►
From YouTube: IETF98-SPRING-20170328-1300
Description
SPRING meeting session at IETF98
2017/03/28 1300
A
B
B
B
B
Thank
you.
So
back
in
Co,
that
was
a
slide
with
blinking
red
on
it.
We
had
missing
IPR
answers.
We
had
a
lot
of
today
and
we've
managed
we
had
set
quite
an
aggressive
target
and
we've
managed
to
reach
it.
So
thank
you
to
the
working
group.
Thank
you
to
the
reviewers
who
made
comments
during
the
last
call.
Thank
you
to
the
authors
who
have
responded
quickly.
B
B
And
you
should
read
Adrian's
draft
on
the
definition
of
soon
it's
only
the
beginning.
Why?
Because
authors-
and
this
is
this-
concerns
you
please
be
aware
of
the
different
reviews
that
will
be
returned
to
you
on
my
ESG
or
from
different
I
like
tonight's.
The
last
call
comments,
I
etfs
call,
but
also
all
the
IFC
editor
of
faith
and
so
on,
so
that
we
move
smoothly
and
don't
forget
to
thank
alcohol
who's,
getting
a
load
on
his
shoulder
to
treat
these
documents.
B
So
with
that,
the
working
group
nearly
doesn't
have
any
more
any
working
on
document.
Only
very
few
of
them,
we've
listed
to
which
are
ready
for
last
call.
We
believe
and
the
author's
too
so
any
cash
segments
and
I
saw
younger
as
a
third
one,
which
is
not
mentioned
on
the
slide,
which
is
the
SR
conflict
resolution
which
we'll
be
taking
a
decision
on
that
on
that
document.
But
we
expect
it
to
be
ready
in
its
current
shape,
to
go
to
last
call.
B
So
I
will
do
it
next
steps.
So
since
we've
Wow,
we
are
approaching
the
end
of
our
milestones.
We
have
and
we
will
have
sent
most
all
of
our
working
group
documents
to
a
history.
So
it
start
to.
It
is
time
to
start
thinking
about
the
next
steps.
We
are
opening
the
discussion
today.
We
plan
on
continuing
it
on
the
list,
but
we
would
welcome
to
see
it
happening
today
already
on
the
mic.
So
it's
really
open.
Should
we
read
chapter?
Should
we
close?
Should
we
leave
for
a
while?
B
D
Joseph
of
orange,
my
point
of
view
is
that
we
have
now
will
pro
implementations
of
segment
all
team,
but
there
is
very
few
deployments
it's
just
starting,
so
we
don't
have
or
your
operational
feedback
on
how
it
works
on
a
live
network.
So
there
may
be
some
things
to
change
in
the
future.
So
clearly
closing
the
rocking
hope
now
may
be
too
early.
E
Maybe
we
should
wait
for
the
next
presentations
about
segment,
writing
policies
and
the
metric
programming
and
see
if
those
two
drafts
fits
into
the
current
charter
or
if
they
require
recharging
or
extending
the
charter
of
the
group,
but
clearly
there
is
still
quite
a
few
things
to
do
in
spring.
So
for
me
it
has
to
be
trust.
F
Hi
I'm
gonna
return
already
so
I
agree.
Closing
the
worker
who
at
this
point
is
something
we
shouldn't
do
I
think
the
two
options
are
we
charter
or
go
to
sleep
I?
Remember
that
this
working
group
was
charter
originating
mostly
to
do
the
use
cases
and
the
architecture,
but
all
the
extensions
are
being
done
somewhere
else,
right,
lsbf
and
everywhere
else,
so
at
minimum.
If
there's
nothing
else,
we
need
to
stay
open
until
at
least
all
those
extensions
get
published.
It
may
be
either
that
it's
short
term
and
be
longer
term.
F
G
So,
while
I
pretty
much
done
with
mpls
three
sixes,
he
had
to
come
all
the
possible
issues.
Implementation
feed,
but
we
haven't
even
started
in
most
cases
so
again
to
reiterate,
have
been
set.
Closing
has
definitely
too
early
and
v6
feedback
should
come
into
the
group
and
result
in
more
work,
and
thank
you
a.
H
Cylinder,
even
without
recharter
and
there's
a
few
documents
that
are
proposed
when
even
working
group
documents,
for
instance,
the
spring
conflict
rebel
resolution
that
we
took
into
this
working
group,
because
it
was
a
generic
across
all
the
protocols
and
entities
assigning
labels.
So
you
know
that
would
need
to
be
done
in
there
are
additional
I
would
call
them
use
case
documents
as
well
for
post,
so
even
yet,
even
even
before
we
charter
there's
plenty
to
work
during
the
retarding
process
and
we
we
actually
took
all
those
details.
H
B
I
I
B
D
Okay,
so
the
first
presentation
is
about:
how
do
we
build
non-protected
paths
using
segments
mounting
so
in
sales?
We
present
you
the
first
version
with
a
couple
of
scenarios,
couple
of
theoretical
possibilities
to
visit
with
non-protected
path,
and
the
goal
of
this
draft
is
really
to
find
a
single
standard
solution.
So
all
implementation
will
implement
it.
So
we
will
ensure
the
deployability
of
this
kind
of
use
case.
D
So
globally
we
have,
we
identified
five
different
possibilities
to
build
this
non-protected
path,
so
the
first
one
is
to
use
only
adjacency
seed
that
are
not
protected,
so
I
just
insisted
that
advertised
with
the
be
flag
and
set.
The
second
scenario
from
a
theoretical
point
of
view,
was
to
use
adjacency
seed
combined
with
Street
SPF,
not
see.
