►
From YouTube: IETF102-RTGWG-20180719-0930
Description
RTGWG meeting session at IETF102
2018/07/19 0930
https://datatracker.ietf.org/meeting/102/proceedings/
A
Good
morning
crowd,
so
today
we
have
slightly
different
from
urban,
usually
so
I'm,
not
just
presentation
of
the
presentation.
What
a
slug
both
discussion
afterwards,
but
two
different
are
questions
being
around
me,
needs
to
be
done
number
of
presentations
and
then
we'll
have
a
given
flow,
which
is
again
number
of
presentations
and
discussion.
Activist
I
will
encourage
you
to
participate.
A
B
C
B
Okay,
thank
you.
The
first
one
is
Britain
is
fighting
to
seal
separated
engine
architecture.
It's
another
is
introduced,
cos
penis,
Motoko,
so
first,
so
we
so
my
presentation,
you
know
two
blocks.
First
thing
is
introduce
architectures.
Another
is
introduced
to
help
and
first
I
want.
You
mentioned
why
we
need
to
house
the
architecture
of
separate
the
PNG,
because
in
previous
at
the
meeting
we
have
arrived,
opposed
several
documents
to
introduce
and
interface.
D
B
Separate
gbng,
but
we
didn't
have
some
document.
The
two
describes
the
function
another
to
introduce
relationship
on
versus.
She
has
serious
pins
related
a
document,
so
our
cheers
suggest
to
variety:
to
propose
this:
an
informational
document
to
introduce
eight
so
far
away
in
just
making
web
upload.
One
government
made
the
poses
information
and
how,
to
rate
her
to
M
stands
relationship
between
my
adventures
with
a
spirited
determined
and
what
is
that?
B
We
don't
have
strong
opinion
to
standardize
this
document,
because
there
is
some
information
now
and
also
the
architecture
of
still
separated
PNG
have
already
published
by
because
the
related
working
group
is
a
P
344
here,
every
Chester,
the
Entertainer
Senate
and
help
the
leadership
Anderson
makes
he
knows
as
soon
as
pious,
architectural,
real
concerns
it.
Sir
thicker,
so
control.
B
B
The
CSP
portico,
and
which
should
be
weighted
operator,
implemented
into
the
detention
of
requirements
and
to
design
lightweight
Patoka,
which
is
port.
The
separate
Avengers
control
channel,
which
can
hear
of
the
chip
interaction
in
to
work
off
different
widgets.
The
ones
here.
I
also
noted
that
it's
a
awesome.
B
D
B
D
B
Option
is
through
the
semester,
important
code
and
it's
us
and
then,
when
a
user
dial,
the
user
can
Darko
cervix
wrong
and
same
period,
send
users
with
a
series
if
he
can
raise
use
a
process
application
and
what
general
greatest
users
requirements.
Services
I
will
answer.
No
such
acts
as
an
available
resource
said
it
will
generate
celery
and
our
country
able
to
attend
century
European
to
relevance.
A
dynamic
user
management
and
anime.
D
B
To
ratify
the
graduation
performance
ensue
as
any
in
the
way
to
benchmark
one.
If
two
are
assessed
in
children
can
communicate
ways
to
enter
ping
was
new
Tokyo,
it's
another
eight
to
right.
Besides
a
function
of
two
years,
I
work
here,
another
that's
a
contributing
and
centralized
management,
user
or
souls
such
husband
and
the
cosines
are
tries
to
rupees
no
address
Philippine
and
economic
times
are
people
to
dogs
as
a
user.
If
you
are
interested
says
you
can
reveal
voltage
here
on
all
download
is
a
demo.
Have
a
terminal.
B
D
B
G
J
I
J
L
I
E
E
I
M
D
I
N
N
M
I
N
I
I
N
I
N
I
I
I
N
O
A
F
Okay,
also
a
longtime
attendee
at
the
broadband
forum,
I'd
actually
say
the
the
problem
space
that
frames
the
solution.
Space
is
a
little
bit
bigger
than
just
the
bng.
We
have
been
looking
at
fixed
mobile
convergence,
which
is
mapping
wireline
access
onto
the
5g
core
net
result
of
that
is
is
one
of
the
operator
aspirations
or
requirements.
Is
that
the
existing
legacy
kit
that
today
talks
to
be
NGS
can
be
inter
worked
with
the
5g
core,
which
implies
in
a
cups
interface.
There
will
be
a
staggering
amount
of
opportunity
either
for
engineering.
F
An
absurd
degree
of
engineering
duplication
across
the
solutions,
so
what
I
would
suggest
is
is
folks
at
least
take
out
a
look
at
what's
publicly
available
related
to
the
FMC
work
and
think
about
that
in
this
context,
because
of
the
tend
to
frame
the
solution,
space
and
proceed
to
save
the
industry,
an
awful
lot
of
pain.
So
thank
you.
