►
From YouTube: IETF104-SPRING-20190328-1350
Description
SPRING meeting session at IETF104
2019/03/28 1350
https://datatracker.ietf.org/meeting/104/proceedings/
A
A
A
So
two
slides
on
the
joint
progress
since
the
IGF
103,
we
have
draft
MSD
C,
which
has
been
approved
by
iesg
Antron
status
is
our
editorial,
so
it's
mostly
finished
for
for
working
group.
Then
you
have
the
assalam
PS
document
which
we
have
submitted
to
iesg
for
publication,
and
current
status
is
waiting
for
writer
and
then
we
have
NP
dispatch
segment
with
which
we
have
just
adapted
as
a
working
document.
A
For
the
remaining
document
a
duck,
we
have
a
service
six
network
programming
which
is
currently
in
coal
for
adoptions
for
at
least
a
week.
So
please
please
comment
on
on
radio
we're
segment
routing
policy
which
would
be
ready
for
watching
what
Nebraska.
Also
we
have
SRA
owner,
which
also
should
be
ready
for
one
last
column.
A
B
Rs.12,
okay,
so
my
name
is
Daniel
Bell,
Canada
I'm,
going
to
present
similar
rotting
point-to-multipoint,
but
just
before
I
start
just
a
few
points.
Right
now,
one
of
my
co-conspirator
woman
is
giving
the
same
presentation
on
the
pin
working
group.
We
are
collaborating
a
lot
also
with
jeffrey
from
Jennifer
who's,
I
think
he's
in
the
room
somewhere,
yeah.
Okay.
B
B
B
We
are
also
open
and
there's
other
use
cases
for
point
to
any
kind
of
point
to
multi-point
MP
as
multicast
technologies
also
valid
the
goal,
and
the
motivation
is
also
to
simplify
multicast
by
removing
some
particles,
RSP
Tem
LTP,
while
providing
traffic
engineering
similar
thing
Boyce.
These
attributes,
which
also
refer
to
the
signal
routing
policy,
drive
it
also
to
make
the
network
infrastructure
friendly
programmable.
B
Okay,
so
SR
education,
it's
a
point-to-multipoint
segment
that
connect
a
root
node
to
a
set
of
leaves
node
in
the
segment
rotting
domain.
So
it
is
expected
that
this
is
to
be
run
in
the
same
irritating
environment
we
define
two
type
of
point-to-multipoint
segments,
spray
and
tree
said:
stray
point-to-multipoint
segment
enables
a
root
node
to
directly
replicate
a
packet
using
a
sr
path
to
each
leaf.
B
We
rely
on
the
current
asar
policy,
as
I
mentioned,
so
the
SR
education
policy
is
a
variant
of
NSR
policy
as
the
Jeff
as
a
per
se
stated
and
applies
equally
to
spray
or
tree
set
point
to
multi-point
segment
unless
explicitly
specified
it
can
be
provisioned
on
the
root
or
on
via
a
controller,
everything
happens
at
the
contour.
Finally,
it's
defined
the
following
element:
the
root
node.
This
is
the
adhan
of
the
point
multi-point
segment
leaf.
B
So
in
a
natural
SR
Education
use
extensively
PC
everything
happened
at
the
PCE.
It
is
expected
that
it
will
run
in
the
Syngman
writing
environment.
As
I
explained
explained,
a
tree
set
point
to
multi-point
cement
packet
replicate
occurs
at
the
root
node
and
on
the
replication
note
or
worse,
the
leaf
node.
He
could
sue
to
the
drawings.
B
Nothing
happened
at
the
transit
node
or
anywhere.
If
there
is
more
than
one
transit
note,
a
SR
vacation
policy
can
be
Estancia,
'td
and
maintained
in
a
centralized
fashion.
Using
a
path,
competition
element,
as
I
mentioned,
no
Barnaby
I
will
be
used
to
delete
and
update
the
tree.
A
tree
can
be
accentuated
from
a
PC
via
an
account
piece.
A
PGP
te
cetera
local
protection
and
pap
petition
can
be
applied
using
a
cell
policy
compute
computed
at
a
PC.
B
C
B
A
D
D
A
First,
we
take
some
time
is
through
to
see
there
are
comments
on
the
list
and
we
discuss
between
Charlson,
but
I'd
normally
wait
for
the
previous
call
for
adoption
to
finish,
which
is
calling
network
programming.
But
we
will
definitely
come
back
to
you
in
a
few
weeks
itself,
Schumann's,
but
before
the
next
meeting
and.
E
F
G
Okay,
hello,
everyone:
this
is
an
action
I'm
going
to
present
this
signal,
routing
based
upon
it
latency
they
stripped
submitted
to
attend
networking
group
represent
here,
because
that
we
tried,
because
we
are
trying
to
use
Sigma
routing
to
support
on
the
latency.
So
we
would
like
to
solicit,
solicit
comments
and
feedback
from
this
working
group,
and
maybe
I
follow-up
work
to
be
done
in
this
working
group.
G
In
this
way,
you
can
see
that
a
packet,
where
only
is
the
one
cycle
at
the
hop
on
the
pass,
so
you
can
see
that
the
end
to
end
on
the
latency
can
be
very
easy
to
calculate
it.
That
is
the
hobbs
times
they
cycle,
so
you
can
see,
does
Sekulow
feels
very
smart
solution,
but
it
also
has
its
own
limitations,
so
it
is
more
suitable
for
small
networks
where
the
dink
delay
is
trivial
and
the
processing
delay
in
general
is
small.
G
Otherwise,
more
balloons
has
to
be
reserved
to
as
gulp
and
for
each
cycle
in
order
to
make
sure
that
a
secure
F
can
be
can
work
as
designed
so
there's
an
undergrad
in
the
networking
group,
which
is
called
not
skilled,
deterministic
IP
networking
it
is
using
tend
to
be
a
framework
document
that
introduces
several
ways
to
to
implement
cycle
based.
That
is,
a
latency
can
control.
So
you
want
to
know
some
details
to
read
a
draft.
G
So
hairy
description
is
written,
quite
sweet
world.
We
call
it
psycho,
specifically
and
forwarding,
as
I
introduced
in
the
previous
slides
CTO
of
does
not
consider
intelligence,
it'll,
link,
delay
and
other
constraints,
and,
if
just
assume
that
a
packet
will
only
one
like
one
cycle
and
it's
at
all,
but
this
will
not
work
for
a
lab
work
really
link
delay
is
not
trivial
and
the
percentile
a
and
the
other
constraints
are
not
can
be.
G
So
this
way
we
can,
you
know,
direct
how
to
transmit
a
packet
and
on
the
path
and
then,
depending
on
this,
it
can
be
achieved.
Sabino
figure
is
our
example
shows
how
this
circle.
