►
From YouTube: Mesos Containerization WG 08102017
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
B
B
B
Hey
Gilbert,
do
you
have
any
you
have
anything
to
to
share
I
just
slide.
Looking
like
that,
yeah.
B
B
B
B
Layer,
GC
I
just
do
it
in
progress.
Gober.
C
A
B
A
B
You
can
you
can
ask,
you,
know,
didn't
get
a
minute
like
if
you
think
do
you
think
this
is
like
a
blocking
the
production
and
it
will
be
really
nice
to
get
to
a
point
for
it.
Just
like
an
email
to
definite
on
that
and
pink
the
release
manager
on
there.
C
B
Prize
this
is
a
cure
in
progress.
The
volume
ownership
group
cleanup
later
it's
on
this.
So
now
we
we
think
a
little
bit
on
that
will
create
a
structure
on
this
and
I.
Think
your
Lea
is
going
to
do
some
digging
on
like
whether
you
can
use
a
CL,
4,
6
8
PL,
just
to
solve
the
problem
so
that
we
can
be
windows
compatible
potentially
in
the
future.
So
Ilya
is
doing
some
digging
on
back,
so
I
need
to
paint
amount
of
status,
so
I
think
it's
viewing
design
and
in
the
current
actors,
actor.
B
Oh
documentation,
I,
don't
think
we
spend
time
on
this.
Yet
if
not
started
and
I
cannot
find
a
owner
for
this
one,
so
yeah
so
yeah,
sorry,
accurate,
CPU
affinity,
doing
desire
I'm,
not
sure,
like
the
guys
show
up
or
not
okay.
So
by
the
way
you
do
anything
that
you
think
you
are
working
on
right
now,
that's
in
progress,
but
not
up
like
like.
Maybe
you
should
move
this
up,
because
something
that
we
are
working
on.
B
C
B
Maybe
like
a
during
that
process,
maybe
just
document
the
isolator
and
I
just
had
a
document
which
I
think
I
got
this
app
takes
like
two
too
big,
maybe
like
what
we
should
do
is
like.
Well,
we
ask
them
the
features
like
this
one
block,
I
old
I,
think
we
don't
have
a
documentation
for
secrets
just
as
in
documentation,
both
teachers
and
the
block
IO
stuff
will
go
in
there.
Naturally,
yeah.
B
B
B
C
B
E
B
B
Okay,
okay,.
B
E
B
E
C
So
so
here
is
the
currently
nicknamed
a
semantics
we
have
in
bases,
so
we
have
so,
as
some
of
you
guys
might
know,
we
have
nested
continued
support
on
bases
and
if
you
are
not
launching
any
Messick
continuous,
for
example,
you
are
launching
in
as
a
cue
to
continue.
So
it
is
the
top
level
continue.
So
we
have
different
semantics
for
top-level
container
and
all
the
other
Nesta
continues.
So
we
summarized
all
the
names
dates
right
now
we
have
so
in
mazes.
C
We
support
line
Lance,
Bass
Fitness
base
level,
nine
states
and
UTS
and
IP
seen
anything
so
for
secret,
namespace
and
user
name
space.
They
are
not
supported
yet
that
we
can
consider
it
to
support
sleeve
of
namespace
in
the
near
future,
I
guess
the
user
name
States.
We
still
need
more
investigation
on
that.
So
right
now
for
a
regular
level
convenient
the
magnin
space
is
not
shared
and
finesse.
The
continuous
long
name
space
is
also
not
read.
So
all
those
monies
base
they
are
controlled
by
the
Linux
file
system,
isolated.
C
So
let
let's
say
if
you
have
a
Linux
file
system,
isolator
innate,
both
all
your
container,
going
to
have
its
own
knowledge
base.
So
it's
the
same
for
the
pit
namespace.
We
have
the
namespace
slash
pick
either
a
day
and
later
for
opinions
based
isolation.
If
you
turned
it
on
so
you're
going
to
have
your
own
Fitness
is
very
isolated,
and
so
the
network
and
UTS
we
start.
If
I
remember
correctly,
we
buy
these
two
into
the
CNI.
C
I
played
a
so
if
you
have
the
CNI
I
play
the
trendy
on,
so
you
can
have
the
network
and
UTS
men's
days.
So
for
the
top
level
continue
weave.
