►
From YouTube: 2016-NOV-16 :: Ceph Performance Weekly
Description
Weekly collaboration call of all community members working on Ceph performance.
For full notes and video recording archive visit:
http://pad.ceph.com/p/performance_weekly
A
H,
let's
get
this
show
on
the
road
this
week
in
polar
requests.
We've
got
a
couple
of
new
ones.
It
looks
like
there
was
one
for
improving
searcy
calculations
when
you
just
got
a
buffer
of
zeroes,
that's
0
1,
which
is
fantastic
and
then
also
maybe
a
slightly
more
interesting
one
here,
40
copy
buffer
lists-
and
it
looks
like
both
Kiku
and
Matt-
have
been
reviewing
that
and
I'm
hopeful
that
we
can
get
some
micro
benchmarks
to
actually
see
if
this
is
actually
helping
or
not.
A
The
splice
makes
me
a
little
bit
nervous,
but
you'll
see
there
were
a
couple
different
things
that
merged
we
are
now
favoring
metadata
in
booster
cash,
more
than
we
previously
had,
which
I
think
is
probably
a
good
thing.
I
think
some
tested
that
and
didn't
see
a
whole
lot
of
improvement,
but
anyway
about
it
is
probably
good
to
be
figuring
metadata
for
cash
in
general.
Anyway,
there
was
a
couple
of
other
general
blue
store
improvements,
more
work
towards
getting
different
things
into
mem
cools.
A
A
Otherwise,
a
lot
of
other
stuff
that
we
don't
have
a
whole
lot
of
movement
on
yeah
I'm,
hoping
that
we'll
we'll
see
mines
at
a
scale
soon,
and
also
this
slab
containers
one
would
be
good
to
see
okay.
Otherwise,
for
this
week
the
probably
the
only
thing
I
have
to
talk
about
is
there's
a
request
for
using
rocks.
Tv,
calm
families
and
I
went
through
and
basically
just
fixed,
that
up
to
work
with
master
and
then
took
a
stab
at
making
the
calm
families
in
blue
store
and
it
more
or
less
works.
A
For
some
reason,
stuff
is
ending
up
in
the
bee
tree.
That
shouldn't
be
and
commit
start
is
asserting
I,
don't
understand
why
that's
the
case
of
Hannah
stared
at
the
code
at
all
times,
you
don't
see
what
would
how
that
would
be
allowed,
but
clearly
there's
something
going
on
so
I'll
just
have
to
stare
at
it
more.
A
The
bitmap
allocator
actually
seems
to
work
fine
for
rights,
but
we
do
end
up
in
a
situation
when
doing
reads
where
there
is
seriously
check
some
failures.
So
I
don't
know
if
those
two
things
are
related
or
not.
Maybe
they
are,
but
there's
still
some
work
there
to
get
this
working
consistently.
The
good
news,
though,
is
that
well
it
does
work.
We
can
get
some
stats
now
out
of
the
stuff,
that's
getting
populated
in
Roxy,
be
specifically
regarding
compaction
and
at
least
when
we
are
separating
things
out
into
a
column
families.
A
What
we're
seeing
is
that
about
one-third
of
the
compaction
traffic
is
due
to
omap
and
about
two
thirds
is
due
to
metadata
opportunity
as
specifically
so
what
we
were
one
of
the
things
we're
kind
of
looking
for,
and
there
is
how
much
right
ahead
log
traffic
was
was
compacted.
Where
I
had
a
lot
of
compaction
traffic
deserting
us,
we
were
afraid
that
we're
leaking
a
lot
of
stuff
into
a
level
zero
and
potentially
level
one
that
need
to
be
there,
and
at
least
when
calm
families
are
enabled.
It
doesn't
appear
that
that's
the
case.
A
Somnath
has
I
think
seeing
some
evidence
that
may
be
in
the
current
stock
code.
We
are
leaking
a
lot
of
great
head
log,
use
that
that
shouldn't
be
making
it
into
level
zero.
So
that's
something
that
we
probably
need
to
figure
out,
but
the
other
good
news
is
that
when
con
families
are
enabled,
it
looks
like
maybe
we
are
in
general,
seeing
right,
amp
or
bright
and
production
in
rocks.
A
Tv
performance
in
general,
with,
like
small
random
writes,
appears
to
be
higher,
while
the
tests
are
able
to
run,
though
in
general,
we
we
have
more
work
to
do
in
this
area,
but
it
seems
like
it
could
be
a
portent
winner,
foot
boost
or
until
we
can
get
Zetas
scale
or
something
else
working
that
doesn't
have
the
same
kind
of
right.
Aunt
issues
that
we
are
seeing
the
Frog
sticky.
So
any
any
questions
on
any
of
that.
A
A
Okay,
Matt
asked
of
next
week
about
adding
to
the
agenda
RCU
by
dan
I
think
next
week
might
be
taken
up
by
some
guys
that
I
think
it's
from
Intel.
Maybe
that
have
been
doing
some
glue
store
testing
and
they
asked
if
they
could,
if
they
could
present
next
week,
but
I'll
I'll
see
how
long
their
slots
can
go.
If
not
next
week,
we
can
certainly
talk
about
the
Blue
captor.
A
That's
definitely
something
that
that
I
love
to
hear
how
it's
going
so.