M
M
Terminated
on
a
different
router,
so
you
need
things
like
terminating
identifying
the
customer
and
determine
okay.
This
flow
must
be
put
into
our
l2tp
session
router
into
an
MPLS
path
through
some
which
stuff.
The
second
challenge
I
see,
is
keeping
the
state
between
the
control
plane
and
the
users
state
accurate
lenses.
I
M
E
M
Working
say,
ongoing,
work
and
I
think
you
know
in
order
to
have
all
the
necessary
details
and
I
think
it's
your
on
or
what
they
had
mentioned.
We
need
to
work
first
out.
What
will
be
the
solutions?
What
a
video
he
take
sure
that
is
needed
for
such
kind
of
separation
in
order
to
also
know
what
kind
of
information
you
need
in
order
to
the
interface
between
the.
D
B
I
Day
one
it
can
support
pure
layer
two
axis.
If
that's,
what
is
the
priority,
but
at
least
you
know
it
can
be
extended
to
mobile.
For
instance,
if
you
look
at
hybrid
access
today,
customers
have
deployed
it
now.
If
you
have
two
flavors
of
cups,
you,
nobody
use
one
protocol
to
control
a
fixed
interface,
another
for
the
mobile
interface.
So
that's
a
point
where
we
want
a
protocol.
Selection
should
be
such
that
it
should,
in
future
or
even
today,
with
hybrid
access.
Allow
allow
convergence
I.
B
B
I
J
I
M
What
I
wanted
to
also
emphasize
is
that
if
we
want
to
work
on
the
on
the
scenario
where
fixed
axis
is
actually
in
place,
then
we
can
already
use
an
architecture
that
it
has
been
already
defined
in
tier
384.
If
it
is
to
to
also
involve
the
fixed
mobile
convergence,
then
we
need
some
more
convivial.
That
needs
to
be
done
in
VBA.
M
From
the
point
of
view
of
you
know
having
their
cups
kind
of
solution.
That
would
be
work.
What
could
work
for
both?
This
is
something
that
could
be
discussed,
but
it
needs
a
lot
of
time.
So
maybe
we
could
start
work
first
working
on
what
what
is
possible
and
then
extending
that
work
towards
that
mobile
convergence.
K
So
I
want
a
kind
of
second.
The
days
come,
and
this
is
an
opportunity
to
do
something
right,
but
also
this
is
a
danger
to
reinvent
a
lot
of
stuff
that
has
been
solved
by
many
people.
Already.
A
caps
is
not
new.
In
some
environment
we
have
to
make
sure
that
what
we
come
up
with
is
not
an
additional
complexity,
because
the
industry
is
moving
with
5g
to
allow
not
only
you
know,
you
need
single
fix,
not
only
mobile,
but
also
combination.
K
If
we
think
we
can
maintain
multiple
solutions
that
that
will
be
very
tricky
and
very
difficult
for
operators
to
run
with.
If
we
think
going
to
the
first
presenter
that
we
can
replace
a
lot
of
existing
4G
and
acid
moves
on
caps
control
playing
with
a
new,
totally
protocol,
I
think
we're
kidding
ourselves
so
I'm
taking
holidays,
making
it
better
and
working
with
BBF
working
3gpp2
that
the
architecture.
G
This
law
would
being
turned
and
published
by
BBF
already.
So
that's
the
right
place
to
look
on
and,
of
course,
some
of
the
phone
call
extensions.
Whatever
have
to
be
done
here,
and
example.
I
noticed
that
for
the
footprint
the
job
you
just
put
out,
which
one
reference,
the
TR
sweetie
for
right-
that's
the
central
office,
so
I
noticed
that
that
year,
just
put
recently
published
that
did
mention
there,
the
cooks
all
right.
G
So,
but
the
second
friend
that
you
mentioned
e
WT,
three,
eight
three,
eight
three,
seventy
eight,
and
that
is
another
requirement
for
the
have
we
access
but
I,
didn't
just
look
at
it
over
yesterday
and
that
didn't
mention
anything
about
the
curves.
So
if
you
talk
about
the
architecture
of
the
cuffs
and
my
recommendation
is
to
collaborate
with
the
PPF.
N
G
J
So
before
that
there
is
already
work
that
is
solves
a
lot
of
these
issues,
but
my
main
comment
is
that
I
think
that
there
is
a
terminology
skew.
It
looks
to
me
that
what
you're
doing
is
management
plain
to
foreign
plane
instead
of
control
plane
at
everything,
there
was
a
figure
before
that
in
the
control
plane.
Everything
you
had
was
management
management
management
management
right,
so
I
understand
of
this
work
has
been
done
also
in
the
3gpp.
So
there
is
an
RFC
RFC
7426
by
the
but
terminology
in
the
Sdn
layers
and
planes.
