►
From YouTube: IETF103-LSR-20181106-1120
Description
LSR meeting session at IETF103
2018/11/06 1120
https://datatracker.ietf.org/meeting/103/proceedings/
A
A
D
E
F
D
Okay,
so
one
of
the
things
that's
happening
is
that
network
architectures
are
changing
and
lots
of
people
are
starting
to
implement
routers
using
lots
of
littler
routers.
We
started
off
using
multi
chassis
routers
a
long
time
ago,
and
those
multi
chassis
routers
are
basically
lots
of
little
forwarding
engines
wrapped
around
a
network
fabric.
D
What
we're
seeing
is
that
people
are
moving
towards
a
better
topology
where
all
those
little
boxes
actually
are
independent
systems
and
the
fabric
is
actually
becoming
the
inner
the
data
center
fabric.
Now,
what
that
does
do
is
that
blows
out
the
system
level
abstraction
that
we
have
with
a
multi
chassis
router,
leaving
us
with
lots
more
routers
that
we
have
to
manage
we'd
like
to
restore
some
level
of
abstraction
so
that
we
don't
have
too
many
entities
hanging
around
and
in
particularly
in
the
case
of
the
iGPS.
D
D
Levels
three
through
eight
are
available.
The
folks
who
designed
is
is
were
somewhat
prescient
and
they
very
nicely
reserved
more
bits
in
the
circuit
ID
for
us
and
did
not
use
them
right
now.
There
are
only
three
levels:
three
values
that
are
defined
bit.
One
is
level
1
bit.
2
is
level
2.
You
can
turn
on
both
if
you're
a
level
one.
Oh,
it's
l1
l2,
but
there's
six
reserved
bits
for
setting
there
I'm
that
we
turn
those
on
and
make
those
levels
three
through.
Eight.
D
D
The
other
things
that
we
need
to
do
to
make
this
happen.
We
need
a
new
hello
PDU
right
now
we
have
an
l1
iah.
You
have
an
l2
iih.
We
could
go
down
the
path
of
creating
l3
to
LA,
I
H
s,
but
I.
Don't
think
that's
strictly
necessary.
I
think
if
we
just
create
a
new
one,
because
we
don't
want
to
have
interoperability
backwards,
interoperable
interoperability
issues,
but
we
could
create
one
new
one
that
allows
you
to
advertise
level.
3,
3
level,
8
IAH
at
one
time,
I
think
that's
possible.
D
H
G
G
Ok,
it's
supposed
to
be
you
know,
1
or
3
I
mean
we
found
use
cases
for
let's
strictly
so
I.
Don't
know
these
are
all
kind
of
knits.
But
to
me
it's
more
about
defining
how
this
should
be
used.
I
think
we
have
to
think
about
the
deployment
cases
harder.
Like
answer
I'd
like
to
see
more
discussion
about
that,
ok.
E
G
G
G
G
I
H
D
J
A
K
Assess
your
brawn
so
tiny,
the
the
I
think
when
you
started,
you
said
that
you
wanted
to
use
one
to
be
a
sort
of
confederation
that
became
a
became
a
Rooter
and
right.
So
it
would.
There
would
be
some
sense
in
having
one
and
n
as
the
discontinuous
pair.
If
you
were
building,
say
a
level
four
Rooter,
then
its
subcomponents
would
internally
be
in
level
in
their
own
level.
One
so
I
could
see
cases
where
you
might
want
to.
K
K
To
me
seem
to
be
the
original
sighted
one,
which
was
one
becomes
your
inter
the
internal
components
of
the
Rooter
and
it
sits
in
a
higher-order
level.
So
so,
if
using
one
in
the
classic
sense
that
it's
a
black
we
using
today,
then
it
doesn't
make
sense.
But
this
sighted
case
was
that
we
were
building
riches
a
different
way
and
I
think
there
is
a
case,
then,
for
one
being
special.
G
J
D
D
The
way
is
is:
does
levels
is
not
very
pretty
if
you
have
a
level
one
area
and
you
want
to
transit
a
level
one
area,
you
have
to
take
those
transit
circuits
and
push
them
into
level
two,
and
they
show
up
in
the
level
two
link
state
database,
and
if
your
data
center
is
going
to
almost
all
be
used
for
transit,
you
effectively
push
everything
into
level.
Two,
that's
a
scale
problem.