If
it
works,
you
can
see
that
Rebecca
X,
it
is
supposed
just
supposed
to
be
transmitted
from
load
a
through
BC
and
to
Elodie
and
after
the
pass
calculation,
the
Peck
is
supposed
to
be
send
it
out
in
cycle
1
at
node,
a
and
the
in
cycle
or
3
an
don't
be
a
cycle
for
a
lot
C
and
psycho
sex
LOD.
G
G
G
So
by
attaching
or
sight
a
psycho
safe
segment
to
the
package
it
can
specify
which
interface
in
which
psycho
or
package
should
be
sent
out.
So
by
attaching
a
list
of
sites
acumen
to
the
packet,
then
you
can
specify
the
Sun
in
cycle
the
each
load
along
the
path
and
and
result
introduce
any
per
flow
state
at
the
intermediate
intermediate
load
and
request
note
and
the
most
importantly,
the
bundle
latency
can
be
achieved
in
this
way.
G
So
here
it
is
unknown.
Instead.
Example
shows
how
SR
amperes
can
be
used
to
support
CSUF
I
think
the
figure
is
very
straightforward.
A
big
ingress
we
can
use
the
classical
computation
is
more
information
to
specify
the
satellites
and
attach
the
subtlest
to
the
packet
and
the
fact
that
list
will
be
used
by
each
load
on
the
path
to
determine
whether
and
which
which
cycle
should
be
sent
out.
G
D
D
G
I
think
the
cycle
can
be
grow
favor,
but
it's
not
necessary
to
allocate
you
know
for
each
cycle
to
allocate
corresponding
seeds.
We
can
be,
for
example,
fine
interface
based
our
practice.
I
think
is.
It's
may
be
enough
to
allocate
three
sales
for
interface.
Then
we
can
use
the
seeds
in
our
cyclic
way,
for
example,
our
cycle
one.
You
seek
one
cycle,
two,
we
will
see
two
cycles,
reuse
century
and
so,
for
you
said
one
can
use
this
in
a
cyclic
whaling.
H
So
Himanshu
from
sierra,
I
have
a
clarification
question
as
well.
How
would
an
intermediate
known
would
know
that
all
the
head
and
would
know
that
this
cycle,
I
was
not
already
filled-
I
mean?
How
can
you
be
so
deterministic?
There
could
be
other
non
other
traffic
also
going
on
right,
so
it
could
be
very
difficult
to
know
that
particular
packet
will
go
at
the
right
time
at
right
cycle,
because
that
middle
node
would
also
be
doing
something
else
right.
H
G
Cycle
can
only
be
used
by
the
critic,
homelessness
for
other
no
best-effort
traffic.
You
can
use
other
seats,
fix
them
for
the
best
terraces
man
maybe,
but
this
is
the
only
use
for
those
traffic
cricket
records
it
would
in
to
work
with
other
segments
Road
in
traffic
yeah.
You
have
to
differentiate
dead
net
flow
and
non-tender
flow,
so
we
used
the
same
to
differentiate
that
so
that
means
you
have
to
your
different
seat
for
different
traffic:
okay,
yeah
I.
Guess
the
big
girls
are
in
your.
G
D
G
D
G
D
G
Means
there
are
some
arrow
happen
on
that
load.
That's
yeah!
That's
a
no
failure
condition.
So
then
that
doesn't
have
some
requirement
on
the
traffic
pattern.
That
means
the
traffic
has
been
no
transmitted
as
known
as
a
way
it
required.
That
means
for
the
the
rate.
The
packing
rate
is
is
no
before
ways
established
pass.
A
G
I
think
here
we
just
want
to
know,
solicit
more
comments
and
feedbacks
from
the
signor
important
point
of
you
know
whether
it
is
available
not
during
the
enhancement
and
optimizations
and
suggestions,
and
there
may
be
known
as
as
I
said,
inner
side
not
sites
may
be
follow-up
works
to
be
telling
this
watcher
working
group.
That
means
there's
some
no
need
to
some
extensions
today,
IDP,
and
something
like
that.
You
know
to
advertise
this
psycho
segment
can.
G
A
A
G
Yeah
I
think
we
actually,
we
have
considered
the
existing
segment
segment
types
that
we
think
there
is
better
to
know
to
define
new
type
of
segments,
because
this
is
you
know
the
tenor
traffic
has
to
be.
You
know,
field
different
from
the
other
packet,
so
they
have.
You
need
to
have
some
dedicated
segments
to
differentiate
that.
A
G
A
I
Domine
from
Holly
I
think
until
now,
because
this
is
the
resource
to
occur
in
here,
so
the
bounty,
the
latency
is
the
Pahlavi
the
agency.
So
we
think-
and
now
we
first
think
about
this
agency
segment,
but
maybe
later
we
can
think
more
about
other
types
of
the
segment.
If
we
use
the
relate
here
is
the
chemical
and
here
this
is
for
worthy.
A
So
we
have
a
ALPHAN
our
discussion
about
segment
routing
on
service
chaining.
We
have
two
presentation
from
to
draft.
The
good
is
not
ready
to
detail
the
draft
not
to
discuss
the
graph
but
to
provide
an
overview
to
to
initiate
the
description.
So
we
have
two
to
sort
of
six
minutes
one
for
each
draft
and
then
the
main
goal
is
to
discuss
the
subject,
and
then
we
have
20
minutes.
J
One
two
good
so
I'm
from
San,
Francisco
and
I'm,
going
to
present
disrupt
on
on
the
alpha
of
my
cultures.
This
is
the
service
programming
with
segment
roti,
so
first
the
scope
of
this
draft.
In
this
draft
we
define
basically
two
things:
first,
how
to
process
a
service
segment
with
MPLS
sign
with
sigma13
v6,
and
also
how
to
enter
service
metadata
that
would
be
attached
to
the
packet.
We
also
illustrate
into
draft
how
service
segments
can
be
integrated
with
other
types
of
segments
of,
for
example,
underlay
of
topological
segments.
J
But
the
fact
that
a
service
can
be
associated
with
a
segment
is
not
something
new.
The
concept
of
service
segment
has
been
defined
in
RFC,
84,
2
and
there's
a
monitoring
policy
architecture
in
draft
IETF
spring
segmentation
policy
also
allows
for
the
combination
of
service
and
topological
segments
in
the
same
seed
list.
J
This
combination
is
actually
a
seamless
integration
between
the
between
the
segments
and
it
is
important
that
the
service
segments
and
the
other
segment
types
do
not
negatively
impact
each
other
when
combining
the
same
segment
list.
So,
for
example,
the
presence
of
a
service
fragment
in
a
message
list
should
not
make
it
more
complex
from
your
endpoint
processing,
underlay
segment
to
do
their
packet
processing
and
similarly,
the
presence
of
non
service
segments.
So,
for
example,
underlay
segment
should
not
limit
the
capabilities
of
service
segments.
J
J
We
also
define
the
some
proxy
behaviors
former
legacy
services
that
would
not
natively
support
segment
routing,
and
for
this
we
just
have
the
proxy
doing
the
segmentation
processing
on
behalf
of
the
of
the
service
and
delivering
to
that
service.
