►
From YouTube: Leverage LXD/LXC with Kubernetes [A] - Lin Sun, IBM
Description
Leverage LXD/LXC with Kubernetes [A] - Lin Sun, IBM
Do you want to increase the density of your Kubernetes workers? We have run some experiments with Kubernetes workers in LXC containers managed by LXD. Come to hear our experience and challenges during the experiment!
Lin Sun
Lin currently works in the IBM Cloud Platform CTO office where she evaluates different projects and incubates leveraging open technologies to help set the direction of the IBM Cloud Platform! She loves to play with computers and innovate! She is an IBM Master inventor and currently has over 150 patent applications with the US Patent and Trademark Office (USPTO), of which 44 have been issued to-date.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
A
So
so,
let's
start
so,
as
some
of
you
hopefully
have
seen,
one
of
the
things
that
we
recently
announced
from
IBM
is
support
for
kubernetes
as
part
of
our
public
cloud
platform,
which
is
called
bluemix.
So
as
of
about
a
week
ago,
you
can
go
to
IBM
cloud
and
you
can
create
and
provision
kubernetes
clusters
and
run
container
based
workloads
on
our
cloud.
A
Con
work
within
those
environments
and
we're
delivering
that
through
our
cloud
platform,
which
is
a
bigger
place
that
has
over
160
additional
services
that
you
can
take
advantage
of
from
kubernetes
things
like
Watson
for
do
artificial
intelligence
and
machine
learning,
capabilities
around
video
and
IOT
and
analytics
and
data
all
kind
of
brought
together
on
our
platform.
Now,
if
you
look
at
how
we
put
this
together,
there's
a
couple
of
things:
we've
done,
which
which
are
interesting
first,
is
the
way
we
deploy
kubernetes
in
bluemix.
Is
we
manage
the
masters
for
you?
A
So
as
a
user
creating
a
kubernetes
cluster,
you
simply
tell
us
how
many
worker
nodes
you
want
we
provision
and
manage
those
worker
nodes,
but
we
run
your
masters
for
you
in
in
the
cloud
and
actually
we
use
kubernetes
to
run
kubernetes,
and
so
we
have
a
large
humanities
cluster
that
we
run.
That
holds
all
the
masters,
all
the
SPD's
and
all
the
masters
for
all
of
the
kubernetes
clusters
that
we're
handing
out
to
you
guys
to
use,
and
so
we
can
automate
the
management
of
kubernetes.
A
A
We
recognized
a
lot
of
people
are
either
nude
IBM
cloud
or
just
exploring
public
cloud
generally
and
it's
nice
to
be
able
to
come
to
the
cloud
and
create
an
environment
and
play
with
kubernetes
and
understand
the
technology
without
having
to
fork
out
money
up
front,
and
so
we
offer
a
free
tier
as
part
of
that
service
and
that
free
tier
runs
in
our
account.
It's
very
simple
for
you
to
create,
and
essentially
what
you
get
with,
that
free
tier
is
a
Ingle
node
kubernetes
cluster.
So
you
get
a
master.
A
You
get
one
worker
node
that
worker
node
gives
you
two
CPUs
and
about
four
gigs
of
ram,
and
you
can
use
that
single
node
cluster
to
play
with
your
native
guides
and
deploy
applications.
Of
course,
you
can
create
much
larger
clusters
if
you
want
to
so,
you
can
switch
over
from
the
free
tier
to
a
normal
paid
cluster
and
you
can
make
that
cluster
as
big
as
you
want,
and
so,
as
the
cloud
provider.
A
You
know
me
the
person
providing
kubernetes
as
a
service
on
the
cloud,
obviously
there's
a
cost
associated
with
providing
those
three
tiers,
and
so
we
wanted
to
do
some
work
to
look
at
well.
How
can
we
run
that
free
infrastructure
as
efficiently
as
possible?
In
other
words,
how
can
we
drive
up
the
resource
utilization
and
the
density
of
that
environment?
A
So
we
can
offer
it
to
as
many
people
as
possible
to
give
you
a
good
experience,
experimenting
with
kubernetes
without
having
a
very
large
infrastructure
footprint
sitting
behind
that,
and
today
we
do
that
through
a
single
VM.
