►
From YouTube: LSR WG Interim Meeting, 2020-04-02
Description
LSR WG Interim Meeting, 2020-04-02
A
A
B
A
A
F
A
B
B
Okay,
I
think
I
think
I'm
ready
to
go
now.
You'll
notice
that
we
don't
have
any
status
on
doctrine
status,
we're
gonna
I
did
some
slides
and
I'm
gonna
defer
that
first,
we're
gonna
only
have
one
session
and
we're
gonna
try
and
fit
it
all.
In
one
session
we
thought
we'd
just
skip
the
document
status,
but
now
that
we
have
two
we'll
do
it
at
the
second.
During
the
second
session,
the
first
one
is
a
flex
algorithm.
That's
Peter
sanic
is
Peter
I.
G
A
G
G
G
There's
been
two
changes
to
the
draft.
We
added
one
more
constraint,
which
is
an
exclude
SR,
LG
constrain.
This
is
a
new
constraint
added
to
the
flag
saga
definition.
It
allows
the
links
to
be
excluded
during
the
topology
calculation
for
a
particular
flex.
I'll
go.
This
has
been
requested
by
some
of
the
the
users
of
place.
Didn't
come
up,
you
know,
so
this
is
good,
but
it
has
been
requested
on
a
by
day
by
the
real
users,
and
there
has
been
couple
of
them.
Basically,
the
use
cases.
G
You
may
have
disjoint
paths
through
the
network
where
in
one
of
them
you
want
to
exclude
certain
s,
ro
G
and
on
the
other
paths
you
want
to
exclude
the
other
SR
LG.
So
any
srg
failure
of
a
particular
group
is
not
going
to
affect
both
of
the
paths
and
that
way
you
actually
achieve
the
redundancy.
If
you're
sending
redundant
flows,
this
same
thing
can
be
achieved
and
the
nest
needs
to
be
clear.
It
can
be
achieved
by
the
affinities,
but,
given
that
users
are
already
used
a
year
solid-
oh
sorry,
oh
geez!
G
G
So
what
we
from
the
protocol
encoding
via
die,
size
and
OSPF
fat,
exclude
astrology
sub
theory
and
then
during
the
calculation
we
make
a
rule
say
check
if
any
exclude.
Sro
general
rule
is
part
of
the
fat
for
the
algo
and
if
the
rule
exists
and
we
check
that
the
link
is
part
of
the
same
energy
or
any
of
the
ones,
then
we
basically
exclude
the
link
from
the
calculation.
So
it's
very
similar
to
what
we
do
with
the
affinities.
G
So
that's
the
first
change.
The
second
change
is
that
we
discovered
a
small
issue
in
terms
of
leaking
the
fat
in
ice.
Is
the
original
text
in
the
draft
said
that
we
can
leak
the
fact
as
a
part
of
the
router
capability
leaking,
where
this
can
be
done
when
the
originator
sets
the
S
bit?
The
problem
is
that
the
fat
election
uses
as
a
system
ID
and
when
we
even
threw
out
a
couple
of
polity
theory
is
leaked.
Well,
the
system
ID
of
the
originator
is
not
available
anymore.
So
what
we?
G
G
G
We
don't
see
really
a
need
to
do
this
other
way
around
from
l1
to
l2,
because
really
the
use
case
here
is
I
want
to
have
a
single
fat
definition
in
my
network.
I
put
in
my
backbone
and
goes
you
know,
I
can
configure
my
BRS
to
leak
the
fact
to
all
the
other
areas.
There's
no
need
really
to
do
it.
The
other
way
around.
So
we
put
text
in
the
draft
saying
they'll
102
rather
must
not
regenerate
the
fat
from
l1
to
l2.
G
G
That's
those
are
the
two
small
additions
we
added
to
the
draft
the
next
step,
so
this
draft
is
now
almost
three
years
old.
The
working
group
adoption
happened.
You
know
two
years
back,
we
have
multiple
implementations
available.
We
have
some
multi
vendor
interoperability
testing
that
has
been
stable
for
some
time.
These
are
small
details,
nothing
that
you
know
changes
any
fundamental
operation
of
the
Flex
algo.
We
are
considering
to
close
this
draft
and
ask
for
a
working
group
last
call.
We
are
fully
aware
of
the
fact
that
some
other
flex
I'll
go.
G
You
know,
constraints
may
come
in
the
future,
but
I
mean
this
can
be
specified
in
a
small
new
document.
We
don't
need
to
keep
the
the
architecture
document
with
all
the
framework
open
for
every.
So
in
our
personal
feeling
is
we
should
move
with
the
document
and,
if
anything
needs
to
be
added,
we
can
add
it
as
a
as
a
separate
document.
A
Questions
I
yeah,
so
we
got
to
I,
guess
I'm
in
first
I
I
I
went
through
the
the
update
really
fast
and
I.
Don't
claim
that
I
know
a
lot
about.
I
haven't
looked
closely
at
the
fad,
but
the
thing
that
jumped
out
at
me
is
if
the
election
is
being
done
to
Main
wide
but
you're,
not
you
know,
but
you're
using
the
system.
Id
of
the
summary,
the
summarizing
router
right.
It's
changing.
Okay,.
G
So
let
me
make
a
clarification.
The
fad
election
happens
inside
an
area,
it's
perfectly
valid,
to
have
different
definition
in
each
area,
so
you
can
have
different
fad.
428
I'll
go
in
in
one
area
and
a
completely
different
one
in
the
other
it
all
works.
It
doesn't
need
to
be
the
same.
So
there's
no
attempt
to
here
to
make
the
the
election
being
domain
white.
What
the
attempt
is
to
make
it
possible
to
configure
the
thing
on
a
single
few
routers
and
not
necessarily
on
all
or
inside
all
of
the
areas.
A
A
A
H
G
G
The
reason
why
we
are
trying
to
do
prevent
this
is
to
avoid
the
looping
of
a
fad,
because
if
we
allowed
it
to
be
leaked
from
both
l1
to
l2
and
l1,
you
have
to
make
sure
that
we
never
leak,
something
which
has
been
already
originated
in
l2
back
to
l2
and
the
other
way
around,
which
complicates
the
things
we
have
to
add
more
stuff.
To
tell
me,
I
personally,
think
from
the
user
perspective
having
the
leaking
from
l2
l1
is
absolutely
sufficient.
G
G
H
G
You
can
obviously
define
the
fat
inside
every
area
on
top
of
it.
If
you
want
to
have
an
election
being
done
there,
based
on
the
local
one
which
you
define
I
mean
nobody
probably
beats
you
to
do
that,
if
you
don't
want
to,
if
you
only
want
to
define
it
on
single
place,
I
think
it's
very
very
you
know
it's
fair
to
us
assume
it's
going
to
be
done
in
the
backbone,
but
again
it
doesn't
mean
you
cannot
have
your
fat
in
l1
you
can,
but
it
will
not
be
leaked
out.
Okay,.
H
G
H
G
So
the
redistribution
typically
happened
to
prefixes.
We
don't
release
tribute
any
other
stuff.
I
haven't
heard
of
anything
else
being
redistributed
strictly
speaking,
if
you
use
that
word
so
for
me,
it
may
only
sense
to
redistribute
something
which
is
attributed
to
the
prefix,
but
the
fat
I
mean.
Obviously
your
implementation
can
do
this
thing,
but
it
doesn't
need
to
be
specified
in
the
in
the
idea
trap.
It's
a
local
behavior
on
the
router,
where
you
do
this,
if
you
do
it,
nobody
prohibits
you
to
do
that.
H
I
I
My
question
is
whether
it
is
allowed
for
user
to
change
the
value
of
the
metric
in
in
the
flex
angle,
for
example,
I
I
choose
latency
as
the
metric
in
one
flex
angle,
topology
and
the
can
I
configure
the
latency
of
some
link
larger
than
it
should
be,
for
example,
to
avoid
the
touring.
In
some
scenario,
it's.
G
Allowed
I
mean
look,
you
can
change
the
metric
of
the
regular.
You
can
change
the
regular
GP
metric
based
on
many
things.
We
have
thing
we
have
stuff
like
that.
You
are
allowed
to
change
the
delay
metric
on
the
link
based
on
what
you
want.
We
are
not
limiting
what
you
how
you
can
set
the
delay
metric
on
the
link
it's
completely
outside,
whatever
the
value
advertised
we
use
it.
I
G
J
I
K
B
Everybody
hear
me
is
my
mic:
good
yep,
yes,
okay,
okay,
yeah,
I!
Think
it's
good,
especially
that
there's
been
some
implementations.
Interoperability
checking,
I
think
what
we
should
do
is
just
because
it's
hard
to
do
a
poll
over
the
you
know.
You
know
over
the
WebEx.
You
should
just
do
a
poll
see
how
many
of
Reddit
take
some
time
and
take
some
time
discussing.
B
A
A
G
I
mean
I'm,
not
saying
you
need
to
do
you
know
this
week
or
next
week.
It's
just
that
I
don't
want
to
wait,
another
ITF
or
so
because
I
don't
necessarily
see
a
reason
why
we
should
not
move
forward
with
this.
We
are
not
in
rash,
but
I
think
it
has
been
there
for
a
while.
We
have
everything
in
place
to
close
the
document,
so
I.
A
Looked
over
this
slide
originally
and
wondered
about
how
we
could
have
multiple
implementations
with
interoperability
tested
with
something
you
just
added
to
yeah.
It
doesn't
feel
like
this
is
you
know,
we've
been
working
on
this
for
a
while,
so
everyone
should
probably
read
this.
You
know
idea
that
we're
probably
getting
close
to
doing
a
working,
app
last
call
on
it.
A
I
C
You
mentioned
a
couple
things
today
about,
for
example,
the
fad
where
you
think
is
more
appropriate
for
people
to
fruition
it
how
you
confirmational
different
places.
You
know
things
along
those
lines.
You
also
mentioned,
for
example,
in
the
SR
LG,
when
you
were
talking
about
that.
How
you
could
do
very
similar
things
with
the
affinity.
I
think,
is
what
you
said.
C
It
would
be
great
if
there
were
some
operational
considerations
in
the
draft,
at
least
showing
what
you
guys
were
thinking
when
you
wrote
it
so
that
the
future
people
don't
start
asking
well
I
can't
hide
leaked
from
Google
one
too
little
too,
and
it
makes
sense
what
you're
doing
I
think,
but
it
would
be
really
nice
to
provide
some
guidance
so
that
when
people
are
actually
doing
this,
they
understand
oh.
This
is
why,
because
maybe
I
should
do
it
in
the
backbone
and
then
leak
it
from
their
televisions.
