►
From YouTube: IETF102-MPLS-20180717-0930
Description
MPLS meeting session at IETF102
2018/07/17 0930
https://datatracker.ietf.org/meeting/102/proceedings/
A
Confusion
of
the
place,
but
Montreal
is
the
right
place.
We're
about
to
start.
You
have
Jill
helpin
up
here,
helping
me
to
run
the
meeting
since
Nick
couldn't
be
here
this
time
he
promised
to
keep
me
on
track
and
on
time,
we'll
see
about
that.
Okay,
okay,
so
we
have
a.
We
have
the
note.
Well,
it's
very
early
in
the
week,
if
you
haven't
seen
it
before
it's
worthwhile
reading.
A
A
So,
as
I
said,
unique
was
not
able
to
attend
and
you
will
have.
You
will
happen
very
kindly
agreed
to
help
me
share
a
meeting.
There
are
audio
streaming
and
recording.
So
when
you
speak,
please
speak
into
a
mic.
It
doesn't
matter
that
the
five
people
closest
to
you
actually
hear
what
you're
saying
people
on
the
other
net
will
not
hear
you.
If
you
don't
use
the
mic.
You
also
have
the
links
to
etherpad
meet
echo
and
other
things,
and
the
I
asked
a
couple
of
people
to
do.
A
A
A
B
A
Okay,
fine
go
ahead,
so
there
are
a
few
things
that
the
two
things
I
want
to
say
on
ampulla
site.
They
are
false.
We
actually
require
authors
and
contributors
to
respond
to
the
IPR
pause
and
we
are
very,
very
reluctant
actually
go
from
that,
but
in
the
case
where
we
have
where
we
simply
can't
find
a
person.
A
But
if
we
can't
find
someone,
we
have
a
mitigation
plan.
So
it's
not
it's
not
a
blocking,
but
it
will
take
time
if
people
is
not
not
responding
and
then
I
have
a
reminder.
I
have
sent
out
the
mail
I'll
tell
people
that
I
think
the
previous
six
meeting.
If
you
have
a
working
group
document,
you
are
supposed
to
give
us
a
progress
report
on
that
document.
A
A
A
Rc
4928,
but
you
don't
need
to
comment
on
it.
We
just
know,
but
it's
not
that
we
need,
we
can
take
it
in
Lisp,
okay,
we
don't
have
any
incoming
or
outgoing
liaisons
the
doctors
yeah,
and
we
have
one
year
of
C
rc8.
If
372
MPLS
flow
identification
considerations,
we
have
Docs
documents
with
the
is
G,
that's
the
MPLS
spring
and
label.
A
Actually
IC
has
a
probe
that
and
it
should
now
be
with
the
RC
editor
and
we
have
one
new
work,
new
document
since
last
time,
actually
I
think
it
maybe
made
the
week
or
the
last
meeting.
So
that's
the
doctor
at
the
FMP
last
SFC.
That
document
tours
only
again
done.
That's
actually
the
the
coding
window
is
or
every
document
it
isn't
blue
assembly
again.
That's
so
save
questions
that
you
have
on
document
blue
until
it
comes
on
the
agenda.
Okay,
we
have
a
number
document
that
has
been
updated.
A
A
We
have
a
number
in
individual
drafts
that,
if
we
have
listed
here
just
for
information
and
I
think
that
most
of
those
all
of
them
are
actually
on
the
agenda.
There's
one
draft
that,
according
to
the
name,
is
intended
for
the
spring
working
group
and
the
Brona
I
think
we
need
to
discuss
into
which
working
to
protect
document
actually
goes
I'm
having
a
problem
in
doing
in
the
spring.
But
we
should
agree
on
it.
It's
the
dot
equal
spring,
MPLS
ping,
alcove,
the
second
one
from
the
top
it
was
presented
in
the
spring
working
group.
A
A
A
A
Yeah,
those
are
the
existing
document.
Yes
go
on
here
we
have
a
little
bit
of
progress
reporting
since
I'm
a
co-author
or
contributed
a
number
of
them.
I
actually
made
a
list
here,
including
those
documents
and
some
document
that
actually
know
something
about
the
MPS
LPG
on.
We
actually
received
the
law
cipher
our
response.
As
of
yesterday.
It
only
took
us
three
weeks,
so
we
know
we'll
go
into
working
group
last
call
as
we
go
out
of
this
meeting.
A
The
draft
idea
of
Ella
sleeping
bag
multi-part
and
the
draft
idea
templates
are
I
RC
p
FR
are:
are
those
waiting
for
the
working
group
chance
to
write
the
Shepherd's
right
up
if
the
draft
xxx
amplest
mem
m,
LD,
p
mm,
it's
the
last
memory
doing
and
Kishore.
If
you
are
here,
I
would
like
to
speak
to
you
after
the
meeting.
A
That
document
has
been
stalled,
also
tent
out
in
mail
yesterday
and
I
think
it
could
progress
on
draw
xxx,
embolus,
SL
framework
and
draft
ideas
and
less
or
more,
we
don't
have
any
report
and
the
ML
DP
Jiang,
as
I
said
before
we
have
being
is
being
prepared.
Working
groups
called
last
call
basically
just
add
addressing
comments
from
the
young
doctor,
hello.
A
The
author
thought
draft
at
the
attempt
has
summary
FR
r
RC
bt
is
that
that
they
are
entity
to
what
your
thoughts
call
I
will
take
contact
contact
with
people
they
with
them
and
see
what
you
need
to
do,
or
next
one,
the
terror
we
do
that
they
may
know
those
two
death
bass,
young
and
Static
yang,
is
correct
and
the
we
are
getting
nearer
to
working
group.
Gospel
I,
think
that
we
are
will
have
an
updated
version
of
both
prior
to
bankrupt
and
then
solicit
further
further
input.
D
A
E
So,
there's
only
just
one
update
in
this
latest
version:
we
introduced
a
new
stabbing
mode
for
IGP
contacts,
ID
advertisement
in
addition
to
the
existing
proxy
mode
and
alias
mode.
So
in
a
stub
link
mode,
a
contacts
ID
is
advertised
by
a
quest,
router
and
protector
as
stopping
IP
address.
So
it's
a
very
simple
mode.
In
fact,
it's
a
simplified
version
of
the
proxy
mode.
E
So
this
provide
the
ideal
virtual
topology
for
the
path
computation
for
transport,
a
nose
and
the
bypass
tunnel.
However,
a
feedback
we
receive
from
deployment
was
that
if
we
could,
you
know
in
some
simple
networks
or
simple
topologies.
If
we
could
just
advertise
the
contacts
ID
in
the
manner
of
the
stabbing
mode,
we
could
still
compare
the
right
path
for
transport
tunnels
and
bypassed
a
nose.
E
So
with
this
simple
update,
we
believe
the
the
draft
is
mature
and
ready
for
working
group.
Last
call
egress
production
has
been
in
the
ITF
discussion
for
many
years.
We
start
with
pseudo
wire,
equal
protection
and
earlier
layer,
3
VPN,
eQuest
protection.
There
have
been
deployment,
you
know,
implementations
and
deployment
of
both
and
pseudo
y
equals
production
has
been
published
as
RC
8104.
E
E
E
The
framework
the
draft
considers
an
egress
failure
as
a
failure
at
both
the
service
level
and
transport
level,
so
the
framework
addresses
us
protection
and
at
both
levels
the
framework
relies
on
two
rows
of
router.
First
is:
is
the
PR
point
of
local
repair?
This
is
a
router
which
detects
a
u.s.
failure
and
performs
a
faster
route
at
the
transport
tunnel
level.
E
The
second
row
is
called
a
protector.
This
is
a
router
which
receives
rerouted
packets
and
performs
a
second
rearend
which
is
at
the
service
level.
Ok,
the
forwarding
this
kind
of
reroute
is
also
called
contacts
based
forwarding
because
the
forwarding
is
based
on
a
service
level,
a
service
label.
In
the
context
of
the
protector
egress
routers
label
space,
the
draft
introduced
a
new,
a
new
notion
in
the
notion
of
contacts
ID,
which
is
used
to
identify
a
pair
of
egress
router
and
the
protector.
E
E
E
A
G
H
H
It's
we're
pushing
for
okay,
okay,
so
this
is
the
first
one.
Second,
one
I
think
that's
in
the
past
years,
I
think
the
use
of
several
scenes
that
happens.
You
know
I
eat,
so
I
sing.
Afore,
you
got
em
ho,
that's
the
four!
That's
the
you
grass
protection
for
the
tunnel,
for
example
that
has
a
for
the
US
Army
ours
I
think
that
any
cost
a
segment
can
achieve
the
seemingly
similar
purpose.
H
H
E
E
H
G
H
That's
the
in
fact
here:
there's
the
the
some
to
some
user
can
have
that
user
already
the
domain
where
the
label
so
because
you're
here
you
have
the
contacts
or
maybe
for
subversive
a
VPN
there's
a
contact
shared
by
two
local
P,
protecting
each
other.
In
fact,
from
my
point
of
view,
so
that
is
the
whole
EPO
or
the
P.
They
can
share
one
domain
where
the
label
can
achieve
with
a
similar,
of
course,.
E
A
So
before
you
start
on
me,
okay,
if
the
kind
of
little
bit
unusual
layout
of
the
again
the
item
free,
is
up
to
me.
Well,
they
I
thought
of
it.
We
have
to
do
Javanese
discuss.
There
is
a
kind
of
overlap,
at
least
a
level
up,
but
people
need
to
understand
how
they
don't
overlap.
So
we
did
the
discussion
on
the
ampulla
ITF
ambulance
as
we
see
in
London.
That
was
the
kind
of
Cirrus
II
representation,
and
you
will
do
the
Cirrus.
A
Your
presentation
for
deaf
man,
well,
0
0,
a
cool
thing
that
doing
the
let
me
do
the
first
time
or
even
whatever
relation
you
are
in.
You
have
the
right
to
come
and
say:
okay,
I
want
to
do
it
for
initial
presentation
on
this
that
one
and
that's
what
you
happen
here
after
that
Stuart
will
compare
the
two
drafts
and
then,
at
the
end
of
our
ganda
item
free,
we
have
a
kind
of
open
issues
on
the
ITF
MPLS
NFC.
So
that's
the
way
we
set
it
up.
Okay,
okay,.
I
Thank
you,
Laura
I'm,
Andy,
malice,
I'm,
talking
to
you
about
the
MPLS
encapsulation
for
the
service
function,
chaining
network
service,
header,
that's
a
mouthful
and
my
co-authors
are
Stuart,
Joel
and
vim.
So
this
slide
is
a
quick
update
for
those
of
you
who
really
haven't
been
following
the
work
in
the
SFC
working
group.
So
in
ssing
working
group,
the
goal
is
to
put
together
chains
of
service
functions
to
do
processing
on
packets
in
a
particular
order,
and
that's
really
all
you
need
to
know
in
order
to
make
this
work.