Naca
is
configurable.
It
means
it
means
you
can.
If
you
use
the
CNI
operator,
you
can
prepare
to
use
the
host
network
or
the
bridge
network.
So
if
it
is
the
host
network,
so
it
means
the
top-level
container
going
to
share
their
host
network
names,
Network
and
UTS
namespace,
otherwise
you're
going
to
have
its
own
network
namespace.
C
C
State
network
name
space
under
the
top-level
continue
so
for
IP,
simple
semantics
extend
to
the
fitness
base,
so
they
are
each
continuing
have
the
IP
scene
in
states
if
the
isolator
is
can
do,
and
we
also
have
the
nine
states
ledger,
IPS
be
isolated
so
for
the
others
control
any
of
them
from
now
for
the
cygwin
user,
so
by
default
or
the
power
level
continue
are
going
to
share
it
with
the
aging
process
and
all
the
next
becoming
they're
going
to
share
from
the
parents.
Basically,
that's
just
the
agent
namespace
and
I.
C
We
would
like
to
can
change
the
semantics
in
the
future,
because
we
we
have
some
some
conversation
between
G
and
V,
no
and
and
a
couple
other
folks,
and
we
really
like
the
absol
newspapers.
We
want
to
share
the
pignon
space
or
the
eyepiece
in
nz9
space,
for
example.
Then
people
would
like
to
launch
multipass
group
under
one
before
as
a
cuter
and
each
of
those
task
group
they
have
different
Nesta
continues.
C
So
for
our
automatics
it
is
totally
impossible
because
for
some
cases
like,
for
example,
like
Connecticut,
when
people
want
to
run
Connecticut's
on
measures,
so
they
might
launch
a
couplet
in
the
net
to
continue
and
launch
continuously
in
another
left,
pecan
dinner
and
both
of
them
off
then
going
to
be
in
under
the
same
task
group
under
under
the
same
cup
people
executed,
so
the
crew
they
want
to
get
into
this
to
get
into
the
canoe
Trentino.
These
fitness
is
to
to
reach
out
to
how
to
continue
do.
C
At
least
you
should
launcher
docker
container,
but
for
for
me,
and
you
need
the
handle
to
get
into
this
DNI
handle
to
Chrome,
then
new,
the
new
net
one
in
space,
and
so
for
those
cases.
We
want
to
make
the
kid
ninh
space
configurable.
So
it's
similar
to
the
IPC.
We
also
want
to
make
it
a
complete
we
can
configure
about,
which
means,
like
the
the
user,
has
been
an
option
from
the
framework
API
to
specify
I
want
this
container
to
assure
the
namespace
from
the
parent
and
I
can
make
a
sheriff
or
not.
C
So
that's
what
we
can.
What
we
plan
to
change
in
our
next,
in
in
district
and
and
for
the
and
for
some
of
the
others
like
balances
and
net
one
and
state,
we
don't
have
any.
We
don't
see
any
specific
requirement
to
big
change
because
for
Magnus
space
I
believe,
like
every
containers
you
have
is
mountains
days
and
for
net1
and
say,
is
we
as
if
we
are
assuming
all
the
other
nasty
container
shared
and
that
one
ends
days
from
here
and
the
top-level
continue?
C
C
So,
for
now
we
just
make
a
very
basic
for
the
fitness
base,
so
this
is
the
framework.
Api
is
very
simple,
so
we,
the
link
next
info,
is
in
the
container
info.
So
continued
info
can
show
the
the
continued
configuration
for
the
continued
runtime.
So
in
each
of
the
continued
either
we
have
the
Linux
info
and
from
now
the
Linux
info
only
contain
the
information
about
other
Linux
cutter
database.
So
we
decide
to
introduce
a
new
product
to
in
the
in
Linux
info.
Core
shear
pin
ends
days.
C
So
if
it
is
true,
you
can
share
the
pitman
space
from
the
parent
container.
Otherwise,
if
you
chrome
this
new
finding
space,
if
the
fitness
based
isolator
is
enabled-
and
we
make
a
note
in
the
comment
here-
it
is
a
top-level
container
if
you
share
from
the
agent
if
we
make
it
shareable
so
yeah,
the
user
interface
is,
is
pretty
pretty
simple
and
we
do
have
some
other
concern
about
the
security,
because
most
people,
the
most
of
the
user.