That
network
function
the
packet
without
dsl
information
or
hiding
some
howdy-hi
information
from
the
party
that
is
delivered
food
to
the
service.
Then
we
move
to
the
next
segment
isolation.
J
So
these
two
types
of
segments
can
be
a
well
babe.
The
proxy
service
and
the
segmentation
capable
service
can
be
combined
in
the
same
seed
list.
Obviously-
and
this
would
allow
for
for
progressive
deployment
of
acceptable
services
if
service
becomes
as
acceptable-
and
it
was
before,
behind
access
behind
the
proxy,
then
it
is
sufficient
to
deploy
the
new
service
and
we
configure
the
head
end
on
the
entity
that
computes
the
segment
to
use
the
new
segment
instead
of
the
proxy
statement
in
terms
of
metadata.
J
So
this
is
mostly
similar
to
other
proposals
that
have
been
inform
for
service
training.
The
metadata
is
carried
within
the
packet
headers,
it
is
set
by
the
Helen
or
by
an
intermediate
solution
point
that
would
need
to
such
a
metadata,
and
it
is
usually
usable
but
I
intermediate
salvation
point
that
would
need,
but
what's
important
is
that
the
metadata
lookup
is
not
it's
not
automatic.
At
the
enemy
segment.
Routing
end
point:
it
is
triggered
by
the
segment
redeeming
function.
J
J
A
K
Just
want
to
give
you
a
brief
overview
of
a
draft
that
I
put
together,
which
was
basically
a
draft
that
allowed
us
to
do
what
we
would
call
stateful
service
training
as
opposed
to
stateless.
So
what
Francois
kind
of
presented
was
really
a
stateless
implementation
where
we're
using
SRP,
six
source,
routing
capabilities
to
do
the
service
chaining.
But
obviously
there
are
cases
where
we
have
service
chaining
using
the
encapsulations
that
have
been
defined
by
the
SFC
working
group,
which
I
guess
we
can
call
stateful,
because
we
need
a
certain
amount
of
State
at
the
routers.
K
K
So
the
draft
goes
into
the
details
of
that.
I
still
need
to
update
to
add
a
function
that
basically
has
all
the
semantics
that
says
what
to
actually
do
with
the
packet
as
you
strip
off
the
segment
routing
header,
because
obviously
we
need
to
cash
that
we
need
to
be
able
to
push
it
back
on
with
the
pointers,
moved
and
so
forth.
All
of
that
detail
needs
to
be
put
into
the
draft
and
I'll
do
that
before
the
next
IDF.
K
So
really,
the
purpose
of
this
is
really
give
the
combination
of
second
gravamen
essays
for
a
transport
independent
SFC.
But
the
key
point
here
is
that
it's
not
in
competition
with
an
S
r
v6
service
chain,
and
these
are
two
independent
ways
of
doing
this.
So
they're
kind
of
complementary,
so
I,
don't
want
to
get
into
a
one,
is
better
than
the
other.
K
It's
really
going
to
be
dependent
on
what
a
particular
operator
wants
to
deploy,
and
so
next
steps
with
this
is
request
working
period
option
in
spring,
and
we
would
want
to
pursue
this
in
conjunction
with
what
Francois
just
talked
about
and
come
up
with
those
solutions
that
cover
all
at
this
space
and
we'll
continue
to
refine
the
technical
details
in
the
draft
as
we
move
forward
once.
That's
it
any
questions
specifically
on
that
clarification.
Questions
know.
A
Discuss
a
very
general
subject
of
doing
service,
chaining
it
with
Sigma
pudding
and
what
we're
saying
is
for
deployment
scenarios.
So
the
goal
is
not
to
to
choose
one
because
which
media
may
be.
We
didn't
need
different
depending
on
the
different
scenario.
So
before
we
we're
saying
is
one
where
any
section
s
are
coexist
so,
as
is
used
as
transport
from
one
service
for
water
to
another
one
to
the
next
one,
and
that's
the
scenario
described
in
June
draft
in
Section.
Two,
please
correct
me.
A
A
A
Because
you
have
no
better
data
and
you
have
sao
web
service
or
SSL
proxy
and
again
as
far
as
our
error
on
the
first
post
first
one
is
the
same,
but
with
metadata
added
in
the
data
been
specific
structure
native
structure,
so
one
specific
structure
for
SIDS
on
one
was
one
specific
structure
for
srt-6.
So
we
are
not
using
any
search
for
metadata,
but
there
is
a
specific
product.
L
A
M
When
I
drink
from
Nokia
I
just
want
to
complain,
it's
not
a
question
or
a
comment.
I
just
want
to
complement
what
Jim
set
is
actually
those
two
proposals,
because
I'm
actually
a
code
on
both
they
are
actually
having
different
environments
as
a
use
case
right
and
I
think
what
I
they
are
very
complimentary
to
each
other.
One
requires
the
I.
M
Assuming
there
is
no
clock
see
one
will
require
to
have
the
service
function
to
support
Anna's
age,
whereas
in
the
other
case
you
have
to
support
segment
routing
so,
but
you
will
see
that
there
is
I.
People
have
expressed
interest
in
both
of
them,
so
I
believe
that
there
is
need
for
both
of
them
and
they
are
very
complementary
to
each
other.
N
N
We
actually
have
the
choice
here
to
not
have
two
ways
of
doing
the
same
thing
in
the
same
network.
I
have
no
particular
feeling
for
going
with.
Let's
use
n
SH
and
say
that
is
the
way
you
do
all
service
function,
training
and
I
have
no
particular
feeling
saying:
let's
use
that
solving
six
to
always
do
that,
but
I
think
that
there's
this
decision
point.
N
K
K
The
simple
answer
from
my
perspective
is
that
I
have
operators
asking
for
both
and
because
I
have
operators
asking
for
both
then
I'm
gonna
provide
both
so
I
and
because
I
don't
see
them
as
being
they're,
not
fighting
against
each
other
they're
complementary.
So
I
really
don't
personally
have
an
issue
with
with
pursuing
both
both
solutions
that
will
satisfy.
You
know
the
operators
that
are
asking
me
for
this
kind
of
service.
D
E
D
O
If
I
may
comment
Chow's
hopping
from
Ericsson,
whose
name
was
recently
invoked
and
I
actually
agree
with
the
invocation
I'm
not
coming
up
here
to
disagree,
it
is
the
whole
point
of
chartering.
The
SFC
working
group
and
defining
in
SH
was
to
define
an
interoperable
mechanism
to
represent
service
paths
and
carry
metadata
and
interoperable
mechanism
across
transports.
O
Asar
v6
is
one
such
transport
MPLS,
including
segments
routed.
Mpls,
is
another.
Such
transport
Ethernet
is
another
such
transport.
We
can
go
on
and
on
there's
a
lot
of
them.
