►
From YouTube: Kubernetes SIG Node 20200514
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
A
We're
trying
to
do
some
a
little
bit
different
here
to
our
cadence,
the
meaning
is
recorded
for
those
who
can
attend,
which
I
know
is
a
not
a
perfect
time.
I
would
look
to
get
the
according
recording
up
later
this
afternoon
and
hopefully
will
use
this
forum
as
a
way
to
more
quickly
triage
new
things
can
I
hand
it
over
to
Victor.
Who
can
talk
through
how
we
may
be
plotted
out
and
play
that
make
this
work
and
then
we'll
go
through
the
first
presenter.
So
go.
B
Ahead
doctor
all
right:
thanks,
Derek,
hello,
everybody,
yes,
so
this
is
gonna,
be
just
a
you
know:
temporary
set
of
meetings
and
we're
gonna
go
through
some
of
the
caps
in
detail.
I.
Think
today.
The
first
item
that
we
have
is
the
memory
manager
cap.
Let
me
see
if
I
can
pull
that
up
and
share
it
and
we'll
get
started.
A
B
Okay,
can
you
guys
see
what
I'm
sharing
yes?
Okay?
Yes,
so
the
let's
go,
we're
sort
of
the
guideline
so
we'll
use
this.
You
know
over
the
next
few
weeks
to
share
the
idea
as
an
approach
that
we're
gonna
look
at
the
different
caps
and
take
them
back
to
sig
note
so
prior
to
each
meeting.
What
we're
gonna
do
is
look
at
the
agenda
and,
if
there's
no
agenda
before
the
meeting,
we
will
cancel
it
so
the
again
this
is
just
intended
to
be
a
temporary
set
of
meetings.
B
It's
not
an
ongoing
long
series,
and
so
that's
that's.
Why
sort
of
the
ground
rules
are
set
now
at
the
end
of
each
meeting,
we'll
also
look
and
see:
have
we
triaged
everything,
that's
open
or
their
other
proposals,
other
topics
to
sort
of
decide
if
we
need
to
meet
again
and
if
we
do
need
to
meet
again
is
a
week
too
far
out
do
we
need
to
meet
sooner
and
if
there's
no
agenda
for
two
consecutive
meetings
will
cancel
the
form
and
just
go
back
to
the
regular
signal
meetings
and
again
this.
B
The
intent
of
this
meeting
is
to
have
a
higher
faster
velocity
to
look
at
new
caps,
new
ideas
and
not
sort
of
overwhelmed
the
signaled
meeting
with
with
these
in
depth
discussions,
and
it
gives
us
more
time
to
talk
about
these
topics,
sort
of
the
organizer
so
far,
it'll
be
meeting
myself
and
Derek.
J
is
volunteered
to
help
out,
and
so
it's
Kevin
if
there
are
others
that
want
to
help
out.
You
know
please
speak
up,
and
you
know
we
can
add
you
to
the
list.
Are
they
any
questions?
Sort
up
about
the?
B
Yes,
so
I
had
a
chat
with
there.
You
know
we
what
what
Kevin
is
asking
about.
We
created
a
apology,
aware:
scheduler
channel
I
think
it
was
last
week
sometime
at
the
at
that
time.
It
was
more
thinking
about
you
know,
there's
a
lot
of
discussion
and
interest
and
talk
about
Numa,
we're
scheduling,
and
so,
when
we
created
the
channel
that
was
sort
of
the
name
that
was
there
and
I.
Think
what
Kevin's
saying
hey.
Does
it
make
sense
to
have
these
this
topic
here
on
that
channel?
B
Should
we
rename
the
channel
or
should
we
just
stay
in
sig
note
and
I
had
a
brief
chat
with
Derek
and
the
consensus
there
was
to
keep
it
in
sig
note.
If
that's
okay
with
everyone
we'll
continue
with
that
and
I
guess
we
can
keep
the
topology
aware
channel.
Just
for
that
discussion,
which
the
topology
aware
channel
may
involve
other
SIG's.
Also
like
six
scheduling
and
I
think
the
idea
was
to
keep
that
channel
there
and
just
use
sig
note
for
this
particular
discussion.
B
C
C
A
Some
other
context
was
those
in
the
past.
It
doesn't
matter
like
a
working
group
has
like
a
formal
meaning
and
Carini's
govern
instead,
like
they
supposed
to
be
time
down
and
spans
SIG's
and
doesn't
necessarily
own
code,
but
maybe
some
recommendations.