D
So
we
tried
in
the
graph
to
do
an
analysis
of
all
the
different
scenario.
So
this
is
a
kind
of
summary,
but
very
small
details
in
the
draft,
so
we
are
analyzing
via
impact
on
the
label
stack
size
so,
for
example,
the
edges
and
CC
the
only
solution
as
a
huge
impact
in
terms
of
size
of
via
label
stack.
This
must
be
taken
into
account,
as
we
are
aware
today
may
be
limited.
That's
why?
D
Combining
with
whatever
bindings
in
do
not
see
provides
something
which
is
better
now
in
term
of
control,
plane
states
that
we
have
to
maintain.
So
if
we
need
to
advertise
couple
of
more
adjacent
secede,
it's
not
really
a
big
deal
while
edging
north
cedar
is
more
impacting
because
it's
global,
so
each
node
will
have
to
to
install
it.
So
there
is
an
impact
on
the
control
plane,
but
also
in
the
data
plane
and
for
V
bindings,
it's
more
variable.
It
will
really
depends
of
the
intelligence
of
the
controller.
D
D
D
So,
in
this
case
of
you
address
and
secede
and
by
Nasheed
scenario
is
clearly
better
in
terms
of
what
is
the
impact
on
the
computation
of
the
path?
So
if
we
are
using
address
and
secede
only,
we
just
need
to
look
at
the
flag
and
select
the
right.
I
just
insisted,
so
just
look
at
the
B
flags
unset
and
we
can
select
it
for
the
strict
SPF
and
address
and
CCD.
It's
also
a
kind
of
policy.
We
can
select
it
easily
where
it's
more
complex,
it's
using
nodal
segments
with
a
local
policy.
D
Otherwise
we
already
have
existing
mechanism
to
do
to
do
the
job
there
was
one
point
raised
in
the
description.
Is
that
for
most
non
protected
path?
Do
we
allow
ecmp
or
not?
Clearly
it's
a
question
of
use
case?
Sometimes
it
may
be
a
road
sometimes
not,
for
example,
in
our
use
case,
we
want
also
to
avoid
any
kind
of
rerouting
so
have
an
ICMP.
You
want
to
avoid
it,
so
in
this
case
or
lineages
and
sidon
by
in
CDs,
is
authorized
so
globally.
D
If
we
are
look
at
all
the
requirements,
only
the
adjacency
seed
and
binding
see
the
scenario
is
providing
a
full
coverage
of
your
requirement.
Now,
depending
of
the
use
case,
if
you
want
to
relax
some
of
the
constraints
or
telling
that
you
authorized
some
transient
rerouting,
yes,
you
can
use
nodal
segment
scenarios.
D
One
of
our
point
that
is
interesting
to
look
at
that
forms
binding
sheet
scenario,
depending
on
the
protocols
that
we
will
use
to
program
those
binding
seed.
We
may
need
to
already
have
a
protocol
session
to
all
the
nodes
of
the
network.
Let's
say
for
a
let's
take
the
example
of
KISS
app,
for
example,
precepts
client-server
protocol,
so
I
need
to
have
all
my
pieces
to
be
already
connected
to
the
controller
to
be
able
to
program
the
seal.
D
So
this
may
have
an
impact,
for
example,
on
the
controller,
because
we
don't
know
where
we
want
to
program
the
particular
binding
scene.
So
the
better
scenario
will
be
true
for
the
controller
to
be
able
to
dynamically
instantiate
a
protocol
session,
for
example,
when
it
is
needed,
it's
questionable,
for
example
using
BGP.
If
you
want
to
do
it
using
precept,
we
may
have
some
to
change
some
things
in
the
protocol.
D
So
the
conclusion
of
our
analysis,
but
please
let
us
know
if
you
agree
or
not-
is
that
the
best
approach
is
clearly
to
use
only
adjacency
seed
with
B
flags
and
set
it's
the
easiest
one.
But
there
is
a
huge
impact
on
the
label
stack
so
clearly,
if
we
look
at
maybe
in
10
years.
Yes,
that
would
be
cool
because
we
will
have
an
odd
way
to
do
so,
but
in
the
meantime
we
really
think
that
the
transient
solution
may
be
to
combine
it
with
binding
C.
D
D
J
Can
reach
me
co-chair
just
a
side
comment
free
support,
this
kind
of
initiative,
because
it
really
clearly
tried
to
address
interoperability
issue
that
we
may
face
in
deployment,
so
full
support.
Just
side
comment
on
this
pset
thing
you
said
before
there
is
no
change
required
in
the
protocol
to
set
up
a
session
in
the
other
direction
good
as
a
matter
of
identifying
your
being
addresses.
What.
K
J
G
K
D
For
BGP,
as
far
as
I
know
in
terms
of
implementation,
so
implementation
of
already
supporting
the
fact
of
authorizing
a
set
of
peers,
for
example,
range.
So
if
your
PC
address,
if
your
controller
address
is
within
the
range,
will
be
authorized
to
dynamic
instances
decision
without
doing
additional
configuration
on
the
pier.
F
A
D
Okay,
so
next
presentation
is
about
segment,
routing
traffic
engineering
policies,
so
the
kauravas
do
of
the
draft
is
to
give
some
details
on
how
must
be
implemented
a
via
segment,
routing
traffic
engineering,
and
it's
mainly
focusing
and
this
concept
of
segment
routing
policy.
So
we
define
what
is
the
policy?
How
does
it
work?
