►
From YouTube: IETF105-MPLS-20190722-1550
Description
MPLS meeting session at IETF105
2019/07/22 1550
https://datatracker.ietf.org/meeting/105/proceedings/
A
A
We
had
do
top
schedules
when
we
started
that
we
actually
found
out
that
we
could
make
it
through
hours
so
from
having
lost
the
room
to
actually
have
an
awful
discussion.
We
actually
went
to
pretty
squeezed
agenda,
but
during
the
second
session
growth
session
for
only
15
minutes
even
make
sense
so
one
session,
the
next
session
is
cancelled.
A
A
A
With
that
I
think
we're
ready
to
start.
You
have
the
note
well
on
the
screen,
I
say
as
Chris
said
morning,
meeting
I'm
not
going
to
read
through
all
this,
but
it's
important
that
you
do
so.
Please
read
the
note
well
and
understand
what
it
means.
Basically,
it
says
whatever
you
do
here:
it's
a
contribution,
diet,
gap,
okay
and
next
slide.
A
So
the
administrative
things
around
here
we
have
all
your
streaming.
So
when
you
talk
to
the
meeting
you
stand
in
front
of
the
microphone
them
to
pretty
close
because
they
are
set
draw
the
load
if
you
are
presenting
that
is
they
in
crossover,
are
use
the
chip
stand
on
that
or
if
he
slips
do
it
and
when
you
start
speaking
state
your
name.
A
A
You
find
the
online
again
then
slice
at
that
URL.
Next,
like
this,
so
again,
the
bashing
this
is
that
draft
we
intend
to.
This
is
a
status
report
and
the
that
we
attempt
to
discuss
I
think
that
the
preeminent
slot
at
the
end
is
no
doubt
enteric
will
discuss
the
gang
model
graphs
during
the
status
report.
A
A
The
first
one
was
to
use
the
zero
instead
of
a
o
or
if
it
was
the
other
way
around
I
don't
know
so.
This
is
verified
its
editorial
waiting
for
an
update
and
visible
I'm.
The
data
tracker
until
that
object
happens
the
next
one.
This
is
sometimes
you
actually
make
work
for
yourself.
So
what
happened
here
was
that
then
we
are
Sierra
to
publish
the
draft.
A
They
forgot
to
put
the
paragraphing,
so
we
made
in
the
dark,
but
to
put
the
paragraph
back
in
and
that's
now,
they're
also
listed
as
an
editorial
in
author
and
also
available
or
visible
in
the
data
correctly,
the
Aesop's.
We
have
had
two
outgoing
in
new
incoming.
That's
actually
the
opposite
to
what
usually
it
is.
The
ones
are
that
the
two
nations
are
fairly
easy.
A
One
is
say,
together
with
other
working
group,
react
to
responded
on
any
relation
from
I
think
heat
that
it
was
15
on
a
work
done
for
of
TNT
standardization
work
and
the
other
one
was
also
question
from
the
stereo
15
on
a
clarification
on
ambulance
sharing
protection
fairly
straightforward
next
document
status
there
had
been
a
couple
of
metres
of
hidden
talent
in
new
grads.
This
time
you
suddenly
have
free
drafts,
our
sorry,
our
deceased,
where
we
haven't
had
any
new
RHCs.
This
time
we
had
three
new
RFC's,
so.
A
Thank
you,
everyone
that
is
involved
in
authoring,
reviewing
pushing
on
these
documents
so
that
you
got
published.
We
have
documents
in
the
policy
and
askew
the
first
one
was
blocked
because
of
a
Miss
from
the
mr.
F
is
gone.
It's
a
wrong
and
big
cluster
that
the
RCA
doesn't
work
in
previous
now,
and
they
say
that
somewhere
before
Singapore,
all
those
drove
all
those
documents
we
do
in
to
author
48.
A
So
when
that
happens
to
be
prepared
and
actually
respond
quickly,
there
we
have
documents
with
the
is
key:
they
are
progressing
around
the
well
I.
Don't
think
there
is
any
draw
the
real
big
gaps,
though
we
are
waiting
for
a
revised
ID
for
the
biggest
protection
framework.
I
haven't
seen
that
yet,
but
it
should
be
progressing.
A
A
A
Next
thing
we
have
new
individuals,
rafts
all
of
them
or
on
on
the
agenda
and
that
we
have
updated
individual
graphs.
I
have
a
can't
find
any
surprises
here.
There
is
one
error,
though,
because
I've
been
working
this
afternoon,
drop
Shang,
MP,
lisanna
or
Ahmad.
Multicart
is
no
longer
in
the
video
graph
to
see
what
you've
done.
I
accepted
it
just
after
noon,
and
then
we
have
an
issue
that
was
brought
in
yeah.
A
So
it's
we
have
assumed
that
it
is
say
on
the
least
significant
bits,
but
we
will
actually
start
a
poll
to
some
of
the
key
implementers
and
ask
what
we
have
done
and
see.
If
we
have
everyone
doing
the
same
thing
and
once
we
actually
understand
the
situation
out
there,
we
will
actually
write
some
text
around.
A
A
A
It's
actually
not
waiting
for
the
Shepherd
right
after
review.
It's
actually
I
have
that
I
have
a
sensory
request
for
publication
for
this
up
today,
so
should
probably
be
somewhere
else
in
the
list,
and
this
actually
affect
the
or
more
extent
LED
or
or
more
extension,
and
it
should
or
perfect
air
drafts
in
spring
and
each
subject
they
have
or
moderate
salsa.
So
they
should
be
possible
to
progress.
Okay
go
back
the
first
one
be
at
the
directly.
A
A
A
A
A
A
C
So
hello,
but
we
wanted
to
give
a
quick
update
on
the
modeling
that's
happening
in
MPLS
for
p
group.
We
have
a
couple
of
crafts
to
begin
with:
LDP
modeling
ml
d,
p
and
l
vp
for
unicast
and
multicast
the
first
one.
There
is
a
working
group
last
call
the
requests
coming
in
from
offers,
and
we
will
initiate
this
working
group
last
call
for
the
second
one.
