►
From YouTube: Ceph Performance Meeting 2021-07-22
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
All
right,
I
think,
I'm
going
to
just
get
started
here,
we'll
we'll
have
the
core
people
trickle
in
as
they
get
done.
A
I
think
all
right,
let's
see
actually
a
a
number
of
new
prs
this
week
and
updated
prs
for
that
matter
and
close
vrs,
it's
been
busy,
so
there
is
a
new
pr
that
changes
the
way
that
vlog
works
and
I
have
to
confess
I
didn't
look
super
carefully
at
what
this
is
doing,
but
kifu
approved
it
and
asked
for
sam
to
take
a
look-
and
I
cc
on
that-
hopefully
we'll
get
more
feedback.
Hopefully
this
is
okay,
but
in
any
event
that
is
there.
A
Radic
has
a
really
interesting
pr
that
lets
you
generalize.
The
object
store
to
use
an
alien
store.
Previously
alien
star
was
restricted,
restricted
in
crimson
to
using
blue
store,
but
now
it
can
basically
use
any
existing
back
in
object
store,
which
means
we
can
test
both
mem
store
and
maybe
even
file
store.
I
know
that
anyone
will
ever
try
it,
but
but
theoretically
so
the
goal
here
is
to
be
able
to
get
a
better
sense
of
how
much
overhead
we're
seeing
in
alien
store
itself.
A
So
that's
that's
really
good!
That's
really
interesting!
Let's
see
there
is
a
new
pr
for
rgw
tracing.
A
So
that's
good
matt
reviewed
that
buffer
list,
optimization
for
the
c
string,
conversion,
that's
good
and
then
a
pr
that
makes
it
so
that
the
finisher
thread
is
better
about
giving
up
cpu
and
that's
a
little
controversial.
I
think,
but
but
kifu
and
kcr
are
looking
at
it
and
they
look
like
they're
bringing
up
the
right
questions.
So
that's
good
closed.
A
Yeah,
all
of
those
look
look
reasonable
adam's
old
pr
for
changing
some
of
the
behavior,
with
the
way
that
the
blue
store
caches
are
implemented
for
charting
got
just
kind
of
it
was
closed
by
the
stale
bot.
We
haven't
touched
it
in
a
while
I'll
have
to
ask
adam
about
that
and
find
out
if
he
still
wants
to
work
on
that
or
not
beyond
that
lots
of
updated
stuff.
This
optimized
objects
memory
allocation
using
pools.
A
I
want
to
review
that
one
I
haven't
gotten
to
it
yet,
but
that
looks
really
interesting,
otherwise,
not
a
whole
lot
going
on
there,
some
things
that
are
in
testing
that,
hopefully
will
merge
next
week.
A
A
A
Alright,
randy
or
anyone
else
any
any
prs.
I
missed
that
you
know
of.
A
All
right
well,
then,
moving
on
rack
and
I
reddick
sam
and
I
actually
earlier
this
week
had
like
a
one
hour
midnight
powwow
to
discuss
crimson
behavior
in
the
reactor.
A
A
That
being
said,
the
the
next
goal
is
to
implement
multi-reactor,
so
greg
farnham
is
going
to
work
on
that,
hopefully
starting
pretty
soon
he's
getting
up
to
speed
right
now,
but
the
good
news
is
that
we're
we're
seeing
efficiency
already
that's
kind
of
inline
or
better
than
classic.
Sometimes
it's.
You
know
not
a
ton
better
like
20
or
30
percent.
A
Sometimes,
in
some
cases
where,
where
the
the
the
crimson
code
is
more
emphasized
in
in
kind
of
the
the
really
ideal
case
in
radix
tests,
I
think
he
was
saying
that
he
was
seeing
up
to
like
two
or
three
hundred
percent,
so
in
reality
we'll
probably
end
up
somewhere
in
between
those
numbers.
I'm
hoping
that
that
you
know
maybe
we'll
see
between
50
and
100
percent
real
later
on.
A
But
you
know
it's
all
just
kind
of
a
little
bit
guessing
right
now,
but
the
good
news
is
that
it's
is
generally
better
overall,
even
even
when
we're
testing
blue
store
and
that's
exactly
what
we
want
so
multi-reactor
is
the
next
can
be
the
next
big
performance
when
and
it
should
be
major
when
we
can
use
you
know
multiple
reactor
threads
per
per
osd
for
for
c,
star
or
sorry
c
store
that
maybe
matters
less,
maybe
we'll
see,
and
we
have
a
lot
of
decisions
to
make
when
we
get
there.
A
But
for
now
this
is
where
we're
at.
Let's
see
beyond
that,
reddick
I
already
mentioned
had
built
in
this
ability
to
test
mem
store
inside
alien
store,
and
that's
really
neat.
I
need
to
take
a
look
at
that,
while
he's
been
doing
that
I've
been
working
on
adding
in
some
fixes
fixes
just
restructuring
of
the
mem
store
code
based
on
some
stuff.
A
I
did
like
three
years
ago
and
also
a
new
vector
object
class
that
at
least
used
to
be
faster
than
the
buffer
list,
implementation
that
we
have
now
we'll
see,
if
that's
still
the
case,
with
all
the
optimization
work
that
radic's
done
on
buffer
list,
but
at
least
potentially
it
could
be
faster
still.
A
This
kind
of
came
up
as
a
result
of
a
discussion
on
pnfs
that
jeff
leighton
presented
earlier
this
week,
he's
really
interested
in
trying
to
make
it
so
that
we
can
have
clients
talking
pnfs
via
what's
available
already
in
the
kernel
and
then
translating
that
to
cfs
behind
the
scenes
that
might
involve
potentially
like
using
ganesha
and
tying
it
as
a
library
directly
into
the
mbs
or
maybe
making
a
library
version
of
the
mds
and
kind
of
bolting
it
onto
the
side
of
ganesha,
or
it
could
mean
potentially
something
really
different
like
well
yeah.
A
I
guess
I
don't
totally
understand
it,
but
but
actually
changing
kind
of
the
way
that
that
we
talk
natively
in
any
event,
it's
it's
all.
You
know.
Kind
of
contentious
and
different
people
have
different
ideas,
and
if
we
should
do
that
and
how
we
should
do
that,
if
we
should
so
we'll
see
where
it
goes,
jeff's
kind
of
actively
seeking
feedback
on
it
and
kind
of
deciding
trying
to
decide
what
he
wants
to
do
with
it.
A
A
I
don't
know,
zuffs
is
basically
dead
at
this
point,
it
sounds
like
netapp
went
through
like
a
restructure
and
laid
off
a
bunch
of
people,
but
the
idea
itself
maybe
isn't
bad
just
maybe
the
implementation
wasn't
ideal.
So
anyway,
lots
of
discussion
on
that
right
now,
too,
we'll
see
if
it
goes
anywhere.
A
That's
all
I've
got
any
questions
or
comments
on
any
of
that
stuff.
A
All
right,
then,
we'll
open
it
up.
Does
anyone
have
anything
that
they
would
like
to
talk
about
this
week?.