►
From YouTube: IETF-TEAS-20210927-1400
Description
TEAS meeting session at IETF
2021/09/27 1400
https://datatracker.ietf.org/meeting//proceedings/
A
A
Good
morning,
good
afternoon,
good
evening,
welcome
to
the
first
tease
working
group
interim
session
for
this
year.
A
A
This
is
the
itf
note.
Well,
it
sets
the
ground
rules
for
the
way
we
interact
with
the
ietf.
It
applies
to
interim
meetings
as
well.
If
you're
not
familiar
with
it,
please
do
get
yourself
familiarized
quickly.
A
A
As
you
may
have
noticed,
we
do
have
a
personal
change.
We
have
a
new
working
group
secretary,
luis
contreras,
will
be
serving
as
our
working
group
secretary.
Starting
with
this
meeting.
Please
join
us
in
welcoming
him.
We
would
like
to
thank
matt
hartley,
our
outgoing
working
group
secretary,
for
all
the
hard
work
that
he
has
done
in
this
role
for
for
the
past
few
years.
We
would
also
like
to
thank
those
of
you
who
volunteered
to
serve
as
the
working
group
secretary.
A
It's
always
good,
to
see
folks
willing
to
come
forward
and
serve
the
community
in
terms
of
meeting
logistics.
We
are
using
miteko
thanks
for
joining
the
session.
Wyoming
techo
the
integrated
chat
or
jabber
is
available
for
use.
We
will
be
tracking
the
chat
during
the
session.
Please
use
the
notes.itf.org
link,
that's
listed
here
for
minute.
Taking
any
help
with
regards
to
note
taking
would
be
greatly
appreciated.
A
A
Today's
session
is
a
network
slicing
focused
meeting.
The
focus
is
exclusively
on
the
open
issues
in
our
working
group,
adopted
network
slice
framework
document
and
on
the
network,
slice
nba,
modeling
discussion
that
is
still
active
on
the
list.
The
objective
is
to
discuss
some
progress
as
much
as
we
can
towards
resolution
on
these
open
items.
A
This
is
the
agenda
for
today.
Adrian
will
have
a
go
first
and
highlight
the
open
issues
in
the
framework
document.
We
will
attempt
to
address
open
questions
on
the
definition
of
the
service,
the
endpoint,
and
also
look
at
the
realization
architecture.
As
part
of
this
section,
we
will
then
dive
into
the
next
section,
which
is
the
nbi
discussion.
The
authors
of
the
document
that
we
are
currently
polling
for
adoption
would
go
first
and
talk
about
the
rationale
behind
that
modeling
approach.
This
would
be
presented
by
bo.
A
Daniely
would
follow
that
up
with
a
view
on
the
concerns
related
to
the
document
that
is
under
consideration.
We
have
about
20
minutes
after
that
for
open
discussion
on
the
topics
covered.
We
will
use
this
20
minutes
discussion
time
as
we
team
fit.
The
chairs
would
then
hopefully
be
able
to
wrap
things
up
by
summarizing.
How
we
did
with
the
session
objectives
in
the
last
five
minutes
in.
A
Okay
in
terms
of
liaisons
and
communications
from
other
standards
bodies,
we
received
one
communication
from
gsma.
In
august
they
sent
a
pointer
to
a
white
paper
that
they
put
together,
which
discusses
the
end-to-end
network,
slicing
architecture.
They
are
asking
us
to
take
this
paper
into
consideration
and
are
seeking
feedback
on
it.
A
There
is
a
request
for
an
action
to
be
taken.
I
think
the
deadline
is
october
15th.
If
anyone
in
the
working
group
is
interested
in
sending
a
response
to
gsma,
please
to
review
the
white
paper
and
and
propose
a
response
on
the
list,
we
can
then
have
this
discussed
and
coordinated
with
the
isg
and
also
with
the
iab,
if
needed,
because
we
don't
have
a
formal
liaison
relationship
with
gsma.
We'll
get
this
done
by
our
ad.
A
A
As
a
reminder,
we
do
have
an
ipr
process
in
place
to
for
our
working
group.
There
are
two
mandatory
steps
before
accepting
it:
accepting
a
document
into
the
working
group
and
also
before
taking
it
last
call,
we
mandate
all
others-
all
authors
and
contributors
for
of
the
document
to
explicitly
state
if
they
are
aware
of
an
ipr
that
applies
to
the
document
in
recent
months,
we
did
introduce
another
step
in
the
process.
A
Finally,
please
do
use
the
mailing
list
as
much
as
possible.