It
has
undergone
a
yang
doctor
review
cycle
and
they
received
comments.
The
authors
should
be
addressing
the
comments
now
and
get
back
circle
back
to
us.
C
There
are
two
other
drafts
for
modeling
in
yang
as
well.
One
for
MPLS
face
both
are
in
the
yang
doctors
review
one.
We
got
some
feedback
today
and
we
will
be
here.
The
authors
will
be
addressing
via
the
comments
from
the
yang
doctors
and
we
will
proceed
for
asking
Bertie
group
last
call
once
we
have
all
the
comments
address.
C
A
Beauty's
I
think
this
is
the
third
meeting
we
have
this
list,
I
actually
put
it
on
the
mpls
wiki.
You
can
find
it
through
that
URL.
Those
are
drafts
that
was
sent
to
the
working
group,
but
never
progress
and
it's
fairly
unclear
or
most
of
them
if
people
want
to
throw
them
at
progress.
I'm
an
art,
so
I
asked
you
to
look
at
the
list,
see
if
you
have
anything
there.
A
B
E
E
So
that's
the
background.
So
we
have
EPE
CID
the
peer
nodes
that
peer
sets.
It
and
add
Sid
that
are
getting
advertised
from,
let's
say,
SBR
one
to
the
controller,
and
then
this
EPE
Sid
is
getting
used
in
building
SR
enter
and
SR
path
across
multiple
ASIS
or
it's
getting
used
as
a
bottom
label
for
a
for
a
SR
path
up
to
the
SBR.
E
So
what
we
want
to
do
so
in
this
example,
let's
say:
there's
a
switch
between
a
str1
and
str2
and
and
the
link
shown
here
is-
is
broken.
So
in
this
case,
we
would
want
to
do
an
MPLS
ping
and
traceroute
and
be
able
to
detect
the
broken
link
and
the
way
to
do
it
is
the
controller
that
sets
up
the
LSP
knows
what
the
EP
meaning
of
the
EP
said
is
whether
it's
a
node
set
and
what
are
the
associated
links,
whether
it's
a
year
set
seed
and
what
are
the
associated
links.
E
So
the
controller
includes
that
information-
and
you
know
the
headend
can
do
MPLS
ping
and
traceroute
with
those
facts,
and
then
the
facts
validation
will
happen
on
ESB
r1.
So
there
are
three
kinds
of
information
that
we
need
to
validate.
First
thing
is
the
control
information
that
bgp
LS
has
advertised
any
PE
cid
and
it
says
certain
information
and
that
information
has
to
get
validated
when
the
packet
comes
in
the
OM
packet
comes
in.
That
information
has
to
match
with
what
what
has
been
advertised.
E
The
second
thing
is
the
forwarding
state
has
to
match
so
spr
1
says
it
is
sending
traffic
on.
Let's
say:
1
link
the
top
link
between
asbr,
1
and
2,
and
then,
when
the
packet
really
arrives
on
ESP
r2,
it
needs
to
validate
whether
the
incoming
interface
is
matching
with
what
ESP
r1
said.
So
the
way
this
gets
advertised
from
ESD
r1
is
downstream
detailed
mapping
TLV.
This
is
from
RFC
8
0
to
9,
and
there
is
no
change
from
that.
E
So
what
we
are
doing
here
is,
we
are
just
defining
the
facts
for
the
epe
suits
for
peer
nodes.
It
sets
it
and
adjacency
said
a
so
updates
from
last
revision,
so
we,
the
last
revision,
had
peer
note
said
with
a
single
set
of
local
and
remote
interfaces,
so
we
changed
the
definition
and
to
include
multiple
sets
of
local
and
remote
interfaces.
That's
because
the
peer
node
state
could
have
been
advertised
for
a
multi
hop
ebgp
session
and
there
may
be
multiple
underlying
links.
E
So
both
the
links,
though
all
the
multiple
links
have
to
be
included
as
part
of
the
factor
definition.
So
the
definition
has
been
updated
for
this,
and
there
was
a
comment
from
last
presentation
to
optimize
the
peer
set
seed
that
has
been
incorporated
and
the
security
consideration
section
has
been
updated.
So
that's
the
definition
for
the
peer
nodes
it
so
we
have
the
local,
a
s,
number
remote,
es,
number
local
BGP
route,
ready
and
remote
BGP
route
riding.
E
So
that
was
there
is
another
draft
with
with
which
is
very
similar,
similar
approach.
There
are
some
minor
differences
in
the
definition,
and
so
we
are
talking
to
the
authors.
So
beyond
saying
hi,
we
have
the
same
draft.
We
haven't
made
much
progress,
so
we
thought
whatever
update
we
have.
We
could
present
it
this
time
and
it
probably
will
talk
to
them
tomorrow
and
then
see
how
we
can
progress
and
one
more
thing
before
I
take
questions.
I
wasn't
too
sure
this
is
supposed
to
go
in
MPLS
or
spring
I.
F
You
know
you're
correct
one
of
the
con
orange
spring
cochair,
so
yes,
indeed,
there
is
another
draft
which
is
similar
last
time.
It
was
present
in
the
in
the
spring.
There
was
kind
of
an
agreement
to
tomorrow,
so
it
probably
needs
to
be
sorted
out
before
bringing
spring
or
MPLS
to
improve
it
in
him.
Well,
the
definition
of
the
effect
would
be
good
to
be
discussing
spring,
because
this
is
where
art
is
a
designer
and
discuss.
Now,
if
you
choose
a
good
point
in
MPs,
opera
and
fine
with
two
means
of
drafting-
and
please.
A
F
A
G
E
The
switch
is
not
visible
right,
it's
a
layer,
3
connection
between
from
ASPR
one
to
SBR
2,
and
we
have
advertiser
SID,
which,
which
says
it's
sending
traffic
on
those
two
links.
So
let's
say
that
one
is
broken.
So
on
a
SBR
one,
when
the
packet
comes
in
it,
it
looks
at
the
forwarding
plane.