D
D
D
D
We
want
all
of
the
our
area
border
routers
now
to
tunnel
back
to
the
area
leader
and
use
that
as
a
l2,
sir
virtual
circuit
and
that's
going
to
give
us
l2
flooding
within
the
area
and
now
after
that
happens,
our
area
leader
now
has
full
ideas
about
all
of
the
level
2
connectivity
for
this
level.
One
area.
H
D
D
Not
we're
not
using
the
tunnel
to
the
area
leader,
we're
doing
it
a
direct
tunnel
from
entry
to
exit
okay
and
that
can
be
sr
or
GRE
I.
Don't
care,
okay
back
to
advertising
right
information,
okay,
so
the
area
leader
now
has
an
l2
database
that
contains
all
the
l2
information
from
all
of
his
area,
border
routers.
That
gives
us
the
identities
of
all
of
the
adjacency
on
the
air
for
the
area
and
he
can
cobble
all
of
those
up
and
list
those
as
adjacent
to
the
pseudo
note.
D
D
D
Okay,
we
actually
already
talked
about
the
forwarding
problem.
All
of
our
area,
border
routers,
are
full-fledged
l2
we've
got,
connectivity
are
flooding,
connectivity
is
to
the
area
leader,
but
that's
not
where
we
want
to
forward.
So
we
actually
want
to
compute
shortcuts
from
entry
to
exit
and
srr
GRE
I.
Don't
really
care
whatever
folks
feel
like.
Is
you
useful
for
tunneling
today.
D
D
J
Just
I
just
had
one
comment
from
OSPF.
My
experience
with
virtual
links
has
been
that
and
I
realized
this
exactly
of
difference.
You've
solved
the
problems
different
ways,
but
I
always
thought
if
we
ever
did
OSPF
before
I'd
get
rid
of
them
because
of
the
added
complexity.
So
we'd
have
to
really
think
we
wanted
this,
and
we
couldn't,
just
by
you
know,
share
you
can
just
share
links
like
we
are
doing
today
having
a
circuit
be
and
multiples
you
to
handle
the
use
case
of
transit.
D
B
Read
the
draft
I'm
sorry
but
I'm
trying
to
understand
this,
so
you
want
to
be
able
to
transcend
it
changes
a
lot
to
know
the
Ciena.
You
want
to
be
able
to
transit
the
area
over
level
area
and
then
how
do
you
set
the
metrics
for
the
pseudo
node
to
the
rest
of
it
right?
You
want
to
be
able
to
get
in
here
and
get
out
there
there's
a
different
metric
than
get
in
here
and
get
out
somewhere
else
from
the
area.
B
How
do
you
set
this
up
and
if
you
are
not
careful,
if
you
are
inside
in
the
middle
of
the
area
somewhere
and
you
want
to
bypass
this
area
and
if
you're
not
careful
setting
these
metrics,
can
you
ever
go
back
because
you
are
doing
hop-by-hop?
Let's
assume
it's
an
MPLS
hop
by
hop
forwarding
in
the
middle
of
the
lower-level
area.
B
D
H
L
D
D
So
I
don't
see
that
there's
a
problem
with
MT
traffic
engineering.
Obviously,
since
you
can't
see
inside
the
area
it's
abstracted
away
from
you
yeah,
you
can
obviously
do
tooi
within
the
area.
You
can
even
do
te
for
the
transit
traffic
at
the
area
entry,
but
external
traffic
coming
in
cannot
control
its
path
through
the
area.
Again,
this
is
an
abstraction
any.
M
N
Chris
Martin
arista
I
just
wanted
to
maybe
help
clarify
if
you
think
about
what
this
area
I
think
this
goes
to
your
point.
Ac
we're
asking
about
a
use
case.
Imagine
a
you
know,
because
its
own,
though,
and
leaf
spine
router
like
a
multi
chassis
router,
it's
mostly
folded,
Clos
or
clothes
apology.
So
the
leaves
in
this
case
the
hierarchies
come
inverted.
Typically,
the
top
of
the
hierarchy
would
be
at
the
root
you
know,
and
then
you
would
have
you
know
in
a
multi
area
case.
N
The
route
routers
would
typically
be
the
ABR's,
but
in
this
case
the
leaves
are
more
likely
to
be
the
inter
area
connection
points,
and
so
you
need
the
you
know
in
terms
of
abstraction
you're
going
to
be
coming
at.