The
intent
here
was
not
to
go
and
build
a
new
mini
month,
working
group
effort
right
now.
Maybe
the
output
of
this
will
be
hey.
We
should
do
that.
B
C
Have
one
question
before
we
jump
into
this:
do
we
can
you
go
back
to
the
other
document?
I,
don't
think
it's
there
yet,
but
it'd
be
nice
to
get
a
kind
of
full
list
of
all
the
things
we
already
know
about
that
we
wanted
to
discuss
here.
Like
directors
mentioned,
you
know,
there's
a
long
list
of
things,
people
presenting
about
a
bunch
of
stuff
in
sig
note
there
were
never
enough
time
to
dedicate
to
it.
So
it'd
be
good
to
get
that
list
here.
C
B
If,
maybe
even
during
the
call
of
some
folks
already
know
of
stuff
and
I
know,
there's
like
the
well
there's
the
pod
level
single
Numa
node.
If
you
have
a
future
topic
that
you
want
to
talk
about,
maybe
you
can
just
add
it
here
and
we
can
have
a
heads
up
and
maybe
look
at
it
ahead
of
time
and
work
it
into
the
schedule.
Yeah.
C
A
It'll
take
Alex's
presentation
yesterday.
I
think
my
hope
is
that
it
will
look
at
all
the
ideas
and
fatality
and
just
having
a
record
of
things
that
might
have
been
related
in
the
list
sounds
good,
so
yeah
if
Alex
was,
wants
been
a
pointer
to
a
stock
there
from
yesterday.
That
would
be
awesome.
Okay,.
B
D
I
think
this
presentation
will
take
like
20
minutes
so
memory
manager
is
a
new
component
in
that
ecosystem
devoted
to
enable
cross
among
their
auntie
to
memorial,
location
for
parts.
You
know
guaranteed
QoS
class,
and
this
is
the
thing
that
it
was
working
conducted
much
and
yeah
I.
Should
we
should
find
as
many
thanks
to
our
optimal
for
implementing
POC
before
concept?
Okay,
we
can
go
to
agenda
mm-hmm.
D
So,
yes,
that's
that's
agenda
will
present
of
our
role
and
motivation,
then
concept
and
design,
and
here
we
will
focus
on
customer
customer
the
design,
because
in
fact
it
was
out
of
them.
The
most
interest
and
many
people
were
asking
about
this
design.
Then
you
just
go
into
details
of
the
implementation,
and
time
show
also
how
to
generate
to
generate
customer
hints
by
by
some
examples,
then
just
some
variants.
D
So
just
recall,
because
I
think
many
people
are
familiar
with
the
goal
of
our
alignment
just
to
just
to
recall,
so
the
overall
goal
is
to
achieve
continued
resources
across
multiple
Numa
notes.
So
the
encompasses
various
resources,
such
CPU
memory,
huge
values
and
devices
and
in
this
presentation
will
focus
on
you
know,
I.
Remember
a
alignment
also
across
multiple
nodes.
D
D
So
motivation,
motivation:
this
is
justin,
except
from
reference
number
one,
the
reference
number
one.
It
might
be
a
little
bit
outdated.
So
so
I
in
fact
I
will
be
working
still
on
the
cap.
The
two
to
enhance
its
ability
to
readability,
so
yeah
cap
will
be
the
most
up
to
date,
resource
for
for
form
or
another
okay.
So,
back
to
to
the
motivation,
I
will
just
highlight
this
three
point,
so
it
is
for
for
many
important
use.
D
D
Okay,
just
concept
InDesign,
so
our
initial
concept
was
limited
to
the
single
number
policy
and,
in
fact,
to
the
memory
consumption
from
a
single
source.
That
is
single
note
and
naturally
the
question
will
arise.
What?
If
a
pot
requires
formal
Morgan?
In
fact
a
single
luminoled
can
provide,
and
this
gave
motivation
for
crossing
Memorial
patient.
D
E
D
D
E
D
D
D
So
we
we
treat
a
group
of
Numa
nodes
as
a
unit
and
we
configure
configure
such
such
such
units
through
a
new
cube,
light
flag.
This,
the
name
of
the
flag,
is
customer
groups
right
and
yeah.