What
are
the
component
of
his
policy
and
what
are
the
different
methods
that
we
could
use
to
steal
so
traffic
into
this
its
position?
So
it's
completely
data
plane
agnostic,
so
we
can
use
SL,
v6
or
SRM
pls.
D
So
the
draft
is
starting
with
some
generic
consideration
about
traffic
engineering,
but
it's
very
similar
to
what
is
existing
with
rsvp-te
today.
So
with
assad's,
the
architecture
is
for
multi
domain
capable,
so
we
are
able
to
learn
topology
so
also,
multiple
way
we
can
use
pls
AGP,
not
confer
or
whatever
it
could
be
senior
re
armintie
per
area.
Multiple
domain-
oh
yes,
clearly,
straightforward.
D
The
main
point
of
this
draft
is
this
concept
of
segment
routing
policy.
It
identified
the
key
of
this
segment
routing
policies
at
Apple,
which
is
identified
by
viadon,
where
this
particular
policy
will
be
instantiated.
There
is
also
the
endpoint.
So
what
is
the
target
of
a
traffic?
What
is
the
endpoint
for
is
a
destination
of
our
traffic
and
we
have
a
through
a
notion
of
color.
D
This
notion
of
color
will
help
some
specific
traffic
steering,
so
we
may
be
able
to
identify
some
destination
diagram
with
a
specific
color
on
automatically
was
a
destination
will
be
steered
towards
the
V
policy,
so
this
policy
may
be
also
created
in
multiple
ways.
It
could
be
instance,
eighty
through
a
BGP
piece,
AB,
not
conform
or
CLI
any
kind
of
interface.
D
This
is
a
policy,
so
it's
detailed
error.
So
we
have
this
key
of
add
color
on
endpoint
and
within
this
policy,
I
may
have
multiple
paths.
It
could
be
multiple
paths
coming
from
the
same
source
so,
for
example,
the
same
protocol
or
it
could
be
coming
from
different
sources
or
one
policy
coming
from
a
one
path
coming
from
vdp1
over
path
coming
from
P
separately,
depending
on
how
we
are
implementing
it.
D
So
we
have
those
preference
among
the
paths,
so
only
the
path
with
the
best
preference
would
be
for
sure,
install
in
the
forwarding
table
and
weaving
the
path
will
have
so
the
seed
list.
So
the
list
of
seed
to
push
onto
a
unto
the
packet
and
in
case
we
want
to
use
multiple
paths.
We
have
the
ability
to
have
multiple
list
of
seeds
with
different
weight.
If
we
want
to
do
an
equal
cost
paths,
load,
balancing.
D
So
this
notion
of
segment
routing
policy
is
also
linked
to
the
binding
seal,
so
VI
deists
for
each
policy
we
associate
automatically
bindings.
So
this
clearly
will
help
the
traffic
steering.
We
will
see
how
what
would
be
the
traffic
steering
methodologies
just
slide
after
and
globally.
It
will,
as
I
shown
in
the
previous
presentation,
using
a
balance.
It
will
always
allow
to
reduce
the
label
stack
that
we
will
push
into
the
packet.
D
D
So
how
can
we
steer
the
traffic
so
so
first
method
is
the
source?
Will
other
traffic
will
directly
put
the
path
socialists
using
pasture?
For
example,
your
binding
said
so,
for
example,
to
stir
the
traffic
to
Z.
It
will
use
via
JSON
C
segment
from
A
to
B
an
adjacent
segment
from
B
to
H,
and
then
H
will
expand
the
binding
seed
that
he
advertised
and
it
will
expand
it
to
adjacent
CC
to
from
C
to
D
and
G
to
G
and
finally,
we
will
which
we
will
reach.
D
So
this
is
the
first
way
to
use
the
segment
routing
policy,
so
the
source
is
directly
putting
the
corresponding
binding
C.
We
are
also
able
to
use
this
segment
well
in
policy
with
BGP
routes.
So
here
we
can
use
this
notion
of
color.
So,
for
example,
we
are
receiving
a
BGP
route
coming
from
this
on
extra
PC
and
we
are
associating
this
color
of
C.
D
D
Segment
routing
policy
which
has
a
cross
C,
so
it
will
from
a
forwarding
point
of
view,
it
will
do
a
recursion,
for
example,
on
the
binding
seat.
So
my
prefix
10/8
I
will
first
push
the
label
L,
which
was
bound
to
the
BGP
route
and
then
I
will
have
an
indirection
towards
the
binding
seed,
and
this
binding
seed
will
be
bound
to
be
to
read
a
list
of
seed.
So
it's
clearly
a
year
article
fever
scenarios
that
we
are
implementing
in
the
routers.
D
D
Like
we
are
doing
today
with
our
GP,
we
can
also
do
some
class
based
traffic
steering,
so
I
can
configure
on
particular
interface
traffic
steering
policy
to
tell
that
to
particular
forwarding
class.
For
example,
my
forwarding
class
0
must
be
bound
to
the
a
chips
path.
My
forwarding
class
1
must
be
bound
to
a
particular
segment
routing
policy,
so
in
this
case
it
will
be
a
recursion
towards
binding
cid
associated
to
this.
G
G
See
the
value
X
to
Y,
there's
some
recursion
possible,
but
it
still
see
you
later
even
away,
so
my
proposal
for
the
future
specifically
would
be
rather
to
decouple
seed
value
from
what
it
means.
You're
overloading
it,
because
you're
saying
this
value
means
that
rather,
we
should
be
able
to
ask
someone
else.
What
does
this
value
really
mean
to
provide?
Do.