That's
why
we
wanted
an
interoperable
mechanism
to
carry
this
information,
creating
a
different
mechanism
to
carry
this
kind
of
information
that
does
not
use
nsh
does
not
meet
this.
It
fails
the
Charter
that
we
as
a
community
spelled
out
as
desirable,
carrying
nsh
in
SR.
V6
is
entirely
fine.
O
I
expected
I
wanted,
but
taking
an
SH
apart
and
putting
the
information
into
an
SRH
header
and
some
other
piece
of
TLV.
As
some
another
proposal
did.
There
was
another
tunnel
protocol
that
said
we're
gonna
just
take
the
metadata
apart
and
put
it
in
other
TL
DS,
and
you
can
put
it
back
together.
If
you
need
to
interoperate
excuse
me,
the
whole
point
was
to
avoid
an
end
squared
and
capsulation
problem,
so
I
have
a
problem
with.
P
O
Q
C
Come
from
Maui
a
circle
cause
of
the
stainless
as
a
paste
as
a
draft
I
think
there
are
lots
of
difference
between
these
two
drafts
because,
as
we
know
that
the
solutions
described
in
and
Jamie's
draft
is
kind
of
stifle
solution,
which
means
that
the
SFF
we
maintain
the
information
of
forwarding
and
then
in
this
way
we
do
not
need
to
change
the
advance,
which
means
the
fast
can
be
can
can
support
an
SH
as
it
this
right
now.
So
it
is
a
complementary
solution
for
stateless
SR
based
SFC
yep.
Thank
you.
N
It's
agent
again
and
I
think
stateless
may
well
be
the
absolute
key
to
all
of
this.
So
so
Joel
is
right
that
the
SFC
working
with
charter
to
do
SFC,
but
Spring
has
also
been
chartered
to
do
service
training.
It's
risen,
I'm
clutching
my
laptop.
It
says
that
one
of
your
one
of
your
work
items
is
source
relative,
stateless
service
training.
So
unless
the
ADEs
want
to
tell
us
that
there's
a
screw
up
in
the
chartering
and
one
of
them,
one
of
the
shelters
has
to
be
fixed.
R
Daniel
Daniel
Canada.
My
point
of
view
in
this
is
a
goal,
is
to
simplify
the
network,
but
I
think
that,
for
example,
Jim's
traffic
helps
with
inter
ability
of
existing
vnf
that
already
support
an
SH.
You
can't
just
carry
them
across
so
being
able
to
have
both
is
a
good
use
case.
So
I'll
do
it.
Joel
talked
about
the
SSU
working
group,
I
agree
with
any
units
part
of
the
charter
of
spring
and
having
that
ability
to
interrupt
like
this
is
I.
Think
it's
a
goose
useful
for
a
useful
solution
for
us.
I.
A
You
say
that
we
have
a
solution
for
service
chaining,
it's
called
NSA.
True
we
don't
have.
We
don't
need
a
second
one
but
I
think
at
the
point
where
we
have
to
one
with
the
generic
any
search
on
one
with
an
impedance
on
kodi
neeya
and
you
know,
because
you're
zero.
So
my
question
is
whether
you're
fine
on
not
with
the
7ps
doing
size
chaining,
given
that
it's
already
done
that
the
PS
that
a
pen
can
be
used
for
service
chaining.
Oh
you
disagree
with
us
on
penis
or
soul.
N
N
A
A
A
N
P
G
N
M
Mandrax
Nokia,
one
more
thing
actually
is.
If
you
look
at
the
end
of
the
day,
I
think
if
you
look
a
lot
of
the
vendors
here
or
transport
and
and
and
doing
things
in
data
center,
which,
as
a
vests
and
stuff
like
that
I
think
the
biggest
the
biggest
implication
of
this
of
multiple
solutions
is
going
to
be
the
value
added
services,
elements
which
need
to
interpret
the
headers
right
and
they
need
to
take
into
account
an
his
age
segment.
M
I
they
are
living
in
more
than
the
application
like
firewalls
vendors,
load-balancing,
vendors
and
stuff,
like
that,
so
they
need
to
support
this
and
I
think
it's
also
a
question
to
those
people
is:
are
they
willing
and
are
we
having
sufficient
information
from
them
that
they
are
okay?
With
all
of
these
approaches,
which
you
are
putting
to
the
table,
I
think
that's
an
important
consideration,
but.
E
M
No
I
think
that's
an
industry
problem
at
the
end
of
the
day,
because
people
need
to
support
and
implement
and
I
see,
most
of
the
people
here
which
are
discussing,
or
mainly
people
who
are
doing
the
transport,
the
encapsulations
and
and
they
are
more
working
in
that
area.
But
it's
the
application
area
where,
where
you
actually
see
that
there
is
a
need
for
adoption
of
these
type
of
things,
and
my
personal
view
is
that
Ana's
age
adoption
in
that
world
has
been
slow
in
general
yeah
I.
R
Think
in
some
ways
I
will
concur
with
Jo
with
wim.
It's
a
at
the
end.
What
I,
like
is
I,
have
a
simpler
network,
but
if
we're
talking
about
carrying
metadata
I
like
more
more
work
being
done
on
what
actually
can
emit
using
metadata
and
Clara
get
across
can
simplify
the
work
and
create
the
value
and
add
value
to
my
network
and
that's
what
I
think
at
some
point
there
are
will
be
multiple
implement
implementations
of
encapsulations
I'm
sure
somebody
will
invent
a
new
one,
some
really
soon.
R
We
should
not
have
that
they've
been
around
this,
but
how
to
create,
which
is
still
a
problem
for
a
lot
of
providers
in
the
world,
creating
chains
and
come
and
create
they're,
adding
by
components
together
if
you're
able
to
start
working
more
on
the
the
use
of
metadata
and
those
kind
of
mechanisms.
I
think
this
would
be
more
valuable
to
me
at
some
point.
Actually.
K
K
K
A
L
K
I
mean
my
personal
view
is
it
makes
an
Adrian
might
hit
me,
but
it
makes
no
sense
to
carry
metadata
in
an
MPLS
data
plane
it's
too
difficult,
but
in
SRV
six,
then
we
have
the
tail
V's
at
least
currently
in
the
SRH,
but
I
want
to
make
sure
that
any
TVs
that
are
service,
chaining
related
are
not
different
to
the
ones
that
are
used
in
the
SFC
nsh
environment,
because
otherwise
we
get
into
a
mess.
We
need
to
make
sure
that
those
are
coordinated
and
and
big
because
they're
the
same
TLV.
These
basically.
S
Just
interrupts
turn
so
few
comments.
First
ala
and
I
concur
with
what
we've
said.
The
industry
outside
of
ATF
is
very
slow,
so
I'm
reading
report
by
Softnet
published
February
this
year,
where
they
try
to
research,
for
they
could
you
signature
out,
do
traffic
city
into
B&S
and
their
conclusion?
No,
because
that
not
enough
V&F
that
actually
can
do
it.
