►
From YouTube: Ceph Performance Meeting 2018-11-01
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
C
B
Well,
that's
good
I,
don't
I,
don't
think
we're.
Gonna
have
too
many
people,
though,
because
we've
got
folks
out
on
holiday.
Today,
though,
I
think
it's
gonna
be
a
small,
and
actually
this
might
be.
The
group
were
five
minutes
past
the
hour.
So
there's
there's
a
lot
of
movement
on
PRS
this
week,
which
is
really
good.
So
I
was
we
were
talking
about
earlier,
there's,
actually
a
variety
of
new
ones
and
I
think
Radek
was
driving.
A
lot
of
this
he's
got
three
new
PRS
this
week,
which
is
crazy.
B
Now
that
he's
got
PR
for
it,
I
think
I
want
to
apply
that,
along
with
some
of
the
other
stuff,
that's
been
done
recently
and
tried
to
look
at
some
of
those
OMAP
tests
again,
even
if
they're
they're
a
little
flawed,
because
they're
making
objects
for
everyone
with
frado's
bunch,
maybe
it'd
be
really
quick
just
to
rerun
it
and
see
if
it
helps
and
I
think
it
might
so
yeah
all
this
bufferless
stuff
would
be
good
anything
that
anything
that
touches
in
code.
I
think
is
going
to
have
a
that
improves
in
code.
B
B
Overview,
I,
guess
I.
When
I
looked
at
this
it
yeah
I,
don't
know
I
think
say:
just
was
asking
a
lot
questions
bug,
say:
I'm,
not
sure.
If
he
really
knows
what's
going
on
either
what
else
this
one
by
Erik
I
guess
he's
doing
new
things.
Looking
at
the
IO
throttler
yeah
I
wasn't
I
wasn't
even
aware
he
was
still
working
on
this
stuff.
So
that's
good.
What
else.
A
Think
I've
actually
started
being
picked
up
by
a
student
from
Gru's
America's
birth
IPR.
Up
to
let's
look
at
CS
career
is
a
very
ranch
where
it
ended
up
pool.
A
B
B
But
for
anyone
that's
here
that
is
the
author
of
any
of
this
I
guess
you
know
if
it's,
if
it's
still
worth
looking
at,
you
know
feel
free
to
bring
it
back.
But
it's
it's
nice
that
we're
starting
to
clear
out
some
of
this,
this
old
stuff
that
just
have
languished
for
a
year
or
more
sometimes,
what
else
updated.
B
This
little
every
day
when
I
don't
have
any
idea
about
other
than
that
they
seem
really
excited
about
it
and
there
was
good
ideas.
So
it's
always
always
nice
to
see
people
excited
about
what
they're
doing
that's
good.
B
There's
this
really
old
PR
from
Adam
is
Adam
here,
no
he's
not
I
object
and
that
got
updated.
That
was
like
over
a
year
old,
I
think
and
just
in
the
last
month
now
it's
been
getting
new
new
commits
and
new
discussion,
so
we'll
see
where
that
goes,
but
that's
good
and
then
the
adaptive
startle
one
that
we
were
just
talking
about.
So
that's
about
it
any
other
thoughts
about
any
PRS
from
folks.
B
D
E
Were
like
this
before
years,
been
I
have
2011
I
think
until
now
we
do
be
trying
to
be
kept.
We've
gained
diverted
and
other
stuff
like
trying
to
use,
vector
and
and
other
things,
but
this,
but
this
is
crazy,
but
I
think
that
but
I
think
one
doesn't
need
a
hanger
ball
pointer,
I
think
one.
It's
a
other
location
for
this.
D
Yeah
naming
things
is
hard,
but
with
respect
to
atoms,
objector
PR,
if
I
understand
right,
he's
almost
ready
to
push
another
update
and
I
think
that
it's
pretty
much
ready
at
that
point
other
than
testing
and
benchmarking,
but
it
does
add
kind
of
a
new
libretto
ste
in
her
face
and
who's
calling
it
ray
dos
unleashed.
But
that
would
be
like
the
kind
of
public
API
and
if
people
are
interested
giving
that
part,
a
review
I
think
would
be
good.
E
D
A
I,
guess
that's
what
that's!
What
I
keep
you
from
working
actually
is
so
splitting
at
the
C++
versus
C
bindings
that
we
can
increase
the
s?
Oh
for
the
C++
library,
without
affecting
the
C
library.