Could
we
go
to
the
next
slide
because
in
the
next
slide,
I
think
this?
This
is
more
apparent
example
with
more
than
one
node.
D
So
what
will
be
the
effect
of
such
partitioning
in
into
into
sync
groups
and
customer
groups?
So
the
effect
is
that
group
are
zero
and
two
free
will
be
used
for
reports.
That's
require
single
meditation
and
the
group
want
to
will
be
used
for
pods
that
both
are
not
limited
to
single
medication
and
have
higher
memory
requirements
beyond
single
numeric
loops.
C
D
D
D
D
Ok,
so
there
is
another
part,
because
the
groups
must
be
disjoint,
because
if
the
groups
of
a
wrap,
this
will
not
ensure
that
memory,
location
and
Houston
explanation.
If
a
pair
of
groups
overlap,
the
more
or
less
doses
of
some
group
can
be
pre
or
constant
by
another.
Go
so
that
the
other
group
will
starve.
D
D
So
the
first,
my
Heinz
Miss
Potter
fiction
is
done
by
cubelet
and
I
will
just
read
a
cubed
can
evict
a
number
of
spots
and
reclaim
memory
after
a
fiction.
Threshold
is
exceeded
by
containers
and
in
the
context
of
memoriam
manager.
We
use
a
heart,
a
fiction
threshold
because
indeed
design
memory
manager.
D
Future,
which,
in
fact
involves
also
this
distort
okay.
We
can
go
to
the
next
slide
to
the
I
think
the
second
behind
is
out
of
memory
killer,
which
is
done
by
terminal
project
system,
and
here
what
should
be
taught
is
that
a
lot
of
logic,
killer,
kills
processes
with
the
accordance
to
out
of
memory,
score,
ranking
and
low
priority
content.
D
E
E
We
are
losing
this
field
in
welcoming
our
Shara
interfaces,
so
the
couplet
knows
about
it,
but
it's
not
propagating
this
request.
Memory
down
was
stuck
so
what
can
help
Colonel
to
reallocate
or
expect
what
process
will
collect
or
use
some
memory?
That's
one
scenario
second
scenario:
instead
of
like
a
weekly
and
like
Keeling
with
routers,
even
well,
like
you
best
afford
ones
like
which
we
consider
other
lowest
priority,
where
our
possibilities
in
Linux
kernel
to
migrate
pages
forcibly.
E
So
if
your
route,
if
you
have
assisted
me
in
privileges,
you
can
say
move
pages
from
for
this
particular
process
from
out
of
this
memory,
node
to
another
one,
it
might
not
fit
into
current
apology
manager
a
year.
We
do
everything
on
the
admission,
but
in
general,
like
this
dynamic
reallocation
of
workloads,
it's
possible
grab
more
than
three
ways
of
solving
eyes.
E
D
E
Possible
in
just
a
bit
more
work
and
a
bit
more
extending
to
interfaces
in
which
your
eye
level,
so
you
know
CIA
level
like
down
from
Warren,
C
or
compatible
runtimes.
We
can
get
that
information
and
actually
from
sig
routes,
version
one
we
can
also
get
for
each
camping
or
how
much
memory
gives
a
very
inch
nap,
our
nominal.
So
you
you
can
get
to
some
areas
where
they
can
dynamically
Schaaf
into
a
part
of
a
memory
foam
or
for
particular
container.
E
D
Okay,
just
back
to
the
presentation,
it's
just
a
record.
How
do
how
the
world
looks
like
without
more
manager,
so
the
commissary
that
we
bet
when
it
wants
to
admitted
about
it.
Query,
stupid,
imager
and
topology
mother
gathers
topology
hints
from
a
number
of
resource
managers.
Then
hinter
merchants
continue
elevated.
D
It
means
that
resources
are
pretty
reserved
and
then
finally
became
deployed.
Justin
looks
in
the
chart
yeah.
We
can
go
to
the
next
slide,
which
introduce
ya
memory
manager,
I'll.
Just
repeat
what
I
thought
in
the
beginning,
just
memorize
a
new
component
in
queue
at
the
coastal
ecosystem
proposed
to
enablement
motivation.
D
It's
a
new
hit
provider
for
topology
manager
and
hints
are
intended
to
imitate
preferred
memory
and
reproduce
affinity,
which
means
the
memory
for
a
container
either
to
a
single
or
a
group
of
notes
and
by
design.