So
the
fact
that
we
are
having
a
discussion
here
trying
to
figure
out
whether
we
compare
apples
or
apples
to
oranges
is
good
and
Industry
clarification.
S
What
is
it
we
are
going
to
do
in
order
to
do
anything
with
it?
So
we
really
need
to
provide
clarity
and
proceed
with
it
and
also
let
our
peers
outside
of
networking
industry
know
what
is
it
we
are
going
to
do
and
what
we've
taught
them
today.
Otherwise,
this
vicious
circle
us
talking
here.
They
are
not
doing
anything.
A
E
E
If
two
here
on
the
mailing
list,
if
there
is
I,
think
it'd
be
good
to
also
get
some
debate
around
what
around
what
we're
going
to
do
for
metadata
I
think
what
Jim
said
makes
a
lot
of
sense
to
be
able
to
say:
let's
define
for
us
a
v6
and
then
it's
HS
common
set
of
tlvs
I
would
like
to
see
some
statement
that
we're
not
going
to
try
and
do
anything
more
in
MPLS
SR
MPLS.
If
that's
the
the
consensus,
because
I
also
tend
to
agree
that
is
it
kind
of
fool's
errand
that
way.
C
S
A
Yeah
one
comment
is
a
appears
that
the
plane
is
defined
by
the
MPs
walking
woman,
not
not
by
Sprenger,
actually
see
I've
already
defined.
A
document
which
has
been
approved
by
the
working
group
on
is.
He
is
J
if
I'm
not
incorrect,
and
so
what
you
have
to
find
a
way
to
carry
a
limited
set
of
Kennedy
I
would
say,
but
I,
don't
we
and
what
would
know
better
I.
N
K
A
I
think
we've
made
progress
in
in
defining
the
problem
space
in
thermo-firm,
stateless
on
Ted
for
size,
Channel,
that's
pretty
clear,
I
think
for
everyone,
so
the
coexistence
of
srn,
etc,
I
think
constructors
and
in
nobody's
against,
and
if
we
want
to,
if
we
want
to
do
stateless
service
chaining,
which
is
part
of
our
charter,
think
we
were
going
to
work
on
that.
So
we're
still
going
to
to
wait
for
a
few.
A
K
A
T
T
T
The
proposal
here
is
that
there
is
harder
friendly
because
it
has
fixed
offsets
for
the
transmit
and
receive
counters
and
I.
Don't
have
to
worry
about
the
existing
womp
delay
measurement
message
there.
There
is
also
one
defined
for
authenticated
mode,
and
there
is
another
really
report
that
lets
use
to
identify
this
message.
There
is
an
LM
TLB
proposed
in
a
in
another
draft,
so
we
will
have
some
discussion
with
with
Greg
so
into
n
s,
our
policy
for
asylum
pls.
T
T
T
T
T
There
is
some
update
media
to
for
the
two,
a
measurement
defining
the
written
part,
for
example.
Some
there
was
some
comments,
review
comments
on
the
mailing
list
for
from
Greg
for
p3
MPs.
So
we
would
address
that
as
well
as
for
the
authentication,
the
utility
protection,
they
welcome
your
review
comments
and
suggestions.
T
U
U
U
T
U
J
U
I
I
think
that
you
need
to
state
it
explicitly,
but
again
it's
in
in
Ryan
with
Roc
5357
that
it
can
use
only
dynamic
port
numbers.
But
again
there
are
negotiated
between,
because
I
think
that
you
mean
that
saying
that
this
is
applicable
to
t1
pride
but
t1
part
is
not
standard.
It's
informational,
appendix
though
RFC
357,
and
because
of
that
there
is
no
interoperability
between
implementations,
yeah.
U
I'm
familiar
with
their
several
till
employ
implementations
and
there
exists
their
industry
technical
reference,
how
to
use
t1
provide
for
performance
measurement,
but
the
concern
is
interoperability
between
different
vendors
that
try
to
use
T.
One
part:
that's
why
we
have
effort
work
on
stamp
which
been
cleared
for
the
working
group
last
call
and
we'll
go
for
ad
consideration
and
an
eyes
G,
so
it
doesn't
have
control
protocol.
As
you
mentioned,
it
has
still
v4
extensions.
It
already
has
direct
measurement
till
we
defined
and
it's
much
more
extent,
extensible
additional
advantage
is.
U
U
T
There
is
no
success
is
only
for
this.
One
is
for
that
one.
So
again,
this
is
the
same.
It
uses
the
12
message.
Format
with
the
same
IP,
UDP
header
and
the
ports
are
configured
on
both
sides,
so
it
would
work
with
the
swot
stamp
as
well.
It
would
work
with
trump
as
well.
It
will
work
with
om
as
well.
U
T
If
you
look
at
this
packet,
this
is
this
is
the
Trump
packet
you
have
IP
UDP
header.
The
payload
is
from
Tom.
Only
difference
here
is
that
UDP
port
is
configured
by
user
instead
of
a
signal,
but
other
than
that
there
is
absolutely
no
difference
the
packet
on
the
wire,
but
in
your
first
that
you,
if.
U
T
A
U
A
U
A
V
Hey
Roger
sorry
I'll
make
it
quick,
I
think
useful
work,
clarification
question
I
had
was
it
it's?
It
am
I,
correct
conditions,
defining
probes
and
response
in
the
SR
context
where
the
probes
and
responses
are
not
sent
inside
the
user
packets.
Rather,
these
are
dedicated.
Packets
did
I,
get
that
right
so
say.
T
V
T
T
T
T
So
we
have
done
some
updates.
Based
on
the
review
comments,
we
have
received
renamed.
The
draft
to
reflect
the
scope
is
sixty
to
seventy
four.
With
this
editorial
changes,
there
was
a
request
to
elaborate
on
the
in
band
performance
measurement.
Also
added
some
text
for
the
ecmp
support.
That's
required
for
SLM
pls
gal
does
lack
the
support
of
a
CFP,
so
we
would
need
to
use
a
entropy
label.
For
example,
there
was
a
comment
from
C
were
to
add
the
rated
part
information
which
which
is
still
pending.
T
So
welcome
your
comments
and
suggestions
and
the
RFC
success
is
74.
There
are
many
implementations,
so
we
do
like
to
request
working
app
adoption
I.
Think
a
question
was
raised
last
meeting
that
should
this
be
in
a
spring,
will
MPLS
but
other
than
that
I
think
we
are
from
our
side.
We
we
do
request
the
adoption
greg.
T
There
are
many
questions
that
ask:
how
does
the
path
segment
identifier,
that's
defined,
for
s
are
applicable
to
p.m.
how
do
you
do
Law
School's
I
mean?
Where
does
that
go
hi
border
bidirectional
measurement
for
sr?
What
about
doing
the
link
measurement?
Do
you
use
the
adjacency
see?
How
does
it
work
so
there
are
many
questions
so
drop
it
it's
an
informational,
so
it
just
goes
and
said.