I
They've
defined
a
new
header
format
called
the
network
service,
header
or
nsh.
So
what
this
draft
is
talking
about
is
how
to
carry
packets
that
include
the
nsh
over
an
MPLS
infrastructure
and
the
SOC
work
group
defines
two
particular
kinds
of
of
boxes,
shall
we
say,
or
when
network
nodes,
one
of
them
is,
is
an
SF
which
is
a
service
function.
The
service
function
is
something
that
does
work
on
a
packet.
Then
there's
the
SFF,
which
is
a
service
function.
I
I
All
I've
done
is
added
a
bit
of
color
to
it,
so
the
red
nodes
that
you
see
the
s
FS,
those
are
service,
function,
forwarders
that
are
MPS,
enabled
and
so
they're
able
to
take
packets
that
that
actually
have
an
MPLS
a
label
stack.
Then
the
network
service
header
through
the
service
function
chain,
as
you
can
see
here,
from
the
source
to
the
destination
in
a
particular
order
over
an
NP
s
infrastructure.
I
So
this
right
here
is
the
details
of
the
encapsulation
to
make
this
happen.
So
if
you
see
on
the
picture
on
the
right,
you
have
the
NS,
H
header
and
payload
that's
standard
from
the
SOC
working
group
and
it's
meant
to
be
transport
independent,
which
means
it
can
work
on
any
particular
type
of
transport.
We're
just
defining
the
transport
in
this
case,
to
be
an
MPLS
infrastructure,
so
to
be
able
to
carry
it.
It's
very
simple
looks
just
like
an
MPLS
VPN
in
that
you
have
a
single
label.
I
That's
interesting
is
that
this
supports
both
normal
MPLS
label
swapping
or
you
can
take
that
that
those
transport
labels
and
may
have
an
MPLS
SR
for
a
second
trotting
stack.
So
it
supports
both
regular
labels,
swapping
and
label
stacking
as
well
and
and
in
that
particular
case,
it's
compatible
with
draft
Bouchard,
SOC
and
SH
SR,
which
was
presented
yesterday
at
the
spring
working
group.
So
one
of
the
things
that
we
have
to
talk
about
is
ecmp
ecmp
forwarding
through
an
MPLS
infrastructure,
may
or
may
not
be
desirable
for
particular
flow
of
packets.
I
So
the
way
that
we
take
care
of
ecmp
is
that
the
first
nibble
of
the
network
service
header
was
actually
designed,
as
and
as
a
ecmp
protection
mechanism
in
that
and
make
sure
that
that
first
nibble
never
looks
like
a
four
or
a
six,
which
means
that
routers
that
perform
ecmp
automatically
on
packets,
that
look
like
ipv4
or
ipv6
won't
perform
ecmp
on
these
packets.
But
in
some
cases
you
do
want
to
use
the
CMP.
I
In
those
cases
there
MPLS
already
has
native
mechanisms
to
allow
a
CMP
to
occur,
both
the
entropy
label
and
the
flow
air
transport
label.
The
authors
are
still
talking
about
which
we
would
recommend
for
this
particular
case.
We
haven't
quite
come
to
a
firm
conclusion,
yet
om
is
also
important
and
for
both
MPLS
and
SSC
SSC
has
its
own
OM
mechanisms
at
its
layer
that
are
defined
in
IRC
8300,
so
I
recommend
you
go
look
at
a
300.
If
you
interested
in
om4
SFC,
we
may
also
want
to
use
om
at
the
MPLS
layer.
I
In
that
case,
MPLS
already
has
its
own
OM
mechanisms
such
as
gal,
for
example.
Now
Stewart
is
going
to
be
talking
at
length
with
the
comparison
between
this
draft
and
draft
IHF
MPLS
SFC,
which
has
a
similar
name.
I,
just
have
a
very
quick
introduction
to
the
comparison
on
this
draft
transports:
SFC
packets,
with
the
network
service
header
between
SSS
or
an
MPLS
infrastructure.
It
supports
all
the
features
of
SFC
because
it
has
the
NS
h,
including
per
packet,
a
metadata
which
is
included
in
that
header.
I
Now,
draft
IP,
FN
+
FCC
uses
the
MPLS
label
stack
to
logically
represent
the
network
service
header
for
interim
deployments
that
don't
support
the
NS
h
in
an
MPLS
infrastructure.
So
there's
no
network
service
header
in
the
packets
in
draft
IAT,
FM,
+
s
FC
and,
as
a
result,
there's
no
per
packet
metadata.
The
only
metadata
that
you
can
possibly
do
is
per
flow
metadata
and
that
uses
either
control
plant
extensions
or
a
new
MPLS
special
purpose
label
that
carries
the
metadata.
I
It
have
also
been
asked
how
this
compares
with
draft
shoo
plaid
spring
s,
our
service
programming,
which
is
also
presented
yesterday
in
the
spring
working
group.
So
this
draft
again
transfers
SSE
package
with
the
nsh
supports
both
traditional
label
swapping
and
sr
MPLS.
As
I
said
before
now,
this
has
the
usual
MPLS
state
in
every
node,
because
it's
regular
MPLS,
so
there's
the
Lib
and
everything
else
at
every
LSR
when
you're
doing
label
swapping-
and
this
is
really
intended
for
SOC
infrastructures
with
MPLS
and
the
SSC
nsh
is
present
in
every
packet.
I
Now
draft
shoe
clad
is
quite
different
and
it
tended
to
support
generalized
service
programming
in
segments
routed,
domains
so
segment
routing
only
services
are
associated
with
SIDS,
and
this
is
actually
a
bit
more
general
and
more
powerful
than
what's
happening
in
SSC
and
that
services
could
be
more
than
just
the
service
functions
as
defined
by
the
FCC
working
group.
Draft
shoe
clad
worked
with
both
segments,
routing,
MPLS
and
sr
v6.
I
It
does
not
support
MPLS
label
swapping
because
there's
no
MPLS
State
and
the
routers
it
only
supports
the
label,
stacking
and
and
the
nsh
is
available
for
use
by
the
use
of
ATL
V.
So
the
next
steps
for
this
draft,
we
have
some
work
to
do
on
the
for
future
study
items
such
as
the
ECM
P
and
control
plane,
and
since
I
wrote
this
draft
we've
actually
had
more
progress
on
the
control
plan.
I
We've
been
working
with
the
authors
of
draft
IETF,
best
nsh
BGP
control,
plane
and
we've
already
determined
that
that
draft
already
has
everything
that's
needed
to
support
this
draft.
So
we
really
don't
actually
need
to
make
any
changes
to
the
best
draft
to
support
this
draft
area.
That's
ready
all
done
and
there
may
be
other
control
plan
options
as
well,
and
we
want
to
start
working
towards
adoption
here.
Any
questions.
J
J
This
is
you
know,
it's
something
we
need
right.
You
cannot
carry
nsh
over
an
MPLS
network
without
an
encapsulation.
We
need
it
so
good
and
one
point
of
interest:
the
this
SFF
label
advertisement
we've
been
looking
at
the
BGP
control,
plane,
work
as
andy
says,
and
as
luck
in
the
script
would
have
it,
we
already
support
this
function,
so
we
need
to
add
a
pointer,
but
then
we're
done
all
right
pointer
to
the
document.
Yes,.
A
K
Comment
on
the
on
the
control,
plane,
I
I,
think
the
the
draft
supports
it,
but
it's
not
explicitly
written
right
right.
How
to
do
that.
So
I
think
there
needs
to
be
a
bit
more
text
on
how
we
would
support
that
with
the
control
plane.
But
at
the
semantics
are
there
yeah
I
think
it
needs
a
bit
more
text
to
actually
clarify
that.
Okay,.
K
L
Sff
the
service
functional
forwarders
are
built
using
existing
MPLS
rooters.
It's
designed
to
provide
a
migration
migration
assistance
to
an
nsh
environment,
making
it
easy
to
introduce
service
function
chains
in
today's
networks,
the
MPLS
layer
has
to
under
and
the
whole
of
the
service
chain
to
make
this
work.
So
the
service
the
chain
is
fundamentally
built
into
the
MPLS
layer.
L
Draught
malice,
MPLS,
FCC
encapsulation
is
designed
to
carry
nsh
packets
over
an
MPLS
network
between
sfx,
so
the
MPLS
layer
does
not
have
to
understand
service
function
chains,
it's
just
acting
as
a
transport
between
entities
that
do
understand
the
nsh
and
it
operates
really
like
a
style
of
MPLS
VPN,
the
there's
a
label
in
there
that
vectors
the
packet.
Once
it's
arrived
at
the
node
into
the
subsystem
that
understands
the
nsh
or-
and
there
may
be
more
than
one
of
those
so
they're
like
chalk
and
cheese
people
question
the
use
of
this
expression.
L
L
The
draft
idea
of
MPLS
SFC
understands
and
knows
the
service
chains.
Draft
malice
in
SOC
encapsulation
understands
how
to
deliver
packets
to
an
SFF,
but
not
what
to
do
next,
neither
can
do
the
job
of
the
other.
So
what
here's?
What
I
think
we
should
do?
Each
graft
has
a
different
purpose
and
is
applicable
to
a
different
network
scenario.
There's
no
conflict
between
the
drafts.
Each
graft
needs
to
be
considered
on
its
own
merits
independently
and
for
its
own
purpose.
L
Graft
IETF,
MPLS
SFC
is
a
working
group
draft.
It
should
continue
on
that
that
path,
the
normal
publication,
the
draft
malice
is
not
yet
a
working
group
draft,
but
I
think
he's
actually
simple
enough,
that
we
should
adopt
it
soon
and
then
both
drafts
should
follow
their
natural
progress
to
to
publication,
and
that
really
is
the
the
end
of
the
story.
A
H
L
L
H
handler
understands
the
whole
service
function
chain
system,
it
e.
It
is
the
thing
that
understands
how
to
do
the
SFF
function.
It
knows
how
to
send
the
packet
off
to
the
service
function,
actual
sort
of
itself,
so
all
we're
doing
is
we're
just
transporting
an
NS
h
over
a
an
MPLS
network
and
then
handing
the
packet
off
to
something
that
understands
how
to
parse
and
NS
h.
Okay,.
M
Nice
to
jeonggi
show
are
we
I
just
want
to
make
the
comment?
Let's
not
overcomplicate
this.
Yes,
there's
a
concept
in
networking
called
layering
that
we
seem
to
have
forgotten,
but
you
know,
let's
understand
it's
still
there.
This
draft
simply
says
I
get
to
an
end
of
an
MPLS
stack.
I
have
something
I
need
to
know
what
the
heck
it
is.
This
label
tells
me
it's
an
Anna
sage
packet
yep.
That's
it
suppose
that
that's
it!
