►
From YouTube: 2020-03-24 :: Ceph Crimson 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
C
Come
here,
let's
start
and
last
week,
I
I
was
there
I,
had
a
discussion
with
a
very
easy
and
Stan
regarding
to
the
owner,
the
discard,
unload
implementation
industry
information
and
also
sparked
a
filler
when
we,
when
crimson
avati,
tries
to
shut
down
its
still
to
do
the
sell,
because
a
pretty
before
for
its
rate
before
reference,
the
issue,
which
was
a
fixed
last
couple
days
ago
and
I
said
from
it.
I
also
spotted
a
memory
leak
caused
by
all
note,
somehow
I
didn't
to
record
it.
I
will
come
during.
D
C
C
Sorry,
it's
a
if
you
a
neighbor!
If,
if
you
to
build
community
with
a
sin
or
with
we
look
to
make
you
type
because
debug
you,
you
will
enable
a
sin
and
when
community
shut
down
it
will
each
time
we
identify
a
identify,
a
leak
because
a
under
the
info
was
not
properly
released.
Somehow
I
could
have
pasted
the
ear.
Oh
okay,.
D
E
A
A
D
A
Try
to
figure
out
still
trying
to
figure
out
the
problem
is,
is
just
that
this
was
pretty
much
my
first
time
familiar
with
some
of
the
creams
and
source
code.
I'm
still
learning
pretty
much
everything
and
or
be
much
step
step
source
code
in
general,
oh
yeah,
it's
pretty
much.
What
I'm
going
to
be
the
plan
for
today
is
he'll,
be
focusing
on
that
and
that's
pretty
much.
It.
E
E
And
surf
extract
our
traction,
so
I
think
if
we
can
get
how
to
implement
it,
the
people
have
traced
a
part
of
the
variable
P.
Then
we
can
use
a
common
algorithm
for
all
of
the
people
as
treating
our
code,
so
it
just
introduced
system
as
a
method
of
all
common
prefix
TR,
the
prefix
and
suffix
Trenton
traction
and
some
measure
to
keep
compression
and
they
treated
the
PP
fast
reactors
ring
so
I
think
our
key
is
a
long
string
right.
F
We're
specifically
talking
word
so
I
think
I
think
people
would
benefit
from
going
back
and
reading
older
papers
on
database.
Indexing
I
can
try
to
put
some
of
those
together,
but
the
short
answer
is
no.
You
don't
have
to
treat
them
as
a
string,
for
instance,
if
you're
constructing
a
database
there
were
a
database
index
purely
on,
let's
say
a
four
byte
integer.
Well,
then,
you
can
store
that
you
can
do
comparisons
on
keys
using
integer
operations.
You
needn't
do
string
comparisons.
F
F
E
F
It's
so
you
have
you
seen
the
oh,
no
tree
design,
yes,
research
shared
it
with
everyone.
Okay
to
the
very
beginning
of
that
document
has
a
page
or
two
explaining
why
the
key
is
what
it
is.
Oh.
E
F
G
G
G
B
F
Yep
I've
been
mostly
going
back
and
forth
three
things
in
on
the
Anoat
redesign
I
think
we
have
three
pretty
well
thought-out
concepts
here.
So
I
think
we
should
discuss
this
at
that
towards
the
end.
I
think
we're
making
progress
I
think
all
of
these
essentially
address
the
problem
with
ensuring
that
we
can
do
compare
is
that
that
we
can,
in
the
event
that
the
crush
I
have
does
a
good
job.
We
should
be
able
to
find
it
the
key
we
need
without
doing
any
string.
F
Comparisons
will
stop,
do
once
drink
a
person
PM
to
make
sure
it's
the
right
one,
but
that's
unavoidable
so
I
think
that's
the
really
important
property,
so
I'm,
happy
they're
corrupted.
This
I
think
is
further
discussions
to
decide
what
we
want
to
do
for
a
prototype
and
then
go
for
it.
I
talked
to
that
I'm
continuing
work
on
the
LP
entry.
C
A
C
D
F
A
F
G
F
D
C
F
Way,
operty
I
think
we
should
weigh
heavily,
which
is
that
all
you
need
to
use.
It
is
like
permissions
on
the
github
repo,
which
are
already
the
necessary
property
for
contributing.
So
in
a
way,
we've
already
felt
that
problem
I,
don't
know
about
Trello.
Maybe
it's
also
easy,
but
I
suspect
that
we
would
have
to
permit
people
to
modify
the
table
separately
well.
D
D
C
F
Can
we
discuss
I
thought
we
would
discuss
the
note
redesign
okay,
cool
I
was.
C
C
C
Just
I
I'm
on
the
weather
yesterday,
so
I
only
skimmed
over
the
earth.
Your
attention,
your
flexible
layout,
fix,
treat
problems,
I
think
I
still
with
you,
because
if
we
can
use
the
fixed
parts
for
the
first
one
levels,
that's
a
big
wing
and
we
can
leave
the
cabo
at
the
bottom
half
if
we
have
to
put
to
pay
the
text
for
it.
C
F
Okay,
so
I
guess
I
wanted
to
lay
out
a
couple
of
like
sort
of
six
points.
They're
happy,
trees,
work
so
who
here
is
super
familiar
with
happy
or
who?
Who
here
is
not
very
familiar
with
healthy
trees,
work
internally.
C
A
D
F
C
F
So
any
particular
node
represents
a
sub,
a
subset
of
the
overall
key
space
order.
