►
From YouTube: MPLS WG Interim Meeting, 2020-04-21
Description
MPLS WG Interim Meeting, 2020-04-21
A
A
We
have
we
have
virtual.
Today
we
are
using
the
ether
pad
to
log
the
attendance.
Please
use
this
link,
we
were.
We
are
notified
that
there
are
two
URLs
for
ether
pad.
We
are
using
this
one
and
there
is
a
note
and
the
other
URL
for,
if
a
pad
to
use
the
proper
one,
please
make
sure
you
sign
in
your
name
in
the
right
one
for
asking
questions.
This
is
a
little
bit
unusual.
A
A
A
A
We've
we've
been
asked
that
to
bring
up
of
the
presenters
that
words
the
end,
bring
them
forward,
specifically
presenter,
five
and
six
would
where
we're
going
to
move
them
to
the
top
based
on
the
request
they
have
to
be
present
in
or
they
have
to
leave
us
earlier.
We
will
accommodate
this.
We
have
seven
present
or
actually,
six
presentations
and
status
group
update
that
the
chairs
will
go
through.
A
A
A
A
A
The
other
documents,
the
working
group
documents
we
have
one
that
needs
a
revision,
the
Refresh
independent
our
CPF-
are
our
authors
are
present.
We
asked
them
to
take
action
on
that
as
well.
We
have
the
RMR
and
pls
erimar
that
again
needs
the
written.
So
if
the
authors
are
present
today,
we
ask
you
to
please
take
action
needed
the
MPS
SFL
framework
document.
We
have
requested
the
the
publication
of
this
document
and
you
know
please
stand
by
and
expect
follow-ups
from
IHG
on
this.
A
We
have
new
working
group
document
that
got
adopted.
This
is
the
LS
beeping
or
SP
v3
and
OSPF
v3,
and
once
we
have
points
there
is
a
little
bit
of
discussion
on
this
document.
So
if
working
group,
you
I
encourage
participants
to
be
engaged
in
that
discussion,
it
has
to
do
with
core
point
application.
A
A
E
A
F
A
A
The
other
documents
they
were
updated,
so
we
expect
a
working
group
report
or
a
status
report
for
the
working
group
documents.
Some
of
them
we've
received
I'll,
go
to
through
those
report
towards
then,
but
as
usual,
if
it
is
a
working
group
document,
we
expect
that
the
author's
to
send
periodic
reports
before
the
meetings,
no
clear,
please
do
that
for
your
documents
recently
expired
documents.
So
these
are
the
working
group
documents,
but
they
got
expired.
A
A
We
have
updated
internet
draft,
some
of
which
are
origins
as
well,
so
have
a
document
that
is
currently
and
in
spring
working
you
the
opposite
of
felt.
It's
also
related
to
MPLS.
So
we'll
hear
more
about
that.
It's
on
the
agenda
now
I'm
going
to
go
through
the
brief
status
report.
What
means
group
documents
that
we
received
the
first
one
it
it
has
undergone
yang
doctor's
review
and
he
da
da
the
comments
were
addressed.
The
step
will
be
a
working
group.
A
Last
call
authors
are
asking
us
to
move
forward
and
we
will
think
that
the
chairs
will
take
the
action
on
that.
The
second
one
I
think
we
spoke
about
it.
Lower
had
pointed
out
that
we
are
at
it's
an
IPR
fall
implemented
in
Paul.
Sorry,
implementation,
Paul
right
now
and
authors
need
to
respond
if
or
anyone
in
the
working
group
that
knows
of
an
implementation.
Please
do
that
as
well.
A
A
A
A
F
A
A
A
F
A
A
It
is
stable
and
ready
for
working
group
last
call.
The
next
step
is
the
for
the
Working
Group
chairs
to
move
ahead.
When
this
working
group
last
called
I
did
talk
about
the
last
document
here,
which,
which
is
tackling
the
code
points,
some
deprecation
for
an
open
update
to
the
code
points
or
LS
beeping
forward
sphere
was
adopted.
We
invite
your
reviews
happen
if,
and
there
is
a
discussion
on
that
on
the
mailing
list.
So
please
contribute
that
if
you
have
any
opinion
on
that
matter,.
A
Next
we
have
the
LS
beeping
registries
update
the
status
of
this
is
we
the
author's,
are
addressing
some
comments.
They
were
raised
by
Adrienne
Thank,
You
Adrienne,
and
next
steps
would
be
to
move
forward
for
working
group
last
call
the
ITF
and
LS
s.
Fl
frameworks,
ions
flow
label,
I
mentioned
that
it
is
in
publication
expected
or
we
are
waiting
on
further
review
from
iesg
and
it's
the
last
one
here.
A
We
have
pointed
out
that
we
have
many
number
of
many
authors
on
the
first
page.
So
if
we're
asking
the
author's
to
discuss
amongst
themselves,
if
it
is
possible,
reduce
the
number
of
authors
and
if
you
think
otherwise,
let
chairs
no-
and
we
will
raise
it
up
with
the
area
director
and
the
next
step
would
be
to
move
ahead
with
working
group.
F
A
A
G
A
G
G
So
a
lot
is
basically
an
extension
to
our
protocol
that
allows
creation
of
until
little
is
faced
empty
unless
we
have
a
steward
of
protocols.