J
As
the
slides
appear,
I'm
Adrian
I
don't
intend
introducing
this
draft.
What
I'm,
because
we've
been
there
before
what
I
want
to
do-
is
flag
up
some
of
the
issues
that
were
raised
on
the
list,
how
we
have
addressed
them
and
what's
still
open
as
an
issue,
and
then
I
will
stand
here.
While
you
throw
things
at
me,.
J
J
So
summarizing
the
use
cases
remember,
the
problem
was
from
memory.
The
first
one
was
was
there
even
any
need
for
this
work
now,
I
think
the
fact
that
we've
adopted
into
the
working
group,
but
probably
over
that,
but
the
aim
was
to
get
the
document
to
be
a
little
bit
clearer
about
what
the
deployment
scenario
was,
and
this
is
not
competing
with
the
work
of
the
SFC
working
group.
It's
not
saying
the
nsh
is
rubbish.
J
What
is
actually
doing
is
saying
that
there
are
brownfield
deployments
which
are
MPLS
capable
already,
and
we
want
to
see
if
we
can
get
some
of
the
functionality
of
the
nsh
into
an
MPLS
network
so
that
existing
MPLS
routers
can
act
as
service
function,
forwarders,
and
that,
if
you
like,
is
a
is
a
bootstrap
to
service
function,
training,
and
will
it
ever
happen
or
will
we
go
straight
to
n
sh
I
I'm,
not
predicting
the
market
on
that,
but
I
wanted
to
enable
this
and
turns
out
it
can
be
done
and
works.
Thank
you.
J
J
So
I
already
talked
quite
a
lot
about
the
brownfield.
The
point
was
to
have
nsh
unaware
SFF,
so
the
forwarders
are
just
MPLS
routers
and
to
highlight
again.
This
is
not
the
end
goal.
We're
not
saying
forever.
Everything
will
be
an
MPLS
Network
we're
trying
to
get
service
function,
training
going
with
what's
already
in
the
field.
J
So
what
we
did
with
this
was
added
some
text
in
the
abstract,
an
introduction
to
try
to
drive
that
point
home
and
we
wrote
section
14
as
a
sort
of
proof-of-concept
strawman
thing
that
says
this
is
a
way
it
could
be
done.
It's
not
a
recommendation
on
how
to
do
it,
but
people
were
saying:
oh,
you
can't
do
it
within
existing
MPLS
routers
and
we
know
we
can
because
we
can.
So
what
I
tried
to
do
is
write
a
here's
a
way
you
could
do
it
essentially
using
the
vrf
model
of
of
labels.
J
J
The
next
thing
must
have
said:
what
are
the
use
cases
for
stacking
land
swapping,
and
so
we
added
section
4,
which
lists
five
specific
cases.
The
first
one
is
the
standard.
It
looks
very
much
like
the
that's
interesting
NHS
I've
got
the
British
Health
Service.
On
my
mind,
the
NS
H
is
mapped
into
two
MPLS
label
entries
and
then
it's
treated
just
like
an
NS
h
at
different
nodes.
It's
it's
looked
up
its
swapped,
so
that's
that's
kind
of
using
the
MPLS
label
swapping,
but
it's
also
very
close
to
the
NS
h.
J
The
session
news
case
was
to
say
we
want
to
do
that,
but
sometimes
when
we
arrive
at
an
SFF
there's
a
choice
of
service
function,
instances
and
normally
you'd.
Let
the
SFF
choose
based
on
load,
balancing
or
time
of
day
or
the
weather
or
or
whatever.
Sometimes
the
classifier
right
at
the
start
will
want
to
control
which
service
function
instance,
a
particular
packet
or
flow,
goes
through.
You
can
do
that
by
having
the
nsh
labels
and
then
underneath
another
pair
of
NSH
labels
to
actually
get
you
into
the
right
SFI.
J
So
that
actually
gives
you
a
need
to
stack
these
news
entities.
The
third
use
case
is
a
hierarchical,
SFC
or
I.
Think
the
draft
is
called
hierarchical.
Nsh
in
the
SFC
working
group
and
that's
essentially
in
SH,
can
catalyst
it
inside
n
sh,
which,
if
you
like,
you
could
call
n
sh
stacking
and
obviously,
when
you
map
that
to
MPLS,
you
end
up
with
stacked
MPLS
label
entities.
J
What
this
achieves
for.
You
is
the
ability
to
concatenate
service
function,
chains,
which
turns
out
to
be
a
really
neat
property
both
for
scaling,
but
also
for
achieving
compound
functions
right.
A
fourth
use
case,
then,
is
is
a
kind
of
reduction
of
that
previous
case.
If
you
imagine
a
chain
two
chains,
both
full
nodes,
long
well,
you
could
actually
have
a
chain,
that's
two
nodes,
long
and
maybe
four
of
them
or
you
could
have
eight
chains
that
are
just
one
node
long.
J
So
you
end
up
potentially
with
something
that
looks
like
a
label
stack
entry
for
each
SFF
you're
going
to
and
that
the
one
that
causes
the
conflict,
because
that
starts
to
bump
into
SR
mpls,
and
what
we've
written
in
this
document
I
hope
clearly
but
willing
to
make
more
edits,
is
to
say
that
SR
MPLS
is
not
in
scope.
Here.
Go
and
look
at
draft
shoo
glad
about
that.
J
J
That
led
to
some
discussion
on
the
list
and
that
the
blue
text
is
approached
from
a
couple
of
emails
and
the
list
I
don't
think
we're
welcome
to
do
it,
but
I,
don't
think
there's
particular
value
in
having
the
debate
about
when
did,
label
popping
and
steering
and
so
on.
First
come
about
because
that
might
just
just
lead
to
lots
of
time
spent
going
through
the
archives
and
and
lots
of
blood
on
the
floor.
Well,
what
we
really
want
to
do
is
look
at
the
ways
forward.
J
J
J
People
wouldn't
notice
second
option
is
to
explicitly
forbid
micro
chains
and
I'm
not
quite
sure
how
that
works,
because
where'd
you
draw
the
line,
there's
a
too
hot
chain.
Too
short,
it
is
a
full
hot
change.
At
some
point
we
would
have
to
decide
what
constitutes
a
micro
chain
and
at
that,
and
in
any
case,
how
would
you
police
it
because
it's
just
a
natural
consequence
of
any
hierarchical
SFC,
and
so
would
you
end
up
forbidding
all
hierarchical
SFC,
which
would
I
think
seriously
constrain
the
function.
Third
option
keep
4.4
but
actually
pointed
out.
J
2's
are
MPLS
as
well
say.
If
you
want
to
do
single
hot
micro
chains,
don't
do
it
like
this
go
off
duress
or
MPLS,
and
a
fourth
option
is
to
say
well
actually,
SR
MPLS
is
more
than
just
the
forwarding
plane.
Sr
MPLS
is
the
whole
control
plane
construct
as
well.
Maybe
highlight
that
a
bit
more
and
and
then
otherwise
leave
the
text
as
it
is,
and
the
fifth
option
is
listen
to
Andy
Andy.
I
For
my
benefit,
oh
sure,
why
we?
Because,
as
you
said,
you
know
it
it's
functionally
identical
to
SR
MPLS,
so
just
include
the
text
which
which
points
to
the
single
hop
micro
chains.
You
know
just
add
text,
section,
four
point:
four,
which
points
to
SR
MPLS
and
then
the
text
that
you
said
that
that
you're
not
doing
as
a
MPLS.
J
M
M
J
J
Different
yeah
so
on
the
rise
as
I
say,
the
the
the
methods
to
set
it
up,
the
control
plate
not
very
different,
but
when
it,
when
you
come
and
look
at
it
on
the
wire,
it
looks
very
similar
and
I
think
what
happened
was
to
two
groups
of
people
or
two
documents,
one
working
from
one
end,
which
was
the
pure
just
single
nsh.
If
you
like,
swapping
one
working
from
the
other
end,
which
was
the
pure
SR
MPLS,
and
they
come
together
in
the
middle
and
and
do
we
document
it
both
approaches.
M
L
I
think
your
second
option
explicitly
forbid
microscope
change
doesn't
really
have
any
intellectual
sort
of
validity
at
all,
I
mean
clearly
you
can
do
it
and
if
you
can
do
it
and
someone
finds
a
need
to
do
it,
they
will
do
it.
So
why
would
we
write
an
RFC
that
we
know
that
at
some
stage
someone
might
write
a
one-line
RFC
to
to
modify
to
allow
something
that
is
perfectly
allowable,
so
I
don't
really
understand.
Quite
what
the
fuss
is
about.
To
be
honest,
the
functionality
is
useful.
L
L
G
J
A
N
A
O
O
Set
ie
TLD
if
it
had
link
protection
in
it,
a
few
editorial
knits
in
the
IANA
section,
we
basically
cleaned
up
the
document,
so
we
received
the
early
allocation
code
points
from
Ayana
and
in
the
original
version
way
back
on
this
document,
we
had
a
section
on
node
protection
which
wasn't
written
up.
So
there
was
a
separate
graph
that
was
published
to
deal
with
some
of
the
details
of
not
protection
and
I'll
be
going
over
that
next
and
the
draft
has
been
stable
for
a
while.
O
O
O
So,
what's
the
basic
motivation
right,
it's
the
base
draft
FSU
inability.
It
introduces
the
concept
of
T
link
labels
which
can
be
stacked.
It
also
introduces
the
concept
of
delegation
labels
and
hops
in
the
base
draft.
So
one
could
create
a
stack
of
labels
and
sort
of
reduce
a
data
plane
footprint
for
our
SOP
LSPs.
It
also
included
guidelines
for
setting
up
such
LSP,
which
do
not
require
any
protection
and
LSP
that
only
asked
for
link
protection,
so
this
draft
is
basically
for
offering
node
protection
on
that
same
transport
must
be.
O
So.
What
is
the
issue?
Why
do
we
need
something
different?
So
if
you
just
take
a
look
at
the
picture
right
so
B
is
the
PLR
DNA
of
merge
points,
so
they
are
next
next
stop
LS
ours.
So
with
T
link
levels,
LSP
is
actually
share.
The
labels,
so
label
100
at
B
is
a.
She
can
be
shared
by
many
LSPs
heading
towards
C,
but
note
protection.
O
The
merge
points
are
different,
so
we
cannot
use
100
for
any
LSP
going
towards
C
for
the
on
that
link,
because
the
next
to
next
stop
was
completely
different,
so
bypass
LSP
is
at
be
to
protect.
These
LSP
is
heading
towards
D
and
E
are
completely
different.
One
is
at
the
top
a
red
line.
The
other
one
is
at
the
bottom
red
line,
so
you
can't
have
two
different
backup
forwarding
actions
for
the
same
label.