It
would
typically
be
like
the
level
one
router
itself
and
therefore
you
have
to
make
the
whole
area
level
to
aware,
and
so
this
abstraction
kind
of
hides
that
going
to
the
point
about
you
know:
traffic
engineering
within
the
area.
Imagine
a
complete
graph.
N
That's
got
full
by
sectional
bandwidth
or
bipartite
graph
for
that
matter,
with
full
bisexual
day
with
you,
don't
care
about
the
topology.
Just
like.
Typically
you
don't
care
about.
You
know.
You
know
we
don't
talk
about
traffic
engineering
between
line
cards
and
routers,
so
the
thinking
would
be
released
initially
that
it's
just
a
complete
graph
or
a
bipartite
graph
with
full
by
sectional
bandwidth.
So
to
your
point,
Jeff,
you
can
just
hide
that
information.
I,
don't
know
if
that
helps
clarify
the
use
case.
Maybe
but.
I
Tony
P
again
so
yeah
I
see
first
layers
of
the
onions
being
peeled.
So
we
took
it,
of
course,
all
the
way
in
PN
I,
and
if
you
do
that
stuff,
choosing
the
ingress
into
the
area
is
something
you
simply
can't
resolve.
You
may
end
up
on
a
bad
ingress,
so
the
only
way
we
could
deal
to
start
with
the
pinna
and
I
were
cranking
back.
What
you
cannot
do
here,
of
course
right.
I
So,
whatever
you
try,
even
if
you
try
to
advertise
some
metrics,
you
summarize
the
stuff
and
you
you
end
up
advertising
this
full
mesh
with
some
approximation
with
sometimes
am
subtitles
right.
It's
inherent
limitation
of
abstraction
abstraction
is
controlled
lying
and
there
is
something
you
you
simply
cannot
change,
correct,
Chris.
N
D
D
J
Wish
I
was
just
gonna,
say:
I'd,
encourage
everybody
to
read
these.
This
is
these
are
definitely
interesting
proposals
and
we
really
haven't
you.
You
know
we
kind
or
the
base
mechanisms
of
the
protocol.
We
kind
of
haven't
really
extended
them
too
much.
You
know
we
had
always
be
8
feet.
3
extended,
L
essays.
That
was
a
big
step
and
then
we
had
some
experimental
things
like
TTC,
but
we
really
haven't
had
so.
This
is
this
is
this
is
definitely
interesting.
J
Ok,
Yuma,
hey!
Go
ahead,
come
on
up
I'd
like
you
to
say
which
I
know
you've
had
this
estándares
track,
I
read
this
and
I
was
saying:
okay,
what's
what's
Tanner
track,
you
can
tell
me
what
you're
standardizing
and
what
you're
just
recommending
out
of
this
or
whether
it
should
be
just
informational.
It's.
E
E
So
we
know
this
multi
topology
for
last
ten
years,
maybe
so
I
implemented
multiple
gene
in
my
earlier
life,
so
but
what's
happening
new
what's
new
here
is
people
are
seeking
to
use
ipv6?
Only
we
have
seen
earlier
before
you
have
e4
and
say
to
people
want
introduce
v6
into
the
network.
Now
people
want
to
use
ipv6
only
topology
they
so
the
use
cases
like
you
know.
People
are
looking
at
III
talk
to
multiple
people,
those
are
in
the
mobile
backhaul
and
yell
3l
3dc
underlays.
E
So,
what's
the
what's
the
issue
and
what's
the
goal
of
this
document
goal
of
this
document
is
to
lay
out
the
nuances
around
the
SS
ipv6.
So
there
are
multiple
ways
you
can
do
this
ipv6
introduction
based
on
your
requirements,
because
we
have
two
ways
to
do
this.
Essentially
so,
but
there
are
caveats
in
each
each
thing
as
I'll
just
explain
that
so,
if
you
are
transitioning
from
ipv4
to
v6,
there's
certainly
there's
no
question
about
it.
E
That's
the
main
point
of
this,
but
if
you
confuse
on
this,
you
have
a
lot
of
problems
so,
but
my
goal
of
this
document
is
to
lay
out
these
various
options
we
have
and
if
any
folks
you
are
trying
to
do
this
and
you
have
any
problems
and
if
you
have
any
inputs,
you
can
feel
free
to
contact
me
next
slide
and
so
I
just
go
back
and
just
to
two
options
we
have
is,
we
all
know
RFC
pi/3
0.
At
the
moment
you
say:
ipv6
is
s.