This
is
control
by
use.
This
is
the
Primo
API
from
the
operator
perspective.
C
I.
We
can
see
most
of
the
operators,
they
don't
want
any
user
to
touch
the
agent
pin
then
state,
so
we
introduce
an
agent
thread
to
control
to
control
the
logic.
So
it's
called
this
allows
sharing
a
jump
in
namespace
and
deposed
to
be
brought
because
so,
if
we
map
this
agent
rack
as
true
so
no
matter
how
the
user
specify
the
user
API
so
top-level
container
you're
always
not
sharing
the
agent
spin
in
space.
So
this
is
the
cinematic.
C
So
if
it's
for
some
security
concern
and
and
right
now
in
our
pin
and
space,
the
logic
is
very
simple:
if
you
we
always
enter
a
parent
in
name
status
and
then
depends
on
this
logic,
this
age
of
back
and
the
framework
API.
So
we
will
figure
out
in
the
isolator.
We
will
figure
out
whether
or
not
the
container
should
come
is
a
new
fit
name
space
or
for
itself,
so
so
Chan
already
implement
that
so,
but
most
likely
I
will
lend
those
patches
that
they
are
tomorrow.
C
D
I
have
a
question
in
the
pretty
buff
API
I
guess
it
doesn't
seem
very,
very
general.
So
do
you
imagine
you
will
keep
on
adding
extra
bully
in
the
fields
to
that
four
different
namespaces
yeah.
C
That's
a
good
question,
so
yes
depends
on
what
we
discussed
here
for
the
for
the
potential
change
in
the
near
future,
most
likely
being
on
we.
If
for
the
change
in
inside
of
the
Linux
info,
we
cannot
add
the
pin
n
space
and
the
IPC
nine
space
talk
about
bill
to
make
it
configure
those.
So
for
the
for
the
focus
option
and
long
term.
As
we
can
see,
we
don't
have
any
plan
to
change
the
semantics
from
that
magnons
based
on
net
1n
space
or
the
UTS
namespace.
D
B
So
James
I
think
the
reason
for
that
is
because,
like
we
already
have
like
a
narrower
info,
which
is
kind
of
like
a
our
way
of
like,
if
you
don't
reckon
our
encoding
is
gonna
join
the
hosts
now
working.
So
we
already
have
some
sort
of
control
in
the
frame
API
to
allow
emoticons
big
tasks.
We
don't
have
to
add
a
new
like
a
feel
inside
and
inactive
employees
specific
only
fit
now
a
needs-based
control
because
it's
go
through
a
different
API,
which
we
can
now
answer.
D
B
D
B
That's
Daniel,
alright,
so
the
executor
were
joined
in
our
ability
like
that
by
week,
yeah
the
DN
a--
isolator
is
gonna,
put
that
particular
into
that
DN.
I
now
work
and
then
for
all
the
nasty
containing
that
on
that
launched
by
the
executor
issue,
just
share
with
the
parent,
because
that's
a
default
behavior.
We
don't
see
a
use
case
where,
like
we
don't
see
a
use
case
yet
at
like
nasty
container
and
wants
to
have
its
own.
Our
interface
I
mean
we
just
don't
see
that
usage.
B
Yes,
just
we
do
have
a
use
case
for
that.
We
can
consider
make
that
computer,
but
not
right
now,
because
I
mean
set
up
network
is
not
just
like
pokin
namespace.
You
have
to
set
up
some
other
like
a
rally
and
like
and
allocate
IP
for
that.
It's
not
just
simple
as
such
as
Cohen
namespace,
and
if
you,
for
that
case,
guy
more
discussion,
I
do
want
to
hear
company
you
his
program.
Hope
I,
don't
see
that
everyone.
E
F
B
I
think
Pina
isolator
can
do
that
thing.
So
I
think
still
like
it's
not
gonna,
be
part
of
the
booting
field
inside
the
Linux
info,
because
you
need
to
set
the
name
and
network
so
that
the
next
container
enjoyed
that
particular
network,
so
should
be
controlled
by
an
hour
ago.
Rather
than
like
a
pudding
flag.
Because,
looking
in
our
insights
to
the
mission,
we
have
to
figure
out
our
guy.