Q
If
the
message
which
I
see
here
is
is
deafening,
is
that
in
BB
F
there
is
a
there
is
a
gap
right
now,
because
they
have
defined
cops
for
BMG
in
class
zero,
but
they
haven't
defined
cops
in
all
the
use
cases
which
a
B
and
G
could
work
in,
and
maybe
we
should
give
a
response
towards
the
liaison
request
from
which
we
got
from
BB
F
to
say,
okay,
why
won't
you?
Why?
Q
We
have
not
in
BB
F
a
full
view
on
all
the
requirements
which
are
required
in
this
space
and,
as
a
result,
I
think
it's
best
that
that
work
happens
there
and
then
once
that
is
more
or
less
in
a
good
state,
is
that
the
xxxx
protocol
extension
and
protocol
to
be
used
would
be
done
here
and
with
the
X
and
T.
That
could
be
my
suggestion.
Express.
L
So
I
think
the
question
of
like
choice
of
protocol.
If
you
look
at
it,
the
3gpp
has
already
gone
through
all
the
discussion
about
whether
to
use
TCP
or
UDP.
If
you
look
at
the
user
plain
requirements
in
general
of
PNG
and
mobile,
you
will
see
that
they
are
actually
one
and
the
same,
and
if
you
look
at
the
constructs
like
what
send
you
had
a
slide
about
where
you
were
identifying
a
session
in
terms
of
policy
enforcement
in
terms
of
collecting
the
statistics,
all
that
work
is
already
done
now.
L
There
was
also
one
question
about
forces
why
we
should
not
use
the
forces,
but
if
you
look
at
the
the
actual
PFC
P
protocol,
it
has
got
all
the
integration
in
terms
of
what
you
need
on
the
gy
interface
it
comes,
and
so
there
is
no
need
to
really
repeat
all
that
work
or
discussion
out
here.
So
what
we
can
do
is
we
can
use
that
the
work
and
leverage
use
that
as
well.
L
A
A
A
Talking
to
be
be
worth
is
another
step,
we're
going
to
take
and
happen
it
gave
and
charges.
Neither
people
on
wealth
cutting
out
for
that
and
I
would
really
like
otters
to
proposal,
to
talk
to
each
other
and
see
how
we
can
find
points
at
coming
across
to
and
start
working
towards
common
solution.
Thank
you.
So
much.
Q
Sexy
gay
Jeff
morning,
everyone,
my
name,
is
Sonia
Campari,
and
this
is
a
new
here.
In
this
document
we
try
to
analyze
some
constraints
of
a
six.
You
know
some
scenarios
and
then
try
to
make
some
extensions
improvements.
Based
on
that,
if
necessary,
we
try
to
provide
a
new
encapsulation
scheme
for
the
current
protocols,
so
the
first
constraint
is
second
consumptions.
Q
Q
Q
Q
Q
Q
Q
Q
H
O
O
O
O
S
A
A
O
O
D
If
you
start
back
any
of
those
topology
flows
that
you
have
there
one
more
example,
this
one
and
you
want
to
move
flow
from
A
to
B
or
many
of
the
notes.
How
do
you
know
what
is
the
link
state
between
if
you
go
to
the
furthest
right
and
on
the
ballot?
What
is
that
like?
What
is
that
in
capacity
and
that,
when
you
move
the
trophy
to
that
way,
the
human
net
or
subscribe
that
way.
N
N
F
Okay,
so
what
the
draft
is
about
is
using
computation
for
the
routing
of
MPLS
segments
or
multicast
segments
to
an
MPLS
based
asari
network
and
how
turtling
using
notes
heads
can
be
used
as
part
of
multicast
reconstruction,
and
we
can
consider
this
both
from
a
distributed
or
a
centralized
control
model.
For
those
of
you
familiar
with
802
that
1aq
shortest
path,
bridging
it
would
be
the
exemplar
of
how
aspects
of
the
distributed
model
could
work.
F
F
Or
economically
feasible,
for
example,
an
existing
kit
or
with
any
sort
of
cheaper
class
of
switching
chips
to
be
used
in
data
centers,
for
example.
So
if
there
was
sort
of
a
continuum
of
solutions
for
multicast
where
beers,
completely
stateless
and,
for
example,
we
considered
Pamela
and
our
VP
as
stateful
as
possible,
this
would
fit
somewhere
in
between
there.
It's
significantly
less
state
itself
in
the
data
plane
than
there
is
for
a
pen
or
ml
VP,
but
it
doesn't
achieve
the
sum
of
the
state
that
beer
does
so.
F
The
new
health
describes,
the
architecture
world
and
the
multicast
trees
are
constructed,
as
the
combination
of
roots
leaves
and
replication
points
interconnected
with
tunnels
and
the
routing
of
trees,
determined
based
on
information
in
the
IGP.
We
do
all
the
multicast
registrations
in
the
IGP
and
therefore
that,
when
combined
with
the
topology
information,
provides
a
node
with
sufficient
information
to
determine
its
worth
in
a
tree
now,
this
gives
you
a
whole
bunch
of
benefits.
You
have
a
minimum
of
messaging
to
converge.