O
You
don't
know
which
traffic
is
going
where
so,
that's
basically
the
problem,
and
you
can
share
that
label
100
only
if
LSP
is
actually
share
the
same
next.
Next,
stop
much
point
so
for
LSP
is
going
from
B
to
C
to
D.
Yes,
absolutely
you
can
share
the
same
label
because
they
can
all
pick
the
same,
not
protecting
bypass
and
similarly
BCE
helps
me
setting
down
that
direction
can
actually
share
the
same
label
so
with
the
procedures
such
come
up
with
in
this
diagram.
O
It's
easy
to
envision,
C
as
being
a
APL
are
just
go
to
one
simple
example:
the
rest
sort
of
very
similar.
So
if
you
look
at
the
labels
between
C
and
D,
348
and
345,
you
have
to
allocate
a
label
for
that
link
for
appear
next
next
stop
radius,
so
where
LSB
is
because
our
SOP
has
you
know
the
ER
on
our
our
o,
every
node
knows
where
the
LSP
is
heading
towards
where
LSP
is
going.
O
So
if
C
is
a
PLR
and
it
knows
and
LSP
is
heading
towards
e
4x
CDE
so
for
all
LSP
is
heading
yr,
D
to
e
you
allocate
a
single
label
and
then
yr
d2h.
You
allocate
a
different
label
so
at
the
next
next
top
radius
level.
The
scale
increases
it's
not
just
for
a
CLE
for
the
number
of
links,
so
the
original
base
draft
had
number
of
labels
allocated
by
c
is
equal
to
the
number
of
links
that
you
have
now.
This
slightly
goes
higher
because
it
goes
up
to
the
next
stop.
O
Next,
next
of
radius
level,
with
respect
to
the
forwarding
action
and
the
basic
primary
forwarding
action
remains
the
same.
You
pop
and
forward
down
c
pops
and
forwards
down
CD
for
back
up
forwarding
action.
You'd
have
two
paths,
two
levels
and
then
push
the
packet
down
or
than
the
not
protecting
bypass
to
the
merge
point.
Similarly,
you
can
see,
for
the
other
examples
are
draft,
gives
you
more
clears
up.
The
picture
gives
you
all
the
labels,
so
this
is
actually
an
example
where,
like
in
the
bass
draft,
we
have
this
concept,
delegation
label
and
hops.
O
This
is
a
very
straightforward
example.
If
there
is
no
delegation
ops
in
this
example,
it's
a
very
simple
example.
It
says:
CDE,
none
of
them
are
delegation.
Hoffs
see
is
a
PL
our
age,
more
easy,
much
more
much
much
point.
What
happens
if
there
is
a
delegation
hub?
So
just
to
recap,
a
delegation
hop
is
a
node
that
actually,
in
the
tree,
is
a
transit
node
that
actually
helps
stack
a
set
of
labels,
because
ingress
is
unable
to
reach
all
the
way
to
the
egress.
O
So
router
in
the
middle
of
the
network
is
actually
stacking
labels
we
went
to
do
the
base.
Drive
talked
about
the
ETL
D
algorithm.
What
is
ETL
D,
how
it's
used
to
automatically
select
delegation
of
so
much
of
what
I'm
talking
about
doesn't
apply
for
explicit
delegation?
This
is
all
automatic
delegation,
so
hops
get
selected
down
the
middle.
O
So
if
you
take
this
example,
we
have
C
is
a
delegation
up,
for
example,
it's
going
to
push
a
lot
of
labels
to
take
the
packet
downstream,
but
to
the
next
delegation
hub,
which
is
e
so
to
CTE.
There's
a
label
stack
so
for
that's
the
protected
node,
so
you
don't
have
to.
If
you
have
to
protect
a
delegation,
half
the
PLR
B
needs
to
be
able
to
push
all
the
labels
that
C
actually
pushes.
O
So
that's
the
base
condition
so
but
B.
If
the
label
capability
is
a
push
capability,
sort
of
all
these
devices
are
actually
very
good,
like
they're,
very
high,
equal
or
greater.
Then
we
can
certainly
push
all
the
labels
that
C
can
push,
but
if
D
cannot
push
all
the
labels
that
C
can
push
because
B
may
not
have
a
depth
push
capability
that
C
here.
O
So
so,
let's
say,
C
is
able
to
push
four
labels
and
B
can
only
push
to
D
cannot
as
a
PLR
push
all
the
labels
that
C
does
so
there
is
knew
we
had
the
gTLD
attributes.
Tlb
in
the
base
dropped
a
little
bit,
so
we
took
another
8
bits
from
it
called
delegation
helper,
so
B
actually
tells
see
how
much
labels
that
can
push
and-
and
it
can
actually
see,
in
fact
the
ETL.
O
The
algorithm
is
actually
change
that
see
how
I
changed
it
such
that
it
uses
the
dhrd
coming
from
the
upstream
hop
to
decide
what
the
outgoing
a
TLD
should
be
so
so
that
C
enables
no
protection
for
the
PLR.
So
that's
the
basic
change
that
was
made
so
the
ETL
d
@c,
basically
company's
computer,
using
a
min
of
the
d
HLT
coming
from
the
previous
hub
and
what
ET
le
it
may
have
naturally
computed.
So,
whichever
is
the
lower
value
it
takes
it.
O
The
forwarding
action,
pretty
much
looks
the
same,
so
C
PLR
would
have
normally
popped
and
forwarded
down
the
link
for
the
back
for
a
bypass
it.
Actually,
perhaps
the
label
I
have
stool
labels
and
pushes
the
label
stack
that
protected
mode
would
have
push
the
the
delegation
hop
would
have
pushed
and
then
sends
it
across
the
bypass.
O
So
that's
so
this
works
and
for
backwards
compatibility
so
either
like
you
know,
the
an
implementation
may
have
implemented
the
base
draft
and
doesn't
support
what
we
are
recommending
in
the
note.
Protection
draft
or
implementation
doesn't
support
so
in
that
situation
rate,
so
the
d
HL
d,
coming
from
b
to
c
c,
doesn't
even
understand
what
dhrd
means.
So
it
does
what
it
does
as
described
by
the
base
draft.
It
sets
the
e
TLD,
as
normally
as
it
does.
B
looks
at
our
ro,
because
B
is
the
one
that's
supposed
to
create.
O
The
label
stack
and
push
it
in
the
bypass.
When
B
looks
at
the
label
stack
and
looks
at
it,
it
looks
very
blodge.
It
simply
cannot
push
such
a
large
label
start,
so
B
can
resort
to
using
link
protection
for
it,
for
it
wants
so
much
of
these
are
easily
detectable
by
other
signaling
plane.
So
BB
can
choose
to
offer
no
protection
if
it
wants
or
choose
to
step
down
and
offer
link
protection
in
these
scenarios.
Q
Q
Answer
now
so
in
a
lesser
I
still
need
to
give
to
the
name.
We
are
working
on
machinery
to
distribute
information
about
what
supported
by
not
by
link
which
includes
amount
of
labour
that
could
be
imposed,
amount
of
level
that
could
be
read.
Potentially
it's
more
information
that
you
could
gather.
Maybe
you
could
use
it
because
we
flood
it
anyway
right,
so
they
must
be
nhdplus
year
old,
yeah.
O
A
A
H
Sorry,
this
Arobin,
you
know,
to
be
honest,
I'm
a
little
secured,
so
Oh
relax,
but
be
sleep
presentation,
so
this
one
so
there's
a
mitotic
I
used
the
pass
programming.
So
in
fact
this
I
have
a
much
relationship
with
disgusted
SFC
walk.
In
fact,
to
be
true.
That's
before
this
one
I
assume
that
how
you
will
present
at
least
I'm
sure
the
extension
header,
because
it's
unnatural
with
the
relationship
with
this
walk,
but
anyway
we
can
icon
kakadu,
so,
firstly,
okay,
so
this
was
I
think
this
is
the
second
routine
had
been
proposed
here.
H
So
these
are
you,
the
Ute,
I'm
sure
programming
capability?
Very
well.
So
this
document,
who
summarized
the
user
cases
under
the
and
is
a
possible
future
of
the
I'm
here
I
mean
okay,
so
his
first
is
a
history
reveal
in
fact
that
I'm
jars
label
stack.
You
know
past
the
years.
The
p-47
routine
is
also
you
exist
here
we
did
stuff
here.
H
We
at
that
time
I
propose
that
at
least
we
have
some
of
these
most
of
the
compressed
cases
that
that
I
use
the
veep
here
or
a
PDP
over
our
DP
over
RC
BT
and
over
the
bypass
faster
route.
It
kind
of
chilled
to
the
fiber
layers
label
yeah,
but
when
it
is
God
who
and
also
that's
who
you
were
history,
we
also
has
some
these
labels.
We
think
that
is
beyond
as
eligibility,
because
at
the
beginning,
we
based
on
this,
the
I'm
sure
label
for
the
reach
ability.
H
So
now
that's
the
I'm
Joseph
hologram,
you
see
you
would
now
develop
a
very
fast.
We
think
that's
they
also
two
important
characters
factors.
The
first
is
that
the
central
control
is
a
table
header,
so
that's
a
user
easier
to
allocate
a
label
for
more
purpose
than
eligibility.
Second,
why
user,
based
on
central
control,
using
easy
to
calculate
Pass
yeah?
So
there
is
a
list
of
the
way
thinking
of
this.
Well,
so
that's
the
we
also
here.
We
mention
that
as
the
I'm
here
as
a
power
Crimean
and
beyond
a
set
of
my
beauty.
H
From
my
point
of
view,
that's
what
I'm
sure
how
the
Crimean
we
have
this
little
layer,
the
first
layer
uses
a
transport
layer.
So
these
are
you
the
most
of
other
eligibility,
there's
the
best
effort
at
her
pass,
and
also
this
the
traffic
engineering
passed
this
away
based
on
the
ceremony
routine
and
now
that
we
think
that's
the
important
work
is
the
service
layer
and
the
network
layer.
This
means
that
you,
the
first
verse
difficult
low.
That's
the
user
will
use
the
I'm
cárcel
able
to
indicate
her
failures
of
the
service
process.
H
That
means
the
user,
along
with
the
specific
prefix
which
indicate
a
specific
of
low
yeah.
So
when
you're
saying
you
think
about
their
transport
layer,
I'm
sure
as
a
programming,
we
think
that's
the
secondary
routine,
but
the
service
always
on
here
the
I'm
here,
the
possible
granny.
So
this
is
what
we
think
this
is
a
new
work.
H
H
This
was
a
service
label
I
single
now
that
we
have
the
entropy
label
and
we
have
the
acidosis
label
or
synonymous
the
flow
label
for
the
OEM
and
also
steering
label,
and
also
now
that
he
used
some
leaves
upon
us,
the
currently
the
label
for
the
networker
slicing,
and
also
we
may
be
the
future.
We
have
this,
the
close
label
with
label
to
indicate
the
coolest
process.