C
B
B
B
H
B
L
I
L
Apologies,
so
a
brief
for
an
introduction
of
a
dynamic
already
and
it
decouples
the
flooding
topology
from
the
physical
topology.
So
the
data
is
defaulted
on
the
pit
of
topology
the
link
stay,
the
proctor
data
is
flooded
on
flooding
topology.
So
if
we
choose
a
sparse
of
graph
of
the
physical
topology
as
the
flooding
topology,
then
we
can
reduce
the
link
state
funding
in
the
dynamic
flooding
draft.
It
listed
a
few
requirements
on
the
flooding
topology
first
and
the
flooding
topology
has
to
include
all
the
system
in
area.
L
Obviously,
everyone
wants
to
receive
the
link
state
updates
and
seven.
The
flooding
topology
has
to
be
by
connected
to
avoid
a
single
point
of
failure.
So
our
psycho
is
a
good
example.
By
connected
graphs.
If
any
node
or
link
fails,
then
we
have
another
path
to
reach
all
the
other
nodes.
The
remaining
flooding
topology
is
still
unconnected.
L
The
flooding
topology
and
is
also
desired
to
have
a
bounded
diameter,
so
the
diameter
of
a
graph
is
defined
as
the
distance,
the
maximum
distance
between
any
pair
of
the
nodes
so
use
psycho
as
an
example.
If
we
have
a
cycle
of
n
nodes,
yes
time,
it
is
in
over
2.
This
means
that
a
link
state
update
has
to
travel
and
over
2
nodes
to
reach
the
father
farthest
node.
L
So
in
our
large
network
in
such
a
long
propagation
delay
may
not
be
acceptable
and
then
flooding
topology
is
also
desired,
to
have
a
balanced
and
no
degree
so
the
node
degree,
and
yes,
the
number
the
number
of
neighbors
it
has
to
flood
to.
So
it
is
an
indicator
of
the
flooding
burden,
our
node,
so
we
in
the
flooding
topology.
We
don't
want
to
overload
any
node
in
the
area
so
used
saenko
as
the
example.
It
has
a
perfect
balance
in
no
degree,
every
node
has
a
degree
of
two.
L
L
Then
our
goal
is
to
find
a
sub
graph
that
covers
all
the
nodes,
and
this
sub
graph
is
composed
of
four
overlapping
small
cycles,
and
we
also
take
into
account
the
diameter
and
the
no
degree
in
the
algorithm.
So
the
resulting
and
sub
graph
from
this
algorithm
has
some
nice
properties
required
by
the
flooding
topology.
It
can
be
used
by
the
error
leader
to
compute
the
flooding
topology.
L
So
here's
an
outline
of
this
algorithm
first
find
a
set
of
nodes
and
edges
to
form
an
initial
cycle.
So
this
is
the
right
cycle
on
the
right,
the
white
cycle
on
the
right.
We
add
this
initial
cycle
to
the
sub
graph.
We
are
computing
and
then
we
expand
this
sub
graph
in
iterations.
So
in
each
iteration
we
add
an
arc
to
the
existing
and
sub
graph
and
this
new
arc
pass
and
has
new
nodes
and
the
new
edges
and
its
endpoint
on
the
existing
sub
graph.
L
L
Here
are
some
details
of
the
aerosol
how
to
find
the
initial
cycle.
So
we
start
by
selecting
a
node
in
the
base
based
graph,
the
physical
topology,
and
that
is
no
the
computer.
No
excuse
me
this
node
can
be
done
node
that
has
the
highest
degree,
and
then
we
do
a
depth
first
search
for
a
limited
number
of
steps.
So
for
this
example
from
starting
node
n
0,
we
find
its
neighbor
and
1
and
then
n
2,
then
and
3,
and
then
we
stop
added
some
steps.
L
This
is
just
to
control
the
diameter
of
the
sub
graph,
the
initial
cycle.
How
do
we
stop?
We
switch
the
gear,
we'll
use
a
press
first
search
to
find
a
shortest
path
from
the
end.
Node
in
the
DFS.
Here
is
the
entry
back
to
the
starting
node
and
0.
So
in
the
example
we
find
that
and
4
is
the
intermediate
step
to
reach
n
0
yeah.
We
find
it
in
every
cycle
and
0
1
2
3
4
yeah.
L
Once
we
have
the
a
new
cycle,
we
can
add
ox
to
the
sub
graphs,
so
the
procedure
is
pretty
similar.
So
at
the
beginning
we
have
thumbs.
The
current
sub
graph
in
this
case
is
just
the
initial
cycle.
Then
we'll
find
a
node
on
this
sub
graph.
We
do
a
DFS,
fun
limited
number
of
steps
and
then
we
do
a
BFS
to
find
the
shortest
path
back
to
the
current
sub
graph.
L
So
in
this
example,
in
this
first
iteration
we
first
to
find
out
the
starting
node
and
0
on
the
initial
cycle,
and
then
we
do
DFS.
Let's
say
we
only
get
two
steps
and
1
and
n
2
and
then
we'll
find
the
shortest
path
back
to
the
any
nodes
on
the
initial
cycle
except
starting
node.
So
here
we
find
up
an
n,
3
n
turn
in
and
+
4,
so
in
each
iteration
we
add
this
ox.
L
To
n
it's
awful,
this
pass
can
be
selected
and
take
into
account
the
no
degree,
and
that
is
a
diameter
so
and
we
can,
we
can
select
the
starting
node
and
by
comparing
the
node
degree
on
all
the
sub
current
a
sub
graph
and
also
considered
us
know
the
distance
to
the
inner
cycle.
So
we
want
to
build
the
ark
path
around
in
your
cycle.
Do
not
go
very
far
away
so
because
that
might
increase
the
diameter
Sara.
Yes,.
B
L
L
Okay,
so
when
selecting
the
end
points,
sometimes
we
need
to
make
a
trade
off,
because
all
the
notes,
probably
under
in
your
cycle,
are
already
used
for
a
few
member
offs
and
arcs.
This
increases
the
node
degree,
so
we
want
to
use
the
nodes
in
other
arcs,
so
we
keep
doing
that
until
we
find
the
other
node.
We
include
all
the
nodes
in
the
sub
graph.
L
If
the
original
and
the
base
Ravi
is
by
connected,
then
using
this
algorithm,
we
can
find
out
by
connected
subgraph,
but
sometimes
the
base
graph
is
not
I
connected.
It
is
connected,
but
not
I,
connected
and
and
in
the
pfcs
stage.
We
are
not
able
to
any
pass
back
to
the
current
a
subgraph.
So
in
this
example,
let's
say
we
have
find
a
sub
graph,
G
Prime,
and
then
in
this
iteration
we
pick
the
note
and
0
and
distance
starting
node.
L
Then
we
do
a
DFS
with
find
n
one
and
two
and
three,
and
now
we
stop,
but
in
the
BFS
stage
we
want
to
find
a
shortest
path
from
n
3
back
to
the
and
current
sub
graph.
There's
no
way
to
go
back,
so
we
can
modify
the
algorithm
to
just
discuss
the
last
DFS
node,
so
we
search
shortest
path
from
the
second
to
the
last
and
two
back
to
the
Supra.
Unfortunately,
in
this
case
we
couldn't
find
any
pass
either.
L
So
we
discarded
the
n
2,
then
we'll
only
left
there
owning
one
node
left
and
one.
So
then
we
know
that
and
0
and
when
this
edge
must
be
in
them
so
graph,
otherwise
the
graph
will
be
is
connected.
So
we
added
this
to
this
edge.
It's
cutting
edge
to
the
sub
graph,
and
then
we
start
the
algorithm
again.
L
L
So
the
answer
crowd
funded
by
this
algorithm
has
some
practice.
First,
it
includes
all
the
notes
and
second,
it
is
by
connected,
because
you
can,
you
can
prove
this,
probably
by
mass
mathematical
induction.
So
the
initial
cycle
is
about
connected
graph
and
then
we
add
arcs
with
two
different
endpoints
lending,
the
under
the
sub
graph.
L
So
this
will
limited
the
size
of
the
new
cycle
and
also
D
in
each
year
steps
we
selecting
the
starting
node
of
the
our
path.
We
can
select
those
close
to
the
inner
cycle,
so
the
arcs
are
build
around
in
your
cycle
instead
of
going
further
further
away.
So
thus
the
sub
graph
also
has
a
parent
can
have
a
balance
to
no
degree.
When
we
select
in
turn
starting
node
of
each
arc,
we
can
select
a
node
with
lower
degree,
so
there
might
be
a
trade-off
between
the
diameter
and
at
node.
B
L
So,
let's
go
through
quickly
goes
through
the
example
of
a
complete
graph
with
ten
nodes.
So
I
didn't
draw
all
the
connections
between
this
among
this
to
ten
nodes,
but
all
of
the
not
connected.
So
we
use
the
Maxima
steps
in
DFS
as
three,
so
we
go
three
steps
and
a
fan
of
way
back
so
first
we
we
can
select
any
node
in
this
graph
as
the
starting
node.
Let's
say
we
select
n
0.
Now
we
go
three
steps
away:
fun
and
1
and
2
and
n
3.
L
We
stop
and
then
find
a
shortest
path
which,
because
only
nodes
are
connected
to
each
other,
so
we
just
go
back
to
n
0.
So
the
white
cycle
is
the
initial
cycle
we
find.
Then
we
can
note
down
the
initial
cycle.
Let's
say
we
pick
and
1,
and
then
we
do
three
steps
in
Tibet's
and
4
5,
6
and
now
I
find
our
way
back.
So
this
and
the
endpoint
can
be
either
n
0
+
2
+,
o
+
3.
L
This
cannot
be
n1,
because
if
we
end
on
in
1,
which
is
the
same
as
the
starting
node,
then
we
have
cutting
node.
So
it's
not
by
connected
okay.
Let's
say
we
end
on
in
0,
so
this
yellow
mark
will
beat
that
first
off,
we
add
it
to
sub
graph.
Now
we
find
we
started.
A
second
iteration
will
find
a
starting
point,
say:
n
3,
and
then
we
go
three
steps:
9
8
7,
and
then
we
can
back
to
any
node
on
the
white
or
yellow
except
n3.
L
Now
we
include
all
the
nodes
in
the
graph.
We
stop
the
Amazon,
and
this
is
the
sub
graph
we
find
for
the
flooding
topology,
so
we
eat
this
sub
graph
only
has
12
and
address
compared
to
45
in
the
base
graph,
but
the
data
cost
that
we
have
a
larger
diameter.
So
the
diameter
of
this
sub
graph
is
fall
versus
one
in
the
base
graph
and
the
node
degree
is
the
30
parents
to
either
have
one?