Light
like
LD,
be
RSVP,
BGP,
Lu
and
so
on.
Each
protocol
has
a
different
set
of
characteristics
right,
yes,
its
own
strengths
and
weaknesses.
Lbp,
for
example,
is
good
at
keeping
state
minimal,
and
but
it
can
only
create
SPF
LS.
Please
RSVP
is
pretty
good
at
traffic
engineering,
but
you
know
it
has
other
than
square
problem
in
the
transit.
G
Elif
is
a
very
simple
protocol
because
it's
based
off
of
art,
which
is,
and
if
it
is
its
own
weaknesses
like,
for
example,
its
scaling,
hey
that's
what
I
love
is
because
of
its
simplicity
is
applicable
in
two
major
deployment
scenarios
oneness
in
the
MPLS
data
center,
that
let's
say
there
is
a
server
that's
doing,
providing
multi-tenancy
right
and
in
order
to
provide
multi-tenancy,
the
server
needs
an
underlay
network.
We
have
a
host
of
underlay
technologies,
MPLS
being
one
of
them.
G
If
the
server
chooses
MPLS
as
underlay,
then
it
needs
a
way
to
create
the
MPLS
tunnel
right.
It
needs
to
speak.
Some
protocol
to
the
network
to
create
the
tunnels
and
they'll
have
is
an
ideal
choice
between
server
on
the
network
use
cases
in
the
access
network.
If
the
broadband,
the
server
is
allocating
labels
to
subscribers
and
one
of
the
DSLAM
sitting
somewhere
in
the
access
network
needs
to
get
the
destination,
it
can
user
love
to
acquire
the
label.
G
G
G
If
the
question
the
question
is,
how
does
the
server
acquire
the
label
pointing
right?
Typically,
the
server
does
not
speak
routing
protocol
towards
the
network
right.
It
may
have
a
BGP
version
running
in
it
or
two
as
a
controller,
that's
mostly
for
exchanging
overlay
routes,
a
part
that
needs
to
be
addressed.
This
once-over
also
required.
G
So
at
the
high
levels
of
the
protocol
works,
traditional
are
one
of
the
nodes
in
the
network
sends
out
a
request
and
says
I'm
trying
to
reach
this
particular
IP
destination.
What
has
this
IP
address'
any
give
me
them.
What
has
this
IP
address?
Can
you
give
me
the
MAC
address
for
this,
and
whoever
wants
that
I
clear
this
response
right?
That's
traditionally
simple!
G
G
Operation
is
pretty
similar
to
our
and
allow
the
client
sends
out
a
request
and
says
I'm
trying
to
reach
this
destination.
Let
me
give
me
the
label
instead
of
MAC
addresses
just
give
me
the
label
right,
one
of
the
ill
observer
sitting
in
the
network,
gets
this
packet
and
when
it's,
they
replay
instead
of
the
MAC
addresses
hands
of
the
label,
and
that's
that's.
Basically,
the
nuts
and
bolts
of
this
protocol
design
go
to
the
next
slide.
Please.
G
Let's
zoom
it
a
little
more
in
the
diagram
you
see,
there
are
three
n
nodes,
XY
and
Z,
which
could
either
be
serviced
or
DSLAM
nodes,
and
there
are
t3
routers,
t1,
t2
and
t3
sitting
in
the
border
of
a
network
which
could
either
be
PNG
service
or
routers
right.
If
it's
in
the
data
center,
let's
say
X
wants
to
talk
to
I
want
to
establish
an
underlay,
less
p2
Y.
So
X
start
by
saying
it's
insert
an
airlock
request
and
says:
I
need
to
get
to
Y.
G
Can
somebody
give
me
a
label
for
that
that
packet
reaches
Stephen
because
ARP
is
a
broadcast
packet
even
gets?
It
looks
into
its
tile
database
and
for
a
second,
let's
assume
the
t1
has
a
tunnel
to
Y
so
assigns
a
label
binding
to
the
LSP
to
Y
and
installs
a
route
entry
in
itself.
It
says
if
the
incoming
limpness
it's
a
l1
right,
swap
it
with
l2
and
send
it
to
Y,
and
then
it
sends
L
1
to
X.
G
G
Likewise
in
elmer
when
t1
hears
from
X,
it
also
learns
that
X
is
present,
that
and
automatically
ourselves
a
label
mining
for
X
which
is
sort
of
propagated
into
the
network,
and
that's
how,
in
the
first
place,
Stephen
had
an
LSP
why
it
was
propagated
by
t2
who
is
t1.
Then
why
try
to
do
something
next
slide?
Please.
G
G
For
example,
let's
say
one
of
the
tenants
had
a
few
containers
or
VM
sitting
in
server
and
a
few
more
sitting
in
another
server,
and
if
these
containers
are
VM,
try
to
talk
to
each
other,
then
you
know
the
two
obviously
to
have
an
entirely
tunnel
among
them
right.
So
maybe
one
of
the
possible
triggers
says
if
the
over
layer
out
present
in
the
server
says
I
have
an
extra
pointing
to
that
server.
That
could
be
a
place
where
we,
where
the
host
actually
sings,
should
develop
request.
