►
From YouTube: 2016-07-21 Kubernetes SIG Scaling - Weekly Meeting
Description
Public meeting recording of the Kubernetes Scalability SIG (Special Interest Group).
A
A
The
recordings
from
last
two
meetings
are
online.
I
posted
a
link
in
the
six
Gail's
lacrim.
So
please,
if
you
care
to
take
a
quick,
look,
make
sure
everything's
in
order
I'll
be
trying
to
post
the
recordings
from
these
meetings,
including
today's
relatively
relatively
quickly
afterwards
so
and
I
like
I.
Let
sarah
know
so
she
can
post
it
wherever
it
is.
She
wants
to
post
the
other
meetings
as
well.
So
that's
my
that's.
My
one
bit
of
group
administrivia,
I'm
curious
to
do
a
quick
poll
here.
A
B
A
Alan
anyway,
I
did
I
did
note
in
some
of
the
is
some
of
the
discussion
going
around
boy
txt.
We
check
your
work
on
the
on
the
throughput
through
foot
optimizations
related
related
to
the
also
related
to
getting
the
performance
for
the
affinity
and
anti
affinity
stuff
going.
So
you
could
be
making
a
list.
D
I
did
have
a
question
as
to
who's,
taking
lead
on
I'm
a
Google
side
on
the
NCD
3
stuff,
because
I
keep
on
prodding
individual
issues.
But
somebody
needs
to
probably
be
the
owner
of
all
of
the
salt
recognitions
and
kuba
stuff
and
all
the
little
script
jiggery
that
it
is
incorporating
for
setting
SED
3
by
default.
C
Yeah,
so
basically
we
had
long
discussion
like
this
week
about
it,
and
it
seems
that
actually,
no
one
has
fruit
now.
So
it
seems
that
unless
someone
from
community
will
do
it,
we
it
won't
be
done
for
1.4
like
we
have
a
very
strong
plan
to
do
it
for
sure
format
for
1.55,
but
it
seems
that
we
don't
have
fun
glee
for
1.4.
C
C
To
weekly
or
like
there
are
some
things
that
only
Google
can
do,
and
we
probably
would
find
some
time
to
do
it,
but
there
are
still
a
lot
of
things
that
are
can
be
done
by
community
like,
for
example,
currently
it's
not
possible
to
like,
like,
for
example,
there
is
no
support
for
hcv
free
in
Cuba.
Script
like
this
is
something
that
is
purely
open
source,
and
this
is
not
done
yet
and
like
assuming
that
we
have
it.
C
C
No
I,
it's
it's
nothing
like
very
difficult,
but
it's
like
some
work
to
be
done,
like
it
will
be
probably
day
or
two
of
work
to
ensure
that
everything
works
and
stuff
like
that,
oh
I,
see
I
mean
there
there's
nothing
like
super
fancy
or
super
difficult
there.
It's
just
like
work
that
needs
to
be
done.
A
D
It's
more
than
defaults,
because
once
we
defaulted
on,
we
get
all
the
testing
cycles.
You
know
as
part
of
the
community
work,
so
every
other
patch
and
everything
else
all
the
different
mechanisms
that
go
on
that
a
ton
of
stuff
like
a
good
example
here
is
like
in
abled
http/2
on
by
default,
and
it
caught
a
bunch
of
issues.
So
we
disabled
it
right
so
that
testing
is
pretty
beneficial.
E
So
I'll,
just
for
STD
free,
the
coop
ops
quit
like
I,
would
take
a
look
next
week
and
hopefully
I
will
Oh
Sammy
simple
request,
but
I
fought
for
the
other
thing
for
how
do
I
get
this
thing
integrated
and
with
other
thing
like
it
would
be
great.
I
hope
water
and
some
of
the
guys
could
help.
F
C
C
It's
like
we
don't
want
to
add
to
like
facilitate,
facilitate
any
HDD
free.
Specific
issue
features
until
like
at
least
one
release
with
default
at
City,
free
to
enable
life
roll
back
in
case
of
problems
and
stuff
like
that.