It
has
two
links
so
in
the
downstream
mapping
it
says:
I
have
two
links
and
then
the
ecmp
tracing
happens
for
in
case
of
tree
stump.
E
E
H
I
H
Now
we
signal
routing
every
time
the
defining
an
incident
we
are
finding
need
for
new
FAC,
which
is
a
problem,
because
if
you
look
at
it,
even
the
information
needed
here
how
much
parameter
user
gonna
input
is
way
too
much.
So
I
am
concerned
about
the
complexity.
There
is
involve
in
a
solution
and
we
need
to
step
back
from
chasing
defining
effect
for
every
sip
type,
to
define
effect
that
can
cover
all
sake.
Type
segmenting
is
all
about
simplification,
so
this
is
my
major
comment.
H
A
H
J
E
J
J
J
E
J
K
Okay
hi
good
evening,
everyone
I'm
Mukul
from
juniper
networks
and
I,
will
be
presenting
in
terrorists
om
for
Assad
networks
on
behalf
of
others,
so
the
agenda
of
my
presentation
will
be
first.
I
will
talk
about
the
problem
statement
and
a
few
RFC's
previous
work,
which
we
have
found
which
is
related
to
this
problem,
and
we
will
discuss
what
is
the
issues
with
it.
You
know
with
that
and
our
proposal
how
we
are
trying
to
solve
it,
and
we
have
a
working
example
to
explain
our
proposal.
K
So
Intel
som
problem
is
a
well
known
problem.
We
have
I
think
it
has
been
worked
in
the
past,
but
here
we
are
trying
to
this
problem.
I
am
trying
to
state
is,
is
applicable
more
for
the
Assad
Network,
so
we
have
2a
s
s
s
1,
and
s
2,
and
we
have
a
and
both
the
sides.
We
have
s
our
path
setup.
There
is
a
controller
which
gets
the
BGP
LS
feed
from
is.
K
We
are
from
both
the
s
and
that's
how
he
knows
the
topology
of
both
days
and
and
and
it
can
push
that
information
to
the
ingress
r1
and
also
set
up
LSP
across
yes.
So
in
this
example,
let's
say
we
have
I
mean.
With
the
help
of
controller,
there
is
a
LSP
setup
from
R
1,
R,
2,
SB,
r,
1
s,
p,
r3,
r3
and
r4.
K
K
The
second
draft
77043
it
talks
about
for
the
MPLS
LS
fees,
interest
M
and
it
there
they
have
defined
a
new
TLV
eco
relay
node
relay
address
stack,
TL
v,
which
essentially
puts
the
IP
address
of
the
router,
which
does
the
relay.
But
the
problem
with
this
is
the
packet
for
each
hop
has
to
go
up
to
the
control
plane.
K
So,
basically,
each
hop
we
are
taking
the
packet
up
to
the
control
plane,
which
is
again
not
good.
So
the
proposal
which
we
are
trying
to
do
is
for
the
SR
Network.
We
define
a
reverse
path,
using
a
stack
of
cid
labels.
So
so
we
define
a
new
stack.
Tlv
reverse
path:
label
stack,
TL
v,
which
is
a
set
of
label
stack
of
labels
which
the
ingress
router
can
put
when
the
eco
request
is
generated
and
how
it
knows
is
it
could
be
yr
the
controller
or
it
is
pre
known
by
at
the
head
end.
K
K
We
will
essentially
contain
a
list
of
SIDS,
which
could
be
perfect,
said
decency,
state
binding
sites
and
essentially,
when
the
router
at
the
second
is
when
he
receives
the
echo
request,
while
responding
back,
all
it
has
to
do
is
copy
those
labels
from
the
a
request
to
construct
the
echo
reply.
So
is
so
this
solves
a
problem
that
the
packet
does
not
go
to
control
plane
at
each
and
every
hop.
So
it's
all
data
plane
forwarding
across
says
this-
is
the
packet
format
for
the
reverse
path:
label
stack
TLB.
K
So
this
is
a
working
example.
We
do
have
a
small
prototype
for
this
thing,
but
if
you
take
the
same
LSP,
which
we
had
in
the
first
slide,
let's
say
we
have
set
up.
Lsp
is
starting
from
r1
and
ending
at
R.
For
the
echo
request
we'll
go
with
this
set
of
sets
which
have
the
terminologies.
These
are
the
notes
it
for
our
to
sbr.
K
One
is,
we
are
three
and
node
set
of
p4
are
for
change,
and
for
this
echo
request,
the
reverse
path
will
will
be
the
set
of
labels,
starting
from
our
four
back
to
r1.
So
in
this
example,
the
reverse
path
year.
We
will
contain
this
set
of
labels,
notes
it,
for
s
are
for
let's
say
if
it
if
it
has
to
come,
where
the
sbr
for
then
EP
set
for
SBR
for
is.
We
are
between
ASPR,
4
and
1,
and
then
notes
it
for
r1.
J
A
J
So
ro
mentioned
their
Stowe
discussion
of
BD
directed
and
BFD
directed
used
to
have
portion
of
this
proposal.
But
now
it's
a
part
of
individual
draft
in
spring
working
group
on
draft
near
Skid,
Spring
BFD,
which
proposes
to
use
explicit
list
of
SIDS
for
controlling
the
return
for
PFD.
So
what
you're
proposing
is
already
being
proposed?
Okay,.
K
J
Npos
is
OSB,
thing
is
used
to
bootstrap
D
of
this
session,
but
part
of
the
proposal
is
that
it
can
communicate
the
list
of
SIDS
to
be
used
to
direct
PFD,
but
at
the
same
time
it
mentions
that
the
same
can
be
used
to
direct
response
for
are
there
spring
out?
Ping
is
itself,
so
it's
already
there.
Okay,
so.
K
G
The
the
ping
request
fails
for
whatever,
whatever
reason
between
ASB
r1
and
ASB
are
two
or
three
Oh
in
that
case
is
about
three
right
right.
So
how?
