►
From YouTube: 2019-03-28 :: Ceph Performance meeting
Description
B
B
B
B
A
With
you
I
think,
will
help
a
lot
more
generally,
though
I
think
yeah,
that's
coming,
you
know,
there's
I,
think
there's
gonna
be
a
quite
a
bit
of
desire
to
see
rgw
running
within
you
know.
Maybe
a
couple
cores.
A
Because
it's
kolak
located
with
other
stuff,
you
say
yeah
kind
of
the
whole.
You
know
hyper-converged
setup,
where
you're
trying
to
give
as
many
you
know,
cores
as
possible
to
clients,
and
then
you
know
putting
the
storage
and
some
of
the
memory
on
the
node.
For,
for
you
know,
Saffir
storage,
but
CPUs
are
kind
of
the
the
big
you
know
resource.
Everyone
wants
I,
guess
yeah.
B
Yeah
and
I
do
think
that
just
the
sheer
number
of
threads
that
we
have
our
thread-
pool
defaults
to
512
right
now
and
I.
Don't
think
that's
really
sustainable
yeah,
but
yeah
I
really
think
that
I
mean
in
theory,
I,
don't
think
I
should
GW
should
really
be
CPU
bound
and
almost
anything
except
for,
like
maybe
compression
and
an
encryption.
A
B
A
Yeah
the
to
be
interesting,
I
haven't
used
my
wall,
clock,
profiler
or
Adams
wall
clock
profiler
on
rgw.
If
I
use
it
at
all,
I
don't
know
I
think
I,
maybe
did
it
early
on
when
I
was
first
writing
it.
But
but
that
would
be
interesting
to
look
at
to
just
see.
You
know
where
it's
actually
spending
real
time.
A
B
A
A
Yeah
I
think
just
the
performers
impacted
with
like
nvme
drives
and
RBD
is
huge.
I,
don't
know
how
much
it
affects
you
guys
or
not,
but
you
know
it's.
You
know
for
like
small
IO,
it's
it's
like
a
50%
performance
improvement.
In
some
cases,
though,
you
know
we
can't
it's
hard
to
justify.
Having
that
one
debugging,
you
know
message
or
that
kind
of
impact
on
customers
or
users
yeah.
A
B
A
A
feeling
that
this
is
not
going
to
be
a
performance
win
at
all,
even
for
like
large,
sequential,
writes
or
reads
where
you
maybe
expect
it
to
be
so
far.
You
know
I,
depending
on
the
compressibility
of
the
data.
It
looks
like
you
know:
you're
saving
some
space,
but
I'm,
not
I'm,
not
convinced.
This
is
actually
gonna
really
help
that
much
we'll
see.
Oh.
A
Yeah
Adam
work
I
think
we're
gonna
wrap
this
up
early
there's
only
three
of
us
today,
Sage's
out
and
other
people
are
taking
PTO
and
stuff.
So
anything
that
you
want
right
about.
Oh
my
head,
alright
yeah,
the
only
that
mentioned
to
Casey
was
that
we're
doing
debug
Imus
zero
by
default.
Now,
instead
of
iboga
miss
one
I
approve
all
right,
very
good,
but
I
already
master
I,
don't
know
if
we're
back
porting
it
to
Nautilus
or
not,
but
I
would
be
in
favor
of
it.