So
so,
if
we
like
for
assuming
that
we
enable
like
Etsy,
be
free
in
1.4,
then
we
won't
add
any
features
like
one
like
start
using
any
HDD
free
features
until
at
least
1.5.
So.
C
A
A
We
may
we
may
touch
base
with
you
Tim
as
well
or
or
maybe
with
some
of
the
core
West
guys.
One
of
the
one
of
the
folks
on
my
team
is
doing
some
at
cd2
at
cd3
performance
testing
and
also
doing
some
control
plane
variances
of
control,
plane
testing.
This
is
the
split
at
CD
versus
non
split.
A
TD
configuration
work,
so
we
may
be
contributing
lasting
here
as
well.
One
weep
along
this
line.
A
D
But
all
the
initial
data
we'd
shown
so
far
is
that
from
a
performance
standpoint
with
the
default
configurations,
it
pretty
much
is
just
a
swap
in
replacement.
Most
of
the
performance
issues
were
skiing,
her
not
related
or
the
biggest
bottlenecks
are
not
related
to
add
CD
performance.
At
this
point,.
D
Did
report
like
initial
findings?
We
haven't,
we
haven't
done
our
full
blastings
because
we're
dealing
with
a
rebase
stuff
right
now.
My
goal
is
to
get
this
in
as
quickly
as
possible,
so
we
can
start
doing
some
of
that.
D
We
did
have
data
results
to
report
off
from
device,
mapper
stuff,
I,
don't
think
anyone
most
people,
probably
don't
care
and
they
probably
use
a
room
later
something
else,
but
a
memory,
utilization
or
device
mapper.
As
you
start
scaling
up
to
really
high
numbers,
is
not
trivial
overhead,
so
we're
dealing
with
that
for
them.
Okay,.
B
Yeah,
just
on
some
reporting
on
our
side,
we
have
an
engineer
who's
starting
to
work
on
overlay,
FS
upstream
there's,
just
a
bunch
of
really
tricky
could
all
work
that
needs
to
happen
in
order
to
make
it
100
percent
correct,
but
I
mean,
as
we
all
know,
the
next
VFS
is
that
it's
not
really
about
correctness
is
about
looking
correct.
So.
B
A
Alright,
very
good
one.
One
thing
also
I
think
it
may
come
up
in
the
meeting
related
to
this
proposal.
I'm
sorry,
the
the
product
management
and
feature
proposal.
Work
is
I,
think
there's
going
to
be
a
stronger
push
to
make
sure
that
all
features
have
home
sig.
So
if
there
are,
will
call
this
a
call
for
any
features
that
the
group
here
feels
like
this
thing.
This
thing
should
be
specifically
keeping
an
eye
out
for
I,
think
I,
think
Voight
X,
throughput
work
certainly
would
qualify.
A
D
B
So
there's
some
rough
heuristics
that
have
been
put
on
the
features,
repo
and
really
I,
think
this
round.
We
should
try
to
put
in
anything
that
we
feel
meets
those
rough
heuristics
and
they're
really
rough,
like
Bob
saying
it's
stuff
like.
Would
it
be
in
a
blog
post
or
does
this
change
any
of
the
operational
like
aspects
of
Cuba
Nettie's?
B
If
so,
it's
a
feature,
and
so,
if
you
just
go
to
the
features
reaped
others
like
is
my
thing,
a
feature
thing
or
section,
and
so
angry
she
is
file
and
it
ends
up
being
like
too
noisy.
We
can
go
back
the
other
way
and
it's
just
going
to
take
us
a
few
release.
The
dither
over
like
where
the
sweet
spot
of
a
feature
is
and
the
heuristics
don't
be
updated
as
well.
I
think.
A
I
think
in
the
in
the
short
run,
it's
a
little
air
erring
on
the
side
of
a
bit
too
much
information
made.
Me
will
probably
serve
as
well,
because
I
think
that
the
issues
we've
had
in
the
past
or
this
past
release
were
it
was.
It
was
very
hard
to
figure
out
what
was
going
on
in
some
areas.
So
right
we
may
swing
the
pendulum
here
a
little
bit,
but
it's
probably
not
a
bad
thing.