G
So
the
we
said,
T
1,
we
would
look
into
it's
tunnel
database,
I,
say
in
a
label
to
that
LSP
and
you
know,
create
a
route
that
stitches
the
to
LSPs
right
and
what,
if
in
the
coat
you
are
actually
doing
spring?
In
that
case,
you
might
have
multiple
recipes,
so
this
concept
is
already
supported
by
spring.
It's
called
bindings.
It
essentially
p1
allocates
a
label
and
swap
out
into
the
LSP
going
into
the
core.
G
G
G
What
do
we
need
to
do
next?
Last
time
you
presented,
we,
we
actually
had
some
prototype
implementation
on
line
X,
and
but
it
didn't
go
too
far,
because
at
that
time,
MPLS
supported
linux
was
not
that
great
I
think
things
have
improved
significantly
now.
I
think
we
need
to
refresh
our
implementation
to
catch
up
with
if
instead
of
affair-
and
we
also
need
to
flesh
out
both
protocol
details
as
to
how
the
label
metric
interval
a
B
capability,
etcetera.
However,
all
our
these
property
is
going
to
be
represented
in
the
packet
server
is
multi-home.
G
How
does
it
choose
the
egress
interface?
How
does
it
work
in
a
spring
network?
How
do
we
deal
that
cell
helped
me
to
support
latest
acting,
for
example,
this
binding?
Is
it
good
enough
there's
no
most
of
the
routing
protocols
have
this
blissful
restored,
so
Code
Red
I'm,
the
control
plane
is
allowed
restart
way.
Traffic
is
still
flowing.
They
similar
capability
is
probably
needed
in
a
lot
too.
A
A
B
B
Allow
wants
to
I
think,
keep
it
really
simple:
just
get
IP
to
and
a
label
binding
and
so
I
think
that
works
just
set
the
proto
tied
to
protocol
type
to
ipv6
further
than
ipv4
there's
a
core
point
for
that
in
Ethernet.
So
that
would
work.
That's
this
current
thing.
I
I
know
ipv6
people
probably
are
looking
askance
at
me,
but
that's
I
think
it's
more
suited.
D
A
comment
and
a
question
I
see
I
think
I.
It's
perceived
that
ARP
is
easy
to
extend
and
I
understand
the
reasoning
behind
that,
but
I
think
if
we
really
need
a
lightweight
protocol,
probably
we
need
even
extensions
and
are
beyond
that
right.
I
think.
If
we
really
want
to
go
down
this
path,
is
it
more
better
to
define
a
very
lightweight
protocol,
even
maybe
using
your
PC
or
something
like
that
where
you
basically
request
and
exchange
it
between
two
to
neighbors
and
be
done.
B
The
thing
is,
art
is
a
simple
protocol,
it
does
everything
you
want,
and
one
other
thing
is
that
you
know
in
in
a
situation
like
in
a
data
center,
where
you
have
you
know
potentially
10,000
or
more
servers,
you
don't
necessarily
want
to
know
where
the
other
server
is
and
what
the
bindings
to
them
are.
You
want
to
be
very
prescriptive
about
that,
and
so
our
allows
you
to
you
know
only
get
the
things
that
you
want
because
you're
explicitly
requesting
it
so.
B
Ldp
downstream
on-demand,
except
it's
so
much
lighter
weight
than
LDP
and
doesn't
need
an
underline
lgp.
You
don't
work
with,
so
you
know.
Yes,
we
could
invent
new
protocols
and
you
know
in
many
cases
that's
the
right
thing
to
do,
but
I
think.
If
does
the
job,
then
we
should
just
simply
use
that
but
I
like
inventing
your
protocols
but.
D
B
H
I
assume
that
the
servers
are
point-to-point
link
to
the
first
hop
router
who
will
answer
the
ARP,
then
I
understand
how
the
various
failure
cases
work,
but
if
the
top
of
rack
thing
is
a
switch
and
not
doing
the
art
but
rather
connecting
to
some
other
router,
that's
actually
answering
the
Arps,
which
is
perfectly
valid.
Then
I.
B
In
the
case,
with
a
nitro
point,
a
point
and
that's
I
think
the
most
common
case
in
data
centers,
especially
if
you're
going
to
move
to
MPLS
I,
think
it's
interesting
to
try
to
work
out
some
of
those
cases
but
I'm
happy
to
stay
with
just
doing
the
doing
it.
You
know
on
a
point-to-point,
you
know
a
server
to
switch
connection
with
all
connections
does.
B
A
B
So
this
this
drafters
joint
work
with
wen
Lin.
She
was
looking
at
the
problem
of
how
do
I
deal
with
multihomed,
immediate
client
and
I'm
doing
fast,
reroute
weather.
When
the
sea
was
down
there,
certain
issues
so
and
then
we
started
discussing
this
and
came
up
with
a
more
generalized
framework.
So
next
slide
please
when
is
online.
So
if
we
have
questions,
we
can
definitely
bring
her
in.
B
So
everyone
knows:
Fosse
Road,
it's
it's
a
really
nice
way
of
reducing
packet
loss,
windows,
link
or
node
failures,
and
in
many
cases
this
is
the
reason
why
people
do
buy
MPLS.