No
degree
two
of
three
some
notes
have
degree
of
two
some
nodes
Helsley
and
in
the
base
craft.
L
L
Oh,
we
can
do
a
little
optimization
to
consider
the
lamp
properties
and
the
sudha
note
do
not
need
to
be
included
on
the
resulting
sub
graph,
so
the
flooding
topology
only
needs
to
include
all
the
real
nodes,
and
if
a
suna
node
is
indeed
included,
then
it
actually
are
the
nodes
connecting
to
these
soon
nodes
and
has
a
unique
connectivity.
So
we
can
adjust
the
algorithm
to
take
into
account
this
this
properties
of
lenss,
so
I'm
not
gonna,
go
into
details
here.
In
summary,
we
propose
the
a
version
that
applies
to
any
connected
base.
L
L
I
L
K
I
L
F
F
L
Yeah,
that's
a
good
question.
You
we
focus
on
the
centralized
mode
and
if
you
want
to
use
the
evidence
for
the
distributor
mode,
then
you
need
to
make
sure
that
the
orders,
the
nodes,
the
order
of
notes
and
address
are
all
the
same
on
all
the
notes.
So
the
resulting
sub
graph
will
be
the
same,
but
we
don't
didn't
address
that.
I
A
A
D
This
way,
if
you
want
to
computer
route
the
touch,
it
is
something
in
the
area
and
you
don't
care
what
it
is.
You
can
add
advertise
this
Sid.
This
is
distributed
by
the
area
leader.
It
gets
stuffed
into
the
proxy
LSP
and
then
it
has
to
be
accepted
by
any
of
the
inside
edge
routers,
and
thus
this
is
a
form
of
an
any
cat.
Sid.
D
We
also
changed
a
couple
of
other
things,
so
we
took
the
proxy
tale
of
the
we
changed
the
encoding
so
that
it
has
now
subtil
these,
and
it
now
contains
the
area
proxy
system
ID
that
was
already
there,
and
then
we
added
another
sub
TLV
so
that
all
of
the
nodes
can
pick
up
this
area
segment.
Sid
next
slide,
please
and
those
look
relatively
straightforward.
D
Next
slide,
we
also
found
a
issue
where
we're
having
trouble
distinguishing
what
nodes
and
more
particularly
pseudo
nodes
are
actually
members
of
the
inside
area.
That's
what
we
were
proposing
is
another
top
level
TLT.
This
is
to
indicate
that
any
node
is
an
inside
node
and
this
would
get
inserted
directly
into
pseudo
nodes
as
well.
D
We
also
spent
some
time
fleshing
out
the
text
describing
all
of
the
tail
V's
and
how
they
get
inserted
into
the
proxy
LSP.
You
can
go
through
all
the
text
to
see
all
the
gory
details.
Basically
the
gist
of
it
is.
If
it
pertains
to
external
connectivity,
it
gets
get
shoved
into
the
proxy
LSP
so
that
the
rest
of
level
two
you
can
see
what's
going
on
our
implementation
is
in
progress.
Hang
on.
D
D
D
B
Is
a
Salim
speaking
his
chair?
We
did
a
little
discussion,
you
know
of
the
requirements
for
this
and
you
know
we've
had
this
problem
with
this
in
ttz
having
similar,
but
some
you
know
similar
mechanisms
for
extracting
an
area,
but
you
know
there's
different.
The
drafts
cover
different
details
of
it.
B
We
didn't
get
a
lot
of
support
for
that
on
the
list.
I
mean
you
know,
you
know
people
really
want
in
it,
coupled
with
the
fact
that
we
have
we
have
we
have
a
lot
of
you
know.
We
have
this
IPR
problem,
not
an
IPR
problem.
This.
This
collaboration
problem
with
the
drafts
I,
know
you've
said
that
you
have
have
a
large
customer
that
wants
to
remain
anonymous.
Is
that
still
the
case?
They
want
to
remain
it.
D
G
B
B
D
D
We
spent
some
time
there
finding
the
adjacency
for
matching
rules,
they're
pretty
straightforward.
The
problem
that
we're
trying
to
solve
here
is
about
preventing
cross
branching.
If
we
have
a
miss
configuration-
and
we
are
not
disclosing
full
information,
then
we
get
into
this
annoying
problem
where
we
could
form
adjacencies
between
routers,
a
b
and
c
as
shown
here.
D
D
Miss
configuration
because
now
we
have
to
level
four
areas
interacting
directly.
We
don't
want
those
adjacencies
to
form.
So
this
is
what
we
felt
is
not
sufficient.
What
we're
proposing
instead
is
that
everyone
discloses
full
information.
It's
pretty
much
brute
force
solution,
but
it
does
seem
to
help
things
in
CC
in
the
bottom
picture.
Now
we
end
up
in
the
situation
where
router
B
is
advertising
its
full
hierarchy:
30
40
50
and
when
it
goes
to
form
and
Jason,
see
with
router
C.
You
can
see
that
has
inconsistency
at
level.
D
D
Okay,
so
here's
basically
all
rules.
So
if
you
have
information
about
where
you
are
in
the
hierarchy,
you
must
support,
must
advertise
it.
If
you
do
not
know
what
levels
you
have
supported,
you
can
advertise
a
dummy,
LSI,
LSA,
I
and
we're
just
using
0.
So
you
can
say:
I'm,
don't
know
where
I
don't
know
what
level
8
number
is
going
to
be
there
some
day,
but
we're
gonna
reserve
it
anyway,
and
you
can
add
a
level
by
using
multiple
level,
si
I
at
a
particular
level
and
then
removing
things
as
you
don't
need
them.
D
A
Well,
it
looks
like
I'm
in
the
queueing
and
I
I,
so
I
think
a
picture
it
might
really
help
like
it.
Just
has
really
simple
little
tree
would
help
explain
the
problem
since
we're
we're
actually
not
used
to
dealing
with
this
multiple
hire.
You
know
multiple
levels,
it's
not
immediately
obvious
or
I
think
make
it.
A
But
anyway,
that's
just
a
comment
that
the
would
it
work
to
just
advertise
I
mean
just
to
reserve
this
zero
or
whatever
and
advertise
it
always
just
make
that
a
mandate.
Then
you
don't
have
to
worry
about
the
Opera,
because
I
got
the
feeling.
I
read
the
draft
that
it
was
a
suggestion.
You
know
that,
like
operators
can
figure
it
that
way,.
D
A
D
A
M
M
Yes,
so
Chris
I
just
a
additional
response
on
that.
If
you,
if
you
know,
if
you
know
what
iria
addresses
you
want
to
use
at
each
level,
even
though
you're
not
currently
deploying
say
level
five,
six,
seven,
eight,
you
can
certainly
advertise
them.
You
don't
need
a
dummy.
Well,
if
you
haven't
decided
yet
like
I,
have
no
need
for
level
five,
six,
seven,
eight
right
now,
I'll
just
advertise
zero
and
whenever
I
get
around
to
eating
level,
five
I'll
figure
out
what
number
I
want
in
there.
That's
the
idea.
A
No
I
mean
I
understood
the
transition.
It's
just
I
just
thought
it
was
thinking.
Maybe
if
we
could
take
it
out
of
the
operational
you
know
you
wouldn't
have
to
guess
if
it
was
just
always
set,
but
as
Tony
points
out,
there's
a
trade-off.
You
know
with
the
Telos
peace
pace
of
that
pre
reservation
by
the
way
that
my
comment
was
question
was
as
I'm
working.
It.
A
A
D
B
N
N
N
We
can
read
the
zone
or
whatever
transfer
to
or
abstracted
he
returns
to
presume
note
again
so,
for
this
hand
of
obstruction
from
doing
to
pursue
the
note-
and
we
transfer
through
the
note
5
to
turn
this,
we
have
four
solutions
in
the
draft,
so
we
have
four
solutions
from
OPM
from
view
are
the
from
Isis
print
of
you.
So
this
is
a
one
big
part.
Existing
draft
right
now
you
addition
to
that.
We
also
have
a
solution
for
abstract
zoom
as
the
ages
of
the
zoom
foot
mash.
N
So
we,
if
we
go
from
the
center
of
the
future,
to
and
I
attract
the
zoom
as
ages
full
mass.
So
we
can
also
look
back
wrong,
just
put
mass
to
zoom.
So
what
and
then
we
could
rearrange
it
to
him
and
then
track
back
in
the
zooms
and
except
it
would
mash
so.
Regarding
to
this,
we
solutions
in
the
draft
for
SS
so
for
a
short
soon
to
be
a
ages
for
the
mash.
The
solution
for
OSPF
we
were.
The
party
has
already
become
RFC,
RFC
299.
N
Repeating
the
update
for
the
current
draft,
we,
yes,
the
chin
or
a
number
of
requirements.
So
originally
in
the
draft,
we
have
some
kind
of
requirements
such
as
we
must
be
back,
oh
I'm,
a
backward
compatible,
and
then
we
master
the
support
at
the
least
of
one
more
level
hierarchies
metal
with
scalability.
In
addition
to
that,
we
also
add
some
requirements
such
as
transfer
zoom
to
be
a
resume
note
or
it
is
fully
match,
should
be
smooth.
N
We
ate
that
these
are
two
more
requirements,
and
then
we
also
have
some
more
existing
requirement
in
the
original
draft.
So
I
think
that
we
should
have
some
feedback
from
surface
provided
for
you.
We
should
they
can
provide
more
requirements
or
provided
several
opinions
about
this
requirement
and
I
think
that
would
be
great
because
from
the
mainly
nice
testing
that
the
couple
host
several
providers,
the
their
interests
or
move
forward
for
different
kind
of
structures
and.
N
A
A
The
you
know
of
this
I
didn't
actually
go
back
and
look
at
the
couple
responses
that
we
did
get
about
people
publicly
saying
they
wanted
this.
How
many
of
them
use
knows
PF
as
their
network
right
like
the
solutions?
Actually,
there
is
an
experimental
RFC
right.
Has
it
been
it?
Has
the
OSPF
version
been
implemented
by
vendors
and
is
it
live
in
networks.
N
A
That
I
mean
that's
one
way
to
gauge
the
interest.
You
know
as
the
the
OSPF
one
as
far
as
I
know
hasn't
been
deployed,
even
though
there
there's
an
RFC
I
mean
we're
just
trying
to
dodge
the
bullet.
Here
that
we
have
this.
You
know
the
solutions
are
very
similar.
There
are
differences.
This
was
called
out,
I
couldn't
like
ITF,
104
I,
think
you
know.
There's
the
zone
boundaries
are
different,
there's
some
differences
in
it
and
what's
inside
the
LSP,
this
get
advertised.