E
People
think
about
RFC,
pi,
3,
0,
8
I
see
some
of
the
tarts,
for
example,
DC
and
or
a
draft
when
about
talking
about
ipv6.
They
mentioned
this
draft,
but
I'm
not
clear
what
they
want
to
serve
both
ipv4
or
v6.
This
case
is
a
single
topology
mode,
so
you
compute
one
SPF
and
one
decision
process
you
don't
you
end
up
with
both
v4
and
v6.
That
is
a
key
point
here
and
the
second
thing
is
RFC
Phi
1
to
0
it
is
a
multi
topology.
E
E
So
so
the
if
you,
if
you
are
seeking
to
use
five
3:08,
so
you
have
to
think
about
network
congruence.
If
you
are
only
ipv6,
you
are
now
going
to
touch
before
it's
fine.
You
go
go
ahead,
go
it
with
Phi
Phi,
zero,
eight,
no
issues,
but
if
you
are
transitioning,
you
have
to
be
very
careful.
One
simple
example:
I
showed
here
is
you
have
a
network
with
all
links,
v4
v6,
a
except
one
link
which
is
v4
between
rx
to
RT.
You
have
it
since
he
formed
so
packet
will
be.
E
The
shortest
path
tree
will
go
to
r2
using
that
link,
but
unfortunately
there
is
no
v6.
You
get
traffic
black
hole
there.
Despite
there
is
an
alternate
path:
RX,
r,
one
r
yr,
that's
not
going
to
be
used
because
the
shortest
path
it
is
rx
2,
R
2.
So
this
is
the
this
is
the
black
hole
issue
and
that
the
ways
to
fix
this
is
only
two
ways.
One
is
either
you
enable
v6
there
are
removed
to
empty
because
debugging
this
kind
of
problems
extremely
difficult
in
the
big
networks.
E
So
the
main
source
of
confusion
here
is
it's
a
topology.
Our
address
family,
that's
the
main
source
of
confusion.
The
terminology
is
used,
is
very
confusing.
Mt
ID
2
is
called
reservoir
size
topology.
Maybe
it's
a
renters
family
I,
don't
know
so,
but
but
say.
For
example,
when
you
want
ipv6,
only
it's
better
safe
to
use,
I
use,
MT
ID
to
multi
topology,
then
people
say
I,
don't
want
multi
topology
I
want
single
topology
I
want
ipv6
only.
E
E
E
So
I
don't
know
I.
If
we
can
do
anything
protocol
wise,
we
can
do
anything
to
mitigate
this
confusion,
but
I
thought
to
put
this
a
document
so
that
that
conversation
can
be
easier.
So
people
suggested
me
to
put
in
the
v6
ops
because
a
lot
of
v6
only
stuff
is
happening.
There.
There's
issues
there
how
to
deal
with
it
there's
an
operational
issue.
So
I
just
seek
your
suggestions.
If
you
have
any
comments
or
solutions.
G
Work
this
you
this
is
this
is
this,
is
a
well-known
limitation
of
the
protocol.
Goes
all
the
way
back
to
RFC
1195.
Okay,
there's
nothing
new!
Here!
You
know.
If
you
look
at
Mt
you'll
recognize
that
there
is
no
limitation
on
an
arbitrary
topology.
You
can
support
whatever
set
of
address
families
you
want.
G
Saying
a
given
topology
can
be
used
for
as
many
address
families
as
you
want
what
we
did
in
50,
50
120,
or
what
Tony
did
rather
his.
He
said
these.
These
particular
values
are
reserved
for
specific
use
cases.
It
places
no
restrictions
if
I
want
to
use
mt
ID
10,
and
I
want
to
run
v4
and
v6
with
some
constraints
or
what-have-you.
It's
perfectly
legal
to
do
so.
I'm.
G
I
can
I
could
I'm
willing
to
from
a
deployment
standpoint.
People
don't
fully
understand
this
always
I'm.
Just
saying
this
was.
It
was
clearly
stated
in
1195.
There
have
been
proposals
in
the
past.
Philip
christian
wrote
a
draft
many
years
ago
to
support
tunnelling
between
different
address
families
never
got
a
lot
of
support,
but
this
has
been
talked
about
a
lot.
Okay,
thank
you.
H
J
J
J
H
A
I
Tony
pj
/,
yeah,
Mia
culpa
Zhou
was
deadband
against
putting
reserved
topology
values
right
and
it
was
the
sales
pitch
and
operational
issue
right.
