►
From YouTube: 2017-MAR-01 :: Ceph Performance Weekly
Description
Weekly collaboration call of all community members working on Ceph performance.
http://ceph.com/performance
For full notes and video recording archive visit:
http://pad.ceph.com/p/performance_weekly
B
B
B
A
Ok,
so
the
there's
a
bug
fixed
from
Hanoi
that
one
I
proved
that
should
be
ssible
the
fastest
batch.
That
submerged
can
only
measure,
difference,
I,
think
weren't
doing
our
club,
but
the
codes
on
boy
should
be
the
cheaper
see
the
aio
one
is
in
my
testing
branch.
Double
prime
rib
shortly,
there's
a
whole
bunch
of
total
request
coming
from
the
narratives
folks
that
are
changing
the
boost
or
a
bit
numb
alligator
to
use
wheless
memory
and
clean
it
up.
Some
that's
good!
A
A
A
Yeah,
oh
yeah
I
get
a
read
more
carefully
conversations
kind
of
a
little
bit
of
I.
Suppose
it
hard
to
tell
them
what's
going
on,
but
I
think
I
bunch
of
the
patches
and
early
with
areas
are
sort
of
simple
and
Mawlana
Sheikh.
So
it
might
make
sense
to
this
much
both.
Even
if
you
don't
to
the
go,
all
the
way
see
and
likes
the
only
other
one,
that's
sort
of
low
on
my
end
of
Lilith
right
cash,
I'm
still
waiting
to
get
back
from
the
nokia
people
to
see.
A
B
B
Cool
alright,
let's
see
I
don't
have
too
much
here.
We
did
some
easy
overwrite
testing
towards
into
last
week.
B
There's
there's
some
other
things
in
there
that
that
we
can
test
going
forward
specifically
for
blue
store,
but
the
good
news
here,
I
guess,
is
that
boost
are
actually
doing.
Maybe
tolerably
good
with
ec
with
RVD
file.
Storage
is
not
to
intolerably
good.
The
best
I
guess
is
kind
of
made
to
be
expected,
though
most
likely
folks
will
probably
want
to
be
looking
at
boost
or
when
and
they're.
Considering
our
vdm
uneasy
volumes.
You.
A
B
A
I
think
the
one
one
suggestion
I
had
was
it
might
be
worth
burning
if
you
test
that
do
a
really
narrow
code
like
a
2+2
yeah
I
knew
her
out
to
replication,
because
I
think
that
might
be
a
useful
trade
off
between
performance
and
efficiency.
So
it's
2x
overhead
instead
of
3
I've
ever
had,
but
it
should
be
pretty
close
to
three
ax
four
replications
performance.
B
A
B
You
to
run
those
foot
yeah
cool,
the
only
we
have
here
is
what's
after
dispatch
has
spoken
to
those
that
the
result
rally
rotating
because
it's
we
got
kind
of
some
scattering
in
the
symbols,
because
I
don't
really
know
why
yet
I'm
actually
running
through
the
baby
test,
the
one
expressed
over
there
now
just
to
verify
that
we
see
the
same
pattern
or
not
so
we'll
see
what
those
are
the
flip
those
turn
up,
but
for
now
that's
what
we
saw
patients,
an
additional
pumping
with
just
valerate
ascension
and
ever
think
out
what
about
the
things
is
good
before
so
we'll
see.
B
Maybe
you
learn
the
thing
I'll
mention.
I
don't
have
any
results,
but
I
do
intend
to
test
some
of
the
food
for
things
that
come
in,
especially
regarding
the
bitmap
allocator
blocking
so
are
they
will
try
to
get
a
chance
to
test
out
later
this
week,
but
yeah.
That's
basically
it
for
me
as
well
I've
got
anyone
else
have
anything
they
would
like
to
bring
up
this
week.
A
No,
that's
basically
follow
me
that
I
have
hard
drives
now,
but
I
don't
have
a
box
box.
I
connect
them
to
doesn't
have
the
right
power
supply,
but
I
just
got
some
other
Hardware
in
white
bread,
soy
left
town.
So
when
I
get
back
on,
if
we
can
go,
look
all
that
stuff
up
and
then
I'm
going
to
plug
this
into
a
real
machine
that
I
can
do
real
testing
them
and
it'll
be
my
actual
development
machine
tape.
So
that'll
be
better
okay.
So
next
week.
D
Basically,
a
deja
de
la
king
is
all
is
only
as
half
of
the
world
problem.
The
second
one
is
related
to
extensive
cache
misses.
That
is
in
turn,
alighted
to
how
to
how
the
allocator
operates.
How
is
how
it's
how
it
performs
look
up,
it's
a
has.
It
mean
it
needs
to,
unfortunately,
traverse
very
Travers
structures
are
that
are
scattered
over
world
memory.
Does
the
problem?
I
guess
I
saw.