What
exactly
the
return
path
label
stack
is
going
to
use
because
you're
populating
with
in
this
case,
whatever
that
cyd
you're,
saying
that
this
is
what
you're
going
to
use
right.
But
it
means
nothing
on
that
intermediate.
No,
but.
K
K
G
K
H
L
L
Lithographed
defined
that
year
incarceration
for
empiricism
the
performance
measurements
with
the
ethernet
mac
him.
I
said
to
the
other
two
realities:
little
SL
requires
a
wanker
repeat
of
rotate
packet
to
measure
packet
loss,
also
the
danger
christ,
so
when
both
himself
842
package
tool
relatively
entry
to
alpha
beta
tricycle,
Oh
how's,
the
this
message
requires
the
follow
identification
alpha.
The
measure
that
indicates
ago
flow.
This
is
a
slight
turn
talks
about
how
to
method
ax,
pequeños
and
delay
when
syncope
white
light
the
treasure
to
the
murder.
Total
traffic
of
elope
need
to
be
happy.
L
She
into
overtime
for
packet,
loss
measurement,
coloring
the
measured,
critical
flow
and
flip
the
page
Carla
periodically.
You
know
the
report
is
a
countless
of
each
block
to
the
EMSL
controller
for
peg.
The
delay
measurement.
My
anomaly,
my
packet
within
each
block.
Each
note
report
is
the
time
steps
are
for
possession
each
mark
hi
kiddo
the
EMS,
our
controller,
the
this
inhaler
for
metal.
With
alpha
the
fellow
pays,
the
p.m.
encapsulation
with
the
alternator
McHugh
Meza,
to
my
floor.
Ided
in
the
Kindle
app
owner,
followed
by
wine,
fluoride,
aleppo,
Green
Cross
for
follow.
L
Itlp
t
beat
as
bait
and
so
on
the
novel
or
idealized
arm.
Here's
the
fellow
identification
a
smelting
unit,
unique
and
allocated
by
the
EMS
al
controller
based
on
the
measure,
the
architect
in
states
such
as
ours,
PL,
p,
w.
There
is
a
one-to-one
mapping
between
flow
IT
and
follow
no
no
threat.
The
follow
IP
indicator
level
may
appear
main
times
in
a
lab
stake
with
me:
threat
to
the
floor,
I
team,
a
present,
which
means
the
most
of
my
floor,
I
team,
a
present
again
auntie
Elsa
labor.
L
Let's
take
our
potato
Stefanik
Avenue
domina
falafel,
oh
and
murder.
Two
packet
loss,
ebay,
Tibbets,
SEO,
slit
owner
method
on
telling,
with
the
no
trace
receive
the
packet
with
that
he
said
to
one.
We
may
see
need
to
you
about
the
recent
reception
type
to
the
EMS
al
controller,
as
it
is
the
bottom
of
the
snake
indicator.
L
L
This
message
about
shocked
by
how
come
here,
madam
enter
the
translator,
notes
kind
how
to
our
work
out
based
on
floor
ID,
whether
to
turn
our
waking
is
use
the
unified
on
method
to
measure
the
packet
loss
outlet
daily.
They
use
the
floor,
RT
indicator
plus
Rho
RT.
You
know
the
current
state
packets
are
interval
called
time.
Then
we
powder
tool,
the
EMS
al
control,
a
parametric
with
their,
which
is
a
wiring
family,
the
EMS,
our
controller
new
calculated
packet
loss,
Delhi
and
printer
variation
in
helped
by
half
hour
into
ain't
about.
L
No
I
will
talk
about
how
to
in
company
encapsulate
took
up.
Our
EP
captures
the
floor
RT
before
animal
ministers,
the
MS,
our
controller
need
Napoleon
the
unica
flow
ie
to
measure
to
protect
it,
states
that
I
teach
in
a
network
animate
in
the
healing
grass,
and
that
note
the
in
Christ
knows
he
inserts
the
floor
idea.
The
key
indicator
light
boat
alongside
the
way
the
floor,
I
peel
a
pop.
At
the
same
time,
the
u.s.,
the
u.s.
L
note
said
that
our
PT
and
froggy
values
the
translated
to
know
to
look
up
look
out
for
the
floor.
Id
level
with
high
off
of
the
floor,
ID
indicator
and
the
Preston
Alfred
may
to
the
collected
information
to
the
EMS.
Our
controller,
which
includes
the
colors
out
routes,
are
for
packet
loss
and
cam
steps.
L
L
The
first
way
is
when
the
this
method
applied
to
the
art
speaking
at
the
young
keelson
and
work
the
floor
idea
for
the
last
variety
indicator
and
a
plurality
our
anti-isil
about
we
applied
to
the
rapini
MP
else
in
the
network,
the
floor,
IP
indicator
and
floor
RT
at
the
needs
of
the
whipping
labor
when
use
the
wind
is
measured,
applied
at
all
posts,
Alt,
key
and
Joaquin
in
amperes
in
the
work
they
are
needed
to
different
floor.
Id
values
that
ADA
to
see
through
idea,
planet
to
our
key
and
variety
flow
idea.
L
C
Direct
from
juniper
I
had
your
drop
and
I
sent
you
some
comments
on
the
year
reading
cooked
mature.
If
you
saw
them,
but
basically
to
general
comments,
there
is
a
year.
Another
way
that
the
working
group
is
working
on
is
called
synonymous
flow
labels
that
can
achieve
this
working
yeah,
I'm,
not
sure.
If
it
they
can
a
look
at
it
or
this
approach
you
think,
is
better
than
or
gives
an
advantage.
The
other
of
this
approach
will
introduce
two
labels
in
your
stack.
If
flow
identifier,
labeled
is
special
illegal
yeah.
C
C
L
A
M
A
M
M
So
requirements
is
to
transport.
The
IOM
data
fields
reaches
our
MPLS
and
cap,
so,
basically
carrying
the
way.
I'm
information
like
time
stamps
by
the
data
traffic
to
the
scope
is
using
the
data
fields
already
defined
by
I
ppm
working
group
and
n2n
I.