F
The
network,
because
the
topology
change
is
the
only
additional
piece
of
information
that
any
node
needs
to
determine
its
role.
In
a
multicast
tree,
as
I
mentioned,
we
have
substantially
reduced
data,
plays
data
plane
state.
That
also
mean
these
things
tend
to
speed
up
because
there's
less
stuff
that
needs
to
be
pushed
into
the
fair
by
the
topology
change.
It
reduces
the
bandwidth
requirements
versus
straight
ITP
derived
trees.
I
will
get
into
the
reason
for
that
in
a
moment,
and
finally,
unicast
convergence
will
provide
recovery
for
most
failures.
F
So
I've
got
an
example
tree
here.
To
give
you
an
idea
of.
What's
going
on,
I've
got
a
rate.
I've
got
four
leaves
and
shouldn't
determine
that
there
was
two
addition.
The
replication
points
required
to
serve
that
and
not
all
the
replication
points
or
the
leaves
were
immediately
adjacent.
So
the
algorithm
permitted
us
to
interconnect
those
using
the
nodes.
So
the
net
result
of
this
is
is
that
in
the
network
only
two
additional
nodes
out
of
a
potential
five
from
needed
to
install
state
to
fully
service
the
tree.
F
E
F
Of
something
like
head
end
replication,
where
I
may
have
like
three
interfaces
off
my
node
and
50
leaves
I'm
sending
50
copies
on
the
three
interfaces,
but
I'm
trying
to
get
to
or
what
we're
trying
to
get
to
in
this
draft
is
the
idea
that
I
don't
worst
case.
I
only
ever
send
one
copy
on
any
interface,
but
there
is
a
nice
EMP
friendly
tree
construction,
algorithm
and
the
left
and
serendipitously.
F
Now,
if
you
got
a
sense
of
deja
vu
and
I
see
lots
of
faces
here,
the
brother
this
was
last
presented
in
IETF
97.
At
the
time
it
looked
like
the
consumable
of
multicast
solutions
or
the
comestibles
was
going
to
sort
of
take
over
the
world.
It's
now
a
significant
amount
of
time
later,
and
that
hasn't
really
happened
and
it
doesn't
look
like
it's
going
to
happen
for
the
lower
end
devices.
F
So
it
turns
out
one
of
the
things
that
had
not
real
as
Matt
then,
unless
there
actually
was
a
community
of
interest
that
was
interested
in
this
solution
and
hence
I
had
lots
of
motivation
to
bring
it
back.
The
current
draft
updates
the
terminology
to
align
with
the
current
state
of
SR
MPLS.
It
has
some
editorial
improvements
for
readability.
It's
added
some
motivations.
F
Some
concerns
have
been
expressed
to
me
that
this
doesn't
quite
fit
with
the
current
state
of
MPLS
friendliness
and
I
want
to
double
check
that
I.
Don't
think,
that's
actually
an
issue.
We
will
bring
forward
future
drafts
with
required,
IGP
extensions
and
four
into
working
with
existing
mechanisms
and
we're
going
to
be
pursuing
standards
track,
given
that
the
segment
routing
group
currently
does
not
have
multicast
in
its
Charter.
The
current
understanding
amongst
the
leadership
is
that
we
will
be
taking
this
forward
and
PIM
and
we
are
seeking
PIM
working
group
adoption.
M
F
F
M
Multiple
of
it,
I
didn't
follow
too
close.
I
just
listened
and
then
taken
a
note,
but
you
turn
down
the
relying
on
MPLS
or
the
MPLS
from
today.
Creating
the
MPLS
from
Zeena's
would
be
considered
a
actually
brighton,
brighton,
main
ascent
to
the
MPLS
working
group
and
actually
introduced
that.
So
we
can
get
the.
F
M
F
A
U
U
A
Trap
you
for
20
seconds
so
we're
about
to
spout,
is
we're
on
this
meeting
and
so
very
worried
that
we
won't
help.
Other
people
already
started
this
work
being
on
being
in
here,
perhaps
to
provide
expertise
enough
in
protocols.
So
you
know
both
now
in
their
ability
only
one
solution
to
any
level.
If
you
look
at
classical
picture
what
management,
cruncher
and
data
points
on
management
well
in
service
representation,
we
we
know
how
to
both
service
models.
I
think
we
could
definitely
help
building
a
service
representation
control
plane
were
the
place,
a
lot
of
solutions.
A
U
U
Specific
use
cases
st
Oberon.
One
thing
is
for
st:
really
we
being
there
lots
of
bus
world
and
there's
lots
of
conferences
and
on
the
air
being
a
aug.
They
all
have
projects
on
his
team
would
come
to
IETF.
Most
people
start
with
a
question.
Is
we
don't
need
any
portable
workstation?
So
today
my
talk
is
really
about
to
stimulate.
Some
of
the
discussion
is
the
way
point
out
some
of
the
gaps
and
those
are
just
starting
point
and
some
other
companies
may
have
different
means
and
they
identify
different
extension.
