►
From YouTube: Crimson/Seastar OSD Meeting 2019-01-15
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).
B
C
B
A
B
Otherwise
Amanda
will
mark
the
the
Christie
town
and
then
out,
if
it
hasn't
heard
from
from
the
community
for
a
long
time
and
I,
think
I'm,
not
sure
if
we
want
to
to
enable
support
of
heartbeat
message.
But
if
we
really
want
to
do
enable
credulity
to
walk
along
with
existing
steel,
we
we
will
need
to
enable.
D
B
B
B
B
B
B
B
D
B
B
B
B
B
D
C
Are
we
absolutely
sure
that
we
are
seeing
what
we
think
we
are
seeing?
I
was
last
week,
I
was
taking
a
look
on
the
on
the
profiling
output
from
wall,
clock
and
I'm,
really
surprised,
because
I
made
an
assumption.
We
are
the
sister
based
messenger
is
slower,
so,
presumably
because
of
doing
some
extra
user
user
space
work
but
from
the
traces
I
can
see
that
the
sister
based
messenger
spends
around
40%
of
its
time
in
read,
while
the
alt
pacing
messenger
was
taking
only
20%
of
its
time
in
kernel
space.
C
Oh,
it
seems
contrary
to
what
I
was
thinking.
I
would
expect
that
if
we
are
in,
if
we
are
slower
in
the
Crimson
messenger,
then
it's
because
of
doing
some
unnecessary.
So
some
extra
work
at
user
space
because
they
well
the
absolutely
perfect
zero
overhead
messenger
will
be
only
doing
I
owe
nothing
more.
Just
I
owe
when
the
aisle
resides
in
kernel
space,
but
we
are
using
the
POSIX
dock.
This
means
that
such
perfect
ideal
messenger
would
spend
hundred
100
percent
of
its
time.
Just
in
kernel,
space.
C
D
Reverb
there
are
extra
work
added
by
the
syste
scheduler
for
the
synchronous
task.
Where
code
is
a
synchronous
and
each
each
richer
future
will
go
to
the
scheduler
to
schedule
to
do
the
schedule
and
they
have
the
Ipoh.
Are
they
the
pooling?
So
we
have
extra
work,
but
if
the
request
is
not
parallel,
it's
not
parallel.
It
is
if
the
Sandara
Craster
one
by
one
theory
stereo
though
we
couldn't
get
the
benefit
from
the
Caesar
framework.
C
C
C
Reach
this
collar,
of
course,
it
can
come
from
the
from
the
socket
rate,
but
what
I'm
talking
about
is
the
share
of
the
rich.
It's
money
you're
in
the
case
of
a
messenger.
This
suggests
that
well
I
mean
we
could
be
more
effective.
Maybe
we
are
just
not
consuming
the
full
CPU.
Maybe
you
are
we
have
some.
Maybe
we
are
somewhat
block.
I,
don't
know,
I
would
love
to
see
more
data,
especially
coming
from
pair
of
stud.
D
C
D
C
C
C
C
B
B
C
Maybe
our
reach
are
slower.
Maybe
they
require
more
CPU
because
of
of
Chang
side
of
Chang,
sighs,
I,
don't
know
what
I
would
love
to
see
is
just
the
what
I
would
love
to
focus
is
just
to
explain
why
those
numbers
are
contrary
to
each
other,
I
mean
consuming
providing
less
cups
while
being
more
in
the
reach
Cisco
by
spending
more
time
in
this
collar.
That's
the
death,
the
surprising
part.
B
D
Okay,
so,
and
last
week,
I
after
I
talked
to
mr.
foucha
panda,
though
I
drafted
a
crimson.
Try
the
model,
though
I
switch
out,
that
we
did
in
the
decide
a
the
fire.
Finally,
that
strata
model
for
the
chromosome
we
just
first
took
after
the
whites,
Ryder
OSD.
First
two
tasks
to
make
the
make
the
EOC
tempo
and
we
can
quickly
implement
and
the
tester
data
reduction
for
the
final
design
destroyed
a
model.
So
I
don't
think
we
have
the
final
decision
for
it.
