►
From YouTube: Ceph Crimson/SeaStore 2021-09-01
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
All
right,
let's
get
started,
let's
see
this
week,
the
lba
rewrite
merged,
I'm
working
on
an
assert
failure,
yang
zum
ran
into,
I
think,
also
chun
mei
and
myself.
Now
I
can
reproduce
I'm
just
trying
to
find
the
cause
greg.
B
I
am
reading
code
and
documentation
and
trying
to
find
things
to
grab
on
grasp
onto
I
put
in
a
couple
little
pr's
that
I
from
stuff
I
ran
into
so
those
went
in
quick
thanks
for
merging
guys.
I'm
gonna
do
more
of
that.
As
I
work
towards
being
able
to
look
at
multi-core
utilization.
C
This
is
a
profiling
file,
profiling,
result
and,
and
the
the
the
right
licensing
is
more
worse
than
the
cfosd
with
blue
star
and
the
raider
latency
is
a
little
better
when
the
workload
is
increased.
C
And
the
cycles
per
op,
also
more
worse
than
the
selflessly
better,
the
rate
read
like
and
say:
reader
cycle
per
op
is
close
to
self-esteem
when
workload
is
increased.
C
And
so
for
our
chromosome
sd,
when
the
workload
is
increased,
the
rated
cycles
op
is
reduced.
A
C
Before
the
crimson
ac
used
less
cpus,
but
when
we
used
repeat,
I
think
we
were
used
when
the
condition
conflicted
when
we
used
the
repeater.
It
increased
the
use,
mouse
appeals
and
before.
C
Yes,
when
I,
when
we
use
interrupt
feature,
oh
condition,
no,
yes
interrupted
condition
before
I
only
can
test
one
number
job
and
two
ldaps.
That
is
very
similar,
simple,
simple
test
result
because.
C
C
Uh-Huh
and
yeah
c-stop
blue
star
crimson,
with
sister
and
seth
with
blue
star.
So
that
is
the
performance,
though,
seems
that
we
don't
have
better
performance.
A
C
A
C
A
A
C
store
is
really
and
if
inefficient
currently
like.
Currently
it
doesn't
even
write,
extends
known
to
be
long-lived
to
a
different
segment.
That's
shuihan's
pr,
which
should
merge
shortly
but
like
even
very
basic
optimizations
aren't,
and
yet
it's
essentially
designed
to
do
as
much
garbage
collection
work
as
physically
possible.
C
C
This
is
fl
fl
matrix,
so
it's
a
little
different
from
the
metrics
output,
so
everybody
can
check
it.
A
C
A
Do
you
have
a
setup
that
would
allow
you
to
do
profiling?
That
would
allow
you
to
identify
hot
spots,
areas
of
high
cpu
utilization.
A
That's
not
what
I'm
asking
what
I'm
asking
is.
We
can
use
something
like
oh
profile
or
anything
else
to
get
statistical
sampling
of
stack
traces
during
execution
of
one
of
these
runs
that'll.
Allow
us
to
get
attribution
of
where
c-store
is
spending
its
time
and
that'll
tell
us
what
to
optimize.
So
I
think
that's
the
next
step.
I
can
use
the
perv
to
to
yes
precisely
yeah
yeah
yeah,
so
I
think
that's
the
next
step.
A
C
Modify
some
cbd
tool
to
support
fio
matrix
analysis.
So
that's
all.
B
Hello,
two
main
two
main
things:
first
of
all,
rook
integration
isn't
a
pr
that
adds
the
ammon
host
override
option,
the
mama
building
another
thing:
the
up
sequencer
finally
went
through
technology.
I
haven't
found
any
memory
issues
like
use
after
free
or
basically
lifetime
management
issues.
So
I'm
going
to
send
it
as
a
as
a
pr,
and
now
I'm
doing
through
results
of
the
backhand
technology
run
looks
there
is
a
problem
with
there
is
a
stuck
during
appearing
trying
to
analyze,
trying
to
drill
through
that.
That's
me.
B
A
Welcome
okay,
sort
of
also.
C
E
I'm
happy
I
I'm
helping
eating
to
the
retune
the
intel
return
to
profiling,
the
crimson
on
the
memory
store,
as
you
see
from
trimming's
result
that
there
will
be
high
cpu
utilis
utility
in
the
crimson,
especially
on
the
c
store.
E
So
we
would
like
to
check
what's
happening
there,
but
based
on
my
prior
experience
as
I
I'm
also
working
on
some
storage
project
with
the
customer
with
the
other
routine
programming
on
the
sort
of
the
storage,
there
will
be
high
cpu
utility
in
the
co-routine
design
that
that
will
be
shown
in
the
high
eye
catch
me
and
the
branch
prediction
issues
there.
So
this
may
be
caused
by
the
co-team
contact
switch
or
something
others,
but
I
think
we
will
see
what
happening
on
the
this
door
there.
E
So
when
junaid's
testing
is
ready,
so
we
can
use
the
retune
to
check
the
details
on
james
setup.
So
that's
it.
A
I
think
that
would
be
great.
