►
From YouTube: 2020-01-07 :: Crimson SeaStor OSD Weekly Meeting
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
The
last
week
I
was
working
on
adapting
the
updated
Petri
immunization
to
our
code,
our
also
Street,
by
removing
the
dependency
on
Alice
ale
and
to
adapt
it
to
our
own
needs
for
exam.
Two
added
some
adapter
to
to
use
the
compared
to
operator
which
used
the
three-way
cooperation
to
speed
up
the
lookup
and,
in
addition,
I
think.
The
next
step
is
to
to
try
to
the
first
step
is
to
make
it
I
think,
because
every
step
might
involve
some
some
IO.
B
B
Okay,
so
I've
seen
everybody's
reviews,
I
have
I,
have
fixed
them.
Apologies
for
a
senior
comment
about
that.
I
should
have
changed
it
in
the
original
comments
and
not
make
new
commits
with
the
fixes.
That's
my
bad
and
other
than
that.
I'll
still
continue
to
progress
with
CBT
as
well,
in
working
close
with
the
rather
slob
about
linking
it
about
generalizing.
It
basically
making
feather
like
an
upper
obstruct
base
glass
with
support
for
metrics
that
are
of
interest
to
us
and
that's
what
I'm
working
on
currently
and
yeah.
It
should
be
done
with.
C
D
I'm
working
on
the
problems
I'm
working
on,
we
watch
TV
mechanism,
I,
have
my
desktop.
I
have
a
crimson
serving
the
RB
d
traffic
with
exclusive
lock
Tom
here
is
the
development
branch
of
the
moment,
I'm
cleaning
it
up
also
fixing
some
issues
like
like
that
locking
I
made
a
mess.
If
occasion
the
retried
or
dead,
operate
technology.
D
A
E
E
Forward,
that's
that's
one
thing:
I
have
still
the
bug
in
sister
that
I'm
pissed
off
that
I'm
trying
to
fix
a
week
ago.
So
I
try
to
email
fix
to
the
mailing
list.
Now
the
code
was
changed
and
my
fix
wouldn't
defeat.
The
new
code
and
I
will
redo
it
now
for
the
updated
system
record.
It's
not
it's
not
a
total
artwork,
it's
a
few
few
lines
and
the
problem
is
that
except
they
will
never
terminate.
E
E
It's
related
to
any
chef
that
I've
seen
at
hidden
now,
if
you
noticed
I
made
a
mind
or
cloud-related
fix
now
to
my
PR,
the
next
in
the
socket
and
the
to
make
the
sittest
rain
okay
this
time.
So
no
I,
don't
think
it's
related
to
that.
It
might
be
related
to
the
performance
that
this
I
don't
I'm,
not
sure,
but
there
is
definitely
a
problem
with.
There
is
a
missing
check
in
sista.
If,
if
you
are
shut,
shutting
down
or
listening
socket
except
we
just
hang
there
a
forever.
E
E
A
C
C
A
A
bunch
of
llamo
file
and
it's
used
gigantic
job
builder
tools
to
rend
it
into
a
XML
which
under
send
it
over
to
the
Jaganath
master
in
which
XML
will
be
interpreted
as
the
description
of
a
new
job.
I
think
the
first
step
is
to
to
talk
to
David
other
way
to
to
see
if
we
have
enough
capacity
of
the
bill,
slaves
to
to
run
one
job,
but
I
think
the
pot,
the
bottom
lines
to
make
it
and
all
personal
which
could
be
triggered
by
a
face
right
like
the
earth.
A
The
test
which
had
this
please
write
like
like
this,
and
we
can
trick
it
on
demand.
If
we
had
enough
power
power
computer
in
your
power
to
run
a
test
for
every
for
every
peer,
we
can
enable
it
like
what
do
we
do
with
with
the
make
check
and
the
arm
I'm
thinking
for
jobs.
But
the
first
thing
we
can
add
a
job
into
the
teacher,
okay
and
the
test
it.
Let
me
know
if
you
need
some
assistance
on
it.
E
Thank
you,
I,
first,
one
to
reach
a
state
where
everything
is
running
smoothly
with
a
clunk.
Apparently
there
is
one
caveat.
Yes,
a
lot
of
work
was
done
by
Mark
carbon
on
the
creating
day
on
checking
and
verifying
the
environment,
and
here
he
had
troubles
using
the
clunk
loader
with
what
we
are
doing
so
currently.
Did
you.
E
E
Yeah
I'm
still
following
what
he
has
done:
I'm
still
using
the
kind
of
GCC
linker
I,
don't
know
what
I
didn't
look
into
it
and
I
try
and
at
some
point
to
see
what's
the
problem,
why
we
he
had
problems
with
that
once
everything
is
ready,
I
think
we
push
it
and
try
I
really
try
to
the
microsites
talk
with
David
and
try
to
push
this
into
a
normal
standard.