If
everybody
has
to
choose
the
same
topology
it's
harder
than
he
was
say
in
the
draft.
You
know
this
is
the
use
case.
We
should
have
probably
call
it
the
v6
transition
topology
or
something
or
maybe
have
put
something
in
the
draft,
which
seemed
utterly
obvious
to
us
that
the
topology
concept
is
completely
orthogonal
to
address
family
concept.
E
I
M
You
to
reiterate
and
umas
points,
this
draft
is
not
meant
for
implementers
of
SS,
it
meant
for
consumers
and
we
spent
quite
some
time
with
on
my
building.
This
is
another
company,
there's
so
much
misunderstanding,
so
many
errors.
So
it's
really
to
explain
people
what
it
is,
what
they
should
do
and
I
believe
there's
a
lot
of
misunderstanding
outside
of
this
room:
how
to
use
it.
This
is
intense,
pure
informational.
A
I
A
J
G
So
I
just
make
the
I'm
sympathetic
to
the
deployment
okay,
but
people
can
get
confused
and
it
may
well
be
value
in
writing
deployment
guide
sort
of
thing,
but
this
problem
is
very
clearly
discussed
in
RFC
1195,
okay
from
1992
whenever
it
was
published.
Okay
would
also
make
the
comment
that
well,
if
we
had
thought
about
maltese
apology
back
in
1988,
maybe
the
first
version
of
ISS
would
have
had
it.
We
wouldn't
have
had
two
different
TLV
types
mm-hmm,
but
of
course
you
know
we
didn't
anticipate
that.
H
So
Michael
Abramson
again
I
think
for
you.
You
have
to
write
different
kind
of
document
for
operational
people
who
don't
know
the
bit
in
nanosecond
of
the
protocol.
So
I
think
we
should
have
these
kind
of
documents
to
tell
you.
Okay
here
are
the
bombs
don't
step
on
them?
Okay,
don't
do
that
and
don't
do
that
because
it
doesn't
work
and
you
can't
it.
You
can't
get
that
from
just
reading
protocols,
but.
H
M
E
E
Okay,
so
what
is
this?
So?
This
is
a
new
path,
routing
mechanism
and
I
I
GP.
It
chooses
the
shortest
path
routing
it
gives
the
same
as
PF,
but
the
value
it
brings
is
it
label
stack
Direction,
the
first
value
it
brings
in
the
label.
Text
label
stack
reduction:
NS
are
the
extensions
for
discussed
in
is
applied
draft
because
draft
is
getting
bloated
up
it's
20
pages,
then
you
know
we
put
two
drafts
up
tightly
because
SPF
contents
was
per
V,
2,
V
3,
so
many
encoding.
So
we
split
it
up
two
documents
there.
E
It
covers
the
core
protocol,
extensions,
the
PPR
ID,
that's
a
forwarding
identifier
and
the
path
descriptions
of
the
LV
and
the
attributes
optional
attributes.
That's
the
key
key
for
this,
the
the
same
extension
and
also
it
also
includes
the
SPF
changes
required.
The
small
change
required
at
the
end
of
this
F.
H
E
Installing
the
forwarding
entry
to
get
the
pen
together
dip
into
the
parent
get
the
next
segment
next
hop
and
install
the
PPR
ID,
and
it
also
allows
you
to
get
the
path
concept.
This
is
free,
om,
transparent
OEM,
because
as
and
when
you
remove
the
segment
from
the
SR,
you
don't
have
the
notion
of
the
path.
So
it's
free,
you
don't
have
to
put
extra
labels.
You
don't
have
to
put
MPLS
special
labels
to
give
the
path
motion
next
slide,
please.
E
So
why
do
we
knew
this?
Yes,
so
this
I
touch
base
last
year,
last
idea,
but
again,
I'll
and
summarize
here.
The
first
point:
is
it
reduced
the
path
overhead
on
the
data
plane
you,
you
can
put
everything
on
the
data
plane,
everything
on
the
packet,
but
there
are
some
issues
with
that.
So,
for
example,
you
have
the
hardware
capabilities.
Let's
take
about
SR
the
MPLS
case.
If
you
talk
about
multiple
labels,
you
already
have
without
any
SR
path:
6
MPLS
labels
on
the
packet.
E
You
already
end
up
in
4-5
labels
how
it
is
that
you
have
VPN
label
interests,
label,
most
cases,
2
labels
and
on
top
of
it,
if
you're
smoothly
tunnel
traffic
to
dip
into
the
packet
into
inside.