But
there
are
cases
where
MPLS
may
actually
I'm.
Sorry
Basquiat
may
actually
make
things
worse.
So
we
look
at
some
of
those
situations
and
then
propose
a
solution
using
a
new
special
purpose
label.
Next,
so
here's
a
case
where
this
is
what
when
was
looking
at
I,
have
a
CEC
one:
that's
multicom
to
P
1
and
P
2
and
it's
active
active.
B
So
what
you
have
is
packets,
coming
from
situ
trying
to
go
to
C
1
P
III
can
decide
to
go
to
P,
1
or
P
2
and
put
appropriate
labels,
and
then
P
1
says
hey,
since
on
multihomed.
I
can
actually
have
a
backup
path.
In
case
my
link
to
c1
goes
down
and
I'll
go
here.
This
link,
I
have
or
any
paths
I
have
to
P
2
and
then
a
speedy
two
to
send
the
package
2
c1.
B
Is
this
protection
pass
from
p1
to
p2
c
e1,
v,
RP,
2
and
similarly
P
D
2
does
the
same.
So
this
is
nice
and
this
is
a
ppm
fast
reroute,
and
this
applies
in
the
case
of
an
EDP,
N
or
an
item
again
with
multihoming.
The
evpn
should
be
in
active,
active
mode.
I
PB
p
on,
though
typically
almost
in
active
active
mode,
it
could
apply
to
VPLS,
but
typically
we
pay
less
on
out
for
an
active
active
that
one
active
standby
in
there.
Certain
issues,
if
you're
an
active,
active
next.
B
So
the
the
case
here
is,
if
she
fails.
If
the
CD
fails,
p1
will
probably
see
this
as
a
link,
failure
and
activate
the
posterior,
and
so
it'll
send
packets
to
C
1
V
R,
P
2,
but
P
2
will
see
the
same
thing.
So
what
you,
what
you
get
is
a
packet
goes
from.
P
is
free
to
P.
P
wants
a
low
I
got
to
go
so
the
packet
goes
to
P
2,
then
back
to
P
1
and
just
keeps
going
back
and
forth.
B
J
B
B
This
is
a
case
for
MPLS
transport.
Essentially,
you
have
a
LSP
from
P
1
to
P
4
and
you
set
up
link
production.
So
if
the
link
between
P
2
and
P
3
fails
a1,
then
I
have
a
backup
path
via
p5
and
p6,
and
if
the
link
between
p5
and
p6
fails,
I
have
a
backup
path
via
P
2
and
P
3.
So
normally
what
happens?
Is
packets
go
from
p1?
B
You
know
all
the
way:
P
2
P
3
P
4,
if
I
want,
kill,
they'll,
go
P,
1,
P,
2,
P,
5,
P,
6,
P,
4,
P,
P
3
P
4,
but
if
you
have
a
situation
where
L
1
and
L
2
both
fail,
maybe
because
you
didn't
realize
that
they're
actually
sharing
page
or
maybe
that
information
didn't
get
propagated
to
the
past
computation,
then
you
have
a
situation
where
the
packet
goes
from.
P
1
to
P
2
is
saying
I'm
trying
to
get
to
P,
4
P
2
says
L.
B
B
So,
in
the
of
our
mind
you
can
you
can
do
the
build
on
the
on
the
right,
but
in
the
case
of
Palomar,
I
won't
go
into
it
in
detail.
Essentially,
you
have
two
counter-rotating
LSPs
for
every
destination,
so
in
this
case
we're
looking
at
the
destination
being
node
1
and
if,
if
you're
trying
to
get,
can
you
just
hit
next?
B
G
A
B
B
If
it's
not
zero
and
node
zero
says:
okay,
it's
not
working
I'm
gonna
flip
around
and
go
the
other
way
and
you
go
all
the
way
around
and
hit
node
2
and
node
2
says
all
my
LinkedIn
or
one
is
also
broken.
It
doesn't
know
that
it's
also
so
it
says
my
link
is
broken,
so
it
flips
back
and
so
now
you
have
this
packets
going
back
and
forth
around
the
rain
again.
The
fast
reroute
turns
out
to
be
a
bad
thing,
because
you're
not
ingesting
the
entire
ring.
B
B
So
here's
how
things
would
work
with
without
NF
o
RR
you
can.
You
can
build
this
out,
Eric
yeah
yeah,
so
there's
nothing
unexpected
here.
Essentially,
you
know
we
explain
how
this
works,
but
I
just
put
down
a
few
labels
here:
there's
a
ton
of
label
from
p3
to
p1,
there's
a
ton
of
label
from
p1
to
p2
and
from
E
to
to
p1
and
then
there's
a
via
the
VPN
labels.
So
I'll
assume
that
you
guys
understand
this
and
go
to
the
next
slide.
B
But
what
happens
here
is
p1
says
I'm
actually
going
to
put
the
label
static,
TL,
2,
VL,
2,
n,
f,
RR,
+
FF
are
are
essentially
saying:
hey
I've
already
done
faster
out.
If
you
get
this
packet
and
you
can't
deliver
it
straight
away,
the
way
it
works
is
with
penultimate
hop,
popping
or
whatever
p2.
