►
From YouTube: Crimson/Seastar OSD Meeting 2018-12-18
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
B
D
E
B
C
Depth
in
our
body
with
this
device
ahead
of
you,
mr.
branch,
there
Simic
team
exchange
has
finally
cut
cut
merged
and
before
that,
I
actually
see
the
body
but
based
upon
the
works
in
progress
branch
from
jetty,
and
where
is
the
signal?
Changes
is
the
pipe
right
by
the
enzyme
exchange.
It
was
not
Murphy
yet
so
I
need
to
get
updates
our
branch
once
the
same
exchange
is
an
object
most
upstream
and
also
I
I
reviewed.
It.
C
I
said
there
won't
be
lot
of
pushback
from
from
from
him
the
way
just
a
movie
there,
a
gate
from
from
messenger
to
to
the
connection
there,
there's
no
fundamental
changes
actually
and
I
believe
we
can.
We
can
even
go
further
by
moving
the
dispatcher
from
probably
moving
that
is
beggar
from
in
messenger.
C
C
C
I
think
the
whole
point
of
a
mental
city
under
crimson
did
the
placeholder
and
we
will
surely
thing
Richard
to
do
we
move
on
and
currently
the
the
four
most
important
thing
is
to
two
integrated
eating,
with
with
up
to
the
messenger,
by
punishing
and
and
to
sit
and
to
prepare
the
dessert,
so
it
can
enjoy
in
the
cluster
and
later
on.
We
were
ready
the
facilities
like
object,
store
and
the
services
which
will
provide
the
underlying
storage
and
community
for
structure.
I
share
the
pie,
charts
and
I
pray
to
your
I
agree
with
you.
C
C
I
think
we
we
need
to
discuss
the
case
by
case
on
the
critical
path
which,
which
we
think
is
what
working
on.
We
need
to
start
from
scratch.
I
think
other
books
and
with
the
thing
keeps
a
philosophy
of
a
run
to
completion
and
there
nothing
design
in
mind
instead
of
tried
to
port
existing,
could
pass
to
persist
with
sister
and
on
the
non-critical
path.
C
We
will
try
to
better
leverage
what
we
have
and
if,
if,
if
the
ten
time
allows
we
can,
we
can
42
to
redesign
it
later
on
but
considered,
but
it
all
time
comes
to
is
limited.
So
we
need
to
reuse
it
with
reduce
them
as
much
as
possible,
but
the
query
past.
We
need
to
rewrite
and
rewrite
it
from
scratch.
C
But
I
we
could
but
but
the
things
well.
Actually
we
we
have
discussed
this
when
we
started
the
crimson
project
regarding
to
how
we
share,
we
can
cooperate
on
the
this
project.
C
E
A
A
A
C
A
C
Or
this
we
could,
we
could
suggest
we
could
have
a
dedicated
branch
in
every
port
and
I
know
Co,
coordinate
on
that
repo
without
the
and
prÃncipe
stabilize
that
product
into
the
appear
when
I
design
all
code
is
the
right
and.
C
All
the
workers
and
y'all
pray
for
an
approach
for
Michaelson.
That
said
long
term.
If
look,
we
can't
get
quit
allocated
design
correct.
It
will
undermine
our
long
term
time
ago,
but
I
understand
your
concerns
that
if
we
don't
get
a
good
number
to
prove
that
the
crimson
project
is
promising,
we
can
hardly
attract
the
people
or
keep
the
people
around
Iran.
For
for
this
project,.
E
A
A
C
C
B
Those
numbers
which
possible
to
obtain,
but
only
for
reads
if
he
won't
like
write,
will
be
a
would
be
well
a
bit
unreal
because
without
joining
the
cluster,
we
cannot
basically
do
the
replication
traffic.
It
will
be
comparable
if
we,
if
we
do
right
benchmarking
without
joining
the
cluster,
it
can
become.
Maybe
maybe
it
can
be
comparable,
but
only
with
I
always
do
running
replication
on
fire
running,
replicated
back
enter
and
a
factor
on
replication
factor,
one
not
so
reliable.
As
for
it
should
be
fine.
A
Yeah
to
to
make
able
to
ever
compare
we
we
make,
we
should
make
affection
into
the
system
with
D
as
the
edge
Cochran
like
the
lack
of
the
chromatin
coder.
For
this
I
I
I
agree
with,
but
but
a
mine
point
is
we.
We
should
find
a
bad
method
to
to
speed
up
this
worker,
the
pr
what
I
think
the
P
I
is
a
is
a
speeding
so
long
so.
D
C
Think
Tamara
I
think
it's
a
it
doesn't
matter
where
we
get
the
code
most.
If
we
follow
this
different
model,
you
can
to
be
honest.
You
can
even
Merc
your
the
working
progress
changed.
For
example,
I,
wouldn't
progress
from
the
Crimson
messenger
from
eating
into
your
you
know
your
your
remote
repo
or
your
local
repo
and
continue
working
on
it
and
I
have
a
under
tested
and
I
have
a
terrible
number,
for
example,
if
if
you
could
have
a
working
or
partially
working
OSD,
my
concern
is
that
how
will
this
the?
C
D
C
D
C
C
C
Pressure
from
the
engine
team-
probably
we
can,
we
can
put
together
the
circuits,
are
I'll
pass
faster
without
without
adding
the
bullet
plate.
Work
to
get
the
ulti
join
in
and
allowed
allows
the
kindest
and,
as
the
children
request
right
through
the
toasty.
D
D
D
The
promise
and
use
the
future
forward
a
true
the
promise,
though
one
after
when
the
PT
women
to
the
PG
lock,
we
got
the
promised
future
at
that
time.