So
everyone
gets
essentially
a
VM
that
is
their
worker
node,
and
what
we
wanted
to
explore
was.
A
Could
we
run
kubernetes
workers
for
different
tenants
for
different
users
of
our
cloud
inside
of
system
containers
on
shared
hosts,
so
that
we
could
over
commit
memory
and
CPU
on
those
hosts
to
allow
more
for
each
year
at
higher
density,
and
that's
the
experiment
that
Linda
team
ran
so
I'm
going
to
turn
the
time
over
to
her
and
have
her
kind
of
take
you
through
the
work
that
we
did?
What.
B
Thank
you,
Jason.
Oh
it's
odd,
to
hold
a
speaker.
I
feel
a
little
bit
interesting
when
I
go
to
show
a
demo,
surely
with
holding
the
speaker,
so
one
of
the
key
things
about
the
free
tier
I
really
actually
love
all
of
it
here,
because
it's
like
simple
1,
2,
3
4
clicks
you
kind
of
boom.
You
have
your
freaky
a
cluster
running.
You
only
see
its
kind
of
take
a
while
it
does
take
a
while
to
launch
a
VM
provision
of
VM.
B
So
that's
why
we're
looking
at
this
minimal
cost,
with
a
faster
to
be
able
to
send
up
a
cluster
to
be
able
to
faster
to
tiada?
Also
because
we
have
a
page
here,
we
are
always
take
into
consideration
of
the
migration
scenario
from
the
fruit
here
to
the
plate.
Here
we
definitely
want
to
be
a
very
seamless
experience
in
between
so
the
Jason
kind
of
talked
about
this.
So
it's
really
about
the
density
for
the
fruit.
B
Here
we
want
to
be
able
to
using
memory
as
an
example,
for
example,
with
the
for
coil
eight
core,
eight
gig
of
memory.
We
want
to
be
able
to
run
more
fruit
here.
You
know
get
more
bang
out
of
the
money
we
want
to
reduce
the
cost
we
want
really
to
be
able
to
deployment
faster.
So
you
guys
can
probably
tell
from
over
demo
today
or
freakier,
does
take
five
to
ten
minutes
to
stand
up
because
of
the
VM
provisioning
instead
of
the
cube
master
and
the
worker,
and
all
that
also
be
able
to.
B
We
also
realize
a
very
important
characteristic
of
a
web
feed
here
is
manual
for
freed.
Here
are
actually
I,
though
so
freaky
our
customer
really
love
them,
because
they
can
click
on
a
couple
of
button.
They
get
their
cluster
up
running.
They
can
run
real
popular
example
of
guestbook
and
then
have
it
running.
It's
like
a
fully
blown
Cuban,
any
environment
with
all
our
capabilities
building,
but
then
customers
interested
they
can
change,
and
sometimes
they
sting
up
the
cluster.
They
don't
really
run
the
cluster
with
it
or
they
forgot
their
cluster.
B
You
know
many
of
the
free
tier
cluster
are
actually
I,
don't
out
there.
So
that's
a
very
important
characteristic,
so
we
we've
done
a
couple
of
experiments.
The
first
experiments
we
actually
done
with
we
want
to
run
cuban
ad
workers
in
docker
containers,
so
Jason
talked
about
our
architecture
with
master
being
deployed
on
t
cuban
at
ease.
B
Well,
the
workers
are
being
running
on
the
vm,
so
we
sort
of
a
natural
transition
was
to
be
able
to
run
cuban
ID
workers
in
docker
containers
using
darker
as
the
way
to
package
cuban
ad
workers
and
be
able
to
run
that,
and
it
was.
Secondly,
experimental
was
why
don't
we
just
run
the
whole
cuban
at
ease?
We
were
you
using
Linux
containers
used
by
leverage.
The
next
container
provide
the
isolation
we
needed
offer
for
each
of
the
tenant.
That's
been
using
all
of
Phrygia.
The
third
experiment
to
be
wrong
is
wrong.
B
Humanity
work
just
the
worker
piece
in
LXE
containers
where
the
worker
are
the
master,
still
being
deployed
as
the
rest
of
the
the
rest
of
the
container
service
on
bluemix.
So
that
is
the
work.
The
master
will
be
delivered
deployed.