This
had
been
happened.
You
know
I'm,
just
history
such
as
our
SP
and
also
this
is
the
between
label
to
indicate
a
specific
review.
H
So
this
is
a
service
label.
It
can
be
flexibly
programmed
along
with
the
prefix.
So
here
we
can
say
for
specific
of
prefix.
It
has
multi
label
to
indicate
a
serious
process
for
the
ECM
here
for
the
OEM,
for
the
kills.
You
like
this
way
and
also
you
can
be
severely
into
a
super
safe
Gaetano,
so
they
say
you
the
new
capability
for
the
som.
He,
and
also
this
is
the
architecture.
H
This
is
the
basement,
essential
control,
so
there's
a
including
how
to
allocate
this
is
the
service
label
in
the
I'm
char
Satomi,
and
also
that
you
know
after
this,
after
this
is
the
programming
for
the
service
of
us
and
the
use
of
download
to
the
ingress
node
with
the
prefix.
So
we
think
this
is
the
PGP
is
suitable
for
this
purpose.
Okay,
okay,
so
here
they
say
you
the
picture
to
show
this
one.
H
Maybe
this
is
a
similar,
but
here
we
can
say
this
is
a
download
that
is
a
label,
so
we
transfer
this
to
the
destination
at
that
time
for
specific
media.
So
this
is
only
the
BPM
label,
but
it
is
not
only
the
vision
label
but
with
the
service
label
stack,
yeah
and
also
here.
You
can
also
super
cific
panel
super
safe
unit
on
oh
yeah,
okay.
So
this
is
the
scenario
for
the
salon,
so
the
we
came
here
this
to
typically
all
the
cases.
H
So
those
will
be
the
studio
in
the
prefix
Gaetano
at
the
beginning.
Maybe
it
is
really
the
relevant
for
the
past,
but
a
later,
maybe
this
congestion
happens,
so
the
patient,
the
pc,
initiate
either
we
said,
have
a
new
RSP
or
the
newest
are
passed
and,
as
is
han
based
on
the
PDP
for
the
som
he
p,
so
that
he
is
a
download
and
the
two
super
see
if
I
specify
that
this
is
the
fellow
to
this
right
of
us,
so
they
say
user
to
determine
the
this.
Is
the
ridi
ridi
ridi
back?
H
That
was
super
sweet
you
got,
handle
and
I
hear.
Another
is
the
determine
the
you
see,
I'm
here
with
entropy
label
before
now
that
for
the
entropy
label
will
always
a
calculated
in
the
US
Route.
Her
inverse
wrote
her
based
on
this,
the
IP
destination
address,
and
also
this
was
the
source
address
as
well,
so
these
conserved
to
some
instant
and
consult
the
usual
of
the
of
the
elephant
flow
and
make
it
a
distributed.
But
in
fact
they
usually
because
of
the
ingress
and
water
is
a
lack
of
the
global
wheel
of
the
network.
H
So
this,
and
maybe
also
has
some-
this
is
not
a
reliable
for
the
ECM
heat,
but
I
will
the
central
control
you
have.
The
global
wheel
has
a
global
wheel.
So
then
you
can
calculate
her
super
safe,
shaker
and
hope
you
label
and
the
download.
We
leave
the
prefix,
so
those
of
you
that
can't
be
for
the
better
or
the
optimal
the
ECM
he
in
the
network,
so
they
use
a
multi
term
in
the
in
the
network
for
the
traffic
pattern.
Yeah,
okay,
so
and
also
these
are
the
glasses
of
hades.
H
They
see
the
overview
of
the
progress
of
the
related
work
of
the
s1.
He
be
in
fact
three
and
the
point
five
years
ago.
I,
probably
so.
The
to
draft
line
is
of
the
us
bring
working
group.
Was
versa,
file,
this
concept
and
solution
and
a
same
time
we
specify
this
the
PGP
extension
for
this
purpose.
But
after
that's,
why
he's
almost
act?
H
Stucked
because
I
have
my
sure
other
walking
now
this
what
we
think
it
is
the
right
hand
to
resume
this
walk
the
first
line
we
changing
either
to
the
I'm
cars
of
working
group.
So
we
think
this
is
a
great
home
for
the
solution
and
at
the
second
align
we
can
say
this:
the
we
make
much
progress
in
the
past
eight
years.
We
can
see
this
I'm
here.
H
Service
label
exist
in
the
entropy
and
also
the
Sonoma
synonymous
flow
label
for
the
OEM
that
YouTube
economy,
the
RFC,
and
also
now
that
we
proposed
and
is
the
past
segment,
there's
a
for
the
OEM
or
the
pipe
directional
for
the
SR
walk,
and
also
we
have
the
lip
here.
That
is,
a
label
to
use
the
label
to
specific
VPN
and
also
the
year.
H
So
this
is
a
based
on
this
one,
but
we
cannot
identify
these
the
requirements
of
the
challenges
so
now
that
most
of
the
label
wanted
to
put
it
at
the
bottom
of
the
label
stack.
So
that's
what
we
see
that
that
follow
label
wants
the
bottom
and
also
the
SFC.
They
also
the
bottom
of
the
bottom
of
the
label
stack
so
now,
I
think
here
we
must
that
either
means
our
forwarding
plane.
H
So
we
should
be
the
thought
formula
that
home
or
label
Nessa,
we
think,
should
I
be
defined
in
DES,
I'm,
sure
working
group,
a
second
away.
So
that's
the
user
I'm
here
see
change,
read
or
walk
because
we
take
the
no
as
at
home,
the
only
is
I'm
just
a
little
stack,
but
as
the
SFC
and
as
I
am
so.
That
means
that
between
the
label
stack
and
also
the
another
preload,
you
must
doubly
feel
someone
is
the
metadata.
So
that's
how
to
take
the
metadata
within
the
reason
I'm.
H
C
H
Also,
this
will
a
deal
with
the
BGP
is
to
ensure
as
saying
that
Hannibal
proper
baseline
for
the
IDR,
but
a
later
I
sing
the
road
et
cetera.
They
do
some
work
on
this
one.
They
say
you're.
The
panel
encapsulation
I
met,
who
was
subversive
in
a
tunnel
for
for
the
to
read.
Ii
worked.
This
fellow
was
the
boss.
If
he
that
handle
and
also
here
they
also
to
identify.
A
H
Yeah,
okay,
sure,
okay,
so
that's
all
almost
all!
So
that's
the
I!
Only
the
requirement
are
you
little
user
to
download.
This
is
the
ID
programming.
Now,
that's
not
only
the
I'm
sure,
programming,
okay,
so
summer
is
there.
So
this
is
our
oldest.
You
know:
nets
abortion,
yoga
versus
a
draft
here
to
consolidate
the
solution
with
the
latest.
The
progress
in
the
IETF-
oh
and
also
reverse,
reverses
our
draft
here
of
the
IDR
extension
to
consolidate
of
the
PDP
extension.
So
after
that
uses
sorry
say
two
more
comments
on
feedbacks.
Okay,
so.
H
Q
Q
E
S
It's
all
the
same
clothes
all
right,
yeah
there'll,
be
an
evil
kinda.
Al
mullah
same
comment
is
I'd
like
to
see
what
this
draft
can
actually
fix,
which
other
dress
are
not
actually
trying
to
fix
the
SRS
FC
or
the
draft
zu
clad
or
serving
segment
rotting
policy.
What
on
out
of
all
those
were
documents
is
not
covered,
that
this
is
actually
a
trying
to
address.
If
not
I,
try
to
see
you
out,
it
can
be
merged,
because
so
I
guess
that
we
are
scrubbing.
H
A
T
A
T
A
T
Go
ahead,
thank
you,
and
what
is
the
problem
we
are
trying
to
solve
is
the
case
where,
where
the
network,
where
one
of
the
two
teepees
is
at
least
one
of
the
two,
is
not
capable
to
is
at
the
control
board,
and
we
want
to
find
a
way
to
be
able
to
send
packets
over
the
network
with
the
control
board.
The
reason
is
that
the
latest
draft
that
has
been
approved
recently
bypass
is
recommending
the
use
of
the
control
board
that
we
didn't
observe
wives.
T
This
option
is
not
always
preferred
by
the
operator,
so
we
were
looking
if
there
is
another
way
to
solve
the
problem
without
replacing
the
box.
And
what
is
the
proposal
in
the
rough
that
is
basically
add
is
to
add
a
new
SP
in
the
network
to
place
this
net,
this
new
SPE
in
a
very
close
location
to
TP
one.
So
what
happens
is
that
in
this
situation
you
have
a
one
link
or
a
very
small
sized
networker.
We
expect
to
be
co-located
or
one
of
the
way
at
the
MPLS
layer
where
you
can
control.
T
There
is
no
ICMP
between
the
TP
and
the
SP
and
then
to
change
the
behavior
of
the
SP
to
be
able
to
receive
packets
silver
package
without
control
water
and
send
them
over
episode
of
a
segment
where
a
control
board
is
enabled
by
doing
that.
You
control
no
ICMP
behavior
here.
So
no
problem
not
having
a
control
board
and
you
have
the
control
board
that
protects
against
in
karate
CMP
behavior.
When
the
traffic
flows
through
the
MPLS
Network
the
once
you
have
set
up
this
multi
silent
zero.
T
The
big
assumption
that
we
have
in
the
draft
is
that
the
behavior
of
the
p1
is
is
independent
from
the
fact
that
is
terminating
a
single
senator
or
a
multi
sememster
wire,
as
outlined
in
the
NRC,
is
6
0
73
and
a
few
slides
since
the
0
0
drop
the
describe
the
basic
approach,
so
sp1
is
receiving
a
packet
generated
by
TP
1
without
the
control
water.
The
operation
of
SP
1
are
the
same
as
of
today's.
What
tips
are
available?
Decrement
the
TTL
martin
additional
80.
T
It
insert
the
control
board
before
sending
the
packet
to
TPU
and
in
the
opposite
direction.
It
removes
the
before
sending
it
back
to
DP
1
and
the
VCC
visit.
So
we
are,
you
have
also,
since
you
modify
data
encapsulation.
Also
the
VCC
V
options
available
on
that
you
segments
are
different.
We
assume
for
simplification
in
the
rough
that
we
can
always
use
CC
type
1
over
the
absolute
while
that
has
the
control
water.
T
So
you
limit
the
number
of
options
to
two
to
address
only
to
CC
type,
three
and
four
on
the
on
the
first
bit
of
water
and
in
this
slide
are
all
address
that
tree.
If
you
are
interested
to
type
for
you
can
read
in
rafter
and
basically
the
solution
is
the
same
as
what
the
label
recommend
ETL
and
you
insert
the
ACH,
and
here
is
important
since
the
STC
tie
three
is
that
SP
one
knows
the
TTL
distance.