The
future
is
not
ready,
so
it
returned
and
when
the
pit
locks
are
afraid,
we
use
the
future
forward
to
promise.
So
at
that
time
the
promise
is,
as
in
actual
future,
future
future.
E
C
D
C
D
B
B
D
C
D
E
D
E
C
E
E
That's
it's
at
all.
Like
works
like
the
client,
a
tool
to
control
the
messenger,
messenger
send
messages,
but
one
what
I'm
talking
about
is
a
test
case.
So
I'm
referring
to
three
kind
of
test
case.
One
is
a
sink
to
a
sink
performance.
Yeah.
Another
is
a
sink
to
crimson
performance
and
the
third
case
is
crimson
crimson
performance
and
we
can't
get
the
performance
numbers
of
all
the
three
test
cases
and
the
seed
see
the
result.
I.
C
D
D
D
C
D
D
E
C
D
C
C
C
D
Does
matter
ways
we
do
it
after
it's
a
sister
framework
star,
we
just
do
it
in
washer
is
okay
right.
All
we
do
it.
We
do
it
before
we
start
the
sister
share
ass
rider
on
the
the
cars.
C
D
D
D
C
E
C
D
E
B
Perhaps
data
random
rarely
occurring
data
corruption
during
scrub
final,
it
student.
Well,
the
buffer
list
changes
was
the
primary
shoes
finally
turned
out:
it's
it.
We
had
a
back
for
years,
but
they
just
exposed
it
and,
moreover,
it
seems
we
need
to
decide
what
to
do
with
the
year.
See
cash
removal
before
no
tedious
does
tech
year
see
cashing
well
in
buffer
list
we'd
have
because
we
have
Searcy
caching,
because
of
that
we
have
even
a
lock
right.
B
C
B
B
The
first
question
is
right
about
the
sharding
of
OSD
in
current
design.
In
current
OSD,
the
big
OSD
I
mean
the
OSD
class
has
object
star
inside.
We
have
one
object:
per
OS
d
cloud
per
OS
d
class
instance,
but
in
crimson
USD
we
decided
to
share
two
shirt
basic
on
OSD:
isn't,
isn't
that
mean
we
will
get
we
get
there?
D
B
That's
not
a
big
issue
for
the
intermediary
step:
I
mean
Iran,
you're,
taking
or
blues,
or
something
like
that
and
enclosing
int
and
closing
it
inside
inside
the
alien
world.
But
the
problem
can
arise
later.
If
we
let's
say
we
will
decide
we'll
decide
to
have
or
something
like
that
that
doesn't
require
any
alien
communication.
B
B
E
E
In
my
opinion,
from
dispatcher
class
I
think
it
is
the
same
story
and
I
think
if
just
if
OST
like
was
object,
store
interfaces
is
defined.
We
can
then
decide
if
the
underlying,
like
blue
star
C
story,
is
shut.
It
or
not,
because
currently
monitor
client
is
not
shouted
and
OSD
is
shut
it
and
they
can
both
work
well
under
dispatcher
interface,.
B
C
B
C
B
E
E
B
C
B
C
E
Yeah,
currently
we
can't
assign
OST
to
messenger
because
Asti
is
charted,
but
messenger
is
not
shouted.
So
that's
so
that's
a
major
problem.
B
E
E
E
B
B
B
C
C
B
Outside
of
the
of
the
Crimson
OSD
I
pushed
extra,
it's
the
Shatt
al-arab
ranch;
basically,
they
duplicate.
They
duplicate
key
LRU
structures.
Well,
because
before
that
we
had
three
copies
of
each
ru
entry
and
those
keys
can
be
pretty
costly.
We
can,
we
tend
to
start
with,
tend
to
use
something
like
th
object,
T
for
keys
or
even
well,
structures
things
inside
are
typical.
I
will
say.
B
C
C
So
I
really
I
would
take
a
look
at
your
a
second
before
you
are
facing,
but
I
said
from
indexing.
What
par
you
are
interesting
walking
on
the.
B
B
C
C
So
general
idea
is
to
to
have
a
blinking
and
the
chamber
working
on
account
that
visiting
crew
semester
to
get
in
shape,
I
know,
I
will
try
to,
and
the
player
run
with
the
with
the
course
cokin
corkers
miss
mr.
tree
her
to
get
it
and
marissa
and
long
with
the
book
right
and
the
trimmer
will
try
to
and
try
to
figure
out
how
to.
E
Though
one
of
our
major
target
is
to
get
pretty
performance
number
from
messenger
right,
but
what
I'm
concerned
is
the
buffer
buffer
list.
The
the
implementation
of
buffer
list
may
impact
the
performance.
So
maybe
we
should
investigate
more
on
this
one
because
it
maybe
have
a
major
impact
of
messenger
performance.
C
E
C
Okay,
I
think:
okay,
radio
has
a
point
us
to
put
the
places
we
are
using.
We
are
pending
the
well
pending
the
but
fathers
to
a
bubbly
stander.
Then
then
did
over
to
do
to
underline
packet
and
sock
it,
but
I
think
we
can
working
on
on
between
parallel
and
I
will
try
to
try
to
take
a
look
at
this
area
and
try
to
improve
it.
B
B
Yet
the
temporary
buffer
isn't
so
extremely
cheap.
It's
from
what
I
can
read
from
the
sister
source
feels
like
well
like
some
kind,
it's
familiar,
let's
say:
STD
Street
the
sharing
implementation
of
STDs
drink,
with
the
main
difference
being
love,
atomic
I'll
reference,
counting
till
reference
counting
is
there
we
have
a
deal
eater,
yes
inside
the
temporary
buffer
and
it's
actually
the
reference
counting,
but.