You
need
entropy
label
for
and
the
path
label
eventually
so
for
filing
fortify
labels,
and
if
you
have
Broadcom
chips,
I'm.
Sorry
for
that,
you
cannot
put
more
labels
on
that,
but
maybe
in
future
new
chips
may
be
available.
It
can
do
more.
E
But
if,
if
we
put
services,
if
you
put
everything-
and
if
you
put
in
ring
topologies
everybody
every
label,
you
have
to
describe
it's
going
to
be
10
to
12
watt,
15
I,
don't
know
how
many
labels
one
needs
it
so
MSD
fixes.
Is
this
in
the
sense
like
you
know
it,
at
least
for
the
controller.
It
will
help
you
to
understand.
Okay,
don't
use
this
part
because
I
have
a
small
router
here.
It
cannot
read
these
many
labels,
so
it
can
avoid
that
path
temptation.
E
So
it
is
a
constraint
on
the
PCE,
but
it
cannot
give
the
alternate
path,
which
is
giving
the
similar
traffic
characteristics.
You
are
seeking
to
start
with
it
also,
once
you
reduce
a
packet
overhead,
you
obviously
get
the
line
rate.
There
are
line
rate
issues,
otherwise
some
hardware's
cannot
do
line
rate.
If
you
put
large
overhead
it
in
the
packet
and
for
a
service
6
case,
it
avoids
fragmentation.
It's
a
serious
problem
and
you
hit
this
fragmentation
issue.
E
It
should
be
seriously
will
be
in
trouble
and
the
last
thing
for
the
lot
of
Phi
G
use
cases,
em
IOT
and
you
are
ll
see
packet
sizes
are
small,
it's
40,
bytes
30
to
50
bytes
packets.
You
cannot
put
a
header
tax
of
200,
vice
100
bytes,
for
the
packets.
For
that
environment.
It
is
sensitive
for
some
environments,
it's
not
sensitive.
Yes,
if
it
is
environment,
it
sends
to
then
too
much
packet.
Overhead
is
a
problem,
so
it
also
simplifies
so
a.m.
it
extends
the
data
planes.
E
You
can,
you
add
more
data
planes
and
also
it
unique
fast
gyro.
It
can
be
done
with
this
next
slide.
Please.
So
a
lot
of
people
asked
for
the
how
it
is
backward
compatible
or
how
it
works
with
SR
I
says
it's
complementary.
It's
it's
a
complementary
control
in
work.
It
works
great
with
a
sir.
It's
an
optional
feature.
E
If
you
want
to
use
for
SR,
MPLS
and
SH,
you
can
reduce
the
stack
size
and
you
don't
have
to
use
it
and
if
you
have
to
do
network
programmability
and
if
you
want
to
do
a
extension
it
a
lot
of
stuff
there,
you
can
go
ahead
and
do
it,
but
if
you
are
using
only
for
the
path
information,
if
you
want
to
compress
that
you
can
use
this
optional
feature,
it
works
there
fully.
Backward-Compatible
next
slide,
please.
So
there
was
a
question
on
the
scalability.
E
There
will
be
millions
of
paths,
see
how
thousand
no
network
thousand
links.
You
have
four
million
part
and
paths.
So,
theoretically,
yes,
there
will
be
millions
of
flows,
not
millions
of
paths.
There
will
be
a
few
thousand
parts
in
some
in
some
environment
even
lesser,
so
it
is
simply
scale
is
PPR
path.
Road
is
a
scale
number
of
routes,
a
number
of
routes
into
the
IGP,
so
you
don't
have
to
scale.
The
answer
is
yes,
you
don't
have
to
scale.
Yes,
there
are
possibly
theoretical
possible,
but
you
don't
have
to
skin
that
much.
E
E
So
basically,
what
we
are
doing
is
in
this
instead
of
a
linear
path.
We
call
this
as
a
branch,
so
we
introduce
two
bits:
one
is
source
and
destination.
So,
for
example,
PDE
one
is
a
source
ingress
v.
Pd
v
is
digress
PA.
So
if
you
are
crafting
a
path
PD
on
PD
two,
you
describe
the
path,
but
if
you
just
put
add
two
bits
like,
for
example,
PDE
one
is
a
source
in
the
figure
PD
three
is
a
source
two
and
the
Green
was
the
destinations.