Aren't
you
the
cuban
antes
that
runs
all
the
rest
of
the
masters?
So
let
me
take
you
through
on
the
first
of
the
experimental
we
run,
so
we
actually
it's
funny.
We
actually
got
it
working.
It
was
a
very
smooth
right
floors.
B
We
were
able
to
run
back
to
cuba,
nary
workers
as
docker
image
and
be
able
to
write
as
docker
containers
and
be
able
to
talk
to
you
the
same
master
that
we
we
using
IBM
container
service
in
bluemix,
but
immediately
we
realize
a
big
issue
with
docker
containers
because
they
actually
recall
the
containers
are
privileged
mode.
Some
of
the
cuban
a
new
worker
process
like
flannel,
for
example,
I-
think
the
queue
proxy
is
another
example.
They
recall
the
privilege
mode,
given
the
nature
for
a
freed
here.
B
We
really
want
each
tenant
to
be
relative
isolated,
so
we
don't
really
want
to
give
any
one
of
the
tenants.
You
have
like
privileged
root
access
and
aren't
you
every
single,
every
single
other
docker
container
running
on
the
same
VM,
so
we
can
secure
that
experimental
right
away.
The
the
second
thing
we
talked
about
earlier
is,
we
said
you
look
at
our
Linux
container
and
then
we
find
that's
right
around
the
time
when
Linux
container
also
have
container
hypervisor
release
the
lxd.
B
I
believe
it's
probably
on
version
2.5,
which
point
six
right
now,
so
it
really
provide
a
new
experience
for
lxc
I.
Don't
know
if
you
guys,
you
guys,
probably
familiar
with
LXE
one
thing
I
noticed
with
LXE.
Is
it's
a
little
bit
harder
to
interact
with
with
other
users?
I
feel.
That's
also
why
darker
has
gained
a
lot
of
momentum
by
leveraging
the
Linux
container
technology
made.
It
really
simple.
B
The
way,
interesting
aspect
when
I
start
to
explore
with
LX
b
and
LX
c
is
lxd,
provided
a
key
components,
two
key
components:
why
is
the
demon
that
helps
you
to
be
able
to
convict
the
hypervisor?
The
other
thing
is
the
LXE
client,
which
really
made
it
easy,
as
a
user
for
me
to
interact
with
the
LXE
container,
so
I
can
create
the
LXE
container.
I
can
do
different
management
capability
around
it
very
simple,
and
so
what
I
need
to
talk
through
that
Innova
demo?
B
To
this
is
not
a
rewrite
of
LXE,
it's
laid
by
a
company
called
Cano
Nia,
so
I
think
they
did
a
really
interesting
job,
make
it
available
and
usually
consumed
by
the
user.
You
may
be
wondering
you
know:
what's
the
difference
between
soccer
and
lxd
right,
so
I
would
say
Doc's
really
specialized
with
deploy
applications,
so
you
can
usually
package
that
your
application
lets
docker
images
and
move
it
to
any
other
world.
B
Well,
darker
runs
in
your
clotting
any
of
the
cloud
or
local
environment.
The
lxd
specialize
in
focus
on
deploy,
Linux
virtual
machine,
which
really
serves
our
purpose,
because
it's
provide
a
lightweight,
are
highly
easy
to
reuse,
the
resources
onto
the
VM,
and
it
serves
really
well
purposeful.
Afriqiyah
an
LXE
can
be
quickly
launched
quickly
destroyed
so
on
the
second
experimental
we're
always
running
cuban
eddying
LXE
containers.
B
So
what
we
were
thinking
is
we
want
to
be
able
just
package,
the
manager
and
the
worker
in
long
privilege,
the
LXE
container,
so
the
whole
container,
the
self-contained
and
each
of
the
wealth
free
tier
user.
They
would
each
get
their
own
containers
and
then
they
would
just
use
interacted
with
the
cuban
at
ease
through
cube
CPR
or
the
UI
of
the
cuban
at
ease.
B
So
one
thing
we
quickly
discover,
though,
is
when
you
run
long
privilege.
Alexey
continues.
We've
had
a
lot
of
issues
running
darker
in
privilege
mode.
As
I
was
mentioned
early,
the
cuban
ad
workers,
the
cube
proxy,
recalled
the
privilege
mode
panel
also
recall
privilege
mode.