D
So
at
least
all
the
possible
try
to
model
and
and
the
to
share
our
mission
did
ok,
something
I
me
over
fabric.
So
it
give
me
a
impression
that
I
speak
very
over
fabric
all
as
gauzy
over
fabric,
something
that
was
right
and
running
dedicated
a
task.
But
after
I
take
a
look,
I
searching
from
the
other
night
and
couldn't
gather
the
the
this
kind
of
description
so
only
like
what
we
do
in
the
crimson.
Messenger
was
rather
work
after
a
scepter
and
then
send
the
message
to
other
boss.
D
D
D
Some
architecture
description
about
SPD
case
so
I,
don't
know.
If
there
are
some
special
strider,
we
will
run
just
a
father.
Dedicated
task
only
only
finish
special
tasks.
I,
don't
know
whether
to
Chara
asking
me
to
reference
to
the
SPD
que
mi
mio
I,
discuss
it
over
fabric,
so
decided
that
it
is
already
verified
in
SPG,
but
I
don't
know
exactly,
but.
B
B
D
B
C
Well,
some
time
ago,
when
I
was
working
at
me
on
T's
was
we
are
taking
our
look
on
this
Amelia
thing
called
qat
the
coprocessor
source
crypto
for
compression
that
are
supposed
to
beat
to
offload
some
workloads
in
the
PDK,
slash
SPD
case
scenarios
and
though
I
am
I'm,
making
a
guess
that
the
dose
all
hardware's
use
Miller
approach
to
the
problem.
There
is
a
rigid
device
offering.
D
C
D
C
D
D
D
B
D
A
A
D
C
Okay,
anyway,
it's
it's
a
story
for
state
the
at
least
the
second,
the
second
stage
of
of
crimson
or
first
well,
I
guess
all
we
need
is
just
to
test
the
network
plus
plus
many
star
liars.
That's
the
golfer
first
stage,
acidic
a
maybe
for
second
one
but
I
to
sure
it's
focused
on
I
guess
he
wants
us
to.
He
wants
us
to
help
with
the
architecture
just
to
be
prepared
for
for
consuming.
B
D
B
D
D
D
Articles
read
and
they
run
run
no,
let's
write
a
perk
up,
but
that
there
is
multiples
right.
The
Rancho
completed
for
the
karate
be
and
SPD
key
I
didn't
defined
that
the
lastest
right
amount
of
all
he
has
Polly
documented
and
I
will
check
with
xuxa
what
I
mean
I
mentioned
at
that,
and
then
just
take
a
look
at
a
chromosome
messenger
and
then
I.
D
If,
if
there
is
no
other
other
developing
work
out,
what
we
do
I
will
helper
easy
and
the
German
tool
to
trace
the
crimson
messenger
and
them
I
think
we
should
have
tested
the
multiple
connection
because
and
I
got
the
FL
just
a
web
connection.
They
not
understand
it
request
the
parallel
and
that
will
not
take
advantage
of
the
system
a
framework
if
it
is
send
a
request
area.
B
D
But
I
think
182
io
q
used
in
the
user
space
for
the
aisle
to
the
scheduler
user
space
at
all
scheduler
the
io
q.
In
the
sister.
Can
you
learn?
It's
depend
on
the
device
device
capability
when
you
are
the
machine.
So
if
it
is,
if
the
io
taps
is
not
bigger,
multiple
multiple
of
Scala
tepees
rider
will
share
the
same
io
q.
D
B
D
B
D
D
B
A
And
and
the
little
incentive
the
new
new
new
system
methods
and
I
I
read
the
code,
and
this
visited
the
new
kernel
feature
it
used.
The
I
was
a
Muslim,
submitter
and
and
by
pastor,
are
you
get
a
you
enter
to
to
read
the
renew
borrow
from
the
user
user
species
ooh,
so
the
performance
may
be
better,
but
I
think
this
bootiful
sista
and
the
chromatin
would
be
so
if,
if
do
it
better?
B
A
A
A
A
C
Could
we
kill
those
servers,
just
after
just
after
finishing
the
I/o
test,
I
guess
that
that
I,
the
real
testing
takes
a
lot
of
time,
maybe
I'm
thinking
about
minutes.