E
So
four
paths
can
be
compressed
into
one,
for
example
PD
one
from
source.
If
for
each
source,
expanded
to
each
destination,
so
every
destination
should
contain
a
forwarding
entry,
so
the
forwarding
entry
can
be
established
so
simple
with
simple
bits:
you
can
extend
it
to.
You
can
compress
it
and
if
you
don't
want
to
mean
a
scale,
you
don't
have
to
use
this,
but
you
know
it's
a
simple
feature:
it
can
use
it
and
also
you
can.
The
same
concept
can
be
extended
to
build
the
graph,
and
this
will
even
further
compresses.
E
It
is
that
the
details
are
in
the
draft,
so
the
use
cases
we
are
looking
at
is
actually
low,
latency
use
cases
so
low
latent
low
and
deterministic
latency
use
cases
lot
of
use
cases
the
verticals
these
use
cases
and
requirements
are
covered
in
the
HC
working
group
and
GP
working
group.
This
was
sponsored
by
water,
phone,
BT
and
BASF
folks,
and
this
has
a
lot
of
use
cases
white
is
required
and
we
it
is
applicable
why
the
stack
size
is
a
problem.
Next
slide.
E
Please,
and
there
were
questions
on
what
is
the
relation
with
the
Phi
G
DM
M
working
group
work
and
this
work,
so
I
presented
the
first
draft
DM
empty,
a
transport
aware,
mobility
in
ietf,
one
or
two
in
DM
and
working
group.
So
what
it
lays
out
is
in
the
Phi
G
how
mobility
can
be
achieved
with
the
awareness
of
the
transport
network
in
the
Phi
G
lot
of
small
says:
new
mobility
scenarios
are
coming
up
and
with
the
new
online
interface.
E
So
when
mobility
sessions
are
taking
care
taken,
you
need
to
account
for
the
transport
path
characteristics
too.
If
the
path
is
available
or
not
so
that
is
one
piece
and
also
there
are
other
other
work
going
on
in
the
DMM
to
remove
gtp
in
the
mobile
core.
How
SR,
v6
and
programmability
can
help
there.
That's
a
separate
work
completely
orthogonal.
This
proposal
assume
say
overlay
and
it's
an
underling
mechanism
and
it
doesn't
doesn't
try
to
remove
gtp
or
something
like
that.
E
So
so
in
comparison
test
service,
it's
a
different
work
and
it
can
work
with
SR
v6
and
the
gtp
replacement.
There
are
a
bunch
of
options,
they're
being
discussed
in
DMM,
including
lists
ila
s
or
v6,
there's
so
many
proposals
for
all
these
proposals.
If
you
use
PPR
it
can
hit,
you
reduce
the
stack
size.
That's.
E
L
So
I
don't
know
where
to
start
I
think
you
you
listed
problems
that
you're
trying
to
solve
are
not
the
problems.
So
the
problem
that
you
listed
your
line
by
line
are
not
the
problems,
so
I
think
you're.
Looking
to
retrofit
your
solution
into
something
that
doesn't
exist.
Okay,
the
problem
that
isn't
it
you're
looking
for
a
problem
to
fit
your
solution
and
I.
Think
it's
a
reverse
engineering.
Can
you
go
back
to
your
slide
number
three.
Please.
L
E
L
Is
the
debit
discussion,
but
it
doesn't
it
doesn't
matter
if
the
accounting
remains
good
and
and
I
can
prove
you,
this
okay,
Oh,
nm
you're,
introducing
additional
states
in
the
network
or
additional
program
in
additional
rewrites
in
the
network,
or
whenever
is
going
to
be
more
complex,
so
you
you
think,
is
simplifying
it,
not
simplifying
it's
adding
more!
You
you!
Whenever
you
bring
something
you
keep
saying
simple,
simple,
simple,
but
by
the
time
you're
done
with
it
is
I.
E
E
A
B
Okay,
hello,
everyone
I'm
going
to
present
this
OSPF
extension
for
the
prefix
originator.
Actually
in
okay.
Actually,
my
last
ITM
meeting,
we
present
another
draft
about
the
inter
area,
topology
retrieval
and
in
that
draft
the
sells
a
prefix
sells.
Router
IP
is
considered
as
a
general
useful
thing.
So
this
time
we
modify
the
draft.
Add
more
use
cases
about
this
ID-
and
here
this
is
a
new
draft.
B
Okay,
there's
also
other
possible
use
cases,
which
is.