So
what
we
decided
to
do
is
instead
of
running
the
traditional
way
people
run
cuban
at
ease
running
as
containers
docker
containers.
B
We
decided
we're
going
to
round
the
cube
processes
such
as
the
cubelet
and
the
cube
proxy
directly
on
the
LXE,
so
that
kind
of
help
has
been
immolate
the
problem
with
running
these
as
darker
containers
and
on
top
of
the
LXE
containers.
So
we
were
able
to
walk
around
that
problem.
We
also
reached
a
couple
of
issues
in
the
community.
I
have
a
summary
chart
that
shows
some
of
the
issues
we
find
out
as
we
went
through
this
journal
be
able
to
run
the
cuban
any
processes
directly
on
our
XP.
B
There
was
some
configuration
change
we
had
to
make
in
cuban
aries
was
related
to
out
of
memory
school
and
cuban,
and
he
doesn't
expose
those
configurations,
so
we're
being
opening
pool
requires
working
with
the
community
on
those.
So
one
thing
about
this
experiments
we've
been
happy
with
the
experiments.
However,
we
talked
about
the
migration
scenario.
For
what
purpose
we
want
user
to
be
able
to
easily
migrate
from
the
freakier
to
the
paid
here.
B
Having
cuban,
adding
master
and
worker
running
in
its
own,
a
lie
see
would
mean
at
the
end
of
the
fridge
here
we
would
destroy
the
the
LXE
container,
which
would
destroy
their
master,
destroy
their
worker
and
then,
when
we
move
when
the
user
move
them
to
the
page
here
they
would
kind
of
lose
their
master.
So
it's
kind
of
a
different
end
point
for
the
user
to
point
you.
So
that's
one
thing
you
know
we
were
like
well.
B
Can
we
push
it
a
little
bit
further
which
lends
to
the
next
experiments
we
have
so
we're
going
to
talk
a
little
bit
in
detail
onto
this
one,
because
this
one
we
really
spend
most
of
our
time
on
that
is
we
run
Cuban
ad
workers
in
the
LXE
container.
So
on
this
diagram
you
can
see
we
kind
of
have
so
it's
we
kind
of
have
the
Cuban
edgy
machine,
that's
dedicated
to
run.
B
The
master
is
running
on
the
VM,
so
it
has
like
multiple
master
deployed
as
deployments
running
our
rescue
Bonetti
deployments
on
on
top
of
cuban
ad.
So
we
rely
on
cuban
Eddie
to
help
us
to
manage
those
Cuban
ad
masters
and
then
on.
The
lxd
host,
which
we
also
are
only
experimental
with
the
V
virtual
machine,
is
on
this
host.
We
are
launching
multiple
Alexi
containers
within
each
of
the
LXE
containers,
so
notice
we
mentioned
earlier
we're
running
cubelet
and
the
cube
plus
proxy
as
processes
directly
on
the
Linux
containers,
and
we
also
run
docker.
B
But
these
two
are
not
running
I
stock
a
container,
so
we
also
run
the
auto
which
helps
when
the
cubes
need
to
request
comes
in
to
be
able
to
deploy.
Paz
I
know
that
so
basically,
these
are
cubelet.
Amplitube
proxy
would
be
read
to
back
to
the
master,
so
that's
providing
a
functional
stack
of
cuban
at
ease
with
the
workers
running
our
LXE
containers.
B
So
this
helps
us
with
the
migration
scenario
we
talked
about
earlier,
because
if
users
are
happy
with
our
free
here,
they
tried
and
if
they
are
happy
at
the
end
of
the
day
when
they
decide
they
are
going
to
migrate.
What
is
going
to
shut
down
their
Linux
container
for
this
suite
here
and
then
they
can
provision
the
page
here
like
this
and
mention
their
own
accounts,
so
they'll
be
running
the
Kuban
any
worker
in
their
own
account
and
point
you,
the
same
master.
B
So
this
is
just
a
couple
of
screenshots
on.
Let
me
highlight
so.
This
is
like
the
on
the
left
side
is
the
master
machine
where
we
run
a
bunch
of
masters,
as
you
can
see,
so
all
the
Masters
are
listed
and
they
are
exposed
as
Cuban
IT
services,
with
no
ports
to
be
able
to
provide
access
for
simplicity's.