So
even
if
there
is
the
between
finishing
the
iopa
eating,
the
server
I
would
expect
will
be
quite
small,
maybe
maybe,
in
the
seconds
the
mine,
okay,
we
would
got
some
drift
in
CPU
measurements,
but
still
it
can
be
quite
acceptable
to
verify
what
is
going
underneath.
D
A
C
Maybe
maybe
another
way
exist.
I
first
understood
those
tests
were
made
with
kill
death
equal
to
one
one
connection.
We
cannot
a
client.
We
was
able
to
issue
a
new
connection,
a
new
request
before
finishing
the
previous
one.
Talk
you
depth
equal
one.
If
I'm,
if
I
correct
in
such
situation,
Rupert
is
determined
by
latency.
So
if,
if
this,
our
messenger
has
include,
has
bigger
latency
than
async
messenger.
B
B
C
B
B
C
B
D
C
D
E
A
Okay,
are
we,
are
we
apply
or
suggestion
to
the
understand
the
center
without.
C
Last
week,
I
was
focused
mostly
on
the
on
the
Engel
credit
versus
multiple
threads
in
his
mouth,
basically,
single
OSD
versus
multiple
or
as
the
cases
in
in
decriminalize
due
process.
We
had
a
very
good
discussion,
quite
a
long
discussion
at
the
perf
and
the
weekly
part
meeting,
and
a
lot
of
folks
were
well
step.
Well,
I
guess
they
were
thinking
that
their
visibility
rules
are
adhering.
A
lot
of
people
were
thinking
that
we
want.
C
We
won't
be
able
to
consume
more
CPUs
with
we
than
one
with
the
trims
know
as
the
process
in
the
single
OSD
scenario.
Well,
that
wouldn't
be
the
case
even
because
of
the
crimson
because
of
the
of
the
alien
adapter.
We
would
have
in
crimson
explain
explaining
that
that
the
object
or
interface
allow
us
to
paralyze
some
operations
teams
to
treat
a
lot
of
confusion.
There.
C
C
If
for
the
best
possible
configuration,
if
there
is
an
unnecessary
consume,
alt
object
start
like
blue
star,
then
we
will
go
we've
with
some
kind
of
adapter
having
on
fret
board.
So
in
such
case,
like
saving
booster,
for
instance,
would
require
more
more
course
for,
but
there
will
be
only
single
sister
reactor
in
the
crims.
The
OSD
process.
B
C
C
B
C
B
E
B
E
B
B
But
what
I
think
of
the
my
head
is
that
we
need
to
do
so.
In
first
thing
is
due
to
that
past
s
SMP
one
to
the
to
accrete
crimson
toasty,
because
that
Caesar
is
used
by
biospark
common
common
person
powder
and
another
thing
that
we
currently
in
the
in
the
in
my
PR
I
always
submit
the
post
e
to
the
zero
current
little
chart
before
we
have
any
way
to
to
specific
what,
as
I
design
it
a
different
chord
to
the
theta
application.
There,
though
it's
it's
another
concern
anymore.
E
Admit
to
the
core
zero
right
for
the
to
start
again
wise.
So
so
in
this
logic,
it
submit
to
core
zero
to
start
the
OST
shard.
So
if,
if
the
sister
app
is
he
started
with
with
see,
because
one
though,
though,
so
in
my
in
my
understanding,
there's
no
need
to
submit
because
zero,
if
the?
If
all
you
have
is
limited
to
one
call,
I
will.
B
C
C
B
Ideally,
to
be
done
in
make
a
test
phase,
but
we
can
pass
the
CPU
mask.
For
example,
we
we
can
assign
for
zero
to
two
or
three,
but
one
and
call
core
Core
to
two
or
zero
to
two.
If
we
can
do
this
and
passed
the
CPU
mask
as
a
command
command
lines
option,
we
got
to
be
ideal
and
sure,
and
we
can.
We
can
keep
the
keep
the
criminology
aware
of
these
facts
that
it
leads
online
to
a
certain
OSD
of
sorcery,
a
certain
core.