We
can
also
consider
to
use
this
mechanism
to
collapse
the
inter
area
topology
from
the
Sdn
controller.
This
is
because,
in
some
situations
the
controller
may
only
have
the
PGP
RS
sessions
with
only
one
or
two
routers
in
the
backbone
area.
In
this
case,
it
means
mechanism
to
collect
an
arm
retrieve
out
a
topology
of
the
other
areas
in
a
GP
domain.
So
you
see
kiss-kiss.
This
theory
can
also
help,
and
this
will
has
some
limitations,
and
this
has
been
described
in
the
appendix.
B
Okay,
here
are
the
TLB
format,
which
is
very
straightforward.
We
need
a
new
sub
heavy
to
carry
the
prefix
thousand
router
ID
and
for
OS
version.
Let's
be
tre
will
be
carried
in
the
hospital
version,
extended
prefix,
opaque
RSA
and
for
SPF
version
three.
It
will
be
carried
in
the
II
inter
area,
prefix
RC,
very
straightforward.
Oh
here's,
a
procedure
of
the
operation.
B
Basically,
when
the
ABR
received
the
router
IOC
in
one
in
one
area,
it
will
need
to
generate
the
corresponding
some
races
and
with
this
news
subtly
added
to
the
extended
our
sales
and
then
it
will
add,
word
hi.
This
summary
carry
to
the
other
areas
in
the
on
the
receiver
side,
the
routers
in
other
areas,
who
you
need
to
obtain
this
into
a
social
I'll,
try
the
information
for
different
use
cases,
as
described
in
the
beginning
of
this
presentation,.
J
A
selenium
you
are
speaking
as
we
had
a
lot
of
offline
discussions
of
this,
and
the
changes
were
that
we
removed
the
there
were
different
encodings
and
there
were
some
non
backward
compatible
component
encoding
using
the
existing
assays.
Those
are
removed
and
we
only
kept
the
OSPF
v2
extended,
LSA
and
notice.
We
have
v3
not
extended
those.
We
have
v2
pre
linked
at
buuuut.
Let's
say
no.
J
J
The
butchers
name
soon
yeah
yeah
I,
didn't
butcher
it
alright.
Okay,
soon
I
had
a
use
case
that
he
wanted
to
use
and
it
wasn't
it
didn't
work
for
all
cases.
It
only
worked
if
you
had
all
numbered
links-
and
there
was
some
so
we
didn't
want
that
to
be
normative,
so
we
moved
it
to
an
appendix
those.
Are
the
major
changes?
Yes,.
M
Justin
Sarah
so
himself
works
here.
The
thoughts
about
it
with
regard
to
use
cases,
usually
around
abilities
are
not
meant
to
be
propagated
across
areas
they're
there
and
that's
why
we
G
pls
work
to
get
them
up
to
the
controllers
at
Bill's,
full
multi
area,
multi
level
graph.
So
specifically,
for
that
reason
we
usually
deploy
pair
of
BGP
other
speakers
per
area
level,
not
per
network,
so
I
would
really
change
the
use
case.
Yeah.
B
That
is
and
that's
why
we
pull
this
use
case
into
the
appendix
yeah.
J
Some
of
us
would
have
preferred
to
see
it
removed
after
the
long,
protracted
discussion
on
it,
but
my
compromise
was
appendix
so
we
might
be
discussing
this
one
a
little
more
on
the
list
in
if
we
can,
because
I
think
everybody
agrees,
the
tlvs
are
needed
because
therein
is
is
right
now
the
same
thing:
the
inter
area,
propagation
of
the
source,
a
router
and
I'm,
not
sure
everything
that's
used
for,
but
I
think
I.
Think
it's
a
good
idea
and
you
have
a
BGP
LS
raft
as
well
right.
There
is
a
debate.
J
J
Okay,
I
guess
we're
done
thanks.
Everybody,
sorry
for
all
the
problems
with
the
Chromebook
I
think,
there's
a
couple
things:
we've
learned
that
well
the
one
thing
the
reef
I
think
we
figured
out
that
the
reason
we
couldn't
find
those
some
of
those
presentations
was
those
are
the
ones
that
were
in
PowerPoint
and
those
do
not
show
up
on
the
agenda
page.
You
have
to
go
to
meeting
materials
and
also,
if
you
don't
convert
the
PowerPoint
PDF,
it
makes
it
very
hard
to
step
through
it
using
the
Chromebook.