U
U
Basically,
you
have
ICP
devices
may
be
placed
anywhere
and
you
have
existing
LCpl
to
VPN,
and
now
we
want
to
aggregate
some
other
paths
like
internet
paths
to
make
the
bank
was
more
flexible
and
to
expand
them.
So,
looking
at
this
picture,
it
was
pretty
simple,
like
you
have
two
points
interconnect
by.
U
Whether
the
truth
is
actually
is
much
more
than
that
right.
So
when
you
have
one
network
like
with
hundreds
of
CPAs,
managing
no
can
be
troublesome,
he's
almost
bring
back
to
the
frame
where
they
base
the
mash
interconnection
can
be
very
complicated
among
the
CPEs.
So
that's
probably
one
of
the
reasons.
Even
though
enterprises
started
with
wanting
sd1
by
the
end,
most
of
them
are
pursuing
managed
services,
meaning
service
provider
managed
also
managed
the
sd1
floor,
though
we
call
that
CPE
based
a
VPN,
our
indicator
is
SD
one.
U
Another
use
case
is
really
that
inter
process
today
is
under
pressure
to
transition
to
the
digital
and
with
that
they
have
to
be
able
to
enable
their
applications
in
places
where
their
clients
needs.
They
call
application
portability
and
data
portability,
and
now
the
Redwood
regulatory
needs
and
certain
data
can
only
be
stored.
U
Certain
places
that's
basically
driving
the
values,
so
there's
a
cloud
datacenter,
a
lot
of
those
applications
being
transit,
meaning
not
permanent
enough
chain
build-up,
the
physical
infrastructure
to
those
locations,
so
there
being
lots
of
them
and
actually
they
are
lots
of
events
requiring
to
be
able
to
interconnect
the
floating
point
into
your
existing
VPN
see
the
example.
The
enterprise
today
I
already
have
the
VPN
the
connecting
their
branch
offices.
Suddenly
they
need
a
few
branch
offices
to
be
instantiated
in
a
cloud
data.
U
Simply
you
another
country,
another
place,
and
so
st-1
becomes
a
very,
very
convenient
way
to
be
able
to
bring
those
cps
to
their
branch
offices.
But
if
you
just
talk
about
plain,
simple,
st
worm-like,
I
think
this
example
we
have.
U
1
e
2
of
those
carry
different
applications,
dis
importance
to
you
in
writing
to
kind
of
remove
any
kind
of
path
that
into
that
path.
Well,
when
they
are
close
together,
like
a
you
know,
one
campus,
that's
fine,
but
when
they
across
the
country
a
continent,
the
performance
can
be
unpredictable.
U
So
it's
desirable
to
be
able
to
finding
the
the
closest
PE
for
those
endpoint
so
that
those
endpoint
canyon
happen
to
the
enterprise
existing
BPM
as
quickly
as
possible,
so
to
minimize
them
like
the
last
mile,
basically
make
things
LTE,
or
maybe
a
third
party
pop
it
in
today.
So
that's
the
second
use
case
and
the
questions
really
what
what
is
lacking.
You
know
each
year
right,
we
develop
many.
Many
protocols
definitely
can
kind
of
put
them
together.
U
Eg
use
so
I'm
during
our
deployment
we'll
find
out
some
of
the
useful
part
and
some
of
the
missing
part
like,
for
example,
5512
we
find
is
really
useful.
I
see
85:12
defined
the
specific
Safi
for
tunnel,
so
we
can
utilize
that
to
establish
tunnels
around
CPEs
and
specifically
for
tunnels
well,
but
unfortunately,
that
obviously
being
oscillated
with
a
new
family
in
cap
proposal
so
and
that
proposal
coming
in
cap,
which
is
I,
think,
is
going
through
idea.
I'll
ask
car
already.
U
We
finally
had
some
trouble
issues
where
we
want
to
use
them
to
hd1
and
so
for
run.
This
tunnel
in
capped
rat
require
all
the
tunnel
advertisement
to
be
attached
to
that
and
first
SDRAM
those
tunnels
are
Barbican
sites.
Those
gonna
have
to
be
created
with
other
routes
being
attached
to
them,
and
people
suggest
to
us.
Let's
just
use
the
figure
out
just
like
whatever
figure
out
to
create
a
tunnel.
Oh
that's!
Fine!
If
we
just
one
tunnel,
but
the
problem
we
face
is
for
each
CPE.
We
may
have.
U
U
And
pretty
soon
you'll
find
out,
there
will
be
many
tunnels
needed
and
just
to
managing
those
fake
addresses
can
be
troublesome.
It's
required
deployment,
especially
in
sd-1.
It
is
expected
there,
a
touch
provisioning
for
the
Backson
I
get
up
himself
without
any
touching
of
the
device.
So
many
Gmail's,
so
many
fake
addresses
can
be
a
powerful
for
us
for
just
generalized.