You
know
in
a
couple
other,
but.
A
B
P
F
A
B
That's
what
the
you
know.
We
will
mention
that
the
OSPF
was
in
RFC,
it's
an
experimental
RFC
because
we
really
didn't
especially
there
were
a
lot
of
complexities
with
the
zone,
transformations
combining
zones
and
everything
we
thought
there
was
a
lot
of
corner
cases
there,
and
that
was
that
was
one
thing.
B
B
A
A
C
L
Q
Q
Okay,
so
this
is
pretty
much
an
update
based
on
two
things.
One
of
them
is
that
we
have
very
extensive
discussion
with
Les
and
Peter.
So
thanks
a
lot
for
all
the
time
you
know
and
stuff
that
they
pointed
out,
we
discussed
through
as
options
and
basically
implementation
experience
and
deployment
consideration.
You
know
deployment
planning's,
some
readability
improvements,
so
there
were
some
comments
from
a
couple
of
other
people
about
making
it
a
little
bit
more
readable
Asajj
that
a
couple
of
section
has
been
added,
mostly
for
clarification
and
some
split
for
easier
readability.
Q
Alright,
so
we
described
in
some
detail-
probably
not
enough
operation
without
l1
comments
right.
So,
if
you
run
this
thing,
the
draft
basically
describes
how
you're
running,
if
you
build
a
full
mesh
of
l1
columns
between
the
Leafs
or
basically
ingress/egress
border
nodes,
which
is
the
simple
mode,
that's
kind
of
a
no-brainer,
but
you
can
also
deploy
this
stuff
without
any
tunneling
whatsoever,
which
you
know
has
upsides
and
downsides
defined.
What
you
look
for.
Q
So
it
could
be
we
we
did
something
that
it
could
do.
Okay,
so,
basically,
right
now
we
define
that
the
router
is
strictly
either
a
client
or
a
reflector
or
nothing
at
all,
right
in
the
middle,
and
you
can
only
run
one
client
or
one
reflector
on
the
router
and
we
already
disallowed
the
links
between
reflectors.
Q
That
is
largely
based
on
what
was
happening
in
BGP
route,
reflector
hierarchies
and
when
people
started
to
mesh
reflectors
in
the
same
cluster
and
all
the
cluster
ID
and
then
whatever
it
was
called
the
identifier
of
the
router
in
the
classroom.
So
a
client
cannot
participate
in
multiple
clusters.
There
was
a
discussion,
I
think
mostly
less
thinking
there,
where
we
have
two
strange
things
where,
and
we
already
have
two
today
without
you
know,
I
need
to
acknowledge
you
whatsoever
when
areas
can
merge
when
you
have
multiple
area
IDs
and
they
overlap.
Q
So
we
basically
to
prevent
all
these
merging
scenarios
and
merging
scenarios
of
possibly
clusters.
We
basically
added
something
that
says
that
you
can
only
be
in
a
single
cluster
as
a
client
as
a
reflector
and
that's
fairly
simple,
because
you
have
two
ties
to
42,
which
is
the
capability
thing
and
we
just
allow
a
single
cluster
ID
being
such
a
thing.
Next,
one.
Q
Yeah,
okay,
so
now
it's
flipped,
okay,
so
more
detailed
changes
and
that's
already
pretty
much
the
last
one
is
so.
There
was
long
as
discussion
about
the
cluster
ID,
whether
we
should
couple
it
to
something
like
every
ID
or
other.
You
know
copious
amount
of
different
identifiers
that
we
have,
and
we
bank
went
back
and
forth
with
all
these
area,
emergency
scenarios
and
people
reconfiguring
any
deployment.
How
you
manage
the
stuff-
and
we
decided
didn't
know
we
decouple
the
cluster
ID
as
it's
completely
independent
concept
of
everything
else.
Q
Kind
of
like
would
be
teepees
and
to
prevent
you
know
in
terms
of
like
deployment
surprises.
We
basically
edit
that
the
cluster
ID
should
be
unique
across
the
network,
so
basically,
each
whatever
you're
on
loves
all
goods.
Let's
assume
we
want
in
l1
eius
eius.
Is
this
thing
where
we're
reflecting
all
the
stuff
around
the
cluster
ID
will
be
unique
in
each
of
this
l1
across
the
network
as
operationally.
You
know
if
something
goes
wrong.
Q
You
now
have
this
things
simple
and
all
the
existing
tooling
works
without
you
know
a
lot
of
backward
bending,
so
when
leaking
we
basically
defined
it
all
leaves
should
be
so
all
you
know,
egress
ingress
into
area
should
be
reflective
clients.
It
will
work
with
that.
But
that's
what
we
say
right
and
these
conditions
by
the
way
are
fairly
easily
detected
because
we
flood
around
l1.
So
you
know
who
is
client
and
who
is
reflector
and
who
is
the
justices
and.
Q
Why
that
is
also
done,
because
if
you
leak
l2
into
l1
as
a
leaf-
and
you
know
the
reflector,
then
when
someone
computes
across
the
reflector
adjacencies,
you
cannot
be
egress
anyway.
It's
kind
of
pointless
you're
leaking
right,
because
if
the,
if
the
forwarding
happens
so
if
you
know
the
client,
you
will
have
not
l1
tunnels
to
the
other
leaves.
Q
And
if
you
try
without
panels,
then
you
cannot
be
an
egress
right,
because
then
you
base,
if
you
start
to
do
surprising
things
in
terms
of
forwarding,
because
people
have
outside
of
that
of
the
cluster
compute
everything
through
l2
adjacency.
So
they
expect
you
to
follow
to
an
egress
right.
You
have
to
take
a
certain
egress
for
the
hole
thing
to
become
this,
then
so
the
leaf.
That
is
not
a
client,
even
if
it
shows
the
prefixes,
will
never
be
chosen.
Q
You
basically
have
to
hop
computation
and
then,
if
you
decide-
and
that
is
still
allowed
again
no
practical
consideration,
the
deployment
is
that
the
clients
can
build
straight
l2
adjacencies
between
them
and
the
computation
can
take
that
into
account.
So
that's
also
working
is
expected.
I
think
the
stuff
has
been
brought
to
the
lists
as
I
think.
A
very
kind
of
varnish
problem
definition
I
think
there
was
a
support
by
a
couple
of
people's
family
to
his
solution.
Of
course,
you
know
we
would
like
to
see
this
work
adopted
into
rolling
forward.
Q
A
Hi,
this
is
Chris
ops.
As
a
working
group,
member
I
went
back
over
the
draft
and
I
wasn't
able
to
put
it
all
in
my
head.
It
still
seems
a
little
complex
to
me,
but
I
I'm
wondering
can
you
is
it?
Is
this
draft
trying
to
do?
Is
a
solution
trying
to
to
do
two
to
two
things.
I
mean
it
talks
about
using
l1
as
a
transit
for
l2
yeah.
Q
Q
I
mean
you
don't
see,
any
flog
reduction
algorithm
see,
of
course,
block
reduction.
We
work
over
the
staff,
each
I
mean
well.
So
if,
if
I'm
forced
to
compare
right,
this
is
presenting
a
much
smaller
l2
topology,
so
naturally,
you're
flooding
will
be
far
less
without
any
further
additions
to
the
protocol
or
complex
computations
or
you
know
any
other
surprises.
It's
a
standard
running
is
just
much
less
l2
topology
to
flock
around
and
that's
what
it
is
because
they
are.
One
of
course,
is
contained
to
the
l1.
Q
So,
in
a
sense
it
gives
you
our
traditional
hierarchy,
I
mean
so
we
didn't.
We
didn't
think
through
any
of
this
higher
draft
right.
So
I
looked
at
an
article,
I
think
I
talked
you
know
less
and
appalled
through
the
hierarchy.
You
know
bifurcation
problems,
I
think
the
solution
presented.
There
is
fine
right,
it
has
certain
properties,
but
you
know
it's
only.
Q
The
static
versus
dynamic
surprises
right,
because,
if
the
poly,
if
the
hierarchy
levels
that
renegotiate
that
you
may
end
up,
you
know
tearing
down
adjacencies
again
so
I
think
what
they
have
is
something
like
more
static
but
more
predictable.
So
that's
fine,
as
Tony
previously
pointed
out
the
multi
instant
stuff.
What
we
see
for
all
practical
purposes
can
solve
the
problem
as
well
in
a
kind
of
more
dynamic
and
flexible
way,
but
you
know
that
it's
not
the
dynamic
is
not
necessarily
desirable.
I
think
we
will
need
some
operational
description
hundred
words.
Q
So
this
is
basically
just
traditionally
presenting
a
much
lower
l2
topology,
while
suspending
the
limitation
that
you
cannot
have
the
whole
entropy
of
l1
when
you
traversing
it
yeah
and
lots
of
this,
where
the
thinking
was
going
here
was
driven
by
operational
constraint.
Right
like
tooling,
this
stuff
will
work
without
any
changes,
for
example
the
BGP
LS
right.
You
can
take
the
topology
other
one
out
of
l2
and
do
all
stuff
you're
doing
today
on
the
traditional
tooling,
which
shows
a
large
consideration.
Q
B
Q
Q
B
Another
comment
I'd
like
to
make
this
draft
has,
if
you
read
the
introduction
of
it,
you
can
envision
how
the
area,
proxy
or
ttz
would
be
used
in
the
same
in
a
in
a
data
center
fabric,
because
it
has
the
best
picture
of
exactly
how
you
use
l12
as
transit,
or
you
know,
for
l2
before
I
mean,
if
you
look
at
that
before
you
do
any
optimization.
So
that's
a
good
point.
The
other
question
is
this
is
a
question.
How
would
you
scale
this
given
given
the
cluster
ID?
How
would
you
scale
this
to
multiple?
Q
B
Q
Yeah
precisely
so,
basically,
we
were
well
event.
Whatever
I
mean
the
desired
properties
years
that
you
just
carved,
rupees
make
it
l1
and
l2.
Basically,
nothing
shakes
much
yeah,
except
you
see
much
more
topology
while
still
getting
all
the
you
know,
basically
forwarding
diversity,
12
watts.
Yes,
so
no
there's,
no,
like
nefarious
like
plan
to
build
completely
different
networks.
No,
it's
a
very
practical
problem.
People
just
ran
out
of
air
to
scale
and
we
both
know
that
scaling
deployment
a
shinee's.
You
know
it's
a
hell
of
a
exercise.
Q
So
yes,
of
course
everybody
is
doing
what
they
can
right,
but
this
is,
but
you
know,
if
you
do
whatever
smart
thing
you
do
and
you
tell
people
that
they
have
to
forklift
the
whole.