E
B
Yeah
I
think
like
a
fool
for
things
like
now,
Nate's
based
in
nature,
who
don't
definitely
want,
don't
want
them
to
be
sure,
and
we
don't
want
there
to
be
configured.
Oh
it
doesn't
make.
It
has
to
be
configurable,
I,
think
I'm,
not
sure
like
it
is
there
any
other
namespace
that
we
haven't
have
like
on
this
list.
C
B
I
think
we
cover
everything,
use,
user
nature,
it's
going
to
be
complicated,
so
I,
don't
know
exactly
what
we're
going
to
do
about
using
namespace
it
it's
much
complicated
than
compared
to
other
namespaces,
but
the
rest
I
think
we
are
pretty
clear
what
we
want
to
do
things
like.
Maybe
the
people
didn't
like
the
configuration
bullying
kind
of
info
about
IP
in
a
space,
but
that's
just
as
Gilbert
mention
like
maybe
just
beeped
you
and
now
we're
should
be
controlled
by
the
now
are
in
Co,
the
other
API.
D
B
C
Okay,
yep
Angie
I
have
a
question
so
for
the
for
the
network
and
UTS
namespace,
so
for
the
top-level
executor
continuing
when
the
tally
will
ask
you
to
continue,
share
the
post
network
and
for
now
we
don't
allow
any
mess
to
continue
to
not
using
to
make
them
not
using
the
host
agents
network
namespace
right.
So
now
we
don't
make
it
we
don't
support.
You
know.
B
D
C
C
B
G
A
C
So,
for
now
we
don't
have
any.
We
don't
have
any
sequel,
namespace
isolation.
So
basically
it
means
like
every
containers
seeker
going
to
be
exposed
from
the
agent
level
gonna
be
shared
from
the
agent
and
and
I've
been
nothing
too
much
type
too
much
tie
to
it
to
investigate
the
the
latest
signals
with
support
from
the
creator,
but
from
what
I
learned.
C
B
Hello,
yeah
hearing,
Oh,
Carrie,
okay,
so
I
think
those
illnesses
are
most
likely.
You'll
notice,
a
thing:
growling,
giggles
protective.
If
your
user
program
don't
manipulate
the
group
directly,
most
of
the
applications
should
not
you
to
now
be
able
to
see
manipulates
teacher
anyway.
But
if
you're
running
things
like
the
darker
demon
or
continually
in
projects
that
directly
manipulation
might
have
a
impact,
because
the
root
of
your
seagull
might
change
depends
on
whether
you're
in
a
new
name.
Even
if
it's
or
not
alright,.
A
So
we
don't
have,
we
don't
have
actual
workload
so
manipulate
group
in
the
container,
but
we
do
have
some
kind
of
code
that
read
the
sigil,
be
information,
especially
like
the
CPR
and
stuff
and
smartly
decide.
Member
of
girl,
routines
or
even
concurrency
level
for
gala,
yeah,
absolutely
broken
I.
Think.
B
That
depends
on
I
mean
I
can
barely.
This
is
an
isolator
that
we're
going
to
use.
So
if
you
don't
cover
your
using
that
concern
system
used
isolator,
okay,
but
if
you
do
use
the
isolator
depends
on
the
logic
of
your
application.
With
a
lot
every
application
like
getting
the
figure
is
like
reading
my
clock
self,
see
you
figure
out
the
routes
and
then
get
the
future
of
your
containers.
B
A
C
So
basically
for
the
for
the
broad
IO
for
the
sequel,
I
will
sub
system.
It
can
be.
If
you,
if
you
look
at
the
kernel
dock,
it
is
divided
into
two
parts.
So
one
part
of
the
C
group
it
is
for
the
statistics
for
monitoring
and
another
another
party
spoken
show.
So
user
can
like
ecosan
rail
imitation
number
or
some
other
configuration
into
a
single
file
into
a
bio
bio,
to
sort
of
the
the
broad
aisle
limitation
rate
limitation
and
and
then
in
so
for
now
we
let
me
suppose,
brought
io.
C
We
decide
to
proceed
for
the
statistical
ever
status.
Basically,
it
just
reads
the
the
broad
io
files
from
the
from
the
biases
and
from
the
kernel
file
system
and
then
parse
those
number
each
entry
to
the
photograph
and
expose
that
into
the
resource.
That
is
everything
ty.