It
can
certainly
be
on
an
Opel.
It
could
use
ingress
or
no
balance
or
service
type,
but
for
the
simplicity
of
the
experimental
we
just
choose
our
no
port.
B
You
experience
it
and
on
the
right
side,
you
can
see
on
my
my
lxd
host
machine
where
I
have
a
bunch
of
Linux
containers
running
those
are
as
Cuba
native
workers
and
on
this
one.
So
this
is
so.
This
is
trying
to
show
the
information
of
the
Linux
container.
So
this
is
one
of
the
Cuba
native
workers.
You
can
see
the
memories
or
I
didn't
bring
a
point
for
some
pointing
at
this.
Now
we
can
see
the
memory.
This
is
right.
B
After
the
the
Cuban
era,
worker
has
started
on
the
Linux
container,
so
it's
about
a
less
early,
less
than
100
max.
So
the
memory
is
not
intensive
at
all
and
then
on
this
one
is
right
after
it's
right
after
we
deploy
one
of
our
guestbook
example,
by
the
way
how
many
of
you
have
deployed
guestbook
in
Cuba,
Nerys,
okay,
very
good.
B
Thank
you
guys!
That's
a
big
rescue
wait,
turn
it
off
so
that
we
don't
get
an
egg
coast.
So
let
me
get
rich
get
out
of
the
keynote
sorry
I
being
recently
converted
to
use
keynote.
So
this
you
can
see
I've
been
having
trouble.
Okay,
so
now
I,
finally
get
myself
out
of
the
keynote,
so
I
have
a
cup
of
machines.
So
let
me
let
me
make
sure
I
mentioned
you
use
some
of
my
machine,
so
my
Linux
container
fault
is
183.
B
So
I'm
going
to
you
what
I'm
going
to
do.
Can
you
guys
see
my
screen?
This
is
bad
enough.
Okay,
so
what
I'm
going
to
do
is
as
a
search
into
into
the
machine
and
I
also
have
the
cuban
le
masters
1:00
1:30.
So
what
we
are
going
to
do
is
as
such
into
that
machine
also
and
while
we
are
waking,
let's
queue
of
this
command
to
apply
the
way.
This
is
the
LXE
CLI
we
mentioned
earlier.
One
thing:
what
you
noted
is
they
have
a
lot
of
commands.
B
So
we're
going
to
use
the
list
which
can
show
you
how
many
of
workers
I
already
deployed
you
can
see.
I
have
a
few
deployed.
My
machine
is
kind
of
a
little
bit,
not
a
memory,
but
storage.
Wise
is
a
little
bit
tight.
Also
I
can
properly
deploy
a
little
bit
more,
but
so
far
I
have
three
running,
and
let
me
get
on
to
this
is
the
queue
Bonetti
machine
that
runs
all
the
Masters.
As
you
can
see,
we
got
a
bunch
of
masters
running
like
I
was
mentioned
earlier.
B
B
B
Yeah
looks
like
man
working
is
just
hanging
out
for
me
for
a
little
bit,
I've
heard
early.
The
networking
idol
a
boost
was
not
great,
so
hopefully
it's
back
now,
so
you
can
see
my
work
at
83
is
rally,
so
what
I'm
going
to
do
is
going
to
expect
into
the
workers
383
and
draw
the
batch
command.
So
this
essentially
gives
me
the
same.
Prompts
here
except
this
prompt
died
because
I
had
a
networking
issue,
so
you
can
see
I
have
the
star
Walker
command.
B
B
B
B
All
right,
so
you
can
see
my
guestbook
is
deployed
and
it's
probably
still
standing
up.
So
we
can
get
a
status
on
this
so
notice
how
faster
we
were
able
to
ping
up
a
Linux
King
container
that
runs
Cuban
any
water.
Even
though
we
had
a
little
bit
networking
issue,
but
you
can
see
it's
like
within
a
minute
boom,
your
prospers
ready
and
then
you
have
a
whole
cluster
to
interact
with
to
deploy
our
guestbook
I,
didn't
see
a
lot
of
hands
raised
to
focus
books
so
well,
the
guest
book
is
deployed.
