capacity
and
resource
field.
D
D
The
the
next
one
is
that
currently,
what
happens
is
when
a
pod
lands
on
an
orchestrate
or
no?
No,
the
the
cubelets
will
select
what
device
it
wants
to
be
exposing
to
right
and
then
go
see
your
community's
device
plug
in
and
says.
Can
you
tell
me
whether
the
operation
that
needs
to
happen
or
me
to
expose
this
device
to
this
container,
so
the
cubit
might
respond
with
one
of
with
either
environment
variables?
Are
there
device
those
I
need
to
be
mounted
or
libraries
or
phones?
D
I
need
to
be
mounted
and
then
go
see
the
orchestra
to
the
runtime
and
says:
hey
read
this
container
and
here's
this
here
I
spec
and
I've
already
merged
device
notes
and
the
environment
variables
and
the
felt
and
the
less
responsibility
of
the
Cadis
device
equipment
here
is
just
monitoring
the
devices
for
health
issues.
D
So
what
what
I'm
suggesting
here
is
that
you
take
away
this
like
mounts
environment
this
space
here
where
jublee
goes
to
the
device,
will
get
an
desk.
What
do
I
need
to
do
and
cubed
which
this
go
directly
to
the
runtime
and
pass
down
in
this
year?
I
spec,
hey
I,
need
you
to
expose
such
that
or
not
selectively,
but
in
video
0
or
my
NVIDIA
GPU,
and
then
a
runtime
would
then
just
read
the
spec.
E
D
One
of
the
that
we
thought
about,
and
we
were
discussing
I
think
just
generally,
we
were,
we
were
more
looking
into
using
the
general
audience,
sorry
be
easier,
I
suspect,
but
taking
a
step
back.
The
problem
here
is
that
runtime
classes
are
in
any
specific
and
at
the
end
of
the
day,
we
want
to
try
and
break
down
and
bring
down
Hogan's
for
devices
to
the
event,
so
that
orchestrators
can
reuse
them.
D
So
that's,
that's.
The
real
problem
with
the
device
plug-in
is
that
right
now,
this
this
whole
operation
of
how
do
I
provide
the
same
container,
whether
you're
on
Nomad,
kubernetes,
mezzos
or
even
just
Trinity's,
using
darker
and
darker.
It's
not
a
solved
problem,
and
that's
that's.
The
big
problem
here
is
that,
if
I
spawn
a
container
using
Cuban
eighties
and
not
run
famous
doctor,
it's
like
as
a
plugin
author,
I
have
to
vary
or
I
have
two
different
stacks
rights.
I
can
either
use
my
and
plug
in
here.
D
Well,
that's,
that's
not
really
the
path
that
we
would
like
to
be
taking
when
I'm
asking
a
user
to
have
a
default
runtime
or
even
if
using
this
run,
the
runtime
class
is
not
really
okay,
as
as
an
Nvidia
like
third-party
bike,
plug-in
fender
I,
don't
want
users
to
be
using
a
custom
one
time,
I
want
to
hook
into
the
runtime
and
just
tell
them
here's
what
you
need
to
do
for
my
devices
to
be
available.
I.
B
A
question
about
you
kind
of
touched
on
it
like
you,
don't
want
people,
reprogramming,
FPGAs,
yeah,
yeah
and
same
kind
of
thing
with
your
video,
like
nvidia
SMI,
allows
you
to
control
your
GPU
right
and
change
things
manage
it.
How
do
you
handle
that
with,
like
the
privilege
level,
that
people
have
what
the
access
level
they
have
to
these
devices?
B
D
Really
depends
on
the
use
kings
when
you're
training,
GPU
models,
most
of
the
time
you're
gonna
be
requesting
one
to
eight
multi
knows
model,
and
in
that
context,
if
you
had
the
full
GPU,
the
expectation
is
that
you're
going
to
add
or
you're
going
access
reconfigure.
It's
a
more
recently
we
announced
thank
you
that
was
last
week,
so
you'll
see
that
next
GPUs,
where
he
can
partition
your
GPUs
and
smart
GPUs.