It
the
way
we've
chosen
to
do
ordering
right
so
based
on
how
the
nodes
parent
has
its
pivot
set.
That
is
the
keys
on
either
side
of
this
child.
This
node
is
responsible
for
all
values,
key
to
buy
a
key
between
those
two
values
right,
so
for
any
node
we
can.
F
F
A
A
F
Yes,
so
the
entirety
of
the
design
is
that
we
designate
for
different
kinds
of
note.
The
first
kind
of
node
has
the
entire
key.
It
has
be
sure
to
pool
the
crush
on
the
right
side,
all
the
char,
the
pool
in
the
cross
hash
on
the
left
side
of
the
of
the
node
and
the
name
stays
the
OID,
the
snapshot
in
the
generation
on
the
right
side.
F
However,
as
we
descend
the
tree-
and
this
should
happen-
that's
almost
the
next
level,
because
it's
going
to
be
it's
basically
never
going
to
be
the
case
that
we
have
for
them
a
few
pools
right.
So,
by
the
time
we
get
to
the
next
level,
we
shouldn't
have
everything
in
one
of
those
nodes
should
be
the
same
pool,
so
we
could
omit
both
the
shard
in
the
pool
leading
us
with
the
crush
hash
on
the
left
side.
F
B
C
F
F
Is
no
side
of
the
yeah?
Well,
we
in
this
case
we've
actually
or
we're
not
even
storing
the
sure
in
the
pool,
because
we're
far
enough
down
the
trees
that
they're
not
relevant
anymore.
So
it's
only
the
crosshatch
we're
looking
at
and
if
we
see
two
adjacent
entries
that
where
the
node
we're
actually
looking
up
is
between
them,
then
we
know
they're
in
a
no
possible
set
of
the
suffix
that
could
possibly
change
the
answer.
F
F
F
You
can
infer
the
fact
that
that's
the
two
are
different
means.
You
don't
even
have
to
do
that.
However.
Yes,
if
you
want
to
on
the
right
side
in
the
variable
section
where
you
have
the
value,
you
could
have
a
one
char
prefix
to
that
variable
sized
portion.
That
indicates
which
fields
are
actually
actually
present,
and
that
will
be
not
the
theory
for
compression,
but
it
only
cost
one
Baniyas
and
it's
only
on
the
variable
size
part.
You
only
look
at
it
if
you're
actually
going
to
do
the
lookup
well.
F
B
A
F
B
F
The
creative,
the
idea
in
its
entirety
is
that,
basically,
you
put
these
very
large,
the
the
very
efficient
dick
sized
portion
to
the
left
of
the
node,
and
you
do
whatever
you
need
to
do
on
the
right
side
of
the
node
to
not
waste
space.
It's
okay!
If
it's
the
little
expenses,
because
you're
not
gonna,
look
at
it
mostly,
you
won't
look
at
it
until
you
were
actually
going
to
need
to
compare
strings
anyway.
F
F
B
A
F
Anyway,
that's
that's
the
core
idea,
other
than
that
it
behaves
exactly
like
a
b-tree.
You
do
splits
and
merges
under
the
same
conditions.
There
there
are
places
we
could
add
complexity.
We
could
allow
the
N
equals
2,
N
equals
3
or
leave
notes
to
be
bigger
than
4
K,
which
would
add
some
nuance
to
the
exact
size
threshold
rules,
but
even
that's
not
really
not
complicated.
It
just
changes
the
exact
threshold.
D
D
F
I
guess
what
I
don't
get
is
what
your
design
accomplishes
that
this
one
doesn't
I'm,
biased,
I.
Guess
in
that
I,
don't
I,
don't
like
the
idea
of
nested
trees,
because
I
think
it
will
be
harder
to
read
and
it
will
be
less
flexible.
With
this
approach
we
could
actually
introduce
additional
internal
node
encodings
even
to
existing
stores
would
be
backward
compatible
of
course,
but
it
we
forwards
compatible.
D
F
F
If
you
keep
it
inside
of
the
node,
then
the
B
trees
split,
merge
process
automatically
handles
allocation.
Garbage
collection
for
you
dad
knows
become
too
small
that
if
they
have
too
much
unused
space,
you
naturally
merge
them
into
their
neighbors,
and
then
you
release
the
space
I
can't
Ross.
If
you
choose
to
store
this
ring
somewhere
else,
we
have
to
have
a
secondary
system
to
do
that
allocation
possible,
but.
D
F
That's
incredibly
expensive:
that's
the
problem,
we're
already
paying
a
high
allocation
overhead
here.
This
is
like
allocating
a
node,
B,
trees,
nodes
and
mutating.
It
will
be
probably
the
most
expensive
part
of
any
right,
so
every
additional
mutation
we
add,
multiplies
that
by
that
factor.
So
no
actually
I
think
that
just
writing
down
where
the
notice
will
be
the
most
expensive
part
of
these
operations,
in
other
words,
that
the
lbh
tree
is
going
to
be
the
most
expensive
part.
F
D
F
A
F
D
F
F
F
A
F
F
Anyway,
so
that's
that's:
why
I
find
that
design
to
be
complicated,
we
need
we
would
need
to.
We
would
need
to
write
down
exactly
how
much
space
overhead
it
is
and
how
many
additional
iOS
it
causes
on
the
average
insertion
be
three
times
the
property
that
this
was
sort
of
them
like
a
normal
one
that
us
this
is
a
well
understood
property.
It's
just
logarithmic
right!