D
G
H
H
L
H
And
it's
a
it's
good
companion
and
that
we
really
need
to
have
a
discussion
on
the
idea,
our
list
of
making
that
a
working
group
document,
because
I
think
but
prior
to
this
document,
you
really
couldn't
understand
and
appreciate
how
one
would
use
it
when
this
puts
it
all
in
the
right
perspective.
Right,
yeah,.
E
Answering
to
what
AC
just
raised.
Indeed,
before
that
document,
we
had
the
BGP
SRT
policy
we
removed
from
the
IDR
draft.
Whatever
is
related
to
the
definition
of
an
SR
policy,
and
we
left
in
the
idea
are
in
the
BGP,
draw
only
the
signalling
the
control
plane,
part,
which
is
just
the
BGP
mechanics
and
the
point
that
Jeff
raised
about
CLI
and
how
to
impose
a
given
value
for
a
binding
City.
M
Hey
hello,
everyone
I'm
comrade
I'm
present
this
version:
zero,
zero
draft
service,
its
network
programming,
on
behalf
of
this
long
list
of
authors,
so
so
that
s
our
networks,
programming
is
basically
refers
to
capability
of
an
application
that
allows
an
application
to
encode
any
complex
program
by
a
means
of
set
of
individual
functions
which
are
distributed
all
across
the
service
six
network.
You
know,
example
of
functions,
could
be
the
underlay
LS
says,
or
overlays
or
complex
functions,
English
implemented
within
some
VM
or
container.
So
so
this
draft
is
basically
key.
M
A
services
document
that
describes
a
service,
6
network
programming
concepts,
the
various
functions
which
are
associated
with
those
and
and
the
use
cases
of
an
use
cases
which
use
these
functions
and
those
use
cases
are
illustrated
at
the
end
of
the
document.
So,
specifically,
the
document
it
talks
about
local
state
functions,
transit
behavior
for
a
nodes.
The
control
plane
is
outside
the
scope
of
this
document,
but
we
touch
upon
a
little
little
bit
about
the
control,
pain
and
their
role
while
advertising
or
signaling.
M
These
sets
also
document
talks
a
little
bit
about
conquerors,
insecurity
and,
and
then
we
have
some
use
Casilla
stations
which
would
solve
them
all
of
them
together
to
better
explain
these
functions
so
quickly,
status
on
a
service,
6,
Network
programming.
So
going
back
to
this
slide.
So
this
services
network
programming
has
already
a
larger
wider
community
support.
There's
lots
of
vendors
and
operators
which
are
already
on
board
and
and
working
closely
and
in
future
we
expect
to
have
a
few
more
joining
as
we
we
go
further
and
in
terms
of
implementation.
M
We
already
have
a
few
interoperable
implementation,
which
are
which
have
either
implemented
most
of
these
functions
defined
in
since
document,
or
at
least
some
of
them,
which
are
already
working
and
rest
is
in
progress.
So
we
encourage
you
to
look
at
this
page,
open
software
project
page,
which
already
lists
lists
to
implement
to
implementations,
which
one
of
them
is
VPP
based,
cisco,
funded
implementation,
which
has
software,
which
has
of
the
support
for
all
these
as
our
v6
network
functions
and
another
one.
Is
linux
kernel
based
initiative
to
make
it
a
service
ready
right?
M
So
before
we
go
to
the
function,
so
a
local
is
is,
is,
is
a
set
which
is
an
ipv6
address,
having
a
specific
instruction
bound
to
it.
So
we
we
assume
that
we
expect
that
every
doctor
has
a
table
of
these
sets.
What
we
call
my
said
local
table
and
this
this
table
this
table
basically
captures
all
the
sets
which
are
a
service
exceeds
which
are
explicitly
instantiated
on
that
node.
In
other
words,
the
node,
which
is
the
N,
is
a
node
which
is
it's
a
parent
for
all
these
local
sets.
M
The
the
function
of
this
table
is
basically
to
keep
to
basically
maintain
all
those
local
states.
Instantiated
plus
you
know
their
instructions
which
are
associated
with
those
sets,
as
well
as
any
parameters
that
are
required
to
execute
the
instruction.
A
local
set
could
be
bound,
it
could
be
routed,
but
it
doesn't
have
to
be
routed.
Most
often,
you
know,
which
is
opted
by
a
shorter,
prefix
and
path.
M
So
if
it's
not,
if
it
is
not
out
it,
then
typically
a
set
list
we'll
have
a
routed
set
preceded
by
a
non-local
set,
will
be
preceded
by
a
router
set.
So
that
we
saw
that
finally,
a
toughy
could
be
state,
you
know
now
could
sit.
This
is
no
different
than
local
segments
in
MPLS,
where
you
need
to
have
first,
a
group
of
segments
preceded
by
a
before
the
local
segment.
M
So
so
we
also
in
the
document
also
present
that
you
can
basically
a
divider
set
into
you,
can
divide
it
into
kind
of.
You
know,
inner
to
two
parts.
One
is
the
locator
parts.
Another
is
a
function.
Part
locator
part
is
the
one
which
basically
used
to
route
which
which
is
used
to
basically
route
traffic
to
the
node,
which
ones
that
said
where,
as
a
function,
part
is
really
local
part
and
which,
which
is
which
is
specifies
and
which
is
specifies
that
local
function
that
is
implemented.
M
Another
node,
which
is
located
by
the
locator
field
of
this
said
now,
you
can
you
can
divide
this
locator
and
function.
The
way
you
like,
there's,
no,
no
assumption
or
restriction.
The
draft
that
we
put
you
can
have.