A
A
Our
flows
changes
from
Radek
Nautilus
that
will
go
into
the
new
liberators
I
thought
I,
so
where's
PP
that
I
said:
we've
got
three,
so
he
can
continue
wording
more
changes
to
the
he
was
plus
interface.
Now.
D
E
A
E
E
A
D
D
E
B
C
B
C
C
C
C
B
C
So
I
mean
that's
probably
fine,
because
I
think
anybody
putting
in
new
notes
are
probably
gonna
go
for
higher
than
that.
So
for
IBD,
that's
fine!
That
people
upgrading
you
know
like
I
found
myself.
It
wasn't
obvious
that
I
had
to
try
and
target
that
size.
The
only
additional
from
I
think
some
people
may
encounter
as
if
they're
like
I'm
guessing
rgw,
which
is
got
larger,
rocks
DVD
requirements.
The
next
step
up
from
that
is
probably
going
to
be
300.
Ish
yeah.
B
C
Yeah
I
sort
of
started
looking
at
that
and
every
time
I
sort
of
was
working
it
through
my
head.
Doesn't
it
one
value
just
made?
You
know
the
next
one
be
even
worse.
So
is
it
was
hard
to
come
up
with
a
good
solution
in
unless
you
drop
the
number
of
the
level
multiplier,
but
then
obviously
get
a
lot
of
amplification.
So
there's
no
real
right
way
of
doing
it.
Yeah
I.
B
B
When
you
hit
like
level
3
or
level
4
right,
you
have
to
have
the
entire
level
on
the
database
partition,
otherwise
it
and
the
falling
over
to
the
the
the
block
device
and
figuring
out
that
hierarchy
is
tricky
and
error-prone
and
I
think
well.
Nick.
You
were
just
saying
it
was
hard
to
find
alternate
hierarchies
that
that
kind
of
universally
work.
Well,
if
I
understand
correctly
yeah.
C
There's
two
times:
I
mean
for
myself
personally
I
found
Italian
to
actually
know
what
the
size
was
needed,
because
if
you
add
up
what
the
level
was
meant
to
be
I
was
getting
like
28
or
29
gig,
but
ash
the
time
to
be
30.
So
it's
just
a
bit
of
trial
and
error,
but
then
yeah
you're
correct
trying
to
get
it
to
fit.
Your
actual
data
set.
There's
going
to
be
the
the
next
issue.
If
you're
going
above
level,
3.
B
E
E
A
B
C
C
Can
we
adjust
this
to
sort
of
maybe
suit
and-
and
there
potentially
is,
if
you
maybe
said
like,
if
you've
got
less
than
50
gig,
then
make
sure
level
three
is
adjusted
to
fit
whatever
size
that
is
and
then,
after
that
you
would
then
say:
I
try
and
adjust
level
four
to
fit
whatever
size
you've
got,
but
if
you,
the
only
real
way
to
make
it
more
granular
would
be
to
drop
the
level
multiplier
down.
But
from
what
I
understand
about
how
rocks
TV
works?
C
B
C
It's
just
that
you
know
say
you
then
made
level
two
five
gig
you've
then
got
to
have
50
gig
for
level
three.
Although
you've
got
you
know,
the
next
you've
doubled
the
size
of
what's
fitting
in
level
two
you've
still.
You
know
my
case
where
I
had
that
and
I
start
off
with
25
gig
partition
I've
still
got
no
way,
I've,
actually
using
the
bulk
of
that.
So
so
pretty
what
you
want
to
have
cut
offs
and
then
say
you
know
like
zero
to
ten.
C
B
C
Yeah
I
mean
I,
think
I
mean
initially
I.
Think
the
the
probably
the
biggest
thing
is
maybe
to
have
that
noted
somewhere
in
your
documentation,
because
at
the
moment,
there's
just
of
like
a
flat
percentage,
that's
advertised,
which
obviously
doesn't
take
into
account
these
minimums
of
what
you
should
try
and
target.
Oh.
C
Us
back
there
I
mean
I
did
have
one
crazy
idea
and
I
haven't
done
it
because
it
is
slightly
crazy
and
that
was
using
LVN
thin
provisioning.
Now,
there's
obviously
a
big
risk
about
that.
If
it
ever
does
fill
up
and
you
run
out
of
the
underlying
storage,
but
it
would
allow
you
to
present
say
a
whatever
sized
Rock
device
with
the
SSD.
C
A
A
B
B
B
All
right
cool
anything
else,
guys
sending
anything
else.
Folks
would
like
to
talk
about
this
week.