Am
we
can
look
at
the
hop-by-hop
as
well
in
future
if
there
is
interest,
so
the
end
cap
looks
like
basically,
there
is
an
IOM
indicator
label
and,
and
the
IOM
data
feels
that's
defined
in
the
draft
for
my
ppm.
M
So
there
are
few
methods
that
can
be
used
for
indicator
label
allocation
and
we
can
use
some
feedback
from
the
working
group.
So
one
approach
could
be
that
we
use
an
extended
special
purpose
label.
So
we
are
located
label,
maybe
say
1819:
there
can
be
a
network
programming
label.
That's
allocated
by
a
controller
for
this
and
controller
would
go
and
provision
all
the
end
cap
and
d-cap
notes,
or
there
is
signaling
where
the
cap
node
would
signal
it
to
them
to
the
end
cap,
node.
A
A
M
So
there
is
some
implication
on
housing
because
we
are
changing
the
MPLS
header
and
that
there
are
some
transit
nodes
that
may
use
the
MPLS
header
all
the
labels
for
guessing.
So
now
it
may
include
IOM
label.
There
are
some
notes
that
may
look
for
the
well-known
protocol
type
or
IP
hashing.
So
now
there
is
some
implication
there
as
well,
so
there
is
definitely
some
more
work
and
discuss
and
required
on
the
hashing
part
for
this
draft.
M
M
I
M
J
Miss
kitty
there
might
be
a
problem
that
characteristic
to
MPLS
and
the
Saran
POS
is
that
you
don't
know
where
this
came
from,
so
how
you
differentiate
the
source,
how
you
attribute,
if
you
especially
if
you
use
the
same
label
advertised
by
the
egress
for
all
the
possible
encapsulate
errs?
How
do
you
know
where
the
source
of
this
and
where
the
statistics
that
repeater
Soha
it.
M
J
So
in
one
of
the
possible
scenarios
that
we
have
multiple
domains
from
their
access
to
the
core
network
and
need
to
do
and
to
end
om
and
with
their
incremental
deployment,
for
example,
of
Sorrento
LS.
We
might
have
a
combination
of
a
Sorrento
s,
domains
and
IP
MPLS
domains,
and
then
we
considering
their
scenarios
how
this
can
be
organized.
Whether
we're
considering
two
models.
One
is
a
stitching
and
then
its.
J
There
exist
and
end
path
and
then
each
node
might
have
and
what's
called
as
an
S
path
and
for
their
nesting
domain
or
the
main
which
tunnels
from
s
are
MPLS,
it's
absolutely
a
path,
so
the
end-to-end
path,
identifier,
is
transparent,
so
it's
only
ingress
and
egress
are
dealing
with
it.
So
this
is
a
start
of
this
work,
the
reward
of
be
worked
on
and
there
will
be
new
updates.
We
appreciate
comments
and
your
interest
and
working
together,
I
hope,
I
save
some
time.
J
M
M
I
J
A
So
low
Anderson
I'm
going
to
talk
about
a
set
of
registers,
actually
a
subset
of
the
register
under
the
heading
LS
beeping
traceroute
register
in
the
Ayana,
and
it's
basically
all
the
TLD
and
subtly
reduces
plus
the
message
type
and
return
codes
and
one
more
I
have
had
as
long
as
I've
been
averaging
being
shared.
I
have
had
a
problem
here
by
understanding
the
rationale
between
behind
how
it
was
done.
So
let
me
try
to
explain
so
they
have
a
background.
It
started
out
at
43,
79.
A
A
The
8126
it's
more
or
less
contemporary,
with
the
80
29,
but
it's
rather
different.
There
are
subtle
difference
that
actually
affect
the
registers,
and
when
we
did
the
last
working
group,
RC
81-66,
we
actually
took
what
is
in
80
29
and
actually
put
that
into
the
register.
To
get
the
model
to
work
from.
A
43
79,
it's
a
very
technical,
solid
document
and
it
says
for
the
Ayana
registers
that
it
needs
to
values
from
the
specific
pacification
required
procedure
needs
to
be
done
by
an
experimental
error,
see
this
is
kind
of
an
oxymoron.
It's
not
wrong,
but
it's
kind
of
complicating
things
more
than
necessary.
So
this
is
what
it
looks
like.
Okay,.
A
A
So
why
use
something
that
says
we
can
take
documents
an
entire
world
to
actually
describe
what
we're
doing
for
type
of
ITF
documents,
so
80
29
changes
the
experimental
RSC
in
44
43
79
to
RC,
but
basically
this
didn't
ever
get
into
the
IANA
registry,
but
basically
it
it
has
the
same
problem.
We
have
a
lot
of
any.
We
can
do
any
IT
FRC,
but
it's
the
same
problem.
It's
much
fewer
documents,
but
no
one
actually
indicated
in
the
specification
required
sorry
about
the
animations.
A
So
we
have
one
more
thing
here:
saying
specification
requirement.
Rc
needed
in
note
is
actually
saying
the
same
thing
as
using
the
registration
policy.
The
RC
required
Stuart,
give
it
to
Korea
to
he
came
late,
so
we
can,
without
any
problem
changes
to
a
RAC
required
saying
the
same
thing
as
a
t29
says
here
are
the
list
of
possible
registration
procedures
and
I
actually
come
to
the
conclusion
we
should
use
them.
We
should
not
try
to
invent
our
own,
so
I
proposed
going
from
specification
required
to
RC
required.
A
An
alternative
would
be
to
get
high
def
review.
That
was
actually
exclude
the
quite
number
of
documents,
but
I
don't
think
it's
necessary,
so
the
update
we
have
made
is
actually
aligned
the
concept
with
8126.
We
added
code
point
for
experimental
use,
actually
I've
heard
from
people
that
actually
were
part
of
co-authoring
a43
79.
But
when
use
the
experimental
orosi
needed,
you
allocate
a
experimental
code
point,
but
that's
not
true.
A
You
act
you
it's
a
general
code
point
that
anyone
can
use
it's
just
it's
specified
by
an
experimental
RSC,
and
then
we
are
listing
all
the
registration
procedures,
even
if
they
don't
result
in
any
allocations,
even
if
they
prohibit
allocations.