We
updated
the
suspected
thing
we
going
to
contain
the
glorious
that
information.
If
we
turn
on
this
isolated
and
and
basically
it
is
divided
into
two
part
for
the
statistic-
one
is
the
cfq
another,
it
is
the
store.
C
Turing
so
depends
on
the
kannada
we
introduce
each
few
for
for
the
for
the
CF
cube
back
or
the
end,
the
tottering
steps,
and-
and
if
you
take
it,
if
you
take
a
look
at
the
Royal
tech
buyers,
they
have
multiple
entries.
Most
of
them
and
different
files
has
different
semantics,
so
some
of
them
gonna
be
just
number
to
subscribe
this
sector
or
time
and
some
of
them
going
to
be
have
multiple
entries
and
each
entry
represent
a
device
and
the
device
is
combined
by
a
major
number
and
minor
number.
C
C
And
we're
going
to
depends
on
each
entry
inside
of
the
broad
aisles
that
fire
within
the
path
then
we're
gonna
clip
each
device
and
with
it
it's
a
device,
start
branding,
sterile
status
time
and
IO
service
I
abide
those
information,
and
for
here,
if
you
look
at
the
value
here,
we
defy
the
value
at
an
operation
operation
with
the
specific
value.
So
we
have
different
operation.
C
So
in
those
single
step
for
I'll
stack,
so
the
operation
can
be
rewrite
sink
or
a
sink,
and
it's
from
the
kernel
implementation.
Here
you
are,
if
you
also
summarize
the
total
number
of
the
October
profile
service,
the
I/o
service
or
the
I/o
bytes,
and
it's
also
similar
to
the
sovereign.
So
it
will
also
come
by
with
the
I/o
service
or
I/o
byte
with
different
operation.
C
F
F
B
F
C
Okay,
sorry
guys
yeah
so
yeah.
This
is
the
photograph
I
I
was
just
describing.
So
we
introduced
the
broad
IO
message
inside
of
the
SICU
info
and
for
now
we
don't
introduce
any
specific
photobook
view
for
the
photo
control
for
the
secret
abroad.
Io
control.
So
we
gonna
you
only
introduce
the
pralaya
status
check,
information
and
the
circus-like,
as
you
guys
can
see.
We
divided
it
into
the
CF
q
and
portaling,
and
we
summarize
the
statistic
by
CF
q,
+
CF
q,
recursive
and
torturing,
and
the
statistics
specifics.
The
for
io
statistics.
C
Resource
statistics,
it
means
every
time
we
update
the
resource
statistics.
If
you
have
will
include
the
floor.
Aisle
information
in
the
end
point,
if
you
have
this
freyal
seacoast
assistant,
enabled
and
I
think
I
think
this
is
the
detail,
because
so
this
adds
to
knock
the
user
facing
control.
So
we've
got
me
going
to
introduce
this
prior
control
in
the
future,
so
this
photograph
is
non-negative
to
be
set
from
the
user
side.
C
C
The
other
is
regular
agent,
rack
and
basically
I
am
just
using
the
say,
Alpine
I'm,
just
using
my
daughter
image,
and
the
comment
is
just
equal
hello
to
a
file.
That's
it
very
simple
and
we're
putting
the
image.
So
we
can
touch,
we
can
patch
the
endpoint,
you
can
touch
the
instead
hit
to
employ
and
then
we
can
see
in
resource
statistic.
We
have
the
iOS
their
information
and
for
the
csq
and
the
and
the
starting
from
the
kernel
dot.
C
Only
if
you
have
the
pro
io
enable
only
one
of
them
going
to
be
specified,
that's
the
kernel
semantics!
So
right
now
we
can
see
we
have
the
totaling
step
recorded
in
our
endpoint.
So
we
include
io
service
pipe
and
io
service.
So
it
represents
how
how
many
bytes
be
right
through
the
file,
and
you
can
see
me
we
also
so
so
from
the
photobook
fear
we
can.
We
can.
C
We
know
that
so,
for
example,
for
the
store,
cooling,
we
will
describe
each
device,
we
will
describe
each
device
and
it's
corresponding
IO
service
and
I
will
bite
in
under
that
device,
but
also
because
in
the
prototype,
if
you
summarize
all
the
device
service
per
IP
and
service
and
IO
service.