You
know:
64
68,
/
48
for
your
for
your
location,
locator
and
rest
for
your
function.
M
So
also,
if
there
are
cases
where
you
where
a
function
is
an
argument,
you
can
even
further
divide
I
said
you
know
to
specify
your
arguments
in
the
in
the
least
significant
bits
out
of
your
function,
for
example,
for
the
application,
or
in
your
VM
case
they
may
require
some
arguments,
so
so
so
this
is
what
what
this
framework
in
the
in
the
document
we
have
so
about
the
state
functions
associated
with
all
these
shifts.
In
the
my
set
table.
There
are
certain
well-known
functions
that
we
have
a
little
listed
in
your
document.
M
It's
a
non-exhaustive
list
and
in
future
we
expect
to
have
more
functions
at
it
and
also
documents.
It
specifies
you
know,
for
each
function
on
a
very
high
level,
a
processing
algorithm
when
a
packet
arrives
and
what
you
need
to
do
the
packet.
When
you
process
for
a
given
set,
we
will
not
talk
about
those
of
you
arm
the
today's
talk
and
you're
encouraged
to
look
at
the
draft.
If
you,
you
want
to
look
at
further
details,
so
so
this
table
bus
just
summarizes
this
table
in
the
next
page
table.
M
It
summarizes
all
the
and
local
state
functions
we
have
defined
in
the
document,
so
basically
end
n,
dot,
X
and
n
dot
T.
These
are
the
functions
which
are
topological
functions,
typically
signaled
an
advertised
by
the
IGP
and
they're
forwarding
behavior
is
some
of
this.
Behavior
is
already
specified
and
a
document
draft
submitted
earlier
and
already
being
present.
M
M
We
also
where
n
dot
X
is
like
adjacent
state
function,
where
you
do
your
bit
your
destination
address
and
forward
using
a
using
a
little
L
3
cross,
connect
adjacency,
it
can
be
also
said,
as
is
specified
in
one
of
the
is,
is
a
table
so
so
first,
three
three
functions
are
really
you
know:
port
functions
to
implement
over
TI
l,
fa
and
SRT
policies
and
advertised
by
JP.
The
next
set
of
functions
which
this
table
captures
relate
to
VPN
in
mostly
in
tab
solution.
L,
2,
L,
3
L
to
be
pinochet
VPN.
M
So,
for
example,
for
m
dot
d
t6
function-
this
is
a
this
is
a
function
that
is,
that
is
associated
local
set
in
what
it
is
telling
us
that,
when
you
receive
a
packet
is
expected
to
be
encapsulated
to
an
outer
ipv6
header
and
possibly
and
segment
outing
header.
So
UD
cap
outer
ipv6
header,
as
well
as
sr
h
header
and
then
then
then
do
the
further
further
forwarding
of
the
d
capsulated
packet
after
doing
the
lookup.
M
So
this
is
equal
to
ipv6
l3
VPN
power
of
that
we
know
in
MPLS
similar
version
for
DT,
for
which,
which
will
do
the
same
thing
where
the
l3
VPN
v4
will
be
encapsulated
within
a
service
6,
sir
v6
outdoor
activities,
header
or
header,
plus
a
Suraj
d.
X6
versions
of
these
functions
are
equivalent
to
cross
connect
or
Percy,
as
we
know
in
MPLS
and
but
but
but
in
case
of
a
in
terms
of
decap.
There
is
this
common
functionality.
M
Only
the
forwarding
is
where
the
differences
in
one
case
is
a
lookup
other
cases,
cross,
connect
and
finally,
there's
also
l2
VPN
DK
function.
That
is
listed
here
where,
where
H
encapsulate
packet
is
l2
frame
and
it's
encapsulated
in
SR,
v6,
header
or
or
a
service
in
hundred-plus
SRH
and
the
cap
has
to
happen.
So
this
is
really
eager
the
functionality
that
that
these
functions
are
defining
the
other
set
of
functions.
I
mean
Stefan,
just
went
to
assert
policy
dog
and
we
talked
about
bindings
it
so
n.
M
Dot
b6
is
abundance,
instantiation
of
officer
v6
data
plane,
and
it
is
used
to
implement
basically
a
service
policy
in
our
draft.
If
there's
a
policy
which
needs
to
which
which
requires
us
to
change
source
address.
So
in
those
cases
we
we
define
n
dot,
B,
six
dot,
n
cap
functionality
and,
and
then
this
n
dot
BM
is
a
is
also
binding.
State
instantiation.
M
But
this
is
to
cover
the
case
when,
when
you
are,
we
are
trying
to
have
a
scalable
traffic
engine
policies
which
is
span
across
multiple
domains,
and
one
of
the
domain
is
and
pillars
remains.
So
if
you
want
to
traverse
an
N
plus
domain
is
still
implement
and
to
end
a
scalable
policy,
then
use
nvm
set
function
and,
and
finally
there
could
there
are
couple
more
functions
in
writing,
skip
but
basically
n
dot,
a
s
and
n
dot.
A
are
functions
that
are
for
service
chaining,
which
are
done
through
an
SR
on
our
application.
M
In
both
cases,
in
both
cases.
Important
point
to
note
is
that
these
s
are
unaware
applications
and
they
do
not
need
to
know,
and
you
know
it
cannot
look
into
SRH
adder.
So
in
one
case
outer
IP
layer
is
removed
and
sr
h
is
removed.
Another
case
the
outer
ipv6
header
has
been
updated
with
the
actual
final
destined
for
the
packet,
so
that
this,
this
Serrano
verification
can
process
just
based
on
the
packet
header
rather
than
looking
into
a
Suraj.