As
we
said
before
matter
of
hours,
guaranteed
memory,
location
for
posting
in
guaranteed
us
class.
B
So
I
I
have
a
question.
You
know
right
now.
The
topology
manager
has
single
Newman
ed
policy
and
which
generally
indicates
all
the
resources
are
prefer
to
be
on
a
single
Numa.
Now,
with
the
memory
manager,
it
seems
like
you
are
saying:
okay,
we
can
also
align
conventional
memory
and
huge
pages
with
the
addition
of
being
able
to
allocate
huge
pages
from
multiple
Numa
nodes
and
I'm.
Just
wondering
I
think
you
touched
on
this.
Maybe
the
next
slide,
instead
of
having
the
approach
of
allocating
memory
from
or
huge
pages
from,
multiple
NUMA
nodes.
D
C
D
C
He
said
this
at
the
beginning,
but
I
think
what
he's
getting
at
here
is.
You
know,
obviously,
if
you
can
fit
all
of
the
memory
that
you
that
you
require
on
a
single
namenode,
then
that's
what
you
should
do,
but
if
you
actually
ask
for
more
memory
than
can
ever
fit
on
a
single
uma
node
and
you
don't
have
the
single
neumann
old
policy
set,
then
you
need
some
way
to
be
able
to
allocate
that
for
multiple
nodes.
Yeah.
B
B
D
C
So
I
think
what,
with
all
of
talk
up
until
now,
Victor
has
been
about,
is
how
can
you
even
set
up
the
bitmask
of
where
you
could
allocate
your
memory
from
and
then
obviously
the
policy
will
kick
in
to
figure
out
what
to
reject
and
what
not.
But
you
need
to
be
able
to
set
up
that
bitmask
to
say.
Oh,
this
is
where
I
could
potentially
allocate
that
memory
from
okay
I
got
all
right.
Yeah.
D
Much
about
jeans,
no
more!
Even
our
first
queries,
it's
internal
database,
it's
called
node
map
and
in
fact
it's.
This
is
the
object
that
that's
responsible
for
tracking
the
memory
by
by
by
keeping
counters
on
free
memory
and
reserved
don't
worry,
and
after
driving
to
this
information.
Memory
manager
will
in
fact
provide
two
types
of
hints
single,
no
more
hint
and
cross
no
more
hints,
and
then
then
the
procedure
is
standard
allocation,
so
resources
will
be
preserved
and
then
container
deployed.
D
D
Here
I,
just
this
is
in
fact
slight
on
how
in
generation
is
done
by
my
mother,
I
just
gave
notice.
That's
crossed
him
already
in
Christ
single
lumen
generation.
So
so
we
just
talk
about
Krishna
mwah
in
the
next
slides,
so
memory
manager
will
ensure
that
the
yield
our
team
at
my
site
of
hints
includes
both
single
luma
hints.
D
Okay
and
now
it's
something
important
that
all
canes
that
match
that
this
joint
group
configured
with
a
cross
on
my
lips
flag,
will
have
a
preferred
set
to
true,
and
why
is
like
that?
The
next
three
slides
will
explain
the
straight
so
so
consider
that
we
have
to
approach
another
running
under
either
restricted
or
single
ma
policy,
and
these
two
policies
will
admit
a
container
only
if
the
affinity
calculation
leads
to
the
final
key.
That
is
preferred.
D
Otherwise
what
will
be
container
will
be
rejected.
So
this
is
exactly
why
we
feed
to
progeny
majority
prefer
to
subject
to
hint.
This
is
to
to
achieve
our
primary
goal,
which
is
in
fact
granted
more
education.
So
just
to
sum
up
the
result
of
that
approach,
another
either
provides
the
best
thing.
That
indicates
a
single
design
group
of
notes,
which
is
either
single
one
out
or
multiple
one
out,
and
this
is
sure
that
we
did
define
I
go
okay,
I
think
we
can
move
to
the
next
slide.
D
D
The
group's
us
I've
seen
the
previous
example,
so
0
1
2
&
3,
and
we
can
write,
preferred
hints
only
for
this
disjoint
groups,
so
we
will
have
the
first
hint
will
be
single
Numa
in
preferred
which
match
0
group.
The
second
one
will
be
cross,
no
markings
preferred
which
match
1
2
group
and
the
first.