In
that
context,
when
you
have
access
to
these
devices,
we
have
like
it
depends
on
what
devices
are
mounted
easily
inside
you.
D
B
B
D
Basically,
the
idea
is
then
wait:
I
mean
it's
yeah.
We
don't
want
to
be
providing
all
the
privileges
to
the
users.
We
want
to
be
able
to
do
that
as
part
of
the
hook
life
cycle
and
the
way
they
would
do.
That
is,
for
example,
here
by
specifying
a
container
or
a
runtime
container
hook
in
that
spec
and
when
the
container
gets
or
when
that
gets
involved,
he
would
reconfigure
the
FPGA.
That's
where
that
continued.
E
D
D
A
A
Cdi
plugin
within
the
runtime,
so
that's
the
layering.
So
what
would
be
the
division
of
responsibilities
between
each
of
the
layers,
for
example?
How
how
would
we
hold
I
asked
for
or
share,
let's
say,
ask
for
I
Cuban,
ettus
container
to
say:
okay,
I
need
to
run
a
workload
and
then
how
does
that
go
down
to
the
bottom
level?.
D
That's
that's
a
great
question:
I
think
that
was
the
purpose
of
sy
that
I'd
never
finished
so
I.
Think
just
I'm
gonna
take
out
the
whole
shared
devices,
because
shared
devices
is
really
kind
of
again.
It's
a
device,
specific
problem
and
it's
up
to
how
the
device
implements
its
there
is
some
part
of
how
do
you
expose
it
in
the
resource
management.
But
let's,
let's
take
away
that
topic,
because
it's
kind
of
difficult,
let's
just
focus
on
I-
want
a
single
device.
I
want
to
the
way
today
works
in
in
communities.
D
Is
you
really
just
expose
it
in
your
specification
by
just
requesting
for
that
specific
research?
So
you
go
and
you're
prospecting,
you
add
a
resource
limits.
A
video
comes
like
if
you
Intel
concept,
yeah
Solar,
Flare,
Iowa
seismic,
and
so
we're
going
to
assume
that
your
skin,
you
were
already
found,
and
it
won't
before,
even
that
you
are
going
to
want
to
have
nose
advertise
that
there
are
resources
on
that
and
I
think.
This
is
definitely
where
the
device
plug-in
comes
into
play.
D
You
want
to
be
able
to
deploy
or
dispatch
a
container
on
all
your
notes.
To
figure
out
through
the
PCI
is
there,
for
example,
a
GQ
make
or
like
make
that
I
support
on
that
machine.
Yes,
so
this
is
the
time
that
the
device
plug
in
with
in
start
telling
Hewlett
a
I've
seen
a
jeep.
You
have
seen
a
device
here,
there's
not
the
right
way.
D
I'm
here
you
go
so
and
that's
that
first
step
here
where
the
device
may
be
that
device
plugin
installs
the
driver
may
be
that
device
plug-in
installs
that
so
maybe
that's
that's
the
point
where
this
device
plugins
goes
into.
Our
bike
has
a
mounts
on
at
CCI
and
just
drops
this
down,
so
we
would
take
nvidia,
gpus,
LSB,
CIA
IC
and
a
GPUs
install
the
driver
that
we
have
a
drug.
What
would
be
called
the
driver
container
gear
and
then
drop
the
jobless
speck
at
that
specific
path.
So
this
would
be
that
phase.
D
One
bootstrap
face
to
a
user
comes
in
presents
this
kind
of
cosmic
gets
submitted
to
the
API
server.
Scheduler
picks
it
up
figure.
It
figures
out
which
no,
that
should
be
mine.
On
assignment
note,
it
would
pose
a
spec
from
the
API
server
and
this
is
where
basically
tool.
It
goes
into
this
loop
mode,
where
it's
figures
out
which
device
should
I
be
assigned
to,
so
you
might
be
familiar
with
the
topology
manager
and
other
plugins
that
are
out
there.