B
B
B
So
just
like
just
like
SPD
chaotic
media,
we
need
to
pass
up
CP
mask
to
it,
to
relate
it,
to
King,
to
our
key
to
a
certain
core
I
think
the
same
PI
to
sister,
but
I.
Don't
think
it's
a
big
deal
we
can
always
handle
it.
Yeah
think
think,
think
the
menu
measurement
of
playing
putting
on
weight
in
the
mchavez
face
or
or
in
the
in
the
wrong
time.
Well
add
another
option
in
the
option.
Thought
see
see
it's
a
big
deal.
E
E
E
B
B
E
B
B
E
E
B
I,
don't
think
we
need
to
pen
and
pen,
dare
you
to
win,
but
that
the
feather
word
for
it
is
to
stash
it,
because
we
cannot
make
the
decision
yet
before
we
have
the
number
out
and
I
think
what
we
need
to
keep
it
pairing
in
our
mind
is
that
we
will
probably
switch
over
to
em
twin
mapping.
But
the
foremost
thing
to
do
is
make
sure,
though
we
have
have
a
working
plantation
if
for
for
the
one-to-one
mapping,
and
it
should
perform
well.
B
B
D
E
E
Yeah,
it
fix
a
park
that
with
multiple
connections
it
can
crash,
but
with
this
fix
it
will
not
crash
okay.
E
D
E
E
E
B
E
B
B
Well,
I,
don't
think
to
be
on
par
with
the
the
v2
incrimination,
Basin
messenger
propeller.
We
need
to
need
to
be
good
enough
to
be
comfortable
with,
with
with
with
monitor
and
nosed
Mayo.
For
example,
I.
B
B
B
B
D
B
B
B
D
B
B
Because
in
our
in
an
ideal
world
and
requests
will
be
when
I
request
richest
booth
D,
it
will
host,
he
will
dispatch
the
request
to
the
tip
to
a
Cuban
PG,
but
the
other
dozen
multiple
places
where
the
PG
won't
be
able
to
serve
the
map
under
the
request,
for
example,
if
the
PG
is
not
quickly
yet
or
it's
not
clearly.
Yet
we
need
wait
for
read
a
PG
they're
ready
to
first
several
requests,
and
then
there
are
some
cases
where.
D
B
D
B
B
There
is
case
that
the
odo
toasty
is
done
is
marked
down
and
we
need
to
and
the
the
the
monitor
as
I'm
assigning
annuity
for
for
the
full
action
set
and
we
need
to
the
new
new
primarily
need
to
be
P,
PP
being
formed
with
the
same
information
and
catheter
active
darkling.
Then
it
will
be
able
to
serve
the
request
targeting
this
PG,
but
before
that
now
the
now
the
Ostia
will
be
serving.
That
request.
Am
I
right.
A
B
C
B
About
that,
I
will
try
to
come
up
with
their
Dennis
implementation
of
the
read
past.
Well,
where
we
have
our
only
very
simple
handle
Oh
t.o.p
methods,
which
will
return
a
rhythm
rhythm
father
list
just
to
fulfill
the
request
from
from
the
kind,
and
we
can
improve,
improve
or
make
it
more
complicated.
Based
on
with
on
it,
make.
B
B
B
E
D
C
Just
for
current
or
it's
not
for
sister
well,
actually
it
is,
but
it
it's
using
care.
The
current
they've
always
D
process,
as
and
subject
to
me
to
some
experiments
to
some
to
the
extraction
of
the
of
the
crucial
of
the
fast
IO
path
leads
over
replicated
bucket.
It
doesn't
care
about
easy
polls,
it
doesn't
care
about.
Some
special
cases.
Handling
is
just
the
hot
path
for
for
reads.
D
C
It's
some
kind
of
extension
to
current
OSD
to
focus
on
the
hot
path
of
of
process
of
rich
processing,
because
the
current,
the
general
OS
Dave
general
IO
path
in
current
residue
is
really
really
complicated.
It
just
had
understand
so
I
spent
some
time
and
try
to
sect
all
those
parts
of
the
IO
path
that
are
relevant
for
handling
reads:
just
reached
no
rights
no
or
map
just
reach
over
replicate
the
bucket.