U
V
So
if
you,
if
you
do
the
5512
in
caps
a
fee
versus
using
the
in
cap
attribute,
which
is
what
you
have
in
the
the
draft
which
defecates
the
in
caps
a
fee,
if
you
use
the
in
caps
a
fee,
every
route
that
uses
at
all,
when
you
receive
the
routes,
you
then
have
to
resolve
it
against
the
so
the
regular
update
with
the
the
actual
prefixes
that
map
into
a
tunnel
that
you
then
have
to
locate
the
tunnel
and
resolve
into
that.
Yes,
it's
pretty
complex.
It's.
U
V
It
was
really
hairy
stuff
to
do
and
it
was.
It
was
pretty
complex
to
get
right.
Okay
and
you
had
a
fair
amount.
Sorry
about
that,
my
project,
you
had
a
fair
amount
of
processing,
overhead
and
ordered
result
routes,
and
so
and
then
there
was
all
these
boundary
cases
of
which
one
there's
timing
issues
I
was
which
one
you
received
first
and
then
you
end
up
building
nice
long
lists
of
routes
to
be
reserved
resolved
and
it
certainly
works.
V
We
have
a
working
implementation
of
that,
but
it's
really
complex
to
the
point
where
we
said
we
really
don't
like
this
and
we
prefer,
what's
in
the
draft
to
always
carry
around
the
tunnel
entry
together
with
the
route
advertisements
and
well,
it
means
you
have
more
advertisements
the
processing
and
the
codes
a
lot
simpler.
So
in
terms
of
what
you're
talking
about
I
think
it
is
completely
viable
to
make
it
work.
We
did
it.
If
you
want
the
code,
it's
an
open
source,
I'm,
not
sure
it's
worth
it.
Okay,.
U
V
Not
the
issue
of
5512
being
complicated,
it's
actually
really
straightforward
in
terms
of
to
define.
This
is
one
of
the
cases
where,
when
you
separate
out
the
theoretical
definition
from
the
actual
implementation,
you
miss
complexity.
The
5512
definition
is
really
straightforward.
5566
is
also
really
straightforward
and
it's
the
code
that
gets
really
compliant.
It's
the
amount
of
memory
and
processing
they
have
to
do
in
the
corner
cases:
okay
and
and
whether
you
use
5512
or
define
a
new
Safi
just
for
it's
not
gonna
change
that
okay.
V
Tunnel
in
cap
attribute
attributes
that
you
carry
with
the
actual
rel
eliminates
that
resolution
that
level
of
indirection
it's
a
lot
simpler
to
implement
and
to
maintain
a
trade-off
here
is.
Is
you
have
a
lot
more
updates?
Ok,
so
it
is
certainly
an
optimization
trade-off.
Okay,
do
you
want
more
verbose,
updates
and
simpler
code,
and
really
less
memory
actually
turns
out?
Okay,
do
you
want
that
or
do
you
want
to
optimize
I'm?
Sorry,
look
what
is
more,
most
verbose
from
the
wire,
the
others,
more
compact
and
harder
and
more
complex
code.
V
V
W
Jeff
has
so
this
is
largely
building
a,
but
Lou
had
actually
said.
5512
mostly
was
not
implemented
because
it
was
decoupled,
and
that
meant
that
your
resolution
code
need
to
build
video
love.
The
fact
that
the
information
not
only
is
tight,
I
GP,
like
existence,
but
know
whether
or
not
you
have
tunnel
mappings,
that's
being
received
across
the
additional,
no
Safiye
the
end
caps
draft
they'll
fix
the
rather
that
row
we
are
seeing
this
used
in
our
own
internal
implementation.
Right
now
is
for
the
srte
policy
draft,
which
I
have
to
agree
with.
W
U
U
W
U
A
U
U
They
need
to
security,
to
connect
that
CPE
to
the
existing
VPN,
that's
actually
very
useful
for
them
as
the
right
environment
we
are
looking
at,
but
the
problem
I'm
talking
to
the
others
about
the
intent
of
this
document,
they're
really
looking
at,
maybe
only
once
ap
evening
out
there
so
they're
talking
about
anything,
is
hand
provisioned.
You
set
up
the
IPSec
contribution
of
the
parameters
and
you
have
keys
pre-installed
in
those
places
and.
U
So
basically,
if
you're
talking
about
scale
rise
to
a
deployment
and
there's
a
point-to-point,
no
it
doesn't
scale
well
right.
So
like
just
like
friendly,
you
have
so
many
tunnels
to
manage.
So
you
know
we
break
them
into
a
clusters.
Each
cluster
may
have
some
CPAs
and
they
have
a
flat
turn
around
flat.
Curves
are
interconnected
together
and.
U
Second
party
is
for
their
own
touch,
provisioning
over
CPS.
How
about
they
will
be
able
to
have
established
here
at
session
with
the
controller,
and
then
the
BGP
will
be
able
to
distribute
medium
information.