The
moves
are
already
a
steal
to
the
moved.
B
B
I
brought
this
packet
so
instead
of
ingesting
that
link,
l-3
on
the
packet
gets
dropped
a
beat
and,
of
course,
all
the
extra
unnecessary
and
useless
work
that
p1
and
p2
would
perform
at
the
packet
was
painting.
So
so,
basically,
this
is
a
special-purpose
label
that
can
be
used
in
this
context
to
indicate
that
p1
or
somebody
else
has
already
done
fast,
readout
and
so
a
second
party
route
may
not
be
actually
helpful.
B
B
One
question
is:
should
we
do
a
special
label,
special
purpose
level
of
any
kind
and
that's
kind
of
addressed
in
the
draft?
It's
it's
good
to
have
the
semantics.
You
know
a
because
it's
it's
a
common
semantics,
so
I
don't
want
to
have
a
different
solution
for
VPNs
and
EVP
Aerith
and
MPLS
transport
and
MPLS
rings.
I
think
this
is
an
important
or
a
semantics
that
we
should
have
a
common
solution.
The
second
thing
is
why
a
special
purpose
label
and
not
an
extended
special
purpose
label.
B
He
said
so
so
we
do
request
that
we
get
a
special
purpose
label
we'd
like
to
have
early
allocation
with
label
value
H,
so
that
we
can
prototype
this
and
see
how
it
works,
and
there
are
a
couple
of
questions
you
know
exactly
where
you
put
the
NFF
are
our
label
few?
Should
it
be
the
bottom
label
and
the
stack
in
the
EVP
NK?
So
it
should
be
the
second
label
below
the
tunnel
label
or
no
it
should
there
be
some
other
place.
You
put
it
so
so
I
think
doing
it.
B
Early
allocation
will
allow
us
to
do
some
of
that
next
and
I
realized
that
we
have
a
couple
of
questions
here,
but
I
do
think
that
we
should
have
discussion
of
this
draft
on
the
list.
It's
a
new
draft
and
a
new
idea
and
sort
of
the
next
step
beyond
passing
out,
which
I
think
was
a
great
thing
when
it's
been
around
for
a
while.
We
need
companion
documents
to
signal
the
ability
to
process
an
FF,
our
are
sort
of
like
the
entropy
label
capability.
B
This
would
be
the
NFF
or
our
capability,
and
we
at
least
need
that
in
BGP
and
IGP
and
we
might
be
different
other
things
and
then
there
is
a
question
of
when
do
you
actually
put
an
FF
or
our
label?
What
are
the
kind
of
computations
you
might
want
to
do?
What
are
the
algorithms
might
want
to
use
if
you
use
n,
FF
or
all
the
time
whenever
you've
done
a
farsi
route?
Even
if
a
second
class
we
haven't
heard,
then
you
get
unnecessary
packet
loss.
B
K
You
ready
it's
very
interesting.
What
I
wonder
is
that
have
you
looked
at
how
I
am
I'll
be
can
be
used
to
in
this
scenario,
because
I
think
that,
in
a
scenario
that
you
closely
investigate
it
with
your
home,
so
you
we
might
expect
that
there
is
a
session
between
peas
and
see,
and
then
there
is
a
big
discussion
between
or
not
be
rebut
om
in
the
tunnel
from
p1
to
p2
and
p3
p2
p2.
K
B
I'm
not
sure
what
concatenate
the
path
is,
but
but
I
mean
the
fact
that
cu1
has
failed
is
something
that
both
p1
and
p2
can
find
out
quickly
using
EFT
or
whatever
you're
using
and
now
so
as
soon
as
the
packet
hits
p1
it's
going
to
be
sent
to
P,
2
and
P
2
is
going
to.
You
know,
see
it.
This
a
packet
doesn't
know
where
it
came
from
and
send
it
back
to
be
1.
The
problem
is,
you
know,
especially
if
there's
not
a
direct
link
between
P
and
E
2.
B
You
don't
know
where
it's
coming
from,
so
you
don't
know
if
it's
coming
from
p3
or
p4
or
some
other
P
in
the
network,
or
it's
coming
specifically
from
p1
and
specifically
because
it
was
from
you
know,
a
packet
that
p1
did
a
fast
reroute
on.
So
we
can
kind
of
maybe
invent
some
ways
of
doing
this
with
OAM,
but
that
is
a
much
more
complex
process
and
faster
is
happening
very,
very
fast
in
the
some
point.
B
B
A
Instead,
I
think
I'm
next,
in
the
queue
my
name
is
tarik
with
Jennifer.
So
I
think
this
is
an
important
piece
if
you're
highlighting
treaty
and
indeed
Annie
VPN,
you
know
there
is
a
potential
that
a
Luke
can
occur
in
that
case.
My
question
is,
you
know
if
he
PN
has
has
capabilities
it
to
signal
this
split
Verizon
Group
label,
to
avoid
you
know,
flooded
packets
to
come
back
to
the
access?
A
It's
it's
not
exactly
a
4r
its,
but
it
uses
a
mechanism
to
allocate
a
specific
label
to
ask
the
other
peer
in
the
multi-home
pair
to
drop
in
certain
cases.