M
So
yeah
one
one
point
to
highlight
that
for
this
topological
GP
and
functionalities
or
and
set
functions
that
we
have
special
in
the
document.
There
are
two
variants
of
these
functions
that
we
have
specified
and
these
two
variants
are
really
related
to
to
popping
the
SRH
that
is
on
the
packet
and
popping
means
meaning
cleaning
up
the
SRH
or
top
SRH
and
these.
M
So
there
are
two
flavors
you
have
defined
ultimately,
segment
pop
and
plant
image
segment
pop
ultimate
segment
pop
is
a
flavour,
which
means
that
if
this
it
is
advertised
as
USP
and
functionality,
then
packet
will
come
all
the
way
up
to
the
ultimate
segment
which
owns
this.
This
said
and
SRH
will
be
parted
at
the
last
segment
versus
penalty
might
have
a
pentamer
segment
popping
a
suresh
will
be
pop
at
the
current
image
segment.
M
If
this
set
is
a
PSP
PSP
said
advertised
and
in
the
draft
we
have
kept
it
kept,
it
open
up
two
of
the
implementation
that
how
they
wanted
to
assign
and
control
this
USP
or
ESP
behavior,
or
they
these
variants
can
be
enable
or
disable
together
or
that
can
be
disabled
in
a
neighbor
individually
right.
So
so
also
we,
we
talked
about
transit
note
where
behavior,
so
by
definition
or
transit.
Node
is
the
node
which
receives
a
packet
ipv6
or
a
service
expeced
whose
destination
is
not
one
of
your
local
set.
M
So
basically
it's
really
a
transit
for
that
traffic.
So
so
there
are
four
transit
behavior
we
have
defined.
One
is
a
pure
transit
behavior.
This
means
this
could
be
a
pure
ipv6
router.
All
it
does
is
just
look
at
the
IP
destination
and
forwards
back
here
according
to
according
to
the
destination,
and
that
next
say
are
really
transit,
behavior,
which
which
somehow
impose
sr
v
six
header
or
in
cap
in
a
service
six,
so
t
dot
insert
is
doing
in
cases
where
you
wanted
to
cross.
M
You
wanted
to
change
an
incoming
incoming
packet
and
insert
and
insert
a
service
successor
and
SR.
Ss
re
insert
a
service
6r
our
insert
an
SR
SR
header
on
incoming
a
service,
its
packet
for
the
cases
of
less
s,
service,
its
policy
or,
let's
say
fee
or
didn,
TI
LFA
and
you
receive
a
packet,
and
you
are
already
in
fr
our
mode.
Then
you
have
to
you,
have
to
impose
more
segments
in
case
in
that
those
cases
T
dot
insert,
is
used.
M
The
cases
where
you
cannot
use
a
TSP
or
insert
you
can
still
use
T
dot,
Inc
app,
where
you
can
you
canon
kappa
incoming
packet
within
a
service,
six
header
and
header,
and
finally,
there's
also
l2
version
of
the
tea
in
cap
18
cap
use
case.
So,
as
I
said
earlier,
this
control
plane
is
also
the
scope
of
this
document.
M
However,
we
listed
a
table
where,
where
we
we
have
tried
to
just
just
summarize
and
observe
that
what
protocols
are
used
to
advertise
or
signal,
or
these
these
sit
functions
that
we
have
listed
in
the
document,
so
endo,
n
and
and
dot
XR
n
dot
T
is
a
GP
and
our
GP
is
a
base.
Topological
signal
is
signaling
protocol
for
for
topological
seats
in
in
overdraft
BGP,
an
IP
for
all
VPN
related
states
that
you
have
listed.
We
also
expect
that
that,
on
every
a
service,
six
node
will
will
will
advertise
its
instantiate.
It's
it's.
M
Why
a
BGP
LS
as
well
as
all
its
other
capability
like
can
I
do
insert
and
if
I
can
do
insert
how
many
states
I
can
insert
and
whatnot.
So
all
these
capabilities
have
to
be
advertised
through
bgp
LS,
because
BG
pls
is
over
basic
discovery
service
discovery
protocol
in
this
whole
framework
and
counter
for
the
counting
regions.
Counters
are
also
listed
in
the
document
so
that
you
can
count
on
local
set
traffic
matching
a
non
matching
process,
not
processed
as
well
as
a
traffic
that
is
girona
given
a
service
six
policy.
M
The
counters
are
also
there
and
for
this,
for
the
security
reasons
for
the
security
cases
to
basically
make
sure
that,
within
a
domain
of
trust,
while
we
are
providing
a
service
based
services,
internal
traffic
can
still
use
those
services.
However,
you
wanted
external
traffic
not
to
be
able
to
access
those,
so
there
are
therefore
secure
for
security
rules
that
all
notions
that
we
have
mechanism.
We
have
is
official
in
the
document
that
that
you
can.
You
can
take
a
look
so
finally,
in
the
video
in
the
document.
M
After
all
this
putting
all
this
together,
we
have
a
use
cases
illustration
using
an
example
which,
which
uses
almost
all
the
functions,
that
we
have
defined
earlier,
how
they
will
be
used,
you
know,
or
what
will
be
the
behavior
how-how-how
insertion
will
be
done,
how
dekap
is
expected
to
be
done
and
how
and
how
a
binding
said
is,
is
to
be
used.
So
these
all
these
use
cases
are
at
the
end
of
the
document.
Please
take
a
look
and
provide
us
feedback.