Our
second
is
the
st1
deployment,
the
IP
sonic
entire
IP
implementation.
We
need
to
simplify
that,
because
IP
said
is
assuming
over
CPS
all
the
endnotes.
They
are
traversing
school
entrusted
to
me,
but
in
this
particular
deployment
we
have
is
over.
U
U
Able
to
pass
down
a
path,
the
key
to
their
security,
Association
keys.
So
that's
actually
we
discussed
inserting
our
children,
SF,
distributors
keys,
hoping
since
all
those
CPS
are
managed
by
bgp.
Already
we
can
extend
attributes
may
be
creating
a
new
color
type
to
carry
the
information
and
for
the
BGP,
especially
like
where
we
have
crossed
over
Cpl
CPS.
Sometimes
the
the
route
change.
You
like
the
controller
to
decide.
U
Hey
I,
want
to
reroute
us
to,
for
example,
to
Dallas
I,
don't
really
care
which
CPU
will
actually
take
over
the
traffic,
but
there's
some
time
you
will
need
to
be
able
to
specify
which
terminal
to
use
so
so
we
got
will
need
not
only
endpoint.
We
need
the
coloring
for
different
traffic.
We
also
need
to
raise
distinguish
different
sites
for
the
BGP
distribution.
The
controller
has
to
be
able
to
pass
downn
keys,
the
public
keys
to
each
you
need
to
do
authentication
and
so
basically
we're
changing
the
IP
set
in
a
way.
U
U
For
the
tunneling
cap,
we
think
we
need
a
new
type
I'll
be
able
to
represent
very
sad
distribution,
or
maybe
we
want
to
distribute
to
the
net
right
instead
of
having
the
CPE
to
resolve
themselves,
the
public
and
private
addresses
having
a
controller,
dispose
them
the
map
and
giving
to
them.
So
here,
I'm,
just
back
of
a
better
word.
St-1
have
a
much
better
working
to
use.
V
V
N
V
Why
we
chose
that
approach
as
well,
so
the
main
reason
was
to
basically
begin
the
process
of,
inter
working
between
these
devices
in
a
more
standard,
rather
than
an
ad
hoc
way,
in
order
to
allow
enterprise
users
to
basically
deal
with
emanates
and
situations
where
they
have
been
networks.
Vendor
SD
line
networks
that
they
need
to
interconnect.
X
K
V
A
large
community
of
customers
who
are
building
enterprise
web
networks,
we
get
to
see
the
architects
front
of
those
customers
come
out
and
talk
about
where
they
want
to
take
their
networks
in
the
next
generation.
Both
you
know:
control,
plane,
land
architecture,
security,
visibility,
control
perspective.
V
V
V
V
Okay,
the
picture
you
see
is
our
reference
architecture.
We
spend
a
few
meetings,
basically
hammering
out
what
an
SD
my
reference
architecture
would
look
like.
Then
what
you
see
here
is
sort
of
a
you
interface
level
at
the
top
the
client
orchestration
there
we
call
it.
The
idea
is
that
that
is
the
you
know
the
place
where
you
would
create.
V
Tools
which
would
be
used
to
then
make
API
calls
into
the
infrastructure
to
configure
specific
SD
line
services.
So
one
of
the
things
we're
trying
to
do
with
this
interoperability
working.
We
have
us
to
create
it
standardized
source
definitions
within
the
sd1,
not
protocol
definitions,
but
service
level,
definitions
to
create
models.
V
Level
definitions
that
vendors
can
then
go
off
and
implement
so
that
they
can
create.
When
we
make
these
common
API
calls,
they
will
create
a
service
that
has
a
common
behavioral
characteristic
between
different
st-1
vendors.
Thank
so
you
can
see
interview
we're
standardizing
sets
of
api's
across
the
interface.
V
In
vintage
that,
where
we're
asking
my
managers
and
controllers
could
talk
to
each
other,
we
don't
have
the
specifications
of
what
that
information
that
might
cross
there
might
be.
But
that's
something
we'll
want
to
work
on
in
the
future
as
well,
and
then,
of
course,
some
of
the
interfaces
that
you
see
inside
that
grey
box
are
proprietary.
V
N
V
V
Domains,
each
of
which
has
different
protocols
and
twelve
planes
and
data
frames,
and
we
confine
the
interworking
to
the
border
elements.
The
amount
of
feature
set
and
specification
that
we
have
to
do
is
obviously
much
less
from
a
work
perspective,
so
we're
able
to
get
some
service
level
of
some
service.
V
N
V
N
N
V
V
All
right,
and
just
to
give
you
a
sense,
a
little
bit
of
a
little
bit
of
the
sense
of
how
we're
looking
at
the
network
from
a
multi
domain
perspective.
Okay,
we
see
you
know
in
a
general
model
that
we
would
not
begin
this
client
orchestration
there,
which
we
are
not
really
specifying
when
we
specify
the
API
calls
that
would
be
consumed
consumed.