I
do
want
to
point
out,
though,
that,
like
I
said
before,
there
are
a
lot
of
very
basic,
just
storage,
algorithmic
things
that
c
sort
doesn't
have
yet
one
of
the
more
basic
strategies
for
a
log
structured
file
system
is
to
write
expected
long
live
extents
not
to
the
journal
segment
and
c-store
doesn't
do
that
yet
so
the
segment
cleaner,
garbage
collector
has
to
look
at
every
piece
of
data
written
it's
wildly
and
inefficient.
A
So
that
probably
is
the
first
thing
to
fix
details
of
instruction
cash
business
will
also
be
interesting,
and
I
am
interested
in
seeing
those
results
as
well,
but
there's
lower
hanging
fruit.
I
think.
E
Oh
yeah,
okay,
that's!
That
would
be
very
interesting
because
we
will
see
what
happens.
A
If
that
is
the
problem,
then
I
would
expect
you
to
see
a
lot
of
a
lot
of
time
spent
in
the
garbage
collector
in
the
segment
cleaner.
Specifically,
so
you
should
be
able
to
confirm
or
roll
that
out.
E
Okay
and
another
question
here,
is
that,
as
the
curve
has
leaved
red
hat,
don't
know,
is
there
anyone
from
your
team
to
backfill
his
role.
A
A
Yep,
you
can
talk
to
me
if
you
need
not
someone
on
the
right-hand
side.
B
A
A
Other
people
will
probably
start
working
on
crimson
over
the
next
year,
so
perhaps
someone
else
in
a
more
convenient
time
zone.
I
try
to
be
pretty
responsive
to
emails,
though
so
well,
for
what
it's
worth
anyway,.
E
Okay,
yes,
so
far,
because
I
heard
that,
because
I
just
talked
with
the
incidence
manager
that
intel
will
add
more
resources
on
the
crimson
project,
maybe
in
the
next
quarter
or
sometimes
so
yeah.
That's
a
good
nurse
yeah.
D
Last
week
I
was
mainly
working
on
two
things.
The
first
is
I'm
still
developing
the
the
multi-device
support
versus
door
right
now,
I'm
trying
to
modify
the
behavior
of
segment
cleaner
to
make
it
work
with
multiple
devices.
I
think
I
there
needs
more
sophisticated
modifications
for
it
to
work
correctly.
D
I
think
I
thought
this
too
easy
earlier
than
this
this
month.
Right
now,
I'm
I'm
working
on
the
extended
placement
manager,
pr
I'm
trying
to
implement
a
standard
condition
variable
in
in
crimson,
because
I
think
the
extent
placement
manager
needs
it
needs
it
to
work
correctly.
That's
all
for
me.
A
I
okay,
so
my
two
things
are.
I
would
encourage
you
to
there's
only
one
condition:
variable
usage
in
this
patch
right.
A
You
need
to
send
out
several,
you
need
to
send
out
several
ios
and
you
need
to
wait
for
them
all
to
complete
before
continuing.
D
Where
the
the
condition
variable
is
needed
is
that
when
multiple
conceptions
try
to
write
to
the
same
segment
or
same
set
of
segment,
they
need
to
be
coordinated
to
on
the
edge
of
segment
rolling.
That's
where
we
need
condition
variables.
A
Sure,
okay!
Well,
you
can
certainly
introduce
more
posix-like
unconditioned
variable
if
you
need
to,
but
you
could
also
just
you
know,
add
a
mutex
and
take
it
appropriately.
That
would
be
fine,
too
either
way
for
the
segment
cleaner
thing
you
mentioned.
I
would
encourage
you
to
try
to
break
that
into
smaller
pr's.
If
you
can,
I
think
it
probably
means
breaking
the
segment
cleaner
into
sort
of
pigment,
cleaner
zones
or
something
one
associated
with
each
device.
Something
like
that.
A
If
you
could
introduce
that
as
an
initial
pr
that
might
be
a
good
foundation
or
if
you
can
find
it
simpler,
well,
buggier,
not
buggier.
I'm
sorry
proper
way
to
do
it
in
the
meantime.
That
would
be
fine.
D
Okay,
too
I'll,
I
will
try
to
submit
our
easy,
a
simple,
simple,
simple
implementation
in
the
first
place,
right.
C
B
Just
quick
question
about
lba
manager.
Recently
I
spotted
a
on
one
of
the
builders.
I
spotted
a
fair
to
build
from
source
issue,
which
was
about
iterators
natural
constructability.
B
I
just
just
wanted
to
ask,
because
I
saw
the
misc
cleanup
pillar
just
wanted
to
ask
whether
it
targets
the
problem
or
it
still
needs
our
probably
email
me.
B
From
yeah,
but
I
need
to
need
to,
I
need
to
find
it.
I
will
I'll
provide.
A
D
I
think
I
met
this
issue
also,
and
I
think
this
this
comes
up
when
you
use
gcc
9
and
I
move
to
use
g610
this.
This
error
has
disappeared.
A
Okay,
okay,
I
didn't
want
to
point
out
for
anyone
who
cares
be
start
now
defaults
to
no
damage.
So,
if
you
need
it
to
run
in
the
foreground,
there's
a
crimson
foreground
option.