The
last
one
is
the
rest.
Preferred
is
also
single
single
monomer
hint
preferred
which
match
the
the
third
group
can
I
ask
a
question.
C
Here
so
this,
this
feels
a
little
strange
to
me
just
because
the
the
intention
of
the
preferred
flag
was
to
indicate
whether
that
was
a
minimally
possible
allocation
across
Numa
nodes,
meaning
independent
of
any
allocation
that
have
already
happened.
If
there's
enough
to
satisfy
your
research
request
on
some
minimal
set
of
Numa
nodes,
only
those
ones
should
have
preferred
set.
D
C
Guess
I'm
getting
has
imagine
your
three
pods
come
along.
The
first
one
gets
out
desert
allocation
according
to
hint
one
that
you
have
here.
The
second
one
comes
along
does
HIDs
allocation
according
to
him
three
then
the
third
comes
along
and
the
only
hint
that's
still
available
where
an
allocation
could
happen
is
number
two
I
actually
want
that
pod
to
be
rejected.
C
I
don't
want
it
to
be
admitted
here,
because
I'd
rather
have
it
land
on
a
different
node,
where
it
could
get
all
its
resources
from
a
single
nimma
node,
instead
of
being
admitted
here
just
because
two
other
allocations
have
happened
on
this
note
already.
That
happened
to
be
able
to
get
it
from
a
single
new
node
and
now
I
can't.
G
C
G
To
use
from
the
same
socket
for
the
memory
like
you,
don't
have
such
implication
or
another
variable,
so
we
just
you
have
to
use
the
somebody
shot
possibility
for
the
configuration.
So
if,
for
example,
you
want
to
use
some
specific
moments
for
most
normal
occasion,
they
give
the
users
ability
to
configure
it
before
they
start
tomorrow.
If
you
will
not
configured
anything
like
you
will
you
will
have
fun?
You
always
look
at
the
single
location,
because
we
don't
have
an
indication
publicans
that
prepared
in
case
when
single
luminol
does
not
respect.
G
C
Don't
care
about
it
being
single
or
not
single,
it's
minimal,
meaning
that
if
you're
asking
for
you
know
500
gigabytes
on
a
say,
700
gigabyte
system
and
you've
got
350
on
each
it.
Obviously
the
only
allocation
I
could
ever
do
is
on
those
two
there's
a
bad
example.
Could
I
want
to
actually
have
more
Numa
nodes
to
drive
the
point
home,
the
point
being
that
the
minimal
allocation
there
is
to
Numa
nodes?
C
C
Question
offline:
it's
just
you
know
the
preferred
flag
was
designed
to
be
used
slightly
different
than
this
with
very
specific
semantics,
so
that
someone
knew
that
if
they
were,
you
know
asking
for
a
certain
number
of
resources.
It
was
landing
on
a
node
that
they
were
going
to
be
given
the
minimal
possible
pneumo
alignment
for
those
resources.
Yes,.
E
E
D
D
Right,
you
are
coming
to
the
summary,
and
so
this
this
new
flags
are
proposed
so
manager
policy,
it's
too
and
I
go
disable
matter.
The
preserve
from
Arizona
flag
is
just
reserve
memory
for
system
demons
and
probably
the
most
important
clock.
Is
the
consumer
groups
track
that
we
that
we
just
described
times
it's
just
to
recall,
to
configure
a
giant
group,
sumo
notes,
beautiful
location,
right
and
I
think
that
there
will
be
the
last
slide
which
the
summary
make
next
one
he
could
go
to
the
next
slide,
so
yeah
I've
just
presented
summary.
D
D
B
Okay,
I
think
that
was
a
great
presentation.
Are
there
any
questions
or
comments
or
more
discussions
about
this
I
captured
one
one
item
in
the
meeting
doc
and
it
was
referring
to
that
Kevin
about
the
you
know.
The
hint
number
two
had
the
preferred
that
that
would
spend
multiple
NameNode.
So
just
so,
we
capture
that
and
maybe
can
think
about
it
offline.
Any
other
sort
of
comments,
questions
discussions
about
this.
B
F
My
first
comment
is
really
just
about
phrasing
and
terminology,
so,
instead
of
calling
it
cross,
numa
I
mean
my
recommendation
would
be
to
call
it
either.