This
is
how
we
think
it
should
work,
but
again
any
comments,
thoughts.
You
have
you.
U
T
U
U
U
U
W
Just
it's
fundamental
to
me
that
if
you
want
to
say
that
in
a
network
where
you
specify
the
paths-
and
you
may
have
some
reason
for
specifying
that
you
should
be
able
to
specify
how
you
get
the
packet
back,
she's
nothing
to
do
with
doing
the
measurement
in
that
direction.
It's
that
in
a
separate
routed
Network,
you
may
want
to
lay
out
a
segment
rooted
path
to
send
your
messages
around.
U
It's
similar
than
to
the
concept
of
MPLS
LSP,
being
the
option
of
how
you
define
that
how
to
reply
back.
Yes,
okay!
So
if
you
want
to
for
performance
measurement
for
each
packet
to
specify
how
you
buy
that
I
think
that's
an
overkill.
If
you
want
to
specify
how
you
returned
this
session
back,
you
can
provision
it
do
it
in
each
and
every
packet
defining
each
and
every
packet,
or
you
send
this
packet.
This
way,
you
send
this
packet
that
way,
I
think
it's
asking
for
inconsistencies
in
measurements.
W
W
W
T
T
So
we
presented
this
draft
at
last
ITF
in
both
IP
pm
and
spring.
We
see
the
number
of
good
comments,
so
many
thanks
for
everyone
who
commented
so
the
first
comment
was
that
we
would
have
hard
time
getting
the
IANA
L
to
allocate
the
ports
and
he
was
indeed
true.
We
talked
to
a
number
of
people
and
we
would
need
many,
a
UDP
port,
so
I
think
we
going
with
the
user,
configure
UDP
ports
approach
for
SL
v6
policies.
T
We
are
using
the
segment
routing
header
to
transport,
the
probe
back
as
to
the
tail
end
using
the
end,
RTP
or
end
of
OB
functions.
We
we
address
the
comments
to
align
the
TLB
structure
for
the
return
path,
clarify
the
sequence
number
TL.
We
added
authenticated
mode,
updated
security
system
security
section
as
well
as
the
elaborating
and
performance
measurement,
so
there
is
a
new
TL
v
that
we
are
there
for
authenticated
more
the
four
sequence
number.
T
P
X
We
made
an
observation
that
there
are
really
two
classes
of
SIDS
in
the
world.
There
are
transport
SIDS
that
are
used
to
steer
a
packet
to
the
terminal
segment
of
a
sr
v6
tunnel,
they're
processed
at
non
terminal
segment,
end
points
when
segments
left
is
greater
than
zero.
An
examples
of
these
are
the
SR.
V6
ends
it
in
the
end
exit.
There
are
relative
to
relatively
few
of
these,
and
their
semantics
are
simple:
they
carry
very
very
little
information.
They
can
be
encoded
in
very
few
bits,
in
contrast,
there's
something
called
a
service
set.
X
It
determines
the
behavior
at
a
terminal
segment,
it's
processed
only
at
the
terminal
segment
when
segments
left
equals
zero.
There
are
lots
of
these
examples
are
end,
the
DX
for
end
d,
x6
and
dt
for
and
so
on,
and
they
have
relatively
rich
semantics.
You
need
a
lot
of
bits
to
encode
these
well,
similarly,
an
ipv6.
There
are
two
ways
to
deliver
in
store
to
downstream
nodes.
There
are
routing
extension
headers
that
are
used
to
steer
packets
from
an
ingress
to
an
egress
they're
processed
at
non-terminal
segments.
X
End
points
when
segments
left
equal
is
greater
than
zero
and
they're
well
positioned
to
carry
transport
sets.
By
contrast,
there
are
destination
options.
They
determine
behavior
at
the
egress
they're
processed
at
the
terminal
segment.
Only
when
segments
left
equals
zero
or
when
there's
no
routing
header
at
all
and
they're
well
positioned
to
carry
services.
X
So
it's
pretty
obvious
where
we're
going
with
this
Oh
before
we
go
there.
We
have
to
talk
about
one
more
issue.
The
problem
with
routing
header
is
in
general,
not
just
the
SRH,
but
the
routing
headers
we've
seen
in
the
IETF.
So
far
they
get
very
long.
Typically,
they
have
eight
bytes
of
overhead.
Four
bytes
are
required
by
RFC
8200.
X
Typically,
they
take
another
16
bytes
per
said.
They
encode
acid'
as
a
hundred
28
bits.
So
a
routing
header
with
three
SIDS
is
56
bytes
long.
This
isn't
a
sex
friendly.
Once
once
the
routing
header
gets
beyond,
maybe
five
six
SIDS,
it
just
becomes
impractical
and
they
impose
on
reasonable
bandwidth.
Overhead
short
packets,
no
less
than
500
bytes
are
pretty
common
on
the
internet.
Routing
header
with
three
SIDS
may
become
common,
and
that
means
a
10%
overhead
just
for
routing
extension
headers.
So
where
do
we
go
with
this?
X
We
propose
encoding
transport
SIDS
and
a
new
compress
routing
header,
that's
engraft,
Bonica
six-man,
compressed
routing
header.
It's
the
topic
of
the
talk
today
and
we
encode
talked
about
encoding
service
SIDS
and
a
new
ipv6
destination
option
crafts,
Bonica
6-man
BPM
desktop
we'll
talk
about
that
in
six-man
tomorrow.
X
The
compressed
routing
header
that
we're
talking
about
today
can
be
used
with
the
SRH
use
one
when
it's
appropriate
use,
the
other
when
it's
appropriate.
What
does
this
new
compress
routing
header
look
like
well,
the
first
five
bytes
are
exactly
the
same
as
the
cr8
as
the
SRH.
Then
we
have
two
bits
that
tell
you
how
SIDS
are
encoded.
Sids
can
be
encoded
as
8
16
or
32
bits,
and
let's
say
so.
X
Let's
what
we've
talked
about
that
we've
talked
about
the
comm
field,
each
Sid
maps
to
an
ipv6
address,
either
through
a
table
lookup
or
an
algorithm,
and
that
ipv6
address
is
copied
to
the
destination
address
of
the
ipv6
header,
exactly
like
any
other
routing
header.
What's
interesting
here
is
what's
copied
to
the
ipv6
address
as
a
real
ipv6
address.
That's
not
a
said.
It
has
no
semantics
beyond
the
ipv6
address.
X
So
what's
the
status
of
this
operators
were
expressing
interest?
There
are
a
few
operators
who
are
co-authors
of
the
draft
prototypes
are
under
development,
both
on
the
forwarding,
plane
and
Isis
extensions
to
support
the
CID,
and
we
would
ask
for
wide
review
in
sec
spring
and
six-man
and
would
ask
for
call
for
adoption
in
six-man
questions
comments,
rotten
fruit.
I.