So
this
is
what
the
register
would
look
like
with
what
I've
been
talking
about.
So
we
have
the
standard
standards
actions
we
have
the
RFC
required,
we
have
experimental
use
and
we
have
private
use.
The
last
one
is
properly.
Then
the
private
use
I,
don't
feel
team.
A
O
So
QED
compiler
I
was
probably
on
something
when
I
put
experimental,
RC
required
you're
completely
right.
There
should
be
RFC,
specify
so
I
mean
in
so
specification
required.
It
should
be
an
RFC
which
is
what
I
wanted.
I
don't
know,
why's
that
experimental,
so
I
received
a
part
is
I,
think
the
right
thing
and
then
the
others
are
other
is
right,
I
mean
they
haven't
changed,
I
think
for
the
private
use,
usually
back
in
the
day
the
when
I
was
actually
thinking
about
these
things.
O
J
A
H
H
H
So
this
is
the
desktop
here,
but
it
actually
requires
software
upgrade
for
the
nodes
that
are
deploying
the
network
and
rather
than
one
end
being
working
in
lockstep,
with
any
forwarding
definition
that
is
deployed.
It
started
like
behind,
like
epe,
for
example,
is
deployed,
but
there's
no
fax,
there's
no
NN
and
it
causes
a
lot
of
problems
and
when
we,
when
a
new
fact
type
is,
is,
is
defined
and
it
requires
the
software
upgrade.
H
H
The
main
point
here
is
just
sheer
amount
of
information
that
is
required
by
the
user
or
by
the
system
to
input
in
order
for
us
to
do
a
ping
to
remote
said
to
have
that
thing
set
up
and
it's
a
lot
of
information
and
same
thing
is:
is
that
I
mean
let's
say
we
define
prefix,
said
fact
for
prefix
it,
but
it
doesn't
work
for
flex
algo.
So
it
makes
a
stink
one
step
back
and
then
say
that:
can
we
simplify
this?
H
What
is
that
we
are
trying
to
achieve
what
is
the
data
model
or
for
anis
our
data
plane,
and
can
we
make
it
consistent
definition
consistent
with
the
data
model
of
the
SR
data
plane
so
that
it
works
once
for
all
for
all
the
fact
that
exist
today
and
they're
coming
future?
So
that's
the
problem
statement
to
simplify
the
process.
The
second
routing
is
all
about
simplification
of
network
operation
and
we
have
a
problem
here.
We
cannot
hide
that
problem.
H
The
other
thing
is
also
that,
if
you
look
at
enforce,
I
can
remove
that
that
when
you
have
clear
node
said
you
do
not
know
the
target
ahead
of
time,
because,
let's
say
if
you
go
between
the
node
2
and
then
4
and
node
3
are
peers.
Said
here
notes
said
so
they
share
the
said.
The
decision,
whether
the
message
will
end
up
with
more
number
4
or
no
number
3,
depends
on
load
balance
again:
node
2
and
node
number
1
who's.
H
H
We
have
method
that
okay
for
pereira,
Jensi
just
put
0
so
I
mean
we
have
X
to
get
around
it,
but
we
don't
have
a
solution
in
the
other
drug
the
shader
presented.
We
said:
let's
send
all
the
possible,
there
could
be
many
many
and
then,
and
then
hope
one
of
them
would
be
there.
Is
it
increased
complexity
and
it
makes
when
I'm
really
like
behind
so
solution?
Is
really
look
at
the
SR
get
a
plane,
the
SR
data
plane
simple,
you
have
a
segment
idea
and
that
segment
idealized
by
somebody.
H
This
is
the
common
denominator.
It
did,
it
doesn't
depend
on
the
seat
type.
What's
the
type
is
in
all,
it
does
not
depend
on,
say,
step
back
and
then
simplify
the
facts,
definition
that
works
for
all
sit
type
now
and
that
may
come
in
future
as
well,
since
this
is
build
on
the
data
model,
this
will
work
for
sure
now.
Here's
the
definition
of
the
sit
again.
Yes,
our
CID
and
the
C
designer
for
some
simplification
for
some
use
cases
like
bindings
it.
H
We
also
include
optional,
LSP
endpoint,
which
is
if
I
know,
who
should
be
responding.
I
can
put
it
that
information,
but
I'm
not
going
to
go
this
busy
slide,
but
there
is
an
exercise
that
is
done
whether
this
definition
works
for
all
sit
types
that
we
have
defined
or
we
have
to
define
future
or
it
doesn't
work.
The
answer
is
yes,
I
mean
you
can
review
it.
I
can
come
back
to
it
if
there
are
more
if
there
are
more
time
remaining,
but
I'll
go
and
illustrate
a
few
example
first.
H
So
in
this
example,
we
have
prefix
SID.
Now
here
the
prefix
said,
which
is
for
flux,
I'll
go
0
and
flexible
128.
Current
definition
of
prefix
said,
does
not
work
for
any
alga,
the
flex
will
go
0
and
there
they
will
attempt
to
modify
the
fact,
but
it
goes
through
the
same
problem.
So
how
it
will
be
done
is
basically
mode.
1
was
just
simply
say:
I
want
to
ping
a
label.
H
Edward
ties
binode
r8.
This
is
the
only
information
that
is
carried
now.
We
pick
here
agency
said
because
agency
it
is
difficult.
It
is
for
prefix
F.
So
when
it
comes
to
no
date,
no,
let
will
say
yes,
I
am
the
assignor,
for
this
said,
and
yes,
this
level
match
my
table
and
then
response
goes
back
to
the
originator.
Now
we
take
this
example
and
try
to
address
the
PAL
region,
see
which
is
not
addressed
by
the
RLC
today.
The
better
agency
case
cannot
be
verified
if
in
today,
because
it
saves
agency
is
zero.
H
Zero,
so
here
we
have
ninety
seven,
seventy
eight,
which
is
a
pair
of
agency
between
link,
1
and
Link,
two
between
node,
7
and
mode
8.