Multi
node
or
multi
numa,
cross
numa
has
a
very
negative
way,
has
a
very
negative
connotation
to
it
right
because
it
usually
refers
to
people,
don't
want
to
the
cost
of
cross
numa
traffic
right.
So
I
think
what
you're
trying
to
describe
with
the
cross.
D
B
E
E
A
Yes,
I
think
you
bring
up
a
good
point.
One
of
the
things
I'm
hoping
we
can
do
is
allow
everyone
to
present
their
technical
materials
and
then
maybe
we
can
coalesce
them
all
and
kind
of
judge
them
from
a
UX
perspective
and
I.
Don't
know
if
there's
actually
a
one
UX
fits
all
outcome,
but
I
do
think
it's
it's
something
we
need
to
think
about.
F
F
A
And
I
think
I
would
maybe
I
don't
know
if
there's
a
one-size-fits-all
sysadmin
for
kubernetes
right,
like
if
you're
running
a
managed,
general-purpose,
kubernetes
service
I,
think
maybe
some
of
the
concepts
Alexander
showed
on
Tuesday
signaled
meeting
our
kind
of
low
touch
hints
that
would
work
pretty
well,
maybe
on
any
general-purpose
cube
set
up.
I
do
appreciate
that
there's
like
users
of
kubernetes
who
try
to
make
a
more
purpose,
specific
configuration,
and
so
like
it's
possible.
You
know
a
5g
telco
network
would
be
thinking
that
the
cost
of
this
is
worth.
A
You
know
the
one
one
time
set
up
right
and
so
I
don't
want
to
force
everyone
to
be
a
general-purpose
community's
distributor
when
thinking
about
their
UX
as
much
as
just
thinking
through
the
who
is
the
audience
that
this
is
serving,
and
is
it
serving
that
audience
well
with
the
right
UX
and
if
we
can
get
to
one,
that's
awesome,
I,
don't
know
if
we
will
but
I
think
I
don't
want
to
make
everyone
feel
like
that.
There
needs
to
be
one
and
only
one
UX,
necessarily
because
I
actually
don't
know
that
I'm
curious.
F
I'm,
certainly
not
recommending
getting
rid
of
the
flexibility
and
configurability
I'm
just
more
referring
to
when
you
have
a
configure
option
where
the
value
is
essentially
a
JSON
encoded
array
of
array
of
integers
that
it's
it's
not
it's
not
obvious
right.
What?
What
exactly
that
configuration
value
refers
to?
That's
all
yeah.
E
I
want
to
clarify
my
statement
that
was
like
it's
good
what
we
have
a
comfortably
configurability,
it's
good,
what
somebody
like
like
use
of
it.
My
question
is:
what
like
we
like
to
enable
with
future.
We
should
not
require
everybody
to
do
with,
or
everybody
to
configure
was
anything
so
like
for
majority
of
open,
I,
guess
users,
it
should
be
like
some
sense
will
default,
which.
B
B
There's
these
questions
we
want
to
ask:
have
we
triaged
all
the
open
proposals
or
topics,
and
the
answer
to
that
is,
of
course
no
and
do
we
need
to
meet
again
and
I?
Think
the
answer
is
yes
on
that,
so
do
we
want
to
meet
sooner?
Do
we
want
to
wait
a
week
to
meet
and
go
over
the
pod
level
single
Numa
policy,
or
do
we
want
to
try
to
meet
sooner
and.
F
D
B
A
A
C
B
A
Yeah
I
think
Tuesday
mornings,
fine,
like
I,
said
I'm
really
just
trying
to
unblock
progress
and
I
think
it's
important.
We
hear
everyone
present,
and
so,
if
you
want
me
to
get
into
stay
in
the
normal
cig
meeting
again
in
the
afternoon,
that's
fine
too
I
just
I
want
to
get
through
the
backlog
and
be
respectful
of
everyone's
time
and
energy.
A
B
H
I
just
wanted
to
ask
for
for
an
offline
review
of
the
cap
on
this
meeting
and
I
think
if,
if
you
could
go
through
that,
we
can
discuss,
maybe
the
results
on
the
next
on
the
next
meeting.
I
just
wanted
to
ask
for
a
view
and
not
go
for
the
cap,
because
the
caps
update
is
relatively
small
to
the
memory
manager.
It
is
only
one
policy
and
it's
not
not
much
things
in
it,
so
it
won't
take
long.
In
my
opinion,.