►
From YouTube: Crimson/Seastar OSD Meeting 2018-12-25
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
A
B
A
C
B
C
C
B
A
B
Okay,
there
it's
a
standard
as
to
which
it
how
what
is
the
driver
looks
like
it's
a
ordinary
readers,
client
or
some.
B
A
A
A
A
C
So,
okay
cooperate
bullet
here
from
a
sink
messenger.
To
this
doll,
messenger
is
already
tested
right,
but
if
we
need
to
test
from
crimson
messenger
to
criminal
messenger
or
from
Krenzel
messenger
to
a
synchronous,
a
sink
messenger,
we
need
the
a
fire
engine
works
with
the
crimson
messenger
interfaces.
C
B
B
D
Hello,
I'm
working
together
messenger
with
with
yours,
o
prototype
of
OSD
driver
I,
have
it
buildable.
The
problem
I'm
focused
on
right
now
is
trying
to
bring
the
monitor
client
to
the
shard
that
messenger,
because
interfaces
has
changed,
especially
the
connect
method.
Well,
we
have.
We
have
complications
there,
like
the
connection
cross-reference,
I
posted
some
comments.
D
A
C
C
D
C
C
C
A
D
C
But
the
foreign
pointer
can
be
wrapped
by
a
local
shared
pointer,
though
it
can
be
copied
it.
Only
the
the
the
foreign
pointer
will
be
the
only
one
instance
in
in
the
foreign
core,
but
the
the
local
shared
pointer
of
that
foreign
pointer
can
be
copied
around.
So
that's
why
this
is
design
the
ax
reference.
D
C
C
A
D
D
B
C
C
D
C
But
there
is
issue
here
because
okay
load
balancer,
is
only
working
when
the
connection
when
the
socket
is
accepted
right,
but
our
requirements
needs
to
move
connection
after
the
first
banner
change,
so
the
pier,
so
the
connection
will
know
the
pier
adjusts,
nonce
and
the
nonce
and
port.
That
means
the
fool
PR
just.
C
C
C
B
A
C
B
B
C
B
D
A
D
A
D
A
D
C
D
C
C
So
there
are
two
different
scenarios.
The
first
is
when
the
messages
is
received
from
by
OSD
and
OIT
will
responsible
to
dispatch
mass
message
to
the
corresponding
PG.
It
is
maybe
cross
Cod
and
another
scenario
is,
is
if
PG
wants
to
send
some
message
to
be
some
reply
to
to
to
other
peers.
The
PG
itself
will
will
will
handle
without
the
the
connection
instance
and
chords
and
interface,
and
the
cross
core
will
be
handled
inside
the
connection
itself.
D
Well,
I'm
asking
this
the
shouting
concept.
It
looks
pretty
complicated
to
me
and
I'm,
just
considering
the
mother,
most
Hampton
Terrier
possible
I,
guess
scenario
with
without
any
shouting,
basically
shut,
nothing
with
messenger
and
OSD.
And
finally,
the
final
object
store
implementation
sitting
just
on
one
single
card.
D
D
C
B
D
D
D
Limit
and
of
course
you
can
for
spinner
basically
will
have
you
will
have
any
OS
this
as
you
have
their
site,
but
for
bastard
devices
I
can
be
miss
well,
even
now.
It's
not
unusual
practice
to
see
multiple,
always
these
occupying
the
same
device.
They
are
just.
They
are
just
divided
on
the
partition.
Mother
largest
partition.
D
B
Because
if
we
can
hide
the
share
the
infrastructure
underneath
the
in
case
encapsulated
in
a
layer,
probably
we
can,
we
can
put
on
television
for
later.
C
D
D
C
D
B
Don't
think
it's
a
it's
a
it's
worse
than
what
we
have
in
existing
ot
in
existing
in
this
model.
It's
always
the
managers
are
single,
let's
say
and
objects
devised
over
partition,
assign
it
to
a
toasty.
If
we
have
a
for
like,
for
example,
like
20
devices
or
repetitions
for
for
for
for
each
austere,
we
will
have
a
20
of
them
in
a
single
host.
A
B
D
Entirely
correct
for
the
spinners
case
for
nvm
ease
when
we
can,
especially
for
random
reads
when
we
can
use
when
we
can
leverage
the
multiple
to
POS,
DTP
spreads.
I
expect
we
see,
we
would
need
to
have
more
a
bit
more
or
less
this,
but
the
upper
limit
is
8
times
more,
but
on
average
I
would
expect
maybe
two
or
four
times
more
on
nvme
sound.
Of
course,.
D
D
Don't
if
you,
if
your
operation
is
well
per
eligible
e
OSD,
can
consume
a
lot
of
kasi
few
cars
with
by
putting
the
limit
one
car
per
OSD.
Only
we
will
meet
just
more
or
is
this
saturate
edges
to
saturate
fast
device,
but
I
wouldn't
expect
the
number
will
be
will
be
bigger
than
maybe
eight
or
something
like
that.
That's
that
parameter
I,
guess.
D
D
D
B
D
B
B
Yeah,
yes,
we
see
level
of
the
map
cache
and
to
only
memorize
the
tear
type
between
between
request
map
and
SD
map,
cetera
I
think
we
do
it
for
no
reason.
B
Why
what
what
what
I
can
think
of
is
that
the
contractor?
These
are
probably
two
to
make
sure
we
are.
We
have
good
good,
I'm
standing
and
a
good
designer
what
kind
of
part
of
our
can
be
committed
by
by
Kabhi
both
of
this
approach,
the
single
process,
approach
and
multi-process
approach,
for
example,
the
the
core
shot,
the
core
shot
messenger
is
the
it's
definitely,
sir,
for
the
multiple
process
of
multiple
strategy
that
criminals
teeth.
B
While
the
single
person
before
the
we
migrate
to
the
multiple
course
called
messenger,
we
the
design,
the
design
works
for
the
single
threaded
of
the
messenger,
and
we
and
we
we
need
to
also
to
have
a
clean,
clean,
clean
line
drink
to
them,
to
make
sure
we
we
intended,
which
part
of
them
can
work
for
most
of
them
and
have
a
better
design.
If
we,
we
feel
that
it's
very
complicated
under
this
number
was
why
we
can
switch
over
to
that
single,
ready,
I.
B
C
A
C
A
D
This
would
allow
to
to
very
easily
test
all
parts
that
are
put
there
in
the
product,
just
with
the
shot
that
with
with
single
messenger
implementation,
and
if
we
discover
that
the
additional
costs
related
to
just
the
book
just
being
able
to
handle
multiple
shots,
we
can
quite
easily
squeeze
this
overhead.
I
guess.
C
B
B
B
B
A
Tell
me:
okay,
okay,.
B
The
upside
of
it
of
it
is,
we
have
ecology
here.
We.