So
this
is
the
total
number
for
all
the
devices.
C
C
So,
for
the
future
planned
me
Jason
and
myself
gonna
going
to
work
on
the
gonna
work
on
the
block,
IO
control
support,
and
then
we
need
a
decider
to
introduce
the
user
facing
API
who
allow
you
to
set
the
limitation
array
of
the
I/o
so
with
it
is
not
decide.
Yet-
and
we
will
send
our
decide
that
once
we
once
it
is
purified
either
from
the
working
group
us
from
the
from
from
some
other
factor,
its
many
people
show
interest
on
that,
and
thank
you
guys.
Alright.
D
D
C
C
D
B
Think
we
should
need
to
figure
out
figure
out
this
and
the
document
does.
This
is
a
really
a
shoe
inside
eternal
I'm
show
something
I
think
like
I,
think
I.
Remember
that
guy
that
guy's
from
to
see
yeah.
So
it's
a
goober.
Let's,
let's
take
an
action
item
to
yeah
we
do
we
should
do
our
experiment.
I
think
you
showed
some
reproduce
command.
We
just
rent
out
our
cannot
see
if
it's
an
issue
or
not.
Let's
just
take
an
action
item
object
and
it
just
really
like
a
performance
issue
due
to
internal
limitation.
C
A
So
one
well
one
:
general
comment:
I
have
here
is
I
think
it's
great
to
provide
like
device
level
in
size.
However,
from
the
container
user
perspective,
the
urate
don't
see,
device
number
and
I
don't
think
most
site
operators
are
given
exposing
these
information
to
the
users.
So
what
did
I
mean
if
they
are
going
to
so
I
mean
I?
Think
this
answers
the
question
of
how
much
I
owe
I
owe
traffic
my
container
is
generating,
but
if
they
are
going
to
ask
a
further
question
like
what
do
you
think
what
my
container
is
doing?
A
A
A
For
each
device
to
your
volume,
so,
for
example,
because
I
mean,
from
the
container
perspective
all
this
feeling
with
actually
cell
system
past,
which
a
lot
of
the
mounted
from
different
locations
to
the
container,
they
could
have
the
same
logical
container
path
but
there,
but
in
different
ways
of
long
term,
they
could
be
launching
from
different
bugs
from
different
devices
Oh,
which
has
different
device.
Member
and
different
host
host
path.
B
D
The
furthest
type
to
audiences
for
this
kind
information,
because
yeah
a
platform
operator
audience
we're
very
debonair.
Ops
teams
really
want
to
know
what's
going
on
on
the
host-
and
we
all
have
to
my
house,
and
so
it's
going
the
system
underline
and
user
acts
and
then
being
user
as
themselves
will
know,
and
the
bar
already
those
kind
of
have
slightly
different
requirements
about
how
to
do
projector
information.
C
So
so,
right
now
the
old
isolator
in
in
under
the
signal
system,
each
of
them
can
just
expose
the
rolls
better
from
the
from
the
Seco
files,
so
I
think
for
so
yeah
I
think
it
will
be
more
meaningful
if
it's
so
small
information
like
like
a
device
tab
so
that
the
user
user
can
consume
it
so
yeah,
but
I
don't
know.
Yet
how
should
we
expose
those
information
into
the
matrix?
Maybe
we
need
some
a
separate
monitoring
component,
a
motorized
monitoring
component
I.
Think.
A
The
best
way
to
do
this
is
driving
through
some
kind
of
real
use
case,
see
for
the
map
acres.
Some
real
use
case
that
the
user
container
user
level
really
cares
about
io,
io,
throughput
or
out
logistics
of
their
application
from
the
from
that
point.
If
they
don't
even
have
even
they
don't
even
have
access
to
what
device
number
are
translated
to
how
they
can
still
monitor
the
application
in
a
meaningful
way.
Yeah.
C
B
D
C
So
basically,
I
need
to
test
with
and
I
think
my
my
assumption
is
they
did.
They
have
separate
raw
IO
information
because
they
are
different
device
number
which
combined
of
the
major
and
minor.
So
basically,
we
just
go
to
the
director
and
reach
the
number
from
the
signifier
and
but
but
yeah
I
need
to
confer
with
the
lbm
it's
a
set,
aesthetic
device
and
virtual
device
yeah.