The
whole
network
at
those
size
is
a
complete
no-brainer
and
no
starter
even
forklifting.
A
whole
part
of
the
network
is
a
no
storm,
so
the
the
TCC
draft
says:
okay,
we
can
carve
these
things
and
slowly
migrate,
the
stuff
and
so
on,
but
those
are
real
consideration
right.
Q
A
N
Hello,
everyone
can
you
hear
me
today,
I'm
going
to
present
ICP
4
right
over
here
ability
right
now,
more
and
more
light
will
be
controlled
by
central
controller.
Ok,
next
page
right
now,
more
and
more
light
oak
are
controlled
by
central
controller,
so
the
central
controller,
it's
a
single
photon
field,
Airport
LAN
work,
so
the
controller
reliability
you
are
a
higher
VG
is
very
important
for
light.
N
Ok
when
I
beach,
this
so
for
so
here
we
consider
a
controller
as
a
cluster
controller
cluster
so
from
outside
from
outside
this
point
of
view,
cluster
is
a
single
controller.
From
the
inside
to
point
of
view,
a
cluster
consists
of
four
number
four
individual
controllers.
So
when
Michael
failure
or
failure
happens
inside
the
cluster,
so
those
are
individual
controllers
may
be
splitted
into
separate
control
groups.
After
this
split,
so
each
group
will
elect
as
a
primary
controller
group,
and
then
we
may
have
a
multiple
primary
and
four
groups
controllers
which
will
control
the
light.
N
N
So
when
we
have
multiple
failures
and
then
the
controller's
was
laid
to
the
inside,
then
in
this
case,
so
we
only
elect
one
intent.
I
am
a
controller
in
one
group
to
end
of
it
hides
many
informations
about
a
controller
in
in
its
group,
because
every
in
the
separate
group
well
triple
its
information
to
the
authors.
So
every
separate
group,
where
have
all
the
information
about
the
authors,
so
in
this
case
so
one
exactly
the
one
primary
group
will
be
elected
directly
and
control
the
lateral
correctly.
So
that's
the
idea
here.
So
next
page.
N
When
faith
in
happens,
cluster
is
splitted
into
multiple
individual
controller
groups.
Each
group
we
are
elect
a
controller
and
a
second
controller
or
key
say
the
elections.
A
record
intend
both
because
it
is
not
finally
decided
yet
so
each
each
group
we
intend
a
controller.
We're
enter,
provides
information
about
its
group
and
then,
after
those,
the
information
is
tricky
to
get
to
every
controller,
and
then
you
should
well.
You
lacked
the
final
time.
The
group,
so
here
we've
also
proposed
that
hyper
key
because,
for
example,
we
have
even
policies
to
elect
primary
control.
N
A
group
resemble
one
one
policy.
Your
strategy
is
that
the
biggest
group
will
be
become
done.
I
know
primer
group
for
some
case
to
group
away
may
have
the
same
number
of
controllers.
So
in
this
case
we
have
all
the
purples
and
fiber
routines.
For
example,
we
may
use
a
priority
as
time
allocation
when
we
use
the
older
position.
N
N
N
Please
just
give
example
procedures
for
recovery,
so
we
have
mark
no
failures
and
then
we
submit
a
cluster
into
matter
group.
We
just
kept
the
example.
We
did
the
controller
cluster
into
two
groups.
One
group
is:
why
is
a
CD
and
that
group
is
B
and
then
and
then
we
have
two
groups
and
each
group
just
to
control
us.
So
each
group,
with
collector
intent,
primary
antenna
signal
and
so
on,
for
example,
include
one
a
will
be
in
Telugu
primer
and
then
in
the
group
tool.
N
P
will
be
the
intended
primary
controller
so
for
each
group,
only
the
inventor
private
and
Halliwell
and
applied
to
the
information
about
in
school.
So
after
those
information
at
identified,
each
group
will
enact
real
primary
controllers
here,
because
each
group
will
have
a
number
two
and
then
they
were
using
hyper
recruits.
Here
we
just
used
the
older
position
because
a
a
the
the
primary
controller,
because
he
has
hired
I,
asked
listen,
so
Google
one
will
be
elected
as
a
group.
So
from
one
will
be
you
really
liked
it
I'm
the
group
so
group
who
won.
D
N
D
So
why
why
do
you
have
to
make
this
so
complicated?
It
seems
like,
even
if
you
wanted
just
stood
on
doing
this
in
the
IGP,
you
basically
could
have
each
instance
advertise
itself
and
say:
hey
I'm,
a
controller
and
then
deal
with
this
outside
of
the
AGP.
All
the
election
stuff
should
not
be
involved
here.
You.
N
N
N
A
E
E
It
doesn't
seem
to
me
like
they
need
to
be
baked
into
routing
protocols
and
what
I
discovered
this
morning,
as
you
are
talking,
is
that
there's,
if
you
even
think
that
it
needs
to
be
a
standardized
there's
apparently
in
IETF
RFC
777
called
distributed
node
consensus
protocol,
which,
if
you're
gonna,
invent
a
consensus
protocol,
it
would
probably
be
smart
to
begin
by
looking
at
what
we
already
have
in
the
RFC
said:
I'm
not
actually
recommending
this
thing
because,
like
I
said,
I
just
discovered
it,
but
it
exists
end
of
comment.
I.
N
Reliability,
yes,
I
know
those
kind
of
reliability
were
each
a
number
of
a
number
of
ways.
There
I
think
this
is
a
mess
away
and
then
maybe
single
or
you
see,
we
can
see
that
we
just
to
choose
the
meat,
more
information
using
ITP
and
just
either
one
key
of
e.
We
resolve
the
most
of
the
problem
right.
A
Is
that
there's
five
authors
in
this
draft,
but
there's
no
references
to
any
prior
art
I
mean
this
is
not
a
consensus
and
leader
election.
The
stuff
is
well
studied
and
it's
not
a
good
sign.
If
you
just
come
in
and
you
haven't
gotten
any
references
to
what
people
have
done
before
right
I
mean
we're
not
in
the
job.
Is
this
working
group
I?
Think
of
coming
up
with
you
know,
consensus
algorithms,
so
we're
not
necessarily
the
experts,
so
it
certainly
helps
if
you
have
the
prior
art
listed,
showing
the
justification.
A
A
If
this
were
gonna
go
anyway,
this
this
sort
of
falls
into
a
comma
I'm
gonna
have
later
too,
but
this
it's
even
if
you
do
something
it
would
be
to
advertise
like
Tony
was
saying
to
advertise.
You
know
the
cluster
IDs
I
mean
not
the
cluster
IDs.
The
this.
You
know
the
controllers
and
then
you're
talking
about
basically
using
IGP
as
a
transport
for
information
that
has
nothing
to
do
with
the
IGP,
and
this
is
Gilligan.
This
is
gonna,
be
my
comment
later
on.
A
N
F
F
F
So
here
we
each
other
on
to
introduce
our
thought
and
the
passio
interface
a
shield.
So
next
next
to
put
slide
you,
you
know,
in
our
network,
try
to
use
the
steam
controller
to
get
the
under
topology
of
the
underlying
network,
while
bt
pls
and
currently
the
BTS
counry
deported
the
links
within
event
of
me,
and
it
can't
discrete
distinguish
reporter
link
from
the
internal
links,
0
1
draft.
In
idea
group
of
the
interest
table
tissue
and
the
leaf
draft,
it
defines
the
stop
link
on
our
to
report.
F
The
potter
link
by
the
new
air
are
so
rotor
rotor.
You
know
you
destroy
from
0
to
2,
Tamiya
van
rental
means
100.
Sp
on
our
domain
is
on
ISS,
so
for
rotor
within
OSPF
domain,
for
example,
a
school
can
extract
the
porter
links
according
the
type
at
the
type
indicator,
which
is
stated
in
the
rotor
air
sa
button.
F
Is
s
I
know
such
information,
so
the
rotor,
for
example,
Chiba,
cannot
extract
artistic
act,
motherly
ink
of
Ramudu
other
internal
links,
and
there
are
other
other
scenario
we
want
to
discrete
distinguish
the
edge
leaks
from
the
already
so
forth.
For
some
security
region,
so
you
know
we
want
to
put
some
more
different
policy
in
the
polar
ink.
So
here
we
just
went
to
the
screen
teaching
teaching
with
the
pottery
ink
under
the
edge
link
from
the
our
normal
internal
links.
So
next
next
next
slide.
Please.
F
So
look
our
for
a
solution
is
usually
in
this
slide.
Our
1794
help
define
the
high
period
ipv6
attended.
Each
parity
attribute
flag,
sub
T
elevate
whenever
thises
additional
flag
associated
with
prefix
countries
it's
defined
for
for
flex.
Here
we
went
to
propyl
and
van
flex
named
appeal
for
passive
all
as
first
up
to
an
indicator
that
the
prefix
are
coming
from
over.
F
She
washed
up
links,
so
if
the
edge
rotor
and
tag
this
tag,
the
prefix
Ian
Ian
easy,
who
misses
a
severe
rotor,
especially
the
rotor
at
the
on
the
pitch
pls,
can
easily
extract
the
certain
links
from
the
internal
innings
and
popular
unpopulated.
The
edge
links
in
the
stop
link
layer,
as
defined
in
the
idea
working
group.
So
next
next
slide.
F
F
5050
find
the
link
attribute
the
sub
theory
to
advertises
additional
information
about
the
leak
that
character,
characteristics,
country,
data,
finalists
repeat,
but
we
but
I
have.
We
have
investigated
it
subtly
and
you
can
find
at
least
subtly
already
included
in
the
energy
area,
T
related
to
the
existing
of
the
eyes
neighbor,
but
pastor
cons
in
our
in
our
in
concern.
Although
there
are,
you
know,
eyes
neighbor,
so
we
think
the
it
is
not
a
opportunity
to
put
the
information.
F
F
J
Get
ontological
Cisco
high
agent
so
about
using
prefixes
for
identifying
links,
I,
think
that
is
discussed
on
the
list
as
well.
That
doesn't
seem
like
an
appropriate
way
for
no
signaling
or
identifying
links
or
interests
links
as
as
what
you
know,
you're
looking
for
here,
even
in
OSPF.
For
that
matter,
the
stub,
the
link
type
stub,
doesn't
really
distinguish
between.
J
J
F
A
R
Okay,
first
of
all,
I
think
of
the
viewer.
The
email
list
gives
a
kind
comments
and
suggestions
about
this
draft,
and
next
I
will
introduce
this
dot
so
and
in
the
I
faith
framework,
but
it
gives
overview
II
and
complex
framework
about
the
unpassed
telemetry
tech
techniques,
including
IOM
and
PBT
and
etcetera
next
page.