Workflow.
C
E
E
Now,
please
take
a
look
and
what
I
noticed
in
previous
jobs,
and
here
is
that
having
a
formatter,
the
crank
format
available
is
really
handy
and
release.
There's
a
lot
of
time
when
you
are
doing
refactoring
or
writing
a
large
amount
of
code,
you
fix
something
and
everything
just
stays
the
correct
regarding
formatting.
So
please
take
a
look
at
what
the
results
and
let
me
know
if
you
think
that
we
can
at
some
point,
adapt
the
adopt
this
format
and
just
use
it
everywhere.
On
chrism.
E
Think
it's
possible
because,
if
I
think
our
projects,
there
are
projects
that
are
doing
just
that,
but
it
is
probably
one
can
do
it.
So
take
a
look
and
if
you
want
to
fix
it,
I
have
no
attachment
to
the
specific
where
your
formatting,
just
if
you
want
to
modify
the
the
set
of
directives.
Just
let
me
know
it's
doing
the
comment:
I
don't
care,
but
just
let
us
have
something
and
use
it.
C
G
Okay,
basically,
my
thoughts
is
mix
the
Intel
day,
our
system,
the
assistants,
design
idea
and
under
the
Samuels
design,
and
at
first
we
just
thought
this
approach
may
provide
some
way
to
do
very
light.
Weighted
snapshot
in
in
the
cyst
or
but
during
the
during
the
design
of
this
of
this
of
this
structure.
I
find
that
this
this
this
approach
can
also,
maybe
maybe
it's.
They
can
also
simplify
the
total
on
this
structure
and
I/o
processing
mechanism
of
the
system
and.
G
The
total
structure
at
first
and
please
let
me
do
some
throughout
introduction
about
the
HLC,
which
is
the
hybrid
logical
clock.
Basically,
is
it's
essentially
the
latter
part
o'clock,
but
it
is
character
close
to
the
physical
clock.
So
if
physical
is,
if
the
physical
clock
is
synchronized
through
protocols
like
ntp,
then
we
can
use
the
use.
This
hybrid
logical
clock
to
query.
G
G
There's
no
way
for
them
to
query
the
stories
birth
date
as
a
time
point
in
the
past
and
and
if
we
give
them.
If,
if
we
give
those
clients
such
a
way
that
then
we
can,
then
each
each
right
operation
is
just
a
small
snapshot.
I,
don't
know
if
I'm
making
myself
clear.
So
with
this
approach,
the
snapshot,
creation.
C
G
G
E
G
E
G
G
A
A
G
Yes,
the
main
purpose
of
this
archery
based
design
is
the
first
to
make
snapshots,
create
a
list,
national
creation,
very
light
weighted.
And
then,
though,
the
second
is
that
we
can
enable
layer
or
greatest
clients,
we
just
raise
clients
to
to
read
a
bit.
We
didn't
the
data,
we
read
any
any
block
any
blocks
and
blocks
off
the
day
off
data
without
actually
storing
those
data
together.
G
To
query
multiple
rectangles
in
the
artery,
you
know
the
artery
in
initially
used
to
the
to
store
spatial
data
and
they
and
actually
they
organized
the
leader,
spatial
data
into
humming
boss,
which
is
actually
a
smallest,
the
minimum
rectangular
that
the
founders
pesetas
objects.
And
if
we,
in
this
design,
we
set
one
axis
of
the
rectangle
as
the
extent
range
and
the
other
passes
as
the
HLC
timestamp
range.
So.
G
When
we
these
needs,
when
we
need
to
read
some
some
extensive
data
at
some
time
at
some
time
point
in
the
past,
then
the
read
and
then
this
read
operation
is
actually
an
our
trees
range
search
and
after
a
research
concern
as
reading.
So
we
don't
have
to
organize
those
operations
in
journal
into
a
into
a
continuous
block.
G
G
G
Will
this
approach
then
we
can
we
don't
need
to
clone
objects
when
doing
when
writing
to
and
already
it's
not
shortly.
The
project
because
writes
already
in
the
journal
and
different
rice
doesn't
go
that
way
with
each
other
or
cover
each
other.
So
actually
the
snapshot
in
objects.
Data
is
still
on
the
disk,
even
if
they've
been
modified
in
after
the
snapshot.
G
So
there's
no,
so
there
doesn't
need
to
be
any
clone
objects
operations
and
which
further
means
there
won't
be
any
objects.
Sharing
the
same
extent
of
data,
so
I
think
with
this
approach
we
can.