Did
you
consider
or
are
you
considering,
210
signaling
in,
for
example,
in
evpn
2,
a
label,
not
necessarily
the
special
label?
And
if
not,
why
so.
C
Yeah
I
can
I
take
that
turret.
Okay,
so,
first
of
all
the
yes
I
laid
low
in
the
evpn
is
to
recognize
the
origin
of
the
packet
so
ESI.
Today's
evpn
is
recognize
the
packet
coming
from
ze
instead
of
coming
from
under
the
PE,
so
we
are
trying
to
do
fr
in
this
case,
traffic
is
coming
from
under
the
P
now
from
Z.
So
let's
put
it
this
yes,
I
label,
our
site.
Second
delay
the
questioners.
It's
a
good
question
whether
we
can
have
another
label
to
do
this.
We
also
being
thinking
about
this.
C
It's
very
useful
to
do
it
per
interface,
but
this
would
require
every
eating,
a
settlement
which
can
be
in
the
service
provider
mode,
a
sub
interface
by
thousands
of
certain
interface
to
have
under
the
label.
So
by
by
using
one
general
mechanism
presented
in
Korea's
presentation,
it's
we
think
it's
a
scale.
More
skill
like
a
scale
up,
will
only
need
to
using
one
label
to
do
this.
So
yeah
transfer
your
question.
Yes,
it's
doable
to
have
an
actual
label,
but
to
considering
scale.
C
A
F
H
F
B
Sure
and
I
agree
with
all
that
I
put
to
the
question
now
so
that
we
can
start
thinking
we
can
get
you
know,
expert
review.
We
do
definitely
want
to
progress.
The
draft
as
well-
and
you
know,
from
a
private
implementation
point
of
view.
We
can
do
some
of
these
without
that,
but
I
think
it
would
be
nice
because
some
of
this
code
goes
into
micro
code,
which
is
not
so
easy
to
change.
F
J
J
So
first
I
support
the
idea
of
having
and
he
do
not
reroute
again
label
in
the
stack
having
an
indication
that
said
that
already
done
repair
him.
It
shouldn't
be
done
again
is
something
that
Mike
Shand
and
Eileen
looked
at
many
many
years
ago,
when
we
were
doing
some
of
the
very
early
work
in
this,
and
there
were
several
schemes
that
we
couldn't
use
because
we
couldn't
reliably
indicate
whether
a
repair
it
previously
happened.
So
I
think
a
good
idea
for
us
to
are
they
concerned
about.
J
J
Nervous
about
doing
early
allocation,
however,
because
this
is
an
incredibly
rare
resource
that
we
are
around
using
and
we
should
be
extremely
parsimonious
with
it
until
such
time
as
we
know
exactly
what
the
scope
of
the
label
is
and
what
the
demand
for
it
will
turn
out
to
be.
What
we
don't
I
think
don't
have
is
an
x-ray
beam
eed,
an
experimental
single
label
use
experiments
to
happen
on.
You
were
talking
about
micro,
micro,
T's
by
definition,
changeable,
so
I
would
hate
to
see
a
sort
of
burn
the
label
and
then
make
a
mistake.
E
L
B
True,
the
the
thing
is,
the
probability
of
two
failures
is
generally
your
much
lower
than
probability
of
signal
failure.
Although
the
cases
that
look
like
two
failures
when
they're,
actually
a
single
failure,
like
the
CD
case
or
in
the
transport
case,
it
was
p3
that
failed.
It
looks
like
to
failure,
but
actually
it's
I
just
won.
B
Having
said
that,
yes,
it
does
depend
on
the
ecology
and
a
little
bit
of
type
is
explored
in
in
in
the
draft
where
there
are
actually
three
ways
of
getting
from
P
1
to
P
for
the
numbers
or
the
names
are
different
in
the
draft,
and
so
even
when
you
apply
the
know
for
the
posterior,
because
it's
load
balancing,
if
you
take
a
particular
path,
it's
as
yeah.
This
is
fine.
I'm
gonna
go
there
and
if
you
take
a
different
father
says
no,
that's
not
gonna
work,
so
I'll
drop
the
packet.
B
So
the
thing
is
by
putting
an
indication
of
neutral
for
the
fossil
or
in
the
packet.
You
save
this
potentially
very
bad
situation,
but,
like
I
said
in
the
further
work
we
do
have
to
look
at
techniques.
You
know
algorithms
and
computation
and
stuff,
and
that
allow
us
to
try
to
be
a
little
more
selective
about
when
we
use
it
today.
We
don't
have
it
at
all
and
I
think
that's
pretty
bad
excuse
all
the
time.
That's
slightly
less
that
feel
bad,
so
maybe
there's
a
better
way
of
playing
it.
A
But
I
we
encourage,
you
know
the
participants
to
you
know,
take
it
to
the
alias
and
if
you
have
any
comments,
remarks
or
concerns,
and
let's
follow
up
on
that.
So
next
time,
I'm
presenting
the
PDF
slides
I
can
I
get
an
acknowledgement.
First,
Rakesh
is
next
and
you're
there
as
well
as
you
can
see
your
slides
on.
E
E
E
E
E
There
are
three
measurement
modes
is
defined
in
RFC
six.