T
It
has
between
2
tp2
to
TP
1,
because
the
TTL
is
there
is
the
discriminating
factor
to
distinguish
whether
the
packet
is
a
data
plane
packet.
So
you
need
to
insert
a
control
board
or
the
packet
is
an
OEM
packet
and
you
need
to
insert
the
AC,
and
this
is
covering
a
bit
a
signaling
procedure,
so
the
LDP
signaling
got
to
negotiator,
so
the
sp1
can
negotiate
a
knock-on
toward
the
we
T
p1
and
can
negotiate
control
or
the
user
control.
T
What
we
TP
to
the
the
key
issue
here
is
that
the
changes
that
we
are
requesting
is
on
the
behavior
on
the
protocol,
behavior
of
sp1
from
ep1
and
ep2.
Nothing
changes,
so
you
don't
have
to
change
anything
at
the
control
plane
for
TP
1,
+,
TP
2.
You
have
only
to
change
the
sum
rule
so
to
allo
sp1
to
negotiate
different
control
capabilities
on
the
two
segments,
and
the
same
applies
to
the
VCC
with
type
a'.
So
you
have
to
negotiate
a
difference.
T
Cc
type
sum-
and
you
have
an
initial
rule
here-
not
to
negotiate
the
ACH
TV
types.
If
you
don't
even
are
able
to
translate
a
CC
type
4-
and
this
is
there-
are
some
corner
cases
when
you
have
the
CC
2
3,
&
4,
on
ep1
and
ep2,
that
you
can
enable
CC
type
3c,
a
CH
not
supported,
enabled
and
last
neither
the
solution
is
described
with
1
SP,
but
the
the
operation
of
DSP
is
independent
from
Allah.
How
many
of
silhouette
semis?
Do
you
have
in
your
network?
So
you
can
have
additional
deployment
scenarios.
T
One
is
in
case
where
both
TPS
are
not
capable
to
us
at
the
control
water.
The
solution
is
quite
straightforward.
Just
add
the
two
TPS
and
everything
works.
Every
SP
works
as
define
in
the
raft
and
you
again
enable
send
a
traffic
with
contour
Ward
through
the
network,
and
the
last
case
is
when
you
have
a
more
complex
scenario.
Are
you
ever,
for
example,
NPS
DP
networks?
Well,
you
don't
have
a
CMP.
T
You
can
also
put
the
SP
that
that
insert
the
controller
would
a
little
bit
far
ahead
in
the
network,
because,
as
long
as
you
know
that
your
network
or
multiple
networks
are
not,
are
you
sure
that
there
is
no
ICMP?
It
doesn't
matter
as
long
as
you
can
send
control
noise
EMP,
you
can
send
a
traffic
without
control
water
as
long
as
far
as
you
like,
it's
very
generic
and
you
can
have
as
many
SP
as
you
like
a
to
all
your
chain.
The
the
behavior
is
only
limited
to
this
single
nada.
T
The
more
important
is
to
have
this
type
of
solution
and
the
second
and
what
are
their
capabilities?
What
what
they
can
do
when,
when
you
have
multi
sensor
wire
in
terms
of
CC
type,
a
CB
types
entity
has
configuration,
because
these
are
the
assumption
that
we
have
on
the
behavior
of
TP,
one
that
allows
the
solution
to
work
and,
of
course,
we
have
said
welcomes
any
comments
and
proposals
to
improve.
Thank
you,
Jeff.
Q
And
Josh
so
elegant
solution
to
solve
the
problem.
The
question
assume
your
work
is
driven
by
debt
net
requirements
with
regards
to
control
world.
Yes,
so
yeah
the
requirement
is
this
one
yeah
yeah,
otherwise,
I
don't
think
we
really
care
about.
None
shall
work
today.
The
point
being
part
of
your
network
doesn't
have
control,
work
and
part.
Does
so
part
of
your
network?
Isn't
deterministic
and
I
would
question
whether
it's
good
enough
for
death
net
service
to
provide
network
that
split
into
oneness
and
deterministic
could
be
seen.
Q
T
C
Q
T
T
Deployment
so
where
you
have
this
networker
is
without
control,
world
and
no
ICMP,
but
this
network
is
a
with
control
with
with
with
ecmp.
But
if
you
have
a
service
that
goes
and
and
how
do
you
ensure
that
there
is
control?
What
in
this
partial
yeah
it's
kind
of
problematic
that,
then
you
need
to
basically
apply
the
solution
to
an
existing
SP
the.
So
the
solution
will
work
on
a
different.
A
Q
L
Is
solving
a
historical
problem
right
when
we
wrote
the
we
were
told
by
the
operators
that
not
having
a
control
word.
We
resulted
in
ecmp
of
some
sorts
of
Ethernet
packets
and
this
was
a
problem,
so
we
wrote
a
draft
and
in
the
draft
which
is
now
a
minority,
we
basically
say
really
really
try
hard
to
put
the
control
word
in
place.
L
Now
some
vendors
I've
got
deployments
and
they
can't
actually
change
those
deployments,
and
so
the
question
was:
was
there
anything
we
could
do
for
operators
that
had
deployed
equipment
that
they
could
not
change
to
at
least
minimize
the
effect
on
the
rest
of
their
network,
and
so
the
proposition
is
that
we've
put
a
proxy
there,
but
for
all
new
equipment,
I,
absolutely
think
putting
the
control
word
in
there
for
all
four
or
detonate
I
think
you'd
be
insane,
not
control
word
unless
you
really
were
in
trouble.
So
the
questions
we
leave.
Ok,.
U
U
One
is:
do
you
know
of
any
merchant
silicones
that
support
inserting
and
removing
the
control
word
at
switching
PE
on
in
the
data
plane?
It's
very
difficult,
yeah
it
yeah.
It
would
be
very
difficult
to
support
that.
Secondly,
the
sequence
numbers
you
know
you
haven't
mentioned
whether
the
sequence
numbers
are.
P
Okay,
I
can
get
started
first
now
so
reason
America
string
has
been
discussed
in
this
group
for
quite
a
while,
but
the
focus
before
was
on
unicast
and
today
I'm
going
to
talk
about
a
multicast
aspects
of
it.
There
are
two
slides
the
two
presentations
and
two
drafts.
The
first
one
is
focused
on
our
city
gp2
in
be
eternal.
P
So
let
me
start
with
the
traditional
RC
bt
eton
began,
signaling,
I'm
using
this
example
of
here
that
the
ring
topology
so
for
pedo
Madonna,
you
have
many
leaves
and
for
typically
for
each
leaf
there
will
be
a
one
stop
AOSP.
In
this
example,
we
have
two.
We
have
one
ring
with
our
one
as
the
ingress
and
our
two
three
four
as
the
punitives,
so
our
one
will
set
up
three
subway
osp's.
So
week
is
near
the
ingress.
P
You
will
have
a
lot
of
paths
as
a
resume
state
and
each
sub
USP
optionally,
could
have
his
own
explicit
paths
and,
additionally,
the
actual
states
for
the
tunnel
protection
is
needed
during
your
set
of
a
bypass
tunnels
or
protection
panels.
So
all
those
could
be
optimized
the
way
when
you
have
in
the
case
of
the
RMR.
P
P
But
if
the
ingress
says
I'm
going
to
allow
impressive
leaves
than
the
each
one,
each
node
world
can
decide
by
themselves
whether
it
becomes
a
leaf
and
when,
when
they
do
decide,
so
they
will
send
resumes
G
back
and
so,
but
the
way
that
the
ingress
indicate
that
he
allows
the
impressively
way
is
simply
list
itself
as
a
leaf
and
there's
an
opacity
around
that
went
back
to
itself.
When
it
goes
around
the
ring.
Each
node
can
decide
whether
he
wants
to
join
the
Hinault.
P
Now
for
the
protection,
we
don't
need
any
additional,
signaling
or
state.
We
just
rely
on
the
global
repair,
but
before
the
global
repair
finishes
upon
a
failure,
we
don't
send
their
messages
so
tear
down
a
tonneau.
Instead,
we
were
simply
kind
of
the
traffic
from
the
point
of
a
local
repair
to
the
next
node
in
case
of
ink
failure
or
to
the
next
next
node
in
case
of
node
failure.
I
have
an
example
of
ink
feeder
case.
P
We
have
R
2,
R,
3,
R
4
other
panel
leaves,
and
if
the
link
between
R
2
and
R
3
breaks
then
R
to
simply
kind
of
the
traffic
using
the
ring
us
being
the
other
direction
to
that
next,
node,
R,
3
and
the
traffic
will
continuous
arm.
So
with
that,
we
don't
need
any
additional
signal
for
panel
protection,
funny
live-live
protection.
When
we
use
panel
to
provide
marker
services.
Sometimes
some
services
was
critical
or
the
mission-critical
services,
the
one
limelight
protection.
Typically,
that
requires
you
to
set
up
two
tunnels.
P
Along
the
disjoint
paths
in
a
general
topology,
but
in
in
cursor
RM
are
in
most
situations.
We
don't
leave
laiá
laiá
protection
because
we
can
quickly
turn
all
the
traffic
around
when
we
detect
local
failure.
But
if
you
really
need
live
level
protection
for
your
mission
critic
of
scenarios
as
long
as
the
application
can
remove
the
duplicate
packets,
then
this
can
be
easily
done
by
setting
up
two
sub
LCPs
in
two
directions
that
we
in
each
direction
we
were
which
audit
panel
lives.
P
In
that
case
the
ingress
was
some
presence
and
the
traffic
in
two
directions
and
each
node
each
leaf
will
deliver
the
traffic,
the
doback
traffic
to
the
application.
Who
will
receive
that
your
crickets
and
finally
Mpemba
can
easily
be
done
as
well.
We
just
add
a
label
to
the
past
message
that
is
used
for
downstream,
knows
percent
traffic
upstream.
P
That
object
will
include
a
ring
ID
and
wind
direction,
and
also
the
source
to
leave
sub
USP,
descriptor
least
or
least,
release
the
explicit
lives
or
and/or
the
ingress
ingress
itself.
If
you
want
to
allow
the
increased
alleles
and
past
means,
you
could
also
carry
a
label
for
the
NP
2mp
purpose
I
forgot
to
mention
earlier.
That's
all
this
optimization
are
good,
but
not
necessary.
If
you
don't
care
about
the
the
the
the
advantage
of
this
optimization,
you
can
still
use
the
existing
one
without
any
changes.
P
R
J
J
P
The
same,
it
is
the
same.
No
change
here
is
just
that.
We,
when
you
in
the
traditional
mess,
ignoring
you
include
multiple,
leaves
in
the
same
subway
OSP.
The
operation
becomes
the
sickening
and
processing
it
becomes
quite
complicated.
So
typically,
you
just
include
one.