Y
Y
So
in
existing
SRT
pass
fast
protection,
so
when
one
node
of
fails,
so
it's
option
note:
where
do
the
local
protection.
However,
when
the
ITP
convergence
and
then
this
kind
of
local
protection
will
not
work
so
in
this
draft,
we're
proposing
solutions
to
rot
and
resolve
these
issues.
So
first,
let's
look
at
the
program
in
details,
so
here
in
this
picture
we
have
a
SRT
pass,
so
this
path
will
go
through
nota
N
and
then
the
traffic
will
go
from
node,
a
BP
MMC.
Y
So
when
node
n
fails
so
the
load
P
were
detected,
is
failure
and
then,
where
do
the
faster
route
and
get
around
at
this
failure
point
and
then
send
traffic
to
know
to
see.
So
this
works.
However,
when
IDP
convergence
for
exam
are
not
a
and
not
B,
and
then
at
this
point
the
traffic
where
work
will
be
dropped.
So
this
traffic
interruption
will
continue
until
a
new
SRT
pass
is
computed
and
installed.
So
we
can
see
from
the
p8
from
the
Dirac
a
DP
converges
and
then
half
after
a
data
point.
Y
Then
a
new
SRT
has
installed,
and
during
this
period
the
traffic
traffic
get
lost,
so
we
will
provide
a
solution
to
resolve
this
problems.
To
the
rough
idea
is
that
we
will
do
some
product
extension
and
then
we'll
make
sure
that
even
after
ITP
converges,
the
nodoby
will
continue
in
the
traffic
to
know
to
pee
and
then
note
of
he
will
continue
to
the
faster
route.
So
we
can
resolve
this
dis
problems.
Y
Y
Y
So
the
detail,
extensions
I
give
for
all
PFD
exchanges.
So
basically
is
this
persimmon
we
use
the
existing
kappa
b
GTA
v.
I
just
use
one
bit
so
this
one
bit
indicate
our.
I
can
support
a
forwarding
proxy
for
link
abilities
so,
for
example,
one
p
and
pi
advertise
his
community
and
then
the
also
note
in
Maddog
will
receive
these
kind
of
capabilities
for
examined
over
x.
So
after
not
IX
received
is
this
kind
of
a
capability
for
other
photons
by
P,
which
means
that
P?
Well?
Where
do
the
proxy
forwarding
for
all?
Y
It's
a
dishes
note
so
in
normal
operation
that
note
X
will
submit
the
traffic
to
another
end.
So
after
know
the
end
fails
and
then
even
after
IDP
converge
converges
and
then
note
note,
X
will
continue
sending
the
traffic
attune
tonotopy.
That
works.
You
note
and
then
the
third
approach
note
way
to
approach
forwarding
to
the
not
end
the
failure
load.
Y
So,
in
addition
to
the
this
kind
of
a
simple
case,
we
also
do
the
protein
extensions
for
pending
because
of
or
not
N,
and
that
no
time
we
may
have
a
pending.
So
this
bundling,
we
abandon
s
ID
to
a
list
of
segments,
so
this
information
will
be
advertised
only
to
the
adjacent
node.
This
is
not
advertise
to
the
whole
area.
So
this
we
also
improved
the
scalability.
So
when
so,
when
the
edge
is
not
P
received
inside
nobody,
you
mention
he
can
do
the
process
forwarding
to
these
kind
of
abundance.
Y
Y
Y
S
You're
using
traffic
engineering
for
reason,
you're
deviating
from
trucks
bus,
so
in
this
case
a
would
compute
a
pass
and
you
are
trying
to
protect
it
from
proxy
prospective.
However,
probably
you
wouldn't
use
IGBT
feet
were
out
and
fell.
You
would
use
sounds
like
be.
Multiple
BFD
are
some
other
fast
detection
mechanism
to
notify
a
that
passes,
unavailable
wouldn't
be
than
a
who
should
be
compute
better
pass
if
such
required
that
meets
all
the
constraints
and
provision
it
in
a
way
avoids
and
I
mean
you.
Y
Mean
the
you
so
the
question
is
that
we
need
some
kind
of
all
so
pike.
A
pass
also
mirror
satisfied,
conscience.
S
Y
S
Y
S
Y
Z
S
Z
In
RFC
7770,
you
should
be
using
the
functional
capabilities
and
the
other
one
is
I've,
never
heard
it
called
proxy
forwarding
before
I
and
and
then
I
looked
at
it
and
I
said
oh,
this
is
this
is
a
node
protection
in,
and
so
it
was
kind
of
a
funny
and
I'm,
not
sure
I
haven't
really
since
I
didn't
read
that
off
I
haven't
thought
how
it
fits
with
all
the
existing
mechanisms.
You
know
that
have
been
done
for
years
with
normal
LSPs
or
anything
like
that.
It
seems
like
a
funny
terminology,
yeah.
AA
Beat
up
shellac
Cisco,
so
I
have
no
problem.
If
you
use-
or
you
should
try
to
protect
the
segment's
in
the
packet
core
note
protection
for
a
protection
itself,
but
I
don't
think
that
trying
to
pretend
that
some
seed
exists,
which
is
gone,
is
a
good
idea.
Especially,
you
may
have
a
secondary
failures
during
that
time
and
in
that
case,
I.
Don't
think
how
I
don't
see
how
this
is
going
to
work.
So
the
suggestion
is
or
Jeff
said.
Y
V
Rajeev,
oh
sorry,
so
Jeff
summed
up
first
of
my
comments
already
sold
focus
on
the
second,
so
the
applicability
that
I
was
stretching
to
see
was,
if
n
was
more,
of
bond
brave,
are
between
say
to
rodding
domains
or
or
two
domains
as
her
domains
and
n
so
lose
mode
from
the
path
point
of
view.
V
Y
Y
V
Y
The
mainly
pointed
to
be
protected,
so
we,
for
example,
suggest
that
the
boundary
node
or
egress
node,
and
then
we
need
apart
I,
provide
protection
and
sync
the
trenches,
note
or
mute
a
note.
We
also
need
to
provide
protection
right,
so
I
think
here
we
focus
on
the
on
the
meeting
of
the
protections
I
think
for
the
egress,
no
traction
I
think
we
also
have
some
couple
proposals
we
are
presenting
in
Friday
I'm
writing
tomorrow.
Y
A
Y
Y
A
AB
Hello,
everyone
I'm
Tran
Tran,
shown
from
DT
my
presentation
today,
SRM
pstp
in
the
domain
use
cases
and
the
draft
has
been
presented
in
yesterday.
Npr's
working
group
I
received
many
great
comments
from
NPRM.
We
appreciate
more
comments
from
spring
working
group.
Ok,
as
our
MPs,
we
defined
it
to
transport
profiling,
as
as
our
MPs
as
we
all
known
that
in
SMPS,
lat
works
to
as
our
past
is
unidirectional,
but
in
some
transports
that
roxtor
assault
by
the
earth
by
directional
past
must
be
establish
it.