It
is
only
four
and
applicable
to
SR
policy
or
SR
pass
as
well,
using
IP
UDP
for
one-way
mode,
using
the
6624
in
wood
accents
or
two
week.
We
have
clarified
using
a
control
code
for
two-way
mode,
and
there
is
different
party
LV
as
well.
If
there
are
cases
where
we
want
to
measure
to
a
more
for
bi-directional,
SL
policy,
for
example,
and
look
back
more
basically
carries
lipstick
for
the
forward
and
reverse
path
in
this
case
thanks
the
next
slide.
E
So,
as
mentioned
the
written
part
clearly
for
two-way
mode,
there
is
the
segment
least:
labels
will
be
binding,
sealed
and
we
identified
its
mandatory
ethically
and
if
it's
not
supported,
then
there
is
an
error.
Define
in
six,
which
is
24
for
unsupported,
mandatory
objects
should
be
written,
must
be
returned
next.
One.
E
E
There
is
a
destination
a
test
here
we
define
in
1674
and
the
delay
and
lost
measurement
news
cases
in
segments
routing.
The
intended
node
must
reply
when
only
if
it's
matching
the
destination
address,
otherwise
it
must
return
an
error.
This
is
to
avoid
there
any
miss
calculation
of
matrix
and
affecting
the
segment
routing.
The
routing
decisions
are
made
based
on
this,
so
it's
must
next
one.
E
A
You're
asking
for
proceeding
ahead
with
the
adoption
of
this
document.
I
I,
don't
know
how
to
do
a
poll
of
how
many
read
the
document
on
WebEx,
so
I
I
think
we'll
have
to
take
it.
The
email
list-
and
you
know,
and
and
and
move
forward
with
the
regular
process
and
MPLS
working
group
with
a
review
team,
expect
the
review
team
to
take
a
look
at
your
document
and
review
it
and
in
moving
forward
with
us
I.
A
My
question
to
you:
Rakesh
there
was
this.
This
was
asked
last
time,
I
believe
the
return
path.
Tlv
there
is
a
mention
and
in
other
IETF
drafts,
specifically
in
I,
am
for
this
repair
return
path.
Tlb
are
you
working
with
other
drafts
to
you
know
is
the
possibility
to
unify
this,
or
is
it
or
your
thinking?
They
are
separate
the
return
path,
dl
v,
4o
am
or
b
FD.
I
think
it
was
is
different
from
this
last
measurement.
E
Yeah
this
was
asked
last
time,
so
what
we
did
is
we
made
the
structures
and
definitions
similar
to
the
other
protocols,
but
scope
of
this
graph
is
a
certain
pls.
So
there
are
code
points
that
we
define
here
are
for
SRM
pls,
but
yeah
I
mean
if
there
are
other
drops.
If
you
were,
if
you
ever
extend
this
for
some
other
use
cases,
then
new
code
points
will
be
different
or
that.
K
Nick
thanks
for
mentioning
other
word,
yes,
I,
think
that
being
here
proposed
is
very
similar
to
the
what
in
the
working
group
document
aren't
being
directed
so
where
the
label
step
is
advertised,
and
it's
a
similar
to
the
work
being
discussed
in
the
spring
group.
So
definitely
would
be
good
to
work
together
and
try
to
come
up
with
a
more
generic
solution.
E
E
So
requirements
is
doing
a
performance
measurement
as
well
as
the
liveness
monitoring
in
SR
networks.
So,
basically
in
ads
for
SRM
plss,
a
v6
data
planes
is
focusing
on
this.
Our
MPLS
is
MPLS
working
group
sessom
also
support
Assad
is
CMPs.
Our
paths
and
idea
is
to
run
a
single
protocol
for
both
lightness
and
performance
measurement
in
SI
networks
to
simplify
the
deployment,
as
well
as
reduced
operational
complexity.
E
We
trying
to
eliminate
the
dependency
on
the
endpoint
on
the
remote
side,
no
state
on
the
endpoint.
So
it's
not
aware
of
the
protocol.
Also
the
the
requirement
was,
he
is
much
much
higher
now
and
the
first
a
bit
interval
requirement
as
well.
So
we
try
to
avoid
you
know:
packets,
not
one
get
out
of
a
spot
on
day
on
the
remote
side
to
to
achieve
such
a
scale
and
interval.
E
It
can
also
need
to
add
a
probe
messages
in
future
and
basically
using
the
user,
configure
the
IP
UDP
parts
for
the
probe
messages
next
slide,
so
a
liveness
monitoring,
4sr
policy
and
yes,
our
path,
uses
loopback
mode.
Look
back
mode
is
currently
not
defined.
41
ply
little
stamp
in
in
the
existing
RFC's,
so
this
introduces
this
mode
and
basically
the
probe
messages
are
sent
using
the
segment
list
in
case
our
policy
candidate
parts.
They
are
not
painted
on
the
reflected
node.
Basically,
they
come
back.
E
A
E
Slide
please
so
there
is
an
intense
look
back
more
defined
works.
The
same
way,
look
back
mode
except
there
is
a
network
programming
function,
enable
on
the
reflected
node.
So
when
it
detects
match
in
this
case,
MPLS
label
could
be
special
labeled.