M
B
N
N
B
L
Basically,
the
use
case
here
is
to
support
the
l3
VPN
unicast
services.
There's
an
MPLS
based.
You
will
pale
plain
Network,
so
there
are
two
separate
planes
you
can
see
in
the
diagram
you
and
the
red
planes
and
certain
traffic
has
to
use
the
blue
routing
plane
and
you
should
never
cross
over
to
red
plane,
even
if
the
routing
plane
is
segmented.
L
So
if
the
specific,
if
the
routing
plane
is
segmented,
then
the
specific
traffic
must
be
dropped
and
it
must
not
be
forwarded
on
to
the
other
plane,
and
there
is
also
a
requirement
where
there
are
normal
SPF
based
paths.
They
can
use
a
certain
plane
and
if
the
plane
is
partitioned,
they
can
flow
over
the
other
plane.
So
there
are
two
requirement:
certain
traffic
that
has
to
be
restricted
to
one
routing,
plane
and
certain
other
traffic,
which
is
which
can
fail
over
to
the
other
plane.
L
So
we
have
many
existing
mechanisms
which
can
be
used
to
solve
this
problem,
and
we
have
some
shortcomings
which
won't
completely
meet
all
the
requirements.
We
can
go
over,
go
through
them
in
the
next
slides
one
is
the
node
admin
tax
to
color
each
of
the
plane
separately
and
then
use
some
kind
of
traffic
engineering.
To
achieve
this,
the
second
option
is
to
use
separate
any
cuz.
It's
one
per
plane.
L
So
the
problem
in
this
with
this
approach
is
that
the
when
the
when
the
plane
is
partitioned,
so
as
long
as
the
plane
is
not
partitioned,
it's
possible
that
to
keep
the
traffic
in
the
plane,
but
once
the
plane
gets
its
partition
like
p3
to
pe7
is
down
and
also
pe7
p1
to
p3
is
down.
So
in
this
case
the
traffic
will
fail
over
to
the
blue
plane
so
and
also
the
post
convergence
path
would
fail
over
to
the
view
plane.
L
So
it
does
not
meet
the
requirement
and
it
also
uses
a
complex
traffic
engineering
approach.
You
need
a
central
controller.
You
need
multiple
protocols
that
talk
to
the
ingress
node,
and
so
this
is
fully
satisfy.
The
requirements
of
the
suitcase
second
option
is
the
anycast
it's
so
you
could
assigned
and
anycast
it
to
the
all
the
nodes
in
the
plane
and
then
you
you
could
direct
the
traffic
using
those
any
cassettes
through
the
plane,
routing
plane.
The
problem
is
very
similar
to
what
we
described
in
the
node
admin
tag.
L
So
as
long
as
the
plane
is
partitioned
not
partitioned
you,
probably
the
traffic
remains
in
the
red
plane,
but
when
it
gets
partitioned,
the
traffic
fails
over
to
the
other
other
plane,
the
view
plane.
So
the
the
the
requirement
here
is
that
the
traffic
never
fails
over
to
the
other
plane,
because
there's
critical
data
that
place
flow
that
is
flowing
over
view
plane
and
the
requirement
is
not
to
mess
up
the
view
routing
plane.
So
the
problems
are
very
similar
to
node
admin
tag.
L
So
third
option
is
multi,
topologies
sit,
so
multi,
topology
kind
of
almost
meets
all
the
requirements,
and
there
is
also
additional
benefit
that
you
could
do
complete.
Separate
topology
routing
by
you
assigning
different
IGP
cost
to
link
to
different
end
piece,
but
this
is
empty,
is
perceived
to
be
a
very
complex
operation
operationally
because
because
of
the
configurational
and
operational
aspects,
and
also
additional
protocol
overheads
to
advertise
empty
related
information.
L
So,
similarly,
the
blueprint
planes
are
also
assigned
with
the
new
algorithm
value,
as
well
as
the
new
seat.
So
the
segment
routing
architecture
and
the
tasks
provide
a
very
clean
separation
of
using
algorithms.
So
the
nodes
which
advertise
the
certain
algorithm
they
are
the
only
nodes
that
are
used
for
computation
for
computations
for
that
algorithm.
L
O
O
O
L
L
L
L
L
L
So
here
is
a
slide
on
how
to
do
service
provisioning.
So
here
the
pe7
is
assigned
with
two
IP
addresses
10.10.5.3
r107
and
they
they
have
the
normal
set,
which
is
shown
in
black
over
there
and
and
the
writes
its
routing
plane
sets
which
is
shown
with
red.
So
the
services
we
have
to
service
prefixes,
X,
dot,
X,
dot,
x
and
y
naught,
where
twice
X
dot
X
dot.
X
is
supposed
to
follow
the
normal,
the
normal
SPF
path,
whereas
the.
Why
not?
Why
not?
L
Why
prefix
is
supposed
to
follow
the
red
plane
right,
routing
plane,
so
the
service
prefix
that
that
is
supposed
to
follow
the
red
plane
will
use
the
next
stop
as
the
IP
address,
10,
dot,
10.1,
10.10,
7,
and
so
when
on
the
ingress,
it
picks
up
the
SID
corresponding
to
the
routing
plane.
If
you
see
the
packet
looks
like
100
is
the
VPN
label
and
then
one
six
one
zero
seven
is
the
street
associated
with
router
pe7
and
it
uses
the
SPF
path
within
the
plane.
L
So
when
the
plane
is,
but
if
you
see
when
the
when
there's
a
single
failure
in
the
plane
it
it
still
uses
the
backup
path,
because
there
is
a
path
available
in
the
routing
plane.
