►
From YouTube: Ceph Crimson/SeaStore 2021-12-01
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.
Last
week
I
was
on
vacation.
I
got
some
reviews
done
earlier
this
week
and
sorry,
I'm
gonna
put
together
a
version
of
that
branch
with
the
refactor
we
were
talking
about,
didn't
have
a
chance
to
get
to
it
today,
but
I'll
get
to
it
tomorrow.
B
A
All
right,
chennai
how's
it
going.
D
I'll
have
to
redux
the
fix
some
interruptable
future
objective.
He
meets
in
his
task
environment,
and
that
is
the
background
recovery
and
he
has
already
tested
and
no
back
found,
and
it
means
that
the
problem
is
solved,
so
the
patch
is
merged,
and
when
I
do,
the
technology
testing
still
has
a
proper
problem
and
the
ost
windows
startup,
the
heartbeat
reported
the
ip
address
doesn't
doesn't
match
so
they
got
the
looper
look
back
interface
address
one
127.
D
Point:
zero,
zero
point,
zero
point,
zero
point
one
and
the
peer
reporter
is
that
address
and
his
local
address
is
zero.
Zero
zero
zero.
So
I
still
compare
the
classical
cefu
messenger
and
the
chromosome
messenger
try
to
figure
out.
What's
the
difference
and
no
result
currently
yeah,
that's.
A
But
it
works
correctly
with
normal
without
c
store,
right.
D
A
Yeah,
I
think
you're
right,
I'm
asking
because
I
think
that
test
used
to
pass
so
it
might
be
a
regression.
Have
you
mentioned
it
to
radic.
A
I'm
saying
I
think
the
tester
running
should
should
have
worked
with
blue
store
in
the
past,
because
radik
added
it
to
the
suite.
So
I'm
asking
have
you
asked
radic
about
this
bug.
B
D
B
F
B
Hey
son,
as
mentioned
you,
it's
like
a
to
check
out
the
source
code
from
the
gate,
and
then
I
tried
with
with
that
also
facing
the
issues
in
the
safe
idm.
So
I
updated
those
details
in
the.
B
Okay,
at
all,
I'm
telling
them
cool.
Let's
see.
B
A
C
Oh
last
week
I
was
mainly
working
on
the
two
crash
issues
that
I
met
during
when
evaluating
the
spread.
A
lot
spread
lfs
strategy
on
systole
I've
said
I've
submitted
prs
for
them
separately
right
now,
I'm
working
on
correcting
the
segment
utilize,
the
segment
utilization
calculation
for
spread
lfs
strategy;
evaluating
that
that's!
That's
all
thanks,
cool.
E
A
couple
of
performance
analysis
to
understand
the
raido's
bench
and
the
rbd
fio
behaviors
and
the
result
is
here.
So
let
me
clean
it
a
little
bit.
So
what
I
found
with
ibdf
io
is
that
during
the
test,
the
all
node
3
depths
cannot
be
increased
because
we
only
have
a
limited
size
of
the
image.
So
we
only
have
the
limited
numbers
of
phone
node,
so
we
can't
actually
increase
the
all
node
numbers
and
the
depth
so
that.
E
The
transaction,
the
transaction
complex,
will
be
a
problem
so
that
that
is
why.
I
think
that
if
we
want
to
evaluate
the
performance
or
test
the
performance,
we
should
switch
to
at
least
to
radius
bench,
because
yeah
from
the
graph,
the
radius
bench
will
have
a
considerable
the
insert
and
delete
operations,
which
is
some
levels
more
than
the
ipfio
tests.
A
A
B
B
A
Because
different
pgs
won't
submit
yeah,
so
we
could
do
something
like
force.
The.
A
It
just
makes
the
code
slightly
harder
to
read:
that's
all
it's
not
a
correctness
problem,
so
yeah,
that's
one
strategy.
The
other
strategy
is
when
we
do
want
to
do
random.
I
o
tests.
We
just
also
run
a
raido's
bench
to
create,
I
don't
know
yeah
8,
000
objects
or
something
and
then
run
radius
mesh.
On
top
of
that,
that
would
work
too.
A
C
A
E
Does
that
seem
good
yeah,
so
the
next
thing
the
journal
ad
emerging?
E
E
By
seventy
percent,
and
so
currently,
I
also
did
some
analysis
to
understand
the
current
main
issues.
First
is
it
is
also
in
that
kiosk.
E
A
A
E
Analysis
also
points
out.
The
current
workload
is
cpu
bounded,
not
maybe
not.
I
bounded
yeah
reactor
utilization
is
100,
so
it
may
be
very
beneficial
if
we
can
start
to
optimize
the
less
performance
logics
to
increase
the
efficiency.
B
E
And
the
third
thing
here
is,
if
you
remember,
the
stress
tool,
is
keeping
updated
by
xinyu
in
our
team
and
because
that
is
because
we
are
doing
performance
comparison
between
restore
and
systole,
by
limiting
the
blue
store
to
use
a
single
cpu.
To
compare
the
to
understand
where
we
are
current,
where
we
are
currently
with
the
systole
development,
and
the
final
report
is
not
done
yet,
but
but
I
think
the
result
is
very
promising.
E
Even
even
even
now
the
question
real
cash
is
not
yet
implemented
and
it
is
causing
the
crimson
rsd
to
have
more
than
100
times
more
device
read
than
the
classical
sd,
but
even
that
the
raido's
bench
result
is
still
competitive
between
the
bluetooth
and
system.
I
think
that
that's
a
good
news.