You
have
one
service
be
for
each
leaf
and
in
that
case
descriptor
only
leaves
at
least
one
leaf,
but
here
we
just
distal
all
the
explicit
leaves
and
option
atleast
ingress
itself
in
case
of
interest
implicitly
right.
P
P
P
P
Additionally,
I
want
to
introduce
this
concept
of
services
and
transport.
The
market
service
is
like
an
end-to-end
SG
or
a
star
G
multicast.
That
is
either
in
a
global
table
or
default
instance,
or
in
a
way
in
Vivien's,
which
could
be
MVP
any
VPN
DPOs
things
like
that.
The
next
one
is
the
transport.
Basically,
the
tunnels
that
is
used
to
to
to
transport
package
for
the
marker
serve
services.
This
include
provider
channels
for
MVP,
evpn,
apos
and
GT,
and
when
he
see
achieved
he
was
an
Olympian
procedure.
P
P
So
all
those
are
the
markers
transports,
so
in
case
of
P,
more
ml
DP,
there
is
no
need
for
optimization
at
all,
except
that
panel
protection,
FIM
MVP
could
be
optimized
very
similar
to
so
P
2
MP,
RC,
BT,
PMP
panel
protection.
We
talked
about
earlier
and
and
whether
that
the
Marcus
domain
goes
beyond
the
single
ring
or
not,
it
does
not
matter.
The
reason
is
that
the
pin
joint,
miss
G
or
ml
DP
label
mapping
they
as
designed
they
just
go.
P
He'll
just
get
murmured
as
the
blow
up
strain
towards
the
root
of
the
source
and
whether
you
have
a
ring
topology.
It
does
not
matter
so
there's
nothing
to
be
down
there,
and
here
the
pin
can
be
for
postmark
a
service
or
of
transport,
and
while
the
ml
DP
is
is
basically
the
transport,
then
our
CP,
TP
tobita,
know
where
the
previous
presentation
we
talked
about
that
here.
That
panel
is
for
transport
either
for
medical
services
or
as
the
base
tunnel
in
case
of
multi
ml
DP
over
hearts,
PT
p
Tameka.
P
If
you
have
a
ton
of
their
spends
beyond
the
single
arm
are
we
can
still
do
optimization
over
each
ring
that
I
need
some
help
of
either
using
the
segmentation
procedures
specifically
services,
for
example
Olympian
segmentations,
or
we
can
built
in
some
signaling
enhancements
into
RC,
BT,
P,
2
and
P
to
help
the
optimization
of
its
PTO
in
patan?
Oh,
that
goes
over
multiple
rings.
P
F
A
You
any
questions,
don't
think
so.
Okay,
so
next
up
this
Deborah,
so
this
is
say:
I
would
call
it
an
invited.
Talk.
Yeah
I
have
over
the
last
year,
a
little
bit
more,
actually
being
rather
frustrated
by
all
the
things
that
we
actually
miss
in
draft
that
actually
submit
to
the
ISD
for
publication.
Is
there
everything
from
acronym
expansions
to
statements
that
is
kind
of
not
necessary
in
the
document
actually
flaunting
a.
A
V
V
I
put
the
don'ts
first,
because
there's
only
a
couple
of
them,
it's
don't
you
know,
expect
others
in
the
chain
and
reviewers
and
there's
quite
a
long
chain
right.
You
have
your
working
group
last
call
document,
Shepherd
reviews,
routing,
Area,
Director,
reviewers,
ad
ITF
last
call
other
area.
Reviewers
is
G
or
C
editor,
but
don't
expect
them
to
fix
your
documents,
especially
technically.
V
Maybe
they
do
maybe
they
don't
and
don't
expect
Dianna
to
clean
up
your
Anna's
section.
Don't
hesitate
to
ask
for
help
if
you're
not
sure
what
to
do
or
how
to
interpret
something
or
respond
to
a
comment.
Just
ask
your
chair:
the
document
Shepherd
ad,
any
anybody
you
want!
You
know,
don't
be
afraid.
You
know,
okay,
especially
if
you're
new
Arthur
and
if
you're
an
author
of
a
document
actually
going
through,
don't
go
to
sleep
after
working
group
blasts
call.
It
was
much
more
to
be
done.
V
It's
free
the
author's
to
respond
to
all
the
comments
that
come
up
as
a
document
progresses
and
the
documents
not
gonna
progress.
If
you
don't
respond,
okay,
dues,
so
read
your
document
line
by
line
and
check
the
figures.
The
document
has
evolved
so
often
sections
get
duplicated,
moved
around
figures
or
inaccurately
labeled
and
during
the
chain
reviews
it's
really
dismaying.
When
a
general
reviewer
says
your
figure
is
mislabeled.
You
know
that
we
didn't
even
pick
that
up.
V
Okay,
so
be
very
careful
that
you
yourself
go
through
your
document
and
then,
as
I
said,
ensure
one
or
your
authors
is
gonna
hold
that
pen
it
doesn't
have
to
be
the
same
one,
all
the
time.
Cuz
we
all
take
vacations
so
just
be
sure.
There's
somebody
available
right
to
respond,
because
you
are
the
ones
that
are
the
experts
on
that
document.
You
know
how
it
came
about
often
some
the
questions
come
about.
There
say
well,
why
didn't
the
working
group
consider
this?
My
idea
is
much
better
solution
for
you
and
it's
for
you.
V
You
can
answer
that
history
question.
Well.
Why
did
we
end
up
where
we
did?
There
was
a
educational
tutorial
on
Sunday
just
at
this
past
March,
meaning
it's
what
makes
a
good
internet
draft
it's
by
somebody.
We
know
very
well
and
if
you
don't
know
and
also
Spencer,
it's
excellent
so
make
sure
you
review
it
and
review
it
again,
because
I
cannot
put
everything
in
here
in
what
they
have
make
sure
you
know
your
RFC
style
guide.
So
these
all
just
pointers
just
to
help
you
out
right.
V
Ensure
number
of
authors
is
five
or
less.
We've
been
getting
a
lot
of
pushback
from
the
rest
of
the
isg
I.
Try
to
when
I
really
know
I
can
support,
and
your
chairs
can
support,
having
more
than
five
authors
that
we
try
to
get
it
through,
but
there's
really
a
huge
pushback.
Now
on
that,
so
be
really
careful
on
it.
Even
the
so
and
then
the
other
part
is
abbreviations
acronyms,
you
may
know
them,
but
others.
V
It's
always
picked
up
in
the
different
area
of
reviews
that,
and
they
go
to
this
list,
the
the
editors
privation
expansion
list,
and
they
say:
well,
it's
not
there,
and
even
if
it
has
an
asterisk,
which
there's
very
few,
you
know
you
should
spell
it
out
the
first
time
so
go
on
the
side
of
caution.
Try
to
spell
these
out
and
for
a
lot
of
you
I
think
you're,
not
realizing,
but
there's
a
new
RFC
to
refer
to
other
than
the
2119.
You
should
be
also
referring
to
8174.
V
So
here's
the
paragraph
that
you
should
be,
including
because
again
other
area
reviewers
are
going
to
nitpick
and
say:
oh
you
know,
you
don't
have
the
correct
statement
there:
okay,
security!
This
has
come
up
over
and
over
the
security
folks
have
they
have
gotten
much
more
verbose,
so
make
sure
you
are
just
have
two
lines
and
say:
oh,
this
has
all
been
handled
before
there
was
nothing
new.
They
need
to
know
why.
Your
understanding
is
that
there's
nothing
new,
so
put
a
couple
of
lines
there
and
then
we'll
work
on
it
with
them
to
see.
V
A
Thank
you.
Yes,
one
point:
the
abbreviation
list
is
available
of
the
style
game
style
guide
page
on
the
RC
editor
I
did
the
routing
directory
the
accurate
review
just
the
other
week,
and
I
actually
found
a
couple
of
acronyms
that
I
thought
were
Asterix
in
the
in
that,
in
that
abbreviation
list
and
I
didn't
check
until
very
late
enough.
To
my
surprise,
I
found
they
were
not
so,
thankfully
use
daily.
It's
still
not
here
yeah
daily.
Here,
it's
not
asterisk
in
the
list
so
need
to
be
expanded
and.
V
Actually,
the
reverse
to
you
may
want
to
check
that
list
before
you
come
up
with
a
new
acronym.
We
had
in
debt
net
that
they
came
up
with
an
acronym
that
I
won't
that
it's
famous
and
it's
in
there
with
an
asterisk
so
make
sure
actually
do
the
reverse
check
that
you
don't
come
up
with
an
acronym
that
is
already
on
that
list.
Okay,.
W
So
today,
I'm
going
to
talk
about
mph
extension
header
for
in-network
services.
First,
an
quote
from
ITF
nanhee
on
the
talk
about
IPS
architecture
and
principles.
Is
that
no
idea
is
so
bad
that
it
won't
be
proposed
over
and
over
again?
So
hopefully,
that
idea
I'm
proposing
today
is
no
too
controversial
and
indeed
it
backed
up
by
several
important
use
cases
and
also
it
opened
the
door
for
future
innovations
on
MPs
natively.
W
W
So
due
to
the
extension
headers
now
we
can
support
a
lot
of
new
applications
very
easily
for
security
for
second
row
team
for
in-network
services,
such
as
FC
in
c2,
OEM
and
even
network
programming,
but
ipv6
not
perfect,
because
its
overhead
is
thought
to
be
only
the
base
handler
is
a
fortified
long
and
also
it's
not
very
easy
to
access
the
inner
package.
If
you
need
to
you
need
to
suggest
through
all
the
stacked
extension
headers.
W
On
the
other
hand,
our
MPs
is
very
cute
right,
but
because
it
scooters
and
also
make
it
very
difficult
to
first
to
know,
what's
up,
what's
the
upper
layer
protocols
type?
Also
it's
a
very
difficult
to
encapsulate
new
headers
and
metadata
on
the
room
for
innovation
is
very
tiny
and
also
we
face
a
backward
compatibility
issues.
W
W
Such
services
means
we
need
to
add
some
metadata
or
new
headers
into
the
packet.
This
mechanist
not
seen
by
the
end
users,
but
is
only
consumed
by
the
networks,
to
name
a
few.
Such
applications
like
in
network
I
in
band
and
network
telemetry
or
instituto
a.m.
and
network
programming
and
DDoS
prevention.
There
are
many
more
such
possible
applications
and
the
orphans.
This
application
need
to
be
stacked
together,
which
means
we
need
to
insert
more
than
one
type
of
headers
in
the
in
a
packet.
Also,
we
need
to
support
the
the
backward
compatibility.
W
Also,
the
performance
considerations
are
very
important
for,
for
example,
we
need
to
avoid
unnecessary
label
stack
scanning.
Also,
we
want
to
quickly
access
in
their
packet.
If
you
need,
there
are
ways
to
achieve
that.
