►
From YouTube: Ceph Crimson/SeaStore 2021-10-13
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
That's
good,
I'm
also
still
working
on
the
cache
lru
and
I'm
gonna
start
reviewing
the
the
big
random
write,
one
from
samsung
chennai
how's
it
going.
C
I'm
starting
the
technology
testing
and
I
can't
set
up
the
open
vpn
behind
the
process,
so
I
use
our
team
stronghold
team
has
his
own
tesla
testing
and
I
can
use
use
that
one
to
do.
The
testing
still
build
the
code
and
try
to
run
the
solid
test
switch.
That's
all
cool.
A
Yeah,
that
seems
good.
It's
always
good
to
have
some
testing
and
other
technology
clusters,
never
quite
as
universal
as
we'd
like
it
to
be
great
how's.
It
going.
B
Right
greg.
D
Okay,
so
while
I
was
not
unmuted,
I
was
saying
that
I
spent
some
time
looking
at
still
a
db's
messenger,
which
did
not
have
any
big
surprises
for
me.
In
fact,
it
pretty
much
just
took
every
message
and
made
them
foreign
pointers
as
far
as
I
could
tell
so.
I
don't
think
I'm
gonna
have
to
fuss
around
with
allocation
patterns
or
anything.
B
A
D
Right
yeah,
so
I'm
gonna
start
looking
at
that.
Oh
well,
I
wanna.
I
wanna
look
through
how
we
do
watch
notify
before
I
actually
start
doing
things
then
I'll
start
putting
in
an
interface
between
messenger
and
pgs
and
then
decide
which
one
we
wanna
actually
implement
for
multi-reactor.
First
yeah.
E
A
It
was
testing
what
lfs,
I
remember,
the
project
images
forgotten.
What
the
name
is
sorry.
What
was
the
day?
What
was
that
you
were.
This
is
a
strategy
you
borrowed
from
another
file
system,
something
lfs,
but
I
didn't
get
the
something.
F
I
was
looking
if
there's
a
significant
ride
overhead,
besides
gabi
cleaning,
and
what
proposed
several
prs
about
profiling
and
optimizations
and
those
are
all
merged
and
I'm
looking
through
the
multi
device.
Pr
and
after
that,
I
will
continue
to
work
on
merging
journal
headers
to
reduce
the
right
overhead
and
properly
it
will,
based
on
tremendous
commits
about
general
batching
and
there's
a
tool
that
might
be
interesting
here.
F
It
tries
to
automate
the
process
stats
to
find
out
how
to
stress
queensland
properly
from
the
results
throughput
and
the
reactor
utilization,
and
it
it
is.
It
is
very
simple
and
you
can
start
multiply
those
bench
instances
with
proper
eco
pins.
A
Just
I
I
suspect,
but
I
don't
know
that
the
reason
why
chad
may
didn't
see
a
large
improvement
was
that
that
patch
has
the
effect
of
limiting
the
number
of
outstanding
ios
at
any
one
time
to
one.
So
I
think
you
will
likely
need
multi-batching.
A
B
B
H
Hi
sam
last
week
I
didn't
get.
I
cannot
able
to
spend
time
to
test
that
there
is
a
one
issue
which
I
need
to
raise
the
well
and
I
prepared
some
scenarios,
which
I
shared
with
you
today.
I
already
shared.
H
Me-
and
I
will
this
week
I'll
continue
this
testing.
B
A
B
H
A
A
H
Yeah
exactly
I
just
mentioned
all
these,
which
other
scenarios
normally
we
identified
some
of
the
a
few,
a
couple
of
scenarios
which
are
the
important
for
the
testing
that
we
noted
it
down
and
then
in
the
google
sheet
I
already
shared
it's
like
a
and
for
the
development.
Also.
I
just
mentioned
one
column
whether
what
is
the
status
of
the
development
status.
You
get
a
time,
then
better.
H
B
B
A
Things
but
the
full
test
matrix
eventually
will
be
everything
you
test
in
classic
osd.
Okay,
so
that's
the
long
term
strategy
like
the
immediate
goals,
should
be
pretty
much
radius,
bench
rbd
bench
on
a
replicated
pool
and
then
we'll
start
peppering
in
failure
scenarios.
But
let's,
let's
get
comfortable
with
setting
up
the
cluster
first.