The
application
tree
may
be
not
needed
and
we
can
just
using.
We
can
just
check
the
VHL
seats
hands
down
to
judge
whether
a
journal
record
is
value
or
not.
Oh.
G
A
A
F
F
So
it
is
not
truly
zero
copy
and
we
can
better
configured
it
to
reach
that
if
we
can
enable
the
huge
page,
backhand
and
so
I
was
trying
to
enable
it,
but
the
application
was
hand
during
the
memory
configuration
so
I
was
trying
to
solve
that,
and
the
details
was
updates
in
the
DBT
K
evaluation
document,
and
another
funding
is
that
we
currently
moved
sake
to
another
call.
We
found
another
coin
or
appropriate
right,
but
a
native
stack.
It
is
a
different
story.
C
C
F
Tcp
is
is
controlling
the
like
the
connection,
the
low
level
connection
states
like
TCP
TCP
window,
like
that
it
is
stateful
and
and
the
the
play
and
it
decide
which
socket,
which
call
the
socket
is
placement
and
and
the
place
where
TCP
TCP
lives
is
determined
by
hardware.
Is
s
so
this
means
we
cannot
move
the
socket
to
another
chord
directly
after
the
socket
is
connected
or
or
it
may
be,
very
complicated
to
move
it
to
another
call,
and
maybe
we
cannot
change
the
I
RS
s
hash
on
the
fly
in
the
fly.
F
E
D
C
D
Even
after
even
we
decide
to
switch
to
multi-core
I
think
that
we
want
to
avoid
hopping
between
between
course,
between
CPU
cost
or
cost.
This
will
translate
into
into
putting
a
dedicated
messenger
for
each
let's
say:
set
of
pages
I
mean,
as
in
the
end,
mapping
will
be
implemented
by
extending
the
radius
protocol
to
allow
for
deliver.
C
F
There's
a
problem
is
still
a
problem
because,
for
example,
native
stack:
we
we.
If
we,
though,
we
can't
really
use
one
one
one
mapping
right,
but
we
can
still
create
multiple
as
these
into
multiple
cores
right.
So
so
it
means
that
the
DB
D
K
is
the
the
underlying
DBA
implementation
is
across
horse,
though,
that
the
hardware
will
distribute
the
targets
according
to
Hardware
RSS,
and
we
need
to
find
a
way
to
place
it
into
the
dedicated
core
of
the
OSD
to
in
a
deterministic
way.
F
F
F
D
Before
going
forward,
let
me
ask
a
question:
do
we
in
such
configuration,
let's
say
one-to-one
or
is
the
multiple
one-to-one,
or
is
this
I
presumed
that
each
of
those
always
DS
will
have
basically
a
dedicated
network
interface?
We
can
think
about
that
about
like
lag
dedicated
pair
of
IP
address
and
each
part,
maybe
even
dedicated
IP
address,
but
we
only.
F
I'm
not
very
familiar
with
that,
though,
but
but
I
will
wait
when
I
find
a
solution
possible
solution.
If
it
is
simpler,
I
will
present
it
though
I
think
there
there
can
be
some
minor
changes
of
of
current
products,
bag
and
natives
that
to
support
this
way
without
using
as
RV,
because
it
is
a
visualization
right.
A
C
F
C
F
D
D
C
F
D
C
C
F
C
C
C
F
Okay,
so
the
second
thing
is
ACN
issue
reported
by
Kofu
the
last
week.
I
can
reproduce
that
issue
locally,
but
I
need
to
run
about
1,000
hands
of
unit
test.
This
now
subject
to
to
have
one
issue
and
nearly
200
times
to
reproduce
the
same
issue
with
unit
has
system
messenger.
So
from
the
ACN
log
it
seems
that
C
star
is
pulling
out
deleted
ft.
D
E
Because
I
think
sister
I
had
a
few
changes
in
the
low
level
a
league
of
five
descriptors
I
was
trying
to
cut
I
was
trying
to
follow
the
same
issue
on
when
I'm,
when
I
ran
a
sister.
Only
person
test,
not
ours,
but
sister,
a
as
on
my
laptop
and
I,
have
consisted
consistent
faders
of
their
tests,
with
the
version
that
we
are
using.
Maybe
it
will
help
you
solve
the
issue.
Did
you
try
to.
E
F
Think
the
I
am
I'm
currently
one
two
route
course
why
it
happens
and
then,
if
the
code
is
updated,
it's
just
minor
changed
rights
and
we
I
can
look
at
the
fix
and
this
confirm
whether
it
is
fixed
in
the
new
code.
We
need
to
confirm
that
because
it's
varies
like
one
time
in
a
thousand
hands.
It
is
it's
very
hard
to
find
and
identify.