But
if
the,
if
the
plane
is
partition,
then
the
traffic
gets
dropped,
because
the
read
sets
do
not
see
any
of
the
paths
corresponding
to
the
view
notes.
So
they
don't.
They
don't
program
them
either
for
primary
or
for
backup.
L
So
next
is
the
BGP
DCU's
case,
so
the
so.
This
can
be
further
extended
to
the
insert
to
the
decease
as
well,
whereas
the
DC's
also
is
also
segregated
into
red
and
blue
plane,
and
then
certain
critical
traffic
has
put
on
blue
plane
and
certain
other
traffic,
which
should
be
restricted
to
the
routing
plane,
is
put
on
the
right
plane.
The
way
this
is
achieved
is
BGP.
L
Inside
the
data
center
and
in
the
core
we
have
the
I
GPS,
so
the
routing
plane
separation
within
the
DC
can
be
done
using
BGP,
where
in
the
you
could
put
policies
to.
You
could
also
associate
the
communities
with
PGP
and
to
restrict
the
view
message
is
flowing
view
advertisements
flowing
through
the
red
plane
and
vice
versa-
and
you
can
see
here
that.
L
So
there
is
a
VPN
prefix.
Why
not?
Why
out
Y,
which
gets
advertised
with
the
next
top
as
pe7
and
then
there's
a
BGP
prefix
said
that
gets
advertised
for
this
PE
7.
So
this
PE
7,
the
BGP
terrific
Sadie,
is
used
for
transport
within
the
data
center
and
the
the
VPN
label
is,
is
carried
unchanged
and
at
on
the
edge
router.
L
The
next
next
top
changes
to
p3,
as
well
as
the
labor
changes
to
two
hundred,
and
it
uses
the
right
IP
address
as
the
next
stop,
and
so
the
the
the
part
that
gets
used
within
the
core
is
the
path
corresponding
to
the
red
plane.
The
this
is
the
SPF
path
corresponding
to
the
red
plane
and
similarly
in
the
in
the
remote
DC,
the
next
stop
changes
to
the
edge
rot
of
pe1.
Similarly,
the
label
changes
to
300
there
and
then
again
the
within
this.
L
Within
this
data
center
again,
the
PGP
prefix
set
is
used
for
transport.
So
if
you
really
see
so
on
this
on
this
edge
router,
what
you
see
is
the
VPN
label
300
and
then
the
red
label
corresponding
to
PE
1,
and
then
here,
when
you
come
here,
it
gets
swapped.
The
European
level
gets
wrapped
with
200
and
then
the
new
red
plane
set
gets
pushed.
The
similar
operation
happens
at
this
edge
on
PE
3,
where
the
VPN
label
gets
back
to
800
and
then
red
plane
said
BG
people
fix
it
gets
pushed.
L
L
So
some
of
the
limitations
of
this
solution
there's
an
additional
set
defined.
So
if
you
want
to
call
that
as
a
limitation
and
the
solution
is
restricted
to
this
use
case
like
if
you
not
accommodate
other
key
constraints
like
link,
colors
or
key
metrics
and
such
so
for
those,
you
will
have
to
go.
Go
back
to
the
central
controller
cell
based
solutions.
L
L
So
the
difference
here
is
you:
you
do
not
have
to
maintain
separate
routing
tables
or
you
do
not
have
to
advertise
multi
topology
related
information.
Also
your
configuration
you
don't
have
to
do.
Multi
topology
related
configurations,
it's
just
e
new
IP
address,
along
with
the
new
suit
associated
with
that
complexity.
L
P
Martin
Hana
fraud,
DT
just
a
minor
remark
for
the
anycast
based
solution.
There
was
indication,
PCE
would
be
needed
for
end-to-end
solutions.
I
would
not
agree.
I
am
convinced
that
it
can
do
as
well
without
any
central
controller
if
you
just
design
the
network
in
a
suitable
way.
That's
not
the
main
point.
P
What
appears
to
me
is
that
from
the
use
case
or
application
perspective,
I
see
a
big
overlap
between
your
document
and
what
can
or
could
be
done
with
Stefan's,
not
protected,
so
I
wonder
what
the
exact
overlap
of
use
cases
is
and
if
there
or
what
the
use
cases
are
which
are
covered
by
one
solution
and
not
the
other.
So
if
the
overlap
is
very
big,
it
might
be
interesting
to
actually
draw
in
the
approaches.
If
it's
not,
it
might
be
interesting
to
point
out
what
the
exact
differences
in
use
cases
or
applications
are.
P
E
Hey
Stefan,
so
the
good
message
from
your
draft
that
I
take
is
that.
Finally,
we
understand
that
multi
apology
routing
doesn't
work
well,
so
that's
the
good
news
of
the
day.
Now,
if
the
requirement
is
I,
don't
want
to
fail
over
into
the
other
plane.
That's
only
a
matter
of
your
strategy
in
your
tlf
implementation.
There
are
very
easy
ways
to
achieve
that
without
changing
the
architecture
without
changing
without
extending
the
protocols,
so
I
think
the
requirement
is
simple.
The
salute
is
as
simple
and
here
what
you
propose
goes
way
beyond
what
we
need.
E
L
It's
it's
possible
to
say
using
TI
LFA,
don't
don't
create
a
backup
path
through
the
other
plane
and
such,
but
that
would
need
you
know
configuration
and
you
use
policies
such
as
the
backup
policies.
That
could
say
do
not
fail
over
to
the
other
plane
and
such,
but
still
that
would
need
a
lot
of
configuration.