First,
we
can
just
introduce
a
special
extension
header
label,
so
further
eight
allocated
a
label
are
so
we
can
pick
one
and
we
think
the
use
case
is
seeking
significant
enough
to
deserve
one.
W
The
second
method:
we
can
use
a
to
label
scheme
first
labels,
the
extension
label
with
label
value,
15,
plus
extension,
header
label,
it's
okay,
but
it's
been
even
one
more
label,
it's
not
as
good
as
first
one.
So
the
third
option
is
we
use
FEC
your
FEC
labels
to
indicate
a
one
of
the
label
is
used
to
indicate
the
existence
of
extension
header,
but
if
the
right
is
the
trouble
to
introduce
a
new
label,
but
ASA
actually
complicates
a
control
plane
design.
So
to
summarize,
we
prefer
the
option
one.
W
So
here's
example
our
pact
format,
so
the
first
part
is
a
MPs
label
stack.
You
can
see.
The
extension
label
can
be
located
anywhere
in
this
table.
Stack
and
following
is
a
MPs
label.
Stack
is
a
header
of
extension
header,
which
you
basically
tell
you,
how
many
extent
extension
headers
we
have
and
the
total
mass
of
eight.
It
also
indicates
what's
a
type
of
the
next
header
and
after
that,
as
you
can
see,
a
stack
of
extension
headers.
Each
header,
you
see,
love
format
is
a
very
similar
to
ipv6
extension
header.
W
So
if
we
want
to
support
the
backward
compatibility
in
the
legacy
MPs
Network
and
it
need
to
be
at
the
bottom
stack,
but
otherwise,
if
the
whole
networks
already
upgraded-
and
it
can
be
any
location-
prefer
to
be
close
to
the
top
of
the
latest
at
for
performance
reasons,
and
also
we
need
to
introduce
two
special
next
header
values.
The
first
one
is
known,
which
means
there's
no
more
extent
headers,
and
in
this
case
you
have
only
headers
in
your
MPs
packet,
there's,
no
even
payload,
of
up
layer
protocol.
W
The
second
special
values
are
no,
which
means
to
be
compatible
with
the
original
MPs
design,
because
we
don't
know,
what's
a
actual
payload
type
in
encapsulate
in
amperes
packet,
and
also
we
can
support
load,
balancing
yeah
in
two
ways.
The
first
just
use
our
entropy
label.
The
second
is
we
provide
since
we
provide
a
quick
access
to
the
upper
layer
protocol.
You
can
skip
the
entire
extension
header
and
to
access
up
protocols.
Now
you
can
retrieve
five
Heather,
actually
typos
for
the
load,
balancing
purpose.
It's
very
easy
to
do
that.
W
So
the
conclusion
is
what
we
think
some
insurance
is
a
ubiquitous,
it's
widely
deployed,
and
also
it
has
a
very
low
overhead,
let's
just
because
so
missing.
Heiser
has
a
better
advantage
over
the
ipv6,
but
also
it's
very
difficult
for
MPs
to
support
in
network
services,
especially
when
we
need
to
stack
multiple
service
headers
together
into
a
single
packet.
Whenever
we
propose
is
a
flexible
and
extensible
solution
to
introduce
the
MPS
extension
headers
into
MPs,
and
we
believe
the
use
cases
are
strong
enough,
so
we
see
its
deserve
a
special
label
to
be
assigned
also.
W
H
From
Holly,
so
this
in
fact
the
justice
and
is
the
important
in
your
work.
We
try
to
propose
and
there's
a
framework
of
past
programming
I
mentioned
in
my
presentation.
Since
the
you
know
we
needed
who
carried
carry
this
the
metadata.
Not
only
the
label
stack
so
that
we
musta
need
assembly,
so
the
new
encapsulation
method,
so
I
think
this.
They
see
the
one
Walker
related.
H
We
the
forward
employee,
but
the
further
controlling
so
I
think
that
that's
also
downloaded
a
label
stack,
but
it
because
of
here
you
sund'ys
the
metadata
for
and
also,
for
example,
the
message
either
download
as
a
fellow
ID.
So
that
are
the
big
if
he
not
only
downloaded
the
label
stack
but
also
somebody's,
the
flexible
group
of
the
ID
information.
So
that's
what
the
major
scenes!
Okay,.
Q
G
W
So
definitely
you
need
to
involve
control
plane
to
inserts
extension
headers,
so
it
can
be
any
anywhere
in
the
within
the
network
right.
So
if
it's
a
first
extension
header,
then
you
need
three
actually
user.
This
extension
header
label
into
the
label
stack.
If
there's
a
last,
you
will
remove
the
last
extension
header
that
you
also
need
to
remove
the
extension
header
label.
So
there's
a
detailed
processing
procedure
described
in
the
in
the
draft
which
described
forwarding,
plane.
W
Q
G
A
Just
a
small
comment
from
someone
that
has
to
take
care
of
the
special
purpose
labels.
There
is
not
critical,
but
we
are
the
it's
a
scarce
resource
and
we
should
be
very
careful
about
assigning
them
without
the
careful
discussion
and
I
think
we
need
a
good
discussion
on
the
mailing
list
before
we
actually
continue
down
this
way,
and
you
should
also
look
at
if
you
can
use
the
extended
space.
Q
A
F
Q
A
D
D
Yeah
so
in
this
case
test
that
it
defines
the
effects,
but
the
length
of
the
fact
was
not
very
clear
like
how
should
it
include
the
result?
Field
I
should
not
include
it,
and
this
resulted
in
some
interoperability
issues
identified
in
E
and
T
C,
because
vendors
have
started
implementing
it.
So
now
we
are
starting
to
see
these
kind
of
issues.
So
as
an
example
over
here
now
we
have
ipv4
prefix
and
then
prefix
length
and
protocol.
D
So
should
it
just
be
6
bytes
long
or
should
we
also
include
the
reserved
field,
making
it
eight
bytes
long,
so
this
table
just
clarifies
that
part.
So
what
we
are
saying
is
that
it
should
improve
the
result
field,
because
that
also
conforms
with
effect
definitions.
Let's
say
LDP
be
deep
effects
defined
in
RFC.
Eighty
twenty
nine
so
for
in
this
case,
for
example,
for
a
GP
IP
for
ipv4
prefix
head
the
length
will
be
eight
now.
D
If
I
GP
ipv6
length
will
be
twenty,
it
should
also
include
the
reserve
field
and
then,
lastly,
just
census
a
FAQ.
So
over
here,
the
length
of
course
depends
on
the
type
of
adjacency
that
we
have
parallel:
p4
ipv6
and
whether
we
are
using
course
PFI,
sis
or
any
so
in
this
is
the
table
that
will
outline
the
length
fields
in
that
case,.
D
Yeah
so
I
feel
the
draft
is
very
well
documented.
It
clarifies
the
problem
area
and
provides
a
solution,
and
if
there
is
any
feedback
from
the
working
group,
we
really
want
we're
interested
in
it
and
we
want
to
address
it
and
otherwise
we'd
like
to
call
for
working
group
adoption
we'll
also
send
an
email
on
to
the
working
group.
Miller
I.
D
So
this
was
also
presented
in
spring
yesterday
and
now
I'm
presenting
it
to
the
MPLS
and
as
Laura
mentioned
it
will
progress
in
spring.
So
the
issue
that
in
hand
is
that
80
to
87,
as
I
mentioned,
defines
the
IGP
prefix
8
for
ipv4
and
ipv6.
So
this
prefix
8
is
combined
of
the
prefix
value
the
length
and
the
protocol.
But
now
in
the
LSR
there's
a
flexible
algorithm
draft,
where
a
saying,
ipv4,
prefix
or
a
basic
prefix
can
be
associated
with
multiple
prefix
set
each
associated
with
a
particular
constraint
path.
D
D
It
doesn't
carry
any
algo
identification,
so
prefix,
1
node
1,
of
course,
will
look
up
the
default
prefix
and
then,
if
will
complain
that
local
label
value
of
the
prefix
does
not
match
with
the
received
label,
it
will
result
in
an
error
notification
back
to
originator
result
saying
the
LSP
is
broken
when
LSP
is
perfectly
fine
and
you
can
also
have
a
case
where
a
node
is
not
really
participating
in
the
particular
algo,
because
we
are
only
checking
default
1.
So
we
are
correctly
saying
yeah.
This
is
all
good
when
the
LSP
is
actually
broken.
D
So
the
solution
is
really
simple.
We
have
16
bits
of
reserved
field,
so
we
can
use
the
first
8
bits
of
it
to
carry
the
algorithm
ID
and
because
the
default
is
0.
So
even
if
someone
is
running
an
older
version
or
they
don't
really
care
about
flexible,
flexible
algorithms,
so
they
can
ignore
it.
So
this
is
practical
and
then
also
backward
compatible.
D
So
now,
when
we
send
the
request
to
node
1,
we
care,
in
addition
to
prefix
length
and
protocol,
we
are
also
carrying
the
algo
ID,
so
prefix
1
can
do
the
proper
validation
with
the
prefix,
the
next
soft
and
predicted
label
of
the
particular
algorithm
instead
of
the
default
one
and
return
the
correct
error
code
yeah.
So
this
is
again
we
feel
that
this
is
a
really
simple
solution.
That
addresses
are
valid
gap
and
we
present
it
both
spring
and
mpls.
D
We
will
we
are
looking
for
feedback,
it
was
just
published
just
before
IETF
so
I
know,
a
lot
of
people
would
not
have
had
a
chance
to
go
through
it.
But
if
you
have
any
comments
over
here
or
you
can
post
it
with
a
mailing
list
will
be
very
interested
in
it
and
then,
if
we
don't
find
any
think
obvious,
then
we'll
call
working
group
adoption
in
spring.
U
From
Huawei
I
have
a
question
on
the
previous
presentation,
if
you
don't
mind
which
was
regarding
the
errata
draft,
so
this
is
a
process
question.
Is
that
better
to
do
a
base
document,
or
do
we
want
to
do
it
as
a
document
/e
rata,
which
updates
the
the
clarification
that
is
missing
in
the
original
RFC?
So
some
guidance
to
the
authors
would
be
nice,
whether
it's
better
to
follow
the
best
procedure
or
they
want
to
do
it
in
the
current
approach.
What
they
have
so
I,
don't
think
there
is
a
it.
A
Something
that
fits
with
all
cases
in
this
I
think
we
need
to
think
about
it
and
I
know
that
at
least
Carlos
is
willing
to
discuss
how
to
do
it.
This
is
not
the
normal,
a
Tartar
to
start
with,
so
it's
requires
a
draft.
It
should
be
a
base
or
not
there
I'm,
not
sure,
because
sometimes
when
you
do
these
drafts,
you
actually
pull
down
so
much
extra
work.
So
you
can't
really
do
it.
We
tried
LDP
once
and
the
update
from
what
is
actually
there
is
was
pretty.