D
D
Let's
go
API
server
here,
so
the
device
isn't
come
into
play
in
that
we
do,
but
basically
just
talks
to
the
runtimes
through
the
CRI
here
and
because
of
here
we
would
be
reusing.
That's
this
device,
it's
kind
of
nice
here
where
you
might
you're
you're,
not
going
to
be
conflicting
with
the
existing
device.
D
You
might
need
to
write
a
bit
more
in-flight
formatting
support
here,
but
basically
the
runtime
is
getting
something
along
the
lines
of
docker
run
best
device
and
video
0
or
play
Nvidia
device,
and
then
the
container
image,
and
that's
where
the
runtime
reads-
this
CD
I
fell
figures
out.
Maybe
there's
just
some
static
specification.
It's
just
only
some
devices
that
need
to
mount
so
I'm
gonna.
Tell
you
doctor
here
for
the
runtime
and
an
invidious
case:
there's
definitely
some
devices
that
you
need
to
mount.
There's.
D
Definitely
some
phones
that
you
need
to,
but
there's
also
a
hook
the
hook.
So
when
docker
then
calls
into
container
D,
you
know
intimately
into
run
see
the
spec
that
is
passed
down
here
contains
not
only
these
device
note
books,
but
also
on
these
OC.
I
hopes,
or
so
these
device
now
same
amounts,
but
also
the
Oceania
hooks.
So
at
that
point
runs
he
has
the
spec
and
all
the
information
to
call
the
actual
books
are
the
actual.
D
D
B
D
A
A
D
It's
definitely
just
mostly
a
JSON
file,
or
you
have
a
bigger
understandings.
Your
and
I
wrote
a
aspect
on
it.
Just
describes
the
different
fields,
I
think
just
I'm
not
gonna,
go
over
each
other,
but
in
terms
of
implementation,
I
think
it
really
is
going
to
be
mostly
depending
on
the
runtime.
We
could
expose
some
kind
of
library
and
that
represent
area
that
just
processes,
the
processes,
the
json
and
going
is
a
container
d
run.
C
and
rent
times
couldn't
afford
it.
D
D
A
E
So
question
2
is
d,
so
I
brought
up
the
idea
of
starting
a
working
group
for
this
I
mean.
Are
you
interested
in
doing
this.
D
E
D
D
E
E
Yeah,
so
my
guess
is
that
if
he,
if
you
get
started
with
a
working
group,
I
mean
you
could
actually
have
separate
meetings.
You
know
in
the
working
group
and
then
some
of
the
folks
that
are
interested
in
working
on
the
specification
that
would
actually
you
know,
discuss
more,
did
more
of
the
details
right
so.
E
F
E
Yeah
also,
the
q
Bayesian
processes.
There's
a
presentation
here
in
the
sig
and
there's
I
think
his
review
of
template
and
there
and
it
gets
review
later
by
the
TOC,
and
then
the
TOC
finds
a
person
who
drives
the
due
diligence
and
I
think
everything
turns
out
to
be
okay,
then
it
goes
for
a
vote
and
the
TOC
it's
for
incubation.
So
that's
generally
the
process
right,
so
that
may
be
a
Meakin
filling
a
little
bit
more
details
about
the
incubation
process.
If
she
has
been
a
more
details
now
you.
C
F
E
You
know
the
sig
is
also
happy
to
help
so
yeah
Richard
typically
is
the
TOC
person
that
which
is
how
to
say
the
the
companies
or
organizations
using
the
project
in
production
and
ask
questions
or
something
and
during
the
process,
and
then
you
know,
I
guess
comes
back
with
with
some
feedback
and
also
gets
documented
and
in
the
in
the
dot.
The
official
document.
F
E
D
Sorry
can
I
quickly
ask
a
good
question
yeah.
What
problems
I
have
right
now
is
that
the
Spanish
surrender,
my
like
github
account,
is
there
a
way
to
like
have
some
kind
of
I
want
to
say
more
official
okay,
like
represent
rehear.