Please.
R
Okay,
so
we
found
that
I
feel
is
the
solution
poker
sound
network
domains,
so
the
means
that
we
think
and
I
feel
torment
their
own
past
telemetry
techniques
may
be
selectively
selectively
or
partially
be
enabled
in
different
devices.
So,
let's
draft
we
trying
to
solve
the
problem
to
to
before
we
tell
me
in
April
I
face
application
in
a
gaming
Network
domain,
and
so
we
need
to
find
and
advertised
afraid
not
capability
of
the
devices
in
network
domain.
So,
let's
draft
we
extends
the
IGP
and
the
PTP
else
protocol
to
to
advertise
the
f8
capability.
R
So,
yes,
you
may
ask
the
questions
why
we
need
the
information
advertisements
here,
so
we
list
you
application
example
here.
The
first
one
is
to
avoid
the
leak
of
I,
fixed,
specific
hater
and
amento
data.
So,
as
we
see
here
so
because
I
faced
a
specific
hater
and
mental
data
must
be
removed.
At
the
end
of
the
note
so
avoid
avoids
the
leak
of
I
fixed,
specific
hater.
We
need
to
know
the
end.
R
Well,
EndNote
have
the
I'faith
capability,
so
this
is
the
one
case
we
we
need
to
extend
the
IDP
and
order
be
GPOs
protocol
to
advertise
this
information
and
the
second
application
we
mentioned
here.
As
we
know,
the
ifreet
option
types
have
different
encapsulation
format
and
different
processing
procedure.
R
We
also
listen
here
to
example,
but
the
one
is
traced,
option
types
and
the
left
side.
We
can
that
they
expect
the
rubic
impact.
The
tracing
data
can
be
collected
at
every
I,
am
trusted
note
and
the
data
can
be
a
processed
at
every
tragedy.
Note.
So
for
this
case,
we
want
to
know
the
tread
is
note,
the
better
it's
about
the
trust
option
types
capability.
So
this
is
another
size
of
the
example
here
we
want
to
answer
the
question
why
we
need
to
advertise
if
it's
information
next,
please.
B
B
So
you
would
have
to
you
would
have
to
use
the
information
from
all
the
routers.
You
would
have
to
compute
the
same
pattern
off
if
you're,
using
you
know
best
path,
routing
or
if
you're,
using
some
other
kind
you'd
have
to
compute
that
path.
No,
in
fact,
that
all
the
transit
notes
and
the
capability.
R
Yes,
actually,
for
example,
we
can
look
at
the
second
example.
If
we,
if
we
are
very
intent
to
enable
a
kind
of
I
fate,
option
types
to
a
specific
route
routing
or
the
SR
mystics
routing
we
need,
we
need
to
know
the
head
node
and
the
end.
A
node
has
the
same
namespace
and
the
same
option
types
they
stood.
Suppose
she
avoided
leak
of
header
and
second,
the
in
the
middle.
In
the
middle
note,
I
mean
the
trial
didn't
know.
B
R
R
Directly
exports
and
the
EAM
option
tied
so,
as
we
said,
I
fit
option
types
and
have
the
relationship
with
the
namespace
identifier,
because
the
names
namespace
identify
allowing
devices
to
determine
bad
words,
I
fit
option
types,
it
should
be
processed,
so
we
can
see
a
list
breaker.
We
we
define
the
node
capability.
Information
format
helps
to
build.
One
is
the
names
based
identifier
and
another
one
is
the
type
option
option
types
enable
flag?
R
R
So
here
in
latina
straps,
we
take
advantage
of
the
OSPF
yeses
and
the
bt
pls,
because
the
it
also
defines
LSA
of
TLB
to
advertise
some
router
capability
information
in
this
routing
domain.
So
we
extend
a
new
I
paid,
not
capability,
ter
ways
to
advertise
this
information.
So
here
we
listed
day
this
year
we
for
each
of
the
protocols.
R
S
A
Yeah
I
sent
a
couple
things
on
the
list
on
this,
because
this
is
definitely
getting
into
the
territory
I.
Think
of
advertising
application
stuff
in
a
GPS
I
realized
that
the
telemetry
is,
you
know,
related
to
the
IGP
functions,
but
lots
of
things
are
related
to
the
IDP
function.
You
know
in
that
sort
of
sense,
I
I
mean
my
suggestion
is.
If
I
was,
if
I
was
like
trying
to
make
this
work
with
operators,
I
would
just
implement
it.
I
mean
you,
don't
even
need
to
change
any
protocols.
B
A
Topology
and
what
your
routers
are,
you
know
you,
don't
you
don't
actually
need
this
capability
stuff
advertised
in
a
GPS
you
know
is:
is
OSPF
BGP?
You
don't
need
to
do
any
of
that.
You
could.
You
could
literally
deploy
this
today,
just
you
to
farang
model
and,
and
so
I
think
that
there's
like
maybe
a
first
step
wrong
right,
like
that.
The
first
step
was
to
assume
that
this
was
the
right
way
to
do
it
and
now
the
justification
is
you
know
this
is
really
useful
stuff.
A
Well,
yes,
I
think
it's
very
useful
stuff
I
think
telemetry
data
is
very,
you
know
useful
and
important,
so
I.
My
comments
are
never
and
have
not
been
directed
at
whether
this
is
useful
or
not
it's
just
whether
advertising
application
availability
in
the
IGP
is
correct
and
to
that,
even
if
the
working
group
decides
that
it
is
I,
think
I
think
we
at
least
need
as
a
working
group
and
I'm
saying
as
those
working
to
remember,
but
I
think
we
need
to
use.
Jenna
I
mean
this.
A
R
Thank
You
Christine,
I
I
saw
your
comment
in
the
email
list
and
I
think
yes,
I,
think
not
NAT
calm,
young
yeah,
it's
another
valid
weight,
as
I
mentioned
in
the
email
list,
and
to
configure
the
capability,
I
think
and
export
the
capability.
But
let
let
us
think
another
way
if
we
want
to
dispute
a
SR
routing
based
on
the
hyssop
were
BGP.
R
A
I
mean
this
is
one
thing
that
when
I
was
thinking
about
this
I
I
specifically
thought
that's
why
I
mentioned
the
wrong
first
stop
right,
I
mean
if
you
start
by
doing
everything
and
routing
protocols
and
it's
sort
of
natural
to
keep,
keep
going
and
keep
doing
things
that
routing
protocols
and
what
I
was
saying
is
kid?
Could
you
step
back
and
say
why
does
this
need
to
be
in
routing
protocols
at
all,
and
maybe
it's
not
so
obvious
that
these
capabilities
should
be
in
there
I.
B
S
Our
Christian,
in
fact,
I,
have
two
comments,
so
the
first
comments
are
usually
related
to
your
comments
from
my
pointer.
Will
we
know
that
the
MSD
information
is
advertised
by
a
GP,
so
I
think
that's
the.
We
know
that
the
I
don't
think
reduce
that
I
fit
here,
you'll,
occasionally
information
because
of
this
I
fit
in
situ
om.
We
can
also
see
attributes
belong
to
a
specific
as
our
past,
because
it
will
check
at
least
the
performance
of
the
SR
pass,
so
I
think
I
will
calculated
the
calculated
as
our
pass.
S
We
also
we
also
to
specify
even
the
Incipio
om
is,
can
be
done
for
the
SR
pass.
We
as
our
policy
when
we
calculate
this
information.
We
needed
to
collect
this
information
from
the
network
node
to
understand
the
capability,
also
the
in
situ
om
functionality.
We
filled
it
for
this
as
our
path.
So
from
this
point
of
view
is,
do
you
think
this?
The
I
think
the
capability
used
to
do
is
the
path
the
attributes
I,
don't
think
you.
S
Will
not
a
changes
into
the
I
will
not
a
changes
is
maybe
not
a
change
in
the
past,
but
I
think
that
if
I
understand
the
capability,
Ike
I
don't
know
that
as
her
downloaded
as
our
policy,
we,
this
theater
I
figure
a
capability,
because
you
know
I
will
enable
this.
That
I
feel
functionality
for
super
specific
is
our
past,
but
you
for
the
use
of
the
nose
Alondra
as
our
past.
You
know
some
knows
the
cannot
support
I
feet.
So
these
are
the
I
theta.
This
is
I
hated.
A
Do
so,
basically
that
I
think
my
the
answers
is
where
I'm
looking
for
the
difference
right,
if,
if
you
were
saying
I'm,
gonna
route
traffic
differently
based
on
whether
I
can
trace
it,
I
think
that,
then
you
start
talking
about
your
making
routing
decisions
based
on
this
information,
and
then
it
might
be
long
in
the
routing
protocol.
But
if
you're
not
modifying
routing,
then
it's
something
that's
sitting
above
it's
a
layer,
see.
S
S
Okay,
I
think
we
have
more
discussion
in
the
mainly
a
still
yeah.
This
is
the
first
comments.
It's
just
taking
a
moment,
sir
I'm
not
sure
this
the
process
of
the
use
of
the
property,
because
the
I
say
that
as
they
as
OSPI
food,
you
one
dropped,
I.
Think
it's
okay,
but
I.
Don't
know
why
the
Pete
stated
he
is
also
in
the
dropship
III.
Don't
think
this
is
not
a
traditional
way.
I
can.
S
E
Just
with
regards
to
you
know
individual
versus
working
group
for
after
me,
and
presumably
every
individual
draft
that's
being
presented
at
any
working
group,
is
you
know
the
author's
hope
for
it
to
become
a
working
group
draft?
So
if
you
think
the
ultimate
stage
is
to
have
it
be
a
working
group
draft,
then
you
know
go
ahead
and
guide
your
individual
draft
in
that
direction.
I
don't
think
it's
too
early
to
put
it
into
the
form
that
you
would
want
it
to
appear
in
the
working
group.
Thank
you.
Sorry
for
jumping
the
queue
mm-hmm.
S
B
B
A
O
So
I'm
close
to
Chris
position
here,
leaking
management
play
information
control.
Plane
is
not
a
good
idea.
I,
don't
think
the
capability
advertisement
would
be
used
to
actually
pass
establishment
because,
besides
capabilities,
we
also
need
to
configure
things.
So
you
would
need
configuration
channel
anyway,
configure
namespaces
and
other
options,
so
it
just
doesn't
make
sense
to
discover
things
you
need
to
configure
the
master
works
anyway.
I
really
don't
see,
reason
for
the
stuff
to
be
in
a
GPS
and
comparison
to
miss
D
is
really
incorrect.
Msd
is
directly
used
for
constraint,
pass
computations
a
PC.