V
So
in
the
actual
working
groups
that
work
with
the
OSCE
that
we
have,
we
have
a
number
of
different
task
forces
that
are
doing
work
in
different
areas.
So
we
have
an
authentication
framework.
That's
looking
at
authentication
of
network
elements
between
the
domains,
so
most
Sdn
solutions
have
the
concept
of
authenticated
network
element.
V
You
want
to
know
that
the
elements
you're
talking
to
are
also
valid
in
that
domain
as
well,
so
we're
working
on
the
procedures
and
and
the
interworking
specs,
for
that
there
is
some
work
going
on
on
service
Cheyney
how
to
do
extend
service
change
between
the
two
domains
as
well
as
reachability
exchange,
which
is
basically
how
are
we
connecting
virtual
networks
or
verbs
that
are
created
one
day
or
the
first
that
are
created
in
another
domain,
our
stitching?
How
are
we
stitching
those
together?
V
There's
a
couple
of
proposals
on
the
table
for
how
to
do
that.
Some
are
controlling
the
control.
One
is
controlling
a
controller
based
the
other
one
is
HP
based
OSC
hybrid
cloud.
Api
is
an
or
task
force
that
me
identified
as
something
that
is
important,
and
the
idea
is
your
here
is
that
we
are
seeing.
E
E
V
V
V
For
the
access
network
service
definition,
so
access
selection,
how
do
I
select,
which
line
access
network
to
place
traffic
on
based
on
policy
based
on
SLA
and
performance?
And
then
we
have
the
external
gateways
factor
the
OSE
gateway
spec,
which
is
really
focused
on
the
cross,
the
main
reach
ability
and
segmentation
exchange
between
two
st
mine
domains.
And
then
we
have
an
API
formats,
practical
we're.
Looking
at
what
the
API
formats
we
will.
Support
are
as
well
as
the
service
model
specification.
V
Go
ahead
for
one
more
because
I
think
that
yeah,
so
what
I
wanted
to
show
here
was
you
know
my
hidden
slide,
you're
getting
show
them!
That's
why
and
so.
I
just
want
to
give
you
an
overview
of
sort
of
key
progress
and
results
of
where
we
are
in
terms
of
our
work,
so
the
st1
in
reference
architecture.
As
I
mentioned
the
first
draft
that
that
spec
is
published
and
available
for
review
the
OSP,
the
lessee
path
management
service
api.
V
We
have
actually
a
first
draft
that
the
api
that's
created
in
yang
is
not
really
ready
for
implementation.
Yet
we
do
need
external
view,
people
to
help
us
improve
that
and
make
sure
it's
structured
correctly,
and
then
we
have,
of
course
the
gateway
service,
ose
gateway
service,
api,
where
we
have
again
at
first
cut
yang
modeling.
For
that.
So
we
need
to
go
through
a
set
of
reviews
and
refinements
of
those
models
for
segmentation
and
creation
and
cross-connect,
as
well
as
IPSec
definitions
between
so.
V
L
V
M
V
Could
be
a
Microsoft,
Azure,
AWS
kind
of
architecture
where
the
customers
hosting
a
bunch
of
workloads
there
but
wants
to
have
bring
that
on
to
their
network
access
branch
is
sort
of
a
companion
use
case.
That
is
that
the
idea
of
being
the
same
thing
and
carrying
many
enterprise
customers
are
looking
at
using
chemical
facilities,
part
of
their
land
architecture
as
a
confluence
point
for
cloud.
V
V
B
V
I
V
V
About
that,
I
showed
you
the
basic
use
case,
we've
defined,
but
there's
a
lot
of
work
in
terms
of
developing
a
translation
or
an
adaptation
API,
much
of
which
involves
kind
of
understanding.
You
know
how
how
these
these,
the
the
SDM
devices
a
virtual
devices
official
Network
function
would
be
spooled
up
with
any
csps
looking
for
common
ground
and
common
than
trying
to
Kamala
gate.
Those
API
is
into.
N
V
N
J
V
V
A
However,
the
translation
should
be
done
outside
of
endemic
scope,
so
yeah
I
think
it's
P
to
be
described
in
8300
to
how
services
configuration
and
yeah
I
think
know-how
is
here
in
a
different
routing
working
both
in
net,
not
not
yes,
so
this
is
where
I
see
personally
notice.
What
nutshell
buggers
are
not
member
and
idea.
Remember
where
we
could
potentially
start
out.
N
W
Jeff
house
I
think
a
bit
of
what
you're
seeing
is
that's
much
lack
of
interest.
It's
more
that
ITF
as
a
whole
has
generally
avoided
intentionally
doing
API
work.
Now
we
tend
to
focus
more
on
protocols.
We
tend
to
focus
more
on
the
mechanisms
that
the
API
is
with
instantiate
and
the
yang
stuff
that
we've
done
recently
has
pushed
us.