B
Let
me
just
quickly
show
you
the
yellow
file,
so
essentially
the
first
one
we
deploy
that
has
a
reddit
master,
it's
a
service
and
deployment,
and
it's
fully
in
this
particular
reddit
image,
and
it's
asking
for
100
Meg
for
memory
and
the
guestbook
llamó
file
also
has
the
rabbit
slave
in
there,
which
is
also
deployed
as
as
a
service.
It
also
has
the
front
end,
which
is
we
are
going
to
access
to
a
guestbook
and
notice.
B
It's
deployed
as
the
note
port,
so
we're
going
to
access
through
whatever
no
fault,
Cuban
area
assigned
to
us
and
I
guess
we
have
three
for
the
front
end.
So
let's
go
ahead
and
check
our
status
of
the
guestbook,
and
now
you
can
see
the
front
end
is
three
which
is
nice,
so
we
talked
about.
We
are
using
no
port,
so
let's
go
ahead
and
get
the
note
put
port
number.
So
so
the
number
is
three
to
five
five,
seven.
B
B
Which
would
show
us
the
IP
address,
and
so
we,
essentially
you
can
see
here
from
lxc
in
list
command.
You
can
see
what
is
the
IP
of
fine
283
with
the
walker
we
just
deployed
and
we
just
have
our
guestbook
running
and
you
can
see
it's
mapped
to
this
guy
here.
So
let's
go
ahead
and
the
launch
of
our
Safari
and
I
have
to
go
back
to
find
out
the
number
of
the
pop
no
portals
lines.
So
it
consumes
a
different
number,
every
single
time.
B
So
you
can
see
this
is
a
guestbook
run
in
fully
deploy
with
Cuban
Eddie
and
the
Masters
manage
the
young
Cuban
Airy
machine
separate.
So
it's
kind
of
little
ago
for
experimental.
Well,
we
want
to
be
able
to
have
the
fruit
here
to
be
able
to
quickly
stand
up
cuban
any
worker
have
a
working
cuban,
any
environment
for
the
user
to
easily
interact
with
and
be
able
to
provide
higher
density
and
be
able
to
fast
migrate
to
the
paid
here.
So
let
me
continue
with
our
chart
here.
B
B
Okay,
so
whispers
we
actually
trying
to
capture
a
little
bit
statistics
or
current
freakier
Jason
mentioned,
but
quickly
on
to
poor
and
full
gigabyte
memory.
So
everybody
comes
to
your
cloud
with
the
fridge
here
we
kind
of
forgive
them
this
much
resources
that
it
need
to
be,
and
nobody
else
can
use
it,
but
with
the
way
experiment
you
can
see,
I
actually
run
the
experiment
on
a
pretty
small
machine,
so
I
write
with
a
core
and
eight
gigabytes
of
memory.
B
That's
my
lxd
host!
So
each
of
the
oxp
are
with
cuban
energy
worker
running
we
measured
100
max
page,
it's
the
average
with
140
is
the
peak
and
if
we
run
a
guest
book,
unfortunately,
because
we
have
like
multiple
replicas
of
the
guest
book,
it
has
a
couple
of
services.
So
it's
an
interesting
application.
B
So
it
does
take
a
little
bit
more
memory,
but
at
the
peak
time,
which
happens
about
the
time
when
you
deploy
a
guest
book,
it's
about
1.5
gigs
for
memory
consumption,
but
at
every
judge
thing
is
just
don't
you
like
300
and
800
minutes.
So
it's
a
it's
a
good.
It's
a
good
example
that
you
can
actually
show
it's
running
but
not
consume
a
huge
load
of
memory,
so
Linux
container
also
XP
supports
this
hard
limit
by
memory.
We
haven't
had
a
lot
of
opportunity
to
explore,
but
it
sounds
really
interesting.
B
B
So
that's
kind
of
our
what
you
do
list
to
be
able
to
explore
a
little
bit
on
that
we
can,
based
on
the
memory
consumption
on
my
LSD
how's,
that
we
should
be
able
to
run
turn
Cuban
Artie
Aleksey
workers,
assuming
20%
of
our
workers,
are
highly
used
and
the
rest
of
them
are
idle,
which
matches
a
lot
of
our
three
tier
that's
out
there
today,
and
so
this
is
a
list
of
the
issues
we
mentioned
earlier.