O
H
Okay,
oh
so
I'd
like
to
maybe
measure
another
use
case
of
advertising.
There's
some
capability
information
which
it's
a
seamless,
PFD
I,
think
also
the
seamless
ppfd
can
be
used
it
for
the
IGP
for
the
loudness
detection,
but
it
can
also
be
used
for
motor
hubs,
past
detection.
So
this
is
maybe
these
capabilities
not
that
directly
related
to
the
past
computation
or
the
routing.
O
A
P
P
Otherwise
the
packets
would
be
will
be
dropped
right.
So
so
that's
one
requirement.
That's
there
is
a
draft
for
it
that
step
so
I'm
not
proposing
this
as
a
solution,
but
I'm
just
saying
that
there
is
one
requirement
for
a
sorry,
I'm
fearless
for
n
cap
node
to
know
the
D
cap
is
capable
of
processing.
It.
F
I
think
the
the
media
should
be
can
be
advertised
by
the
IT
protocol,
because
if
you
configure
it,
if
you
configure
its
infamous-
and
why
is
the
net
Cavalia
a
one
twice-
the
art-wise
cannot
know
the
capability,
so
it
cannot
decide
whether
whether
it
can
use
the
high
fade.
So
you
know
okay,
so
I
think
the
in
man
is
the
less
he'll
provide
the
harder
way
to
transfer
some
human
reason
that
he
divides
by
the
IPP
level.
F
A
So
I'm
in
the
queue
behind
you
but
I
would
say:
I.
Don't
I,
don't
understand
that
like
you,
you
certainly
could
query
anything.
You
want
with
you
hanging
right
and
you
can
and
there's
a
capabilities
directly
built
into
it.
So
the
transport
is
there,
I
mean
it's
not
a
GPS.
Don't
have
to
be
used
as
a
transport
to
query
routers
on
their
capabilities.
A
A
That
was
what
I
was
gonna
write.
Something
is
computing,
these
paths
right
so
I
mean
the
the
endcap.
You
know
turning
these
features
on
and
often
configuring
them
you
have.
You
have
to
query
capabilities
to
know
that
you
can
do
that
right.
It
doesn't
again,
it
doesn't
have
to
be
advertised
in
the
IGP.
It's
just
really
easy.
You
know
and
I'll
put
my
chair
hat
on
for
a
minute.
It's
really
easy.
A
If
you
have
a
little
bit
of
information
that
you
want
your
network
to
know
to
say,
hey,
let's
throw
it
in
the
IGP
right,
because
and-
and
you
know,
having
been
chair
for
a
while
I've
seen
these
presentations
many
times
right
and
and
because
it's
so
easy
to
say,
let's
just
throw
it
in
the
IGP
and
it'll
get
distributed,
and
we're
done
and
so
there's
a
real.
You
know.
A
People
often
think
that
that's
a
great
idea
and
we
had
a
lot
of
pushback
in
the
past
in
eius
eius
and
we
even
wrote
and
published
work
as
a
working
group
called
Gienapp
to
specifically
take
all
these
wants
and
move
them
out
of
the
main
routing
protocol
instance.
So
this
is
this
is
not
a
new
ask.
It's
just
that
one
that
we've
we've
pushed
back
against
for
many
many
years,
I.
S
S
P
P
B
Iii
guess
we
should
take
this
to
the
list.
This
is
AC
speaking
as
co-chair.
Also
is,
that
is
the
other
Ned's
is
gonna.
Be
interesting.
With
this
work
is
whether
this
whole
I
fit
anymore.
It
gets
adopted
in
the
management
work
whatever
it
is.
The
OEM
and
she
group
right
you're,
presenting
that
there,
when
they
do
have
an
interim.
R
Yeah
and
I
think
I
thought
not
as
the
narrows
it
has
a
different
the
change
NASSCOM
and
the
existing
protocols.
So
so
can
we
think
less
scenario.
That's
if
we,
if
the
topology
of
network
dynamic,
changing
so
maybe
use
the
IGP,
and
we
do
be
protocol
to
to
collect
the
changing
of
the
note,
as
the
first
I
mean
the
efficient
way
to
to
to
now
the
changing
and
the
capability
of
knows.
R
B
This
is
AC
again
that
wasn't
my
question,
but
the
statement
I
was
making
again
is
the
framework
being
adopted
in
the
ops
area.
Working
group
would
be
sort
of
that
would
be
that
would
have
to
come
before
us.
Putting
you
know
considering
this,
for
the
iGPS
and
and
for
BGP
LS,
your
framework
document
right.
S
We
would
be
constantly
I
think
the
framework
Lisa
protocol
extension
in
the
OTS
a
year
and
also
that's
not
important,
but
now
they
were
also
taken
down.
This
OPR
CWD
is
the
working
group
on
the
travel
over
the
I
theta
walk
because
that's
what
the
informational
I'm
not
for
the
iti
makinia
Warcraft
o.
S
A
H
Okay,
I'm
start
okay,
so
this
is
a
GP
extensions
about
the
same
routing
based
VPN
and
we
are
presenting
update
of
the
existing
draft
and
also
to
simplify
the
solution
which
was
specified
in
the
theater
version.
Here
are
some
background.
First,
the
vaping
PLAs
framework
has
describing
the
ITA
T's.
He
has
been
dropped,
which
is
a
working
group
document.
It
ascribes
a
layered
architecture
and
canada
technologies
in
each
layer.
Two
together
enabled
enhance
between
service
and
water
transport
network,
virtual
networking
in
the
VPN
class
architecture
and
for
the
SR
paced,
vaping
applause.
H
We
have
this
magnesium
metal
playing
extensions,
the
define
in
the
spring
as
sulphur
in
hands
between
draft.
Basically,
basically,
it
defines
mechanism
to
associate
a
Cersei's
with
a
different
set
of
resources
in
the
network
for
the
packet
processing
and
so
that
it's
kind
of
a
resource
aware
sees
could
be
used
to
build
the
resource,
currently
the
virtual
networks
or
/
pass.
It
also
describes
the
mechanism
for
the
creation,
the
packet
forwarding
in
the
SR
based
diversion
networks.
H
For
this
documents,
we
want
to
define
the
IGP
protocol
mechanisms
and
the
necessary
extensions
to
support
is
SR
based
living
class
so
that
a
GP
can
be
used
to
distribute
the
required
information
in
the
network
and
also
to
the
controller,
and
we
try
to
reuse
existing
protocol
mechanisms
to
build
a
basic
or
simplify
the
solution
and
in
addition,
we
also
provide
a
flexible
and
scalable
solution
with
some
more
additional
protocol.
Extensions.
Okay,
next
slide,
please.
H
First,
the
vaping
applause
missing
the
between
class
kind
of
enhanced
the
Vivian
service
which
additional
commitments,
such
as
the
enhanced
isolation
or
performance
guarantee,
and
another
term
is
called
virtual
transport
network
or
VPN
and
basically
VTN,
is
a
virtual
and
entered
a
network
that
connects
that
customer
sites
with
the
capability
of
providing
the
additional
commitment
like
the
isolation
performance
characteristics
which
are
required
by
the
service
customer.
So
with
the
end
speed
with
a
customized
apology
and
a
set
of
other
dedicated
or
shared
a
natural
resources.
H
So
we
can
see
the
relationship
between
the
VPN,
palasa
and
VTN
and
with
physical
network
within
the
implants
is
provided
based
on
the
integration
between
the
overlay
VPN
and
it
and
early
virtual
transfer
network,
all
VTN,
so
VTN
can
provided
the
underlay
for
one
a
group
of
vaping
plus
services.
So
this
is
a
relationship
between
these
terms.
Ok,
next
slide.
H
Ok
here
we
briefly
talk
about
the
mechanism
we
proposed
in
this
high
res
art
as
our
between
empty
draft.
This
drought
describes
how
to
use
the
multi
topology
and
the
existing
is
STR
wheeze
or
sub
theories
to
advertise.
The
attributes
of
a
VPN
English
draft,
we
reuse
the
MT
ID
as
a
control
plan,
identifier
of
a
VT
n
and
it's
a
Mountie
topology
Isis
Markinson
is
eautiful
used
for
the
topology
advertisement.
H
We
can
use
the
ISIS
Admiralty
extensions
to
advertise
to
NPR
C's
or
a
service
X
locators
of
C's
at
the
pathology
level.
In
addition,
we
also
consider
to
advertise
the
patate
topology
teh
abuse
for
each
BDN.
For
example,
we
need
to
advertise
the
bandwidth
for
reserved
for
each
topology
on
a
specific
link.
In
this
case
we
can
reuse
the
maximum
link
pen
waste
of
goe,
but
since
one
link
can
participate
in
multiple
topologies,
so
in
this
case
we
we
don't
need.
We
should
not
advertise
them.
Physical
link,
banner
waste
for
the
port
topology.
H
We
should
advertise
maybe
a
subset
of
the
link
resource
bandwidth
for
a
specific,
the
topology
and
the
some
of
these
others
anyways
for
all
the
topologies
on
one
link
will
be
physical
link
bandwidth.
Another
mechanism
means
we
can
advertise
the
association
of
the
MT
ID,
which
depend
on
the
member
links
in
a
little
bundle.
H
I
think
we
reached
receive
some
comments
from
muralist,
and
actually
we
also
consider
the
air
to
bundle
and
can
be
generalized
to
describe
either
the
physical
member
links
or
the
virtual
member
links.
So
in
his
concern,
it
is
not
necessarily
for
it
to
be
a
little
physical
member
link
next
slide.
Please.
T
H
Summarized
some
comments
and
discussions
at
least
for
about
this
draft.
The
first
one
is:
should,
with
this
document,
be
a
standard
track
or
informational
document.
My
understand
is,
we
are
currently.
We
are
open
to
this
kind
of
question,
because
we,
firstly,
we
want
to
figure
out
whether
we
should
specify
something
which
are
not
existing
the
existing
RFC's
or
we
won't
whether
we
need
some
more
extensions
to
make
this
work.
The
first
day
is
about
how
to
carry
that
attributes,
except
for
topology
level.
H
H
Then,
if
these
they
are
allowed,
another
question
comes
like
I
just
mentioned
in
the
previous
slides.
We
need
to
further
specify
how
to
advertise
the
topology
specifically
attributes
one,
especially
when
one
link
participates
me.
Multiple
colleges
is
something
maybe
not
exist
in
the
existing
RFC's
and
maybe
for
some
specific
T
attributes.
We
need
to
specify
how
to
split
it
for
different
topologies
well,
for
some
other
key
attributes.
Maybe
it
should
be
a
for.