It
will
also
insert
this
real-time
strap
in
the
pro
packet
before
forwarding
it.
So
idea
is
to
you
know,
timestamp
and
forward
the
packet
on
the
endpoint
node.
So
now
you
get
t1
and
t2
and
the
lightness
works
the
same
way.
E
If
using
like
probe
messages,
the
sender
would
add
the
transmit,
timestamp
reflector
would
add
the
receive
timestamp
at
offset
16.
This
is
enabled
by
the
network
programming
function
on
the
reflector
node.
Many
fields
are
not
used
in
this
SH
not
required
next
slide
same
thing
for
stamp
as
well,
so
we
can
skip
it
next
slide.
Please.
E
E
Please
just
like
other
HR
policy
p.m.
this
also
allows
us
to
handle
these
EMP
parts
by
using
the
127
/
8,
guessing
those
destination
addresses
and
also
for
ipv6.
We
can
use
the
label
next,
please
it's
based
on
the
user,
configure
UDP
path,
so
our
controller
or
any
Sdn
can
be
used
to
provision
on
the
sender
and
reflector
side.
Only
thing:
if
you're
using
enhance
mode
then
deflector
side.
There
are
times
at
this.
E
A
E
E
There
is
IP
I
work
being
done
in
IP,
PM
working
group
and
we
just
adding
the
MPLS
encap
for
it
next
slide.
Please,
the
dot
has
been
around
for
a
year
and
a
half.
Now
it
was
in
spring
and
got
mood
to
hear
less
working
group
was
presented
last
time,
so
we'll
only
discuss
the
Delta
since
the
last
meeting
thanks
next
day
and
cap
is
made
generic
to
MPLS
data,
plane
added
procedure
for
hop-by-hop
iom
and
address
various
review
comments.
E
There
are
two
labels,
one
for
the
edge
to
edge
case,
and
one
is
for
hop
by
hop
processing.
If
you
want
all
norsu
processor,
there
is
a
separate
label
font
and
next
line.
Please
there
is
a
another
flow
label,
define
to
indicate
Iowa
meant
go
type.
This
is
to
address
the
ECP
case
for
IP
header.
So
this
way
the
next
four
four
bits
are
0,
0,
0
and
I
piercing
will
ignore
the
not
mess
up
because
of
ipv4
ipv6,
miss
understanding.
E
E
So
it's
it!
It's
we're
asking
an
allocation
from
extended
special-purpose
label,
but
it
can
be
controller
allocated
or
signal
next
slide.
Please
we
can
skip
this,
and
this
is
well
the
next
steps
yeah
and
we
we
looking.
We
saw
you
shut
your
comments
and
suggestions
for
this
trap.
Okay,
thank
you
and
any
questions
on
this.
K
A
M
Yeah,
so
this
chapter
is
the
zeroth
revision
draft
and
we
have
a
several
major
changes.
According
to
the
comments
received
previously
and
the
first
one
is
we
change
from
the
base
special
purpose
labor
to
the
extended
special
purpose
labor
and
the
second
one
is
a
change
of
the
definition
of
the
flu
ID
Leber's,
TC
and
ttio
to
make
them
complaint
with
obviously
cert
here
32
and
the
text
comparison
with
I'm
PRC
ISA,
the
ISA
to
Graham
in
the
last
one
is
I,
am
the
some
text
on
the
security
consideration.
You
know
next
page.
M
So
this
page
issues
rule-based
performance
measurement
and
capitalization
in
a
composite
special-purpose
labor
we
with
extension,
labor
15
plus
the
flow
ID
labor,
indicate
followed
by
through
ID
labor.
The
full
ID
labor
is
used
as
an
PRS.
The
flu
indication,
the
TC
and
TTL
of
the
flu
ID
labor
emitted,
that
of
an
Interpol
labor,
except
that
TC
of
the
fluid
flow
ID
labor
is
a
recommended
answer,
column
making
kalamaki
filled
so
next
page.
M
M
Like
issues
that
the
comparison
mr.
easy
to
om
the
major
differences
between
the
post
solution,
there
are
two
points:
the
first
one
amperes,
the
embedded
performance
measurement,
doesn't
introduce
any
new
hider,
whereas
I'm
just
in
situ
om
needs
a
new
hiding,
which
means
maybe
additional
basic
requirements.
M
The
decided
the
second
point,
an
PRC
in
bad
performance
in
this
measurement,
allows
the
network
note
to
report
the
refund
it.
For
example,
they
calculated
the
performance
which
occurs
Oh
sushi
80
degrees,
spice
excited
to
move
we're
after
amphiaraus
in
situ
who
I
am
the
request.
Leaders
Network
note
to
report
data,
for
example,
the
single
eyes
in
sales
and
the
ego
accident
face
determine
our
social
interior
with
red
packet
latest.
M
So
we
often
standard
security,
configuration
Universal
and
the
major
points
on
the
confidence
region,
including
the
first
appoint
who
I
deliver
indicate
and
a
full
I
deliver,
must
not
be
singled
and
distributed
outside
one
performance
measurement.
Women
and
a
second
point
is
that
to
prevent
it's
carrying
through
I
deliver
from
leaking
from
one
domain
to
another
domain.