In
in
this
example,
the
fact
would
be
choosen
again.
This
is
agency
they
advertise
by
note
7,
ok
and
this
agency
value
is
93
78.
So
this
is
what
is
the
bare
minimum
information
that
is
required
when
the
packet
goes
to
note,
eighth
note
8
and
it
can
optionally
contain
the
end
point,
but
end
point
is
not
mandatory,
but
packet
comes
to
note.
Eighth
note.
H
8
can
then
verify
that
this
is
not
a
local.
This
is
a
rise
by
7
so
see
this
7
is
my
3
neighbor,
and
it
verifies
that
it
is
this
matching
on
the
interface
I
receive
this
packet
on
with
the
pair
region,
see
there
is
a
row
ties
by
that
node,
so
in
this
case
93
78,
so
it
can
easily
verify
from
his
low
local
table
that
information,
so
it
works
for
parallel,
see
it
will
work
for
the
EPE
cases
as
well
and
all
now
this
is
peer.
H
Note
said
it
works
the
same
way
because
the
common
denominator
is
node.
7
is
eric
izing
and
I
jen
Z,
which
is
93
78.
Now
packet
comes
to
node
7
and
it
can
either
go
to
node,
8
or
88.
Either
way.
The
receiver
will
say
that
the
arrow
tiser
is
not
me.
It
is.
Let
me
check
if
it
maelstrom
labor,
does
it
allocated
this
agency?
H
Did
I
receive
packet
on
the
right
interface
or
not,
and
then
it
will
respond,
so
it
doesn't
matter
how
seven
is
doing
is
load
balancing
the
packet
may
end
up
at
node
eight,
and
it
would
respond
if
it
right
interface.
Choosen
packet
may
end
up
no
deity,
yet
it
will
respond
if
his
right
interface
is
choose
him
and
if
his
end
of
the
wrong
interface
wrong
neighbor,
then
it
the
response
will
be
negative
as
well.
So
all
the
validation
would
be
done
and
it
works.
H
So
in
a
nutshell,
we
are
trying
to
simplify
the
segment
routing
FAQ
by
defining
fact,
which
is
goes
back
to
the
common
denominator
of
the
data
model
in
segment
outing.
It
will
do
it's
done
for
ease
of
operation,
because
then
user
don't
have
to
input
all
different
parameters
needed,
and
we
saw
the
complexity
of
all
the
FAQ
and
it
is
defined
such
a
way
that
I
mean
it
would
be
consistent
with
any
future
segments
that
would
be
defined
and
4-iron
registry
I
will
skip
this.
This
was
a
question
lower
from
you
on
the
mailing
list.
H
We
already
discussed
this,
I'm
karis
a
spring
working
group
in
the
other
discussion
that
will
discuss
also
in
amperes
in
spring,
but
the
fact
definition
in
more
especially
spring,
but
we'll
let
chair,
decide
how
they
would
like
to
proceed
with
that.
They
would
like
to
ask
for
feedback.
There
is
some
discussion
that
going
on
with
sure,
though
we
try
to
connect
and
pass
the
prophets,
but
well
not
able
to,
and
then
we
have
meeting
tomorrow
that
we
will
discuss
more
on
this.
K
G
P
H
K
K
1
will
not
know
which
interface
chorus,
so
I
am
talking
more,
let's
say:
BGP
egrets
adjacency
seeds,
so
today
in
BGP
addition,
since
it
when
it
is
advertised
in
the
BBC
pls
update
we,
along
with
our
label,
we
also
advertise
the
if'
index
of
the
router
2
from
where
the
packet
will
go
out,
and
there
is
always
also,
if
already
a
field
which
tells
out
where
what
is
the
if'
index
on
which
node
3
will
receive
the
packet.
So
that
way,
note
1
or
knows
by
a
controller
or
something
it
can
know,
given
a
level.
K
H
Answer
then
yeah.
What
I
said
is
that
node
1
does
not
know
how
node
2
would
load
balanced
any
given
pink
packet
based
on
the
hashing
that
is
putting
in
the
pink
packet
whether
it
will
land
on
node,
4
or
node
3.
It
would
not
know
okay,
so
the
only
way
to
do
is
to
pack
the
entire
list
of
possible
peer
node.
In
fact-
and
this
is
bringing
complexity
for
both
ingress
and
egress,
how
you
expect
the
operator
to
input
this
data?
H
I
mean
it's
just
going
out
of
proportion,
and
it's
too
complex
for
user
to
use
it
be
world
will
for
simplicity.
We
are
asking
not
more
complexity,
I
mean
think
of
it.
I
mean
ICMP,
you
don't
need
any
of
this.
I
mean
people
are
operating,
I
could
be
six
network
for
it
for
ages
and,
and
you
don't
get
into
these
delicious.
H
They
simple,
because
when
the
message
is
received
by
receive
note,
let's
see
if
I'm
not
for
I'm
gonna
check
who's.
The
advertiser
of
this
said,
if
not
me,
okay,
that
means
the
room.
It's
a
remove.
It
is
an
agency,
see
that's
type.
Okay,
it's
the
guy
with
a
tie.
Thing
is
my
upstream
neighbor.
Yes,
what
is
the
value
if
you
did,
he
wrote
I
this
value
for
this
sit
on
the
interfere,
I
receive
the
packet
on.
H
J
H
Edges
that
that's
that's
an
alternate
problem,
that's
not
the
problem,
we're
trying
to
solve.
That's
the
problem
exporting
the
tree.
Ok
and
then
you
can
change
the
load,
balance
a
parameter
and
you
can
achieve
this.
But
but
the
point
your
problem.
The
point
is
that
when
to
respond
back
to
the
tie
with
the
da
snap
DD
map,
it
can
include
the
exert
interface
information
because
it
knows
how
it's
going
to
do
the
load.
E
So
the
comment
that
I
have
is
the
problem
that
you
talked
about
in
the
beginning
of
your
slides
that
we
have
these
EPS
it's
implemented,
but
we
don't
have
om
facts
implemented
for
those
that's
a
real
problem,
but
there
is
different
way
for
solving
it
right.