The
physical
link
should
be
shared
or
reused
by
the
pathology
attributes.
H
H
No,
in
this
case,
we
a
we
think
at
the
layers
for
a
parent
link
of
the
alert
bound
bundles
it.
You
need
to
participate
in
multiple
market.
Apology
want
to
specify
which
subset
of
the
T
attributes
on
this
link
is
reserved
or
allocated
to
a
particular
topology
obediently
we
need
we
can.
Maybe
we
can
reuse
for
generalize
the
later
bounded
mechanism
for
this,
and
the
extensions
are
specified
in
the
next
slides.
H
Okay,
this
one
similar
to
the
multi
topology
based
mechanism.
We
also
proposed
found
the
flux
agro
based
mechanism.
This
is
about
how
to
use
flux,
Argo
and
some
small
extensions
to
learn
a
bundle
to
advertise
Davidian
attributes,
in
this
case
the
flux
algo
identifier,
used
to
reused
as
a
controller
ID
at
the
BTN
and
flux
Argo
is
used
to
defile
private
apology
constraints.
H
Igps
our
map
for
flux.
Our
goal
can
be
used.
Algorithm,
specific
prefixes
are
SRA,
sticks,
locator
of
seeds.
The
other
extension
into
in
this
draft
is
we
need
to
extend
a
little
bundle
to
advertise
the
T
attributes
associated
with
space
each
flash
our
weekend,
so
we
need
to
define
a
new
flag
in
the
little
bundle
attribute,
which
is
the
Cordoba
flag.
H
When
it
is
sad,
it
indicates
that
the
member
links
of
this
little
bundle
of
virtual
links
are
not
a
little
player
to
physical
members.
We
can
generalize
the
little
bundle
for
the
key
attributes
advertisement
and
also
the
association
to
personate
a
specific
word
for
a
physical
member,
linked
with
a
flux
algo.
H
So,
in
the
previous
two
proposals,
the
magnesium
can
provide
a
simplified
solution.
While
there
are
some
constraints,
like
the
numbers
of
circles
or
number
of
the
apologies
are
limited
and
also
if
we
want
to
achieve
more
flexible
combination
of
this
different
I
use
to
build
a
underlay
I
think
we
need
a
more
flexible
and
scalable
solution,
so
is
proposed
in
this
era.
Star
I
sorry
has
we
been
dropped
so
busy
leading
this
draft.
H
Particular
network
nodes
or
links
in
this
traffic
also
try
to
decompose
advertisement
and
processing
of
different
attributes
so
that
for
different
attributes,
we
can
reuse
the
protocol
existing
protocol
mechanisms
if
possible,
and
it
can
also
reduce
the
overhead.
We
can
reuse
some
information
when
this
components
is
shared
by
multiple
regions,
so
that
were
having
advertisement
and
in
computation
can
be
reduced.
H
This
is
the
definition
of
the
BTN.
Basically,
we
need
introduce
a
new
sub
trv
in
the
ISIS
router
capabilities,
trv
to
advertise
the
relationship
between
the
VT
ID
and
the
topology
ID
and
other
optional
attributes.
So
the
ID
is
of
6
16
bits.
Identifier
and
the
algorithm
is
a
8
bit
identifier
to
identify
as
a
normal
algorithm
or
the
flux
algorithm.
We
can
also
introduce
the
sub
zero
waste
in
future.
If
immediate
slide.
H
That
apologized
abused
advertisement,
we
can
reuse
the
multi
topology
of
Lasogga,
because
both
are
can
be
referenced
in
the
definition
of
the
BTN
and,
as
mentioned
before,
both
magazines
can
either
highest
topology
from
the
topology
specific
seeds
or
locators,
and
the
MTR
can
also
spot
to
advertise
topology
specific
attributes
of
this
kid.
Two
mechanisms
can
be
considered
as
options
for
the
topology
information.
As
advertisement,
we
may
also
can
see
there
to
combine
the
multi
topology
with
some
algorithms
when
he's
needed,
I'll
consider
a
useful
here.
H
H
For
the
resource
attribute
advertisement
here,
similar
to
the
flag
or
draft
brussels
extension
dropped,
we
extended
the
arrow
bound
o
TR,
we
to
add
word
highest
attributes
the
attributes
of
either
a
virtual
member
link
or
the
physical
member
link.
In
addition,
we
introduced
a
new
within
ID
sub
theory
to
describe
the
mapping
between
the
vidiians
and
the
member
links.
Basically,
what
membrane
can
be
associated
with
one
or
multiple
with
here's,
so
this
chapter
we
will
be
carried
in
the
little
bound.
A
member
attribution
theory
in
this
slide.
H
H
And
here
we
also
mentioned
another
deplane
mechanism,
which
is
we
can
infer
consider
to
introduce
a
dedicated,
a
vision
ID
in
the
data
plane.
So
that's
the
data
plane
with
the
ID
can
be
the
same
as
the
VT
ID
we
carried
in
the
control
plane
protocols.
Here
we
have
a
optional
encapsulation
defined
in
the
drops.
Imagine
here
just
we
to
use
the
extension
headers
in
the
ipv6
to
carry
the
vision,
ID
field,
okay,
next
line,
okay.
H
H
They
say
the
combination
of
the
existing
theories
and
we
try
to
reuse
existing
encodings
as
much
as
possible.
Some
additional
specification
or
extensions
may
be
needed
to
fulfill
the
required
features.
In
the
last.
We
also
proposed
a
flash
parcel
in
the
scalable
solution,
which
is
to
meet
some
specific
requirements,
but
the
protocol
extensions
will
be
more
than
the
previous
two
solutions
in
either
the
control
plane
or
the
data
plane.
H
A
U
Can
you
hear
me
hi?
This
is
stuff
from
Joe
from
Juniper
Networks
I
have
a
question
on
the
meeting
ID
and
the
multi
topology
ID.
What
is
the
is
there
a
direct
relationship
between
the
two
or?
Are
they
the
same
or
they
that's
question
number
one
and
question
number
two
is
about
the
VTN
idea
again.
Is
it
global
in
the
network?
Is
it
inter
domain
mobile.
H
Okay,
okay
for
the
first
question,
I
think
14-4
has
to
really
have
three
proposed
proposals
here
for
the
multi
topology
based
Mac
isn't
meeting
ID
is
basically
to
reuse
the
existing
multi
topology
ID
as
an
identifier
in
the
control
plane.
So
there's
there's
one-to-one
mapping
between
the
VTA
and
at
the
multi
topology
for
the
flats
I'll
go
it
is
similar
or
we
try
to
reuse
the
flux
of
our
ID
to
identify
the
VTN
in
the
controlling
so
that
it
also
uses
one-to-one
mapping
for
the
third
mechanism.
H
We
allow
end
to
a
mapping
between
the
between
ID
and
the
multi
topology
or
the
flux
ID,
so
that
this
is
a
more
flexible
and
we
don't
need
to
be
constrained.
The
way
that
the
space
of
the
flat
ID
or
the
municipal
ID,
the
r18
ID
can
be,
as
we
defined
here
is,
can
has
a
larger
space.
This
is
for
the
question
one
for
the
question
to
the
BT.
90
is
a
global
identifier.
H
U
H
Yeah
yeah,
if
you
used
like
the
mountain
topology
based
on
magnetism,
there's
some
constraints
like
the
month
one
match
and
also,
if
you
use
it
for
inter
domain,
it
requires
the
model.
Topology
ID
in
different
domains
are
the
same
some
constraints,
but
with
a
solution.
This
is
more
flexible
on
this
part.
Q
So
first,
observations
of
this
El
Cubano
thing
is
kind
of
you
know.
Orange
stripe
herring
here
by
all
means
is
running
of
el
freeze.
Okay,
rather
they're
trying
to
like
make
itself
to
look
like
el
freeze.
The
real
problem,
I
think
is
semantics,
and
the
comment
is
I
mean
I
attribute
that
to
less
but
I'm
thinking
precisely
the
same
thing,
it
is
not
clear
what
is
the
semantics
if
you
have
multiple
topologies
or
whatever
Flex
out,
which
are
kind
of
harder
to
figure
out
because
they
just
wander
around,
but
no
Multi
topology
is
very
static.
Q
Like
I
know,
the
link
is
in
this.
Topology
doesn't
go
away.
What
does
it
mean
when
you
advertise
over
multiple
topologies
traffic
engineering
attributes
right?
What
is
it?
Each
of
them
gets
a
slice
they
kind
of
influence
each
other.
You
know
raise
this
stuff,
you
have
to
define
the
semantics
was
or
what
the
T
attribute
means
in
case
of
market
apology,
because
of
the
simple
consideration,
and
then
we
can
take
all
the
arguments.
Father
from
there
and
again
I
think
the
l2
is
a
red
herring.
You
just
try
to
make
l2
l3
me.
Q
A
A
We
have
a
couple
different
approaches
to
address
that
issue
and
we
have
it
in
a
separate
slot,
well
one
because
we
run
out
of
time
in
this
one,
but
also
to
have
a
sort
of
open
discussion
that,
maybe
you
know,
might
be
different
from
the
mailing
list,
and
maybe
we
can
make
some
more
progress
there
between
the
two
drafts
and
see
if
we
can
come
up
with
something
we
more
get
rough
consensus,
basically
on
what
a
good
way
forward
is.
I
was
hoping
that
the
ICC
RG
was
going
to
meet
on
the
13th
yeah.
A
A
A
So
anyway,
I
was
gonna
say
we
could
guys
homework
as
a
working
group.
We
could
go
to
the
ICC
RG
meeting,
which
is
the
you
know:
congestion
control,
a
research
group
but
I
think
they're.
Not
there
doesn't
seem
like
there
have
an
engine
going,
but
my
point
in
that
was
that
the
stuff
can
be
kind
of
hard
great
to
get
right,
at
least
when
you
want
to
do
it
dynamically,
but
yeah
anyway.
So
we
have
that
meeting
coming
up.
It
should
be
interesting
and
then
AC,
please
sign
the
blue
sheets.
A
Q
We
talked
about
these
resource
stuff,
I.
Think
there's
even
a
more
profound
discussion
here.
Once
you
go
to
the
level
of
complexity
like
this
VPN
class,
which
does
you
know
multi-dimensional
resource
management
across
lots
of
abstraction
easily,
even
one
is
to
stuff
all
the
stuff
into
the
IGP.
Isn't
then
really
a
role
for
something
like
a
provisioning
system?
You
know
in
terms
of
a
controller,
something
centralized
that
counts
all
these
resources.
We
may
be
ill
served
to
try
to
you
know
to
start
to
do
multi-dimensional
advertisement
of
all
these
resource
management.
One
can
argue
right.