We
had
issues
running
with
privilege,
docker
containers
in
LXE.
B
It's
not
fully
working
and
we
are
streamed
always
with
requests
coming,
and
we
had
the
out
of
memories
for
adjustments
as
humanity
hasn't
been
able
to
allow
us
to
provide
a
configuration
out
of
the
box.
So
we
end
up
building
a
Cuban
ID,
our
imaging
solar
cell
for
consumption
and
also
with
Rossi.
That's
why,
in
experimental
you
saw
that
we
actually
push
Cuban
and
he
and
doctor
into
the
Linux
container,
because
we
had
a
little
customization
building
on
the
summary
I
would
say
the
experimental
is
really
interesting.
B
We
thought
it
was
amazing
that
you
we
were
able
to
launch
a
Cuban
Airy
worker,
really
fast,
having
a
functional
stacks,
be
able
to
work
in
and
be
able
to
match
into
the
architecture
we
landed
in
bluemix
to
be
able
to
have
fully
managed
master
and
be
able
to
quickly
migrate
form
the
three
tier
to
the
paid
here.
We
certainly
have
a
lot
of
things
on
our
wishes,
as
always
with
not
enough
time
during
the
day.
So
we
want
to
be
able
to
explore
the
CPU
and
memory
limits
we
mentioned
earlier.
B
We
also
have
a
snapshot,
feature
and
copy
feature
that
comes
with
lxd,
where
you
can
manage
your
Linux
container
much
faster,
so
we
thought
that
was
interesting
in
our
demo.
We
didn't
use
the
US
for
service
clock
up.
We
actually
use
the
environment
for
service
look
up
in
the
guest
book,
so
we
would
like
to
try
the
u.s.
having
the
DNS
all
LG
kubernetes.
We
also
wants
to
play
a
little
more
with
keys
and
certs,
be
able
to
config
that
drawing
so
I
think.
That's
the
end.
Any
questions.
B
B
So
we
didn't
draw
master
kan
components
in
the
Alex
Alex
II
container,
along
with
the
Cuban
ad
workers,
and
that
experiment
went
really
well
too,
but
one
thing
we
did.
We
want
to
be
consistent
with
our
offering
in
bluemix.
Well
we're
on
master,
as
on
top
of
the
cuba
narrative
and
provide
management
capability
around
the
master.
So
we
want
to
provide
a
smooth
migration
scenarios
to
our
user
to
so
that
you
can
always
point
to
a
single
master,
whether
you
are
in
the
freaky
or
the
tapir.
Does
that
make
sense?
Yeah?
C
B
D
B
A
Mean
the
only
thing
I
would
add
is
is
in
the
the
paid
structure,
absolutely
you,
as
the
user
are
paying
for
those
resources
by
the
hour.
So
this
kind
of
desire
to
drive
up
density
and
do
over
commit
is
not
something
you
would
want
when
you're
paying
for
those
resources
right
so
in
the
paid,
clear
paid
clusters.
Architecture
is
the
same,
but
you've
paid
for
those
cpu
memory
resources,
so
they're
dedicated
to
you
and
the
free
tier
I'm
paying
for
it,
and
so
we
ran
this
experiment.
A
A
Yeah
we
haven't,
we
haven't
done
that
kind
of
scale.
Testing
directly
on
the
shadow
mean
it's
fundamentally
going
to
be
limited
by
memory
right,
and
you
know
what
the
memory
footprint
of
those
individual
containers
or
pods
are
so
I.
Think
that's
going
to
be
the
limiting
factor,
especially
in
the
single
node
case,
you're,
just
going
to
effectively
run
out
of
memory
first
before
anything
else
in
the
system
starts
to
break.
A
In
general,
in
most
of
real
world
scenarios
on
a
single
node,
we
wind
up
not
hitting
those
kinds
of
limits
directly
on
the
number
of
containers.
This
is
not
you
know
by
the
time
you
take
to
CPU
cores
and
you
split
it
up
a
thousand
ways:
you're
not
doing
very
useful
work
anymore,
so
it's
more
theoretical
than
practical,
all
right,
all
right,
we're
out
of
time.
Thank
you
very
much
appreciate
it.