So
when
we
define
a
new
set,
we
also
define
the
fact
for
it.
E
We
make
it
mandatory
right
so
that
that's
another
way
of
solving
that
problem,
so
that
those
all
those
upgrade
problems
and
other
things
you
talked
about
that
won't
wouldn't
exist
if
we
had
done
it
this
way,
but
we
didn't
do
it
that
way
and
we
are
in
the
situation
now
at
least
going
forward.
We
can
make
sure
you
know
any
a
new
cell
that
comes
in
comes
along
with
no
effect
for
that.
It's.
Q
E
Not
possible,
but
that's
not
necessary.
You
know
this
information
is.
It
has
been
advertised
in
the
network
and
there's
one
guy
who
knows
this
information,
one
node
in
the
network
who
has
all
this
information
and
that
should
be
capable
of
preparing
the
FAQ,
which
is
totally
incrementation
right.
Well,.
H
H
E
E
H
E
E
E
H
E
E
H
E
E
H
K
Mukul
from
juniper
networks,
one
comment
for
B
for
BGP
ep6.
Those
seats
are
special
in
the
way
that
if
you
talk
about
three
routers
one,
two
and
three
those
labels
are
allocated
at
node,
2,
node,
4,
node,
3
and
node.
3
has
no
knowledge
about
that
label.
Ok,
so
the
flow
chart
which
you
have
right
that
will
fail.
I
know.
K
K
K
C
All
right
from
Jennifer
I
did
review.
Your
draft
gave
you
comments
on
the
list
yeah
in
general
I.
You
know
I'm,
okay,
with
generalizing
the
effect,
but
not
dropping
information
that
this
allow
us
to
do.
Control
table
data
play
validation,
so
I'm,
ok
with
abstraction,
but
with
abstraction
you
lose
data
that
you
need
to
validate.
So
if
you
don't,
if
you
can
find
a
sweet
spot
in
the
middle,
but
do
you
have
the
information
and
you're
generalizing
a
monkey?
But
there
is
a
couple
of
other
comments
also
that
I
gave
yesterday.
H
C
H
H
A
I
prefer
that
we
actually
let
people
talk
to
point
before
they
actually
respond
and
not
interrupting
talking
actually
about
those
Safari
Sandra.
They
do
the
same
thing
this
time
it
was
useful
but
think
about
in
the
future.
We
let
people
talk,
the
people
make
that
point
before
you
try
to
correct
them.
A
A
So
if
I
remember
correctly,
as
it
said,
the
synonymous
flow
label
drafts
and
it's
gone
there
now
I'll
go
back
and
I'm
not
going
to
use
that
slide.
We
have
three
different
drafts
that
the
main
author,
Stewart
told
me
are
stuck
and
I
try
to
figure
out
how
and
I
try
to
write
down.
What
was
the
problem
and
Stewart
says
you
got
it
almost
right,
so
I
kind
of
stepping
back
and
let
Stewart
try
to
explain
what
the
issue
is
and
where
we
should
go
from
here.
D
D
Now
several
corporate
lives
that
go
I
stood
at
the
front
of
one
of
these
meetings
and
introduced
this
control
graft
very
simple
scheme
that
largely
George
and
I
put
together
a
lot
of
work
from
George,
and
it
doesn't
require
us
to
touch
any
of
the
other
control
planes.
So
when
I
stood
up
here,
people
said
no,
no,
no
Stuart.
We
need
an
rsvp-te
solution.
We
need
an
LDP
solution.
Presumably,
we
nee
now
need
a
solution
for
the
iGPS
and
I
have
not
had
time
or
the
funding
or
anything
to
do
this
and
I've
been
requesting.
D
Other
people
help
me
here.
So
I
can
do
one
of
two
things:
three
things
I
suppose
we
could
feel
the
tank
filled
in
here
as
reserved
and
take
a
risk
that
we
got
the
size
right
or
we
could
control
plane
that
I
originally
proposed,
and
if
people
want
another
control
plane
eventually
they
can
write
it
or.
D
D
Right,
so
so
so
this
is
the
this
design
here
actually
has
a
lot
of
Merit
right,
because
it
doesn't
require
you
to
touch
the
real
control
planes.
It
sits
underneath
the
real
control
planes.
And
yes,
if
the
OEM
system
says
it
wants
to
do
this
operation,
then
a
label
is
transfer,
is
sort
of
silently
inserted
to
replace
it,
so
that
actually
has
a
lot
of
Merit
as
a
system
and
no
one
I
don't
and
then
people
I,
don't
think
we
don't
know,
look
at
the
minutes.
People
didn't
object
to
that.
D
A
I've
been
struggling
with
this
for
some
time
and
what
I'm
trying
to
understand?
Is
it
that
way
of
going
forward
with
something
very
similar
to
what
you
have
here,
simple
solution,
but
actually
putting
the
hooks
in
to
actually
allow
anyone
that
wants
to
do
very
much
more
than
want
to
do
that
and
how.
P
P
A
So
that
left
us
with
another
question
but
I
actually,
okay,
actually
we
have
been
taking
these
two
documents
through
the
working
group
and
people
say,
wouldn't
obsess
they
want
to
progress
on.
So
that's
why
I'm
pushing
for
a
solution
that
we
actually
make
them
RFC's.
If
there
are
suggestions
that
while
this
is
not
implemented,
it
will
never
be
implemented.
So
why
should
we
bother
if
that's
the
but.
R
D
D
A
Q
I
haven't
looked
at
the
draft
in
several
years.
Indeed,
I
am
retired
anomaly
anyway.
So
do
we
have
a
requesting
process
ID
in
the
initial
exchange,
because
if
we
did,
then
you
know
somebody
who
wants
it
to
be
requested
through
RSVP
or
LDP
or
BGP
could
make
build
an
implication
that
says:
okay,
ours
he's
going
to
request
this
thing
and
then
the
request
comes
back
to
RSVP.