AB
For
example,
in
MPs
TP
lat
works,
so
the
draft
ITF
spring
amperes
per
segment
and
the
draft
has
been
approved
to
their
working
group
draft
just
now
and
they're
a
different
path.
Segments
to
support
us
are
bi-directional,
has
correlation
so
based
on
the
drafts.
This
document
discourses
third
into
two
main
scenarios
in
SRM
pstp,
let
works
in
this
document.
We
defined
an
energy
and
power
segment.
AB
They're
into
map
has
segments,
and
in
this
scenario
the
s
are
bi-directional
end-to-end
passes
can
be
established
across
her
multiple
domains.
Then
the
there
are
two
models
in
SRM
pstp
into
domain.
The
first
one
is
a
splitting
and
domain
model.
The
mames
are
isolated
and
the
stitching
SR
load.
We
approached
their
list
of
SIDS
for
the
loose
ab
turner
and
for
lasting
in
the
domain
model.
Global
segments
listed
at
the
ingress
SR
and
the
end-to-end
a
pass
as
ID,
where
you'll
across
all
the
domains
in
the
waste
also
discussed
the
SNP
LTP
into
working
wisdom.
AB
AB
So,
let's
look
at
the
SSR
and
ps
t
ps
preaching
in
the
domain.
As
the
figures
show,
there
are
string
as
our
domains
stitching
together
we
will
establish
an
end-to-end
pass,
drone
load,
a
tool
ology
and
the
parts
will
be
divided
into
several
paths
of
several
segments,
and
each
segment
will
be
identified
by
in
by
herb
has
segments
we
defined
to
automate
how
segment
or
into
may
pass
s
ID.
So,
for
example,
exeter
in
at
the
end
interest
load
a
we
will
push
their
a
2x
as
RT
e
and
the
I
pass
a
2x2.
AB
Their
load,
ace,
MPs,
stack
and
Esther
s,
our
MPs
procedural.
There
trust
transy
knows
where
pop
pops
out
at
the
top
label
and
when
their
packets
being
forwarded
to
the
load,
X
and
then
the
I
pass
a
to
X
will
be
top
label
and
at
the
load
X
it
will
do
the
matching
from
there
I
pass
a
to
X
switch
to
there
X
to
Y
srte
and
that
I
pass
away
X
to
Y,
then
at
the
Lord.
AB
Why
did
the
same
with
the
procedure
at
load,
X
and
then
their
packets
of
being
forwarded
after
load
D
and
it
will
owning
their
I
pass
Y
to
the
existed
at
the
top
lab
stack
each
at
low.
Today
it
will
correlate
in
the
forward
forwarding
path
and
the
reverse
parts
bakudan
powder
at
the
load.
Z,
the
label
is
the
I
pass
Y
to
Z
and
it
will
switch
the
I
passed.
You,
the
reverse,
pars
3
to
Y
and
the
Z
2
is
RTE
and
there
I
passed
three
two
one.
So
the
reverse
pass.
AB
It's
the
same.
We
stir
for
impossible
and
they're
a
nesting
inter-domain.
The
difference
is
that
Jo
is
an
end-to-end
path
segments
to
identified
eight
from
A
to
Z
the
editor,
the
interest
alone,
a
we
were
pushed
her
sad
past,
a
2x
and
the
bounteous
ID
x
to
y
fungicide
ey
to
Z
and
their
Valerie.
We
pushed
her
into
in
the
past.
It
is
a
pass
for
short
and
thereupon
esad.
AB
We
used
it
to
reduce
the
depth
of
the
pious
label
and
when
their
packet
being
forwarded
to
the
reload
X,
then
the
bounteous
ID
x
to
y
will
be
mapped
to
the
x
to
x,
RT,
there's
a
path
segment
X
to
Y,
then
epidural
Lord.
Why
it's
the
same
and
they're
at
the
loaded,
Z
juror
will
be
owning
their
a
and
two
and
pass
a
to
t
existed
and
their
the
end
to
I
pass
a
2d.
You
will
be,
will
correlate
the
two
unidirectional
parts,
for
example,
from
a
path
from
A
to
Z.
AB
We
have
switched
to
the
c2
a
pass
ended
the
3/2
Y
X,
R
T.
Then
the
reverse
process
will
be
the
same
procedure
with
the
boarding
pass,
so
we
will
discuss
their
assam.
Here's
T
P
and
NP
STP
into
working
as
a
figure
shoulder
a
there.
Are
s
are
interim
SR
domain
and
their
MPs
to
TP
domain.
They
eclectic
ways
or
with
loes
Porter
loss
in
SR
networks
we
built
as
our
Turner
and
in
MPs
TV
networks.
AB
We
built
an
PST
Peter,
they
are
or
bi-directional
Turner's
we
wanted
to
achieve
and
to
enter
whip
Ian's
service,
and
then
we
we
are
used
in
SR
networks
there
from
a
to
one.
We
will
use
the
SI
the
list
and
the
path
segments
from
one
to
a
to
translate,
pushed
who
stays
there,
editor
load
one
and
after
their
load
a
it
will
be
the
path
segments
of
a
2-1
and
at
the
Lord
a
we.
AB
It
will
switch
the
path
segments
to
their
amperes
label
and
this
is
the
forwarding
path
and
the
reverse,
possibly
the
same:
their
MPs
label.
We
all
switched
a
cuter
path
segments
and
enter
as
are
Turner,
so
this
is
their
into
working
so,
which
are
we
just
er
discussed
the
use
case
about
their
SMPS
TP
into
domain
and
their
their
inter
working.
So
comments
and
the
discussions
are
very
welcome.
S
S
U
U
Well,
actually
impair
STP
can
be
provisioned
with
a
control
plane
to
again.
There
are
control
plane
for
MPLS,
TP
is
gmpls,
and
then
that
they've
been
discussed
so
began.
There
are
different
options
and
they're
in
different
considerations,
but
MPLS
TP
does
have
GM
LS
control
point
that
been
stated
in
ITF
document.
S
U
D
U
Again,
that's
absolutely
legitimate
question
I
appreciate
the
question
so
yeah.
We
need
to
think
about
the
control
plane
aspect
of
this
and
how
this
all
works.
I
think
that
what
is
now
reflected
in
this
that
there
are
controller,
centralized
controller
or
hierarchical
controllers
may
be
used
to
address
distributed
control
plane.
It's
absolutely
valid
scenario.
Well,
let's
think
about
it.
We
need.
A
To
to
Catalina,
maybe
a
flirt,
a
nun
question
from
from
Jeff:
it's
not
clear
for
artists
the
spring
working
up.
What
CSR
has
to
be
profile,
it
is
or
would
be
because
I
don't
think,
we've
ever
discuss
Asylum
pstp.
So
if
you
say
it's
a
profile
of
a
Sheraton,
yes,
they
go
to
to
define
it
so
that
we
know
what's
inside
the
profile
or
outside
your
profile.