Please
be
aware
that
the
mailing
list
is
where
the
working
group
consensus
is
actually
determined,
so
it
always
makes
sense
to
have
these
conversations
happen
on
the
list.
We
don't
know
how
long
we
will
continue
to
be
in
this
online
mode,
so
we'll
try
and
make
the
best
use
of
the
tools
that
are
available
to
us
to
make
progress
in
the
documents
working
group
members
can
request
for
an
online
interim
to
be
scheduled
if
there
is
a
need
to
cover
a
specific
topic.
A
Today's
session
is
a
good
example
of
that,
and
also
note
that
several
other
several
working
group
participants
are
already
making
use
of
the
working
group
webex
to
hold
regular
informal
working
meetings.
You
can
request
the
chairs
if
you
have
such
a
need,
these
meetings
will
be
announced
on
the
list
and
will
be
open
to
all.
That's
all
I
believe
we
have
for
this
lot.
I
will
now
go
over
to
adrian
for
his
presentation.
B
Thanks
praveen,
do
you
want
me
to
present
you're
going
to
present?
What
are
we
doing?
I
can
present
lovely.
Give
me
you
one
second.
So
what
I'm
going
to
try
to
do
is
a
quick
run
through
the
the
chief
open
issues.
B
Next
slide
that
we
kind
of
identified
at
the
last
atf
and
have
been
working
through,
maybe
a
bit
slowly.
B
B
So
network
slice
service-
that's,
interestingly,
screwed
up
its
conversion
from
powerpoint
to
whatever
this
is.
We
have
a
a
top
level
definition
and
that's
almost
agreed
and
I've
quoted
the
text
here
in
the
indented
stuff.
B
With
that
change
made,
I
believe
we
do
have
agreement,
but
there
was
debate
about
one
point
here
and
the
point
was
whether
service
one
network
slice
service
can
contain
multiple
connectivity,
matrices.
B
And
there's
a
little
confusion
here:
okay,
our
definition
of
a
connectivity
matrix
is
not
a
network
with
delivery
capabilities
and
the
way
to
think
about
that
is
a
mesh
network
in
a
mesh
network.
B
B
So
when
you
build
a
a
network
slice
service
request,
you
might
be
saying
I
want
a
set
of
points
to
point
potential
connections
so
that
I
can
build
a
full
mesh
and
send
my
traffic.
How
I
want-
or
you
might
be
saying
I
want
several
point
to
points
and
several
point
to
multipoints
and
one
multi-point
to
multi-point
and
build
it
up
like
that.
B
So
with
that
understanding
of
a
connectivity
matrix.
I'd
just
like
to
ask
if
anybody
wants
to
answer
the
question
on
the
previous
slide,
if
you
get
that
one
pavan
that
that
question
is,
are
we
happy
with
a
service
being
allowed
to
contain
multiple
connectivity,
matrices.
C
I
think,
okay,
what
you're
proposing
is
having
a
connectivity
matrix
per
per
type,
meaning
point-to-point
point
to
multi-point
or
multi-point
to
multi-point.
So
within
the
same
slice
I
one
can
have
a
you
know:
different
connectivity
matrices
with
different
types.
Did
I
understand
you
all
there.
C
Yes,
I
think
this
is
useful
and
indeed
there
could
be
multiple
applications
running
on
top
of
the
slice.
Some
unicast
some
multicast
and
having
the
capability
would
be
useful.
That's
my
opinion.
B
Okay,
so
this
is
good,
because
this
is
what
the
text
in
the
document
currently
has.
So
anybody
wanting
to
object
to
that,
or
in
fact,
if
lewis
wants
to
speak.
D
You
adrian,
my
only
point
here
was
that
also
in
this
definition
of
the
connectivity
matrix,
there
was
the
the
idea
of
only
a
the
the
parameters
for
the
delivery.
No,
I
mean
for
the
transmission
for
the
reception,
so
this
will
be
covered
later,
but
I
mean
my
opponent
here
will
be
okay
for
the
for
having
more
than
one
matrix.
This
is
at
the
end.
It
can
be
realized
in
different
manners
later
on
by
the
a
network,
slice
controller,
but
this
matrix
should
be
complete
and
probably
for
for
being
complete.
B
Okay
got
you
this
isn't
on
my
slides.
I've
made
a
note
and
I
will
set
that
out
on
the
list
and
see
if
we
can
get
agreement,
because
I
think
people
did
have
different
opinions.
B
Right
network,
size,
service
and
service
request
is
going
to
be
the
subject
of
much
more
discussion
later
in
the
mean
in
the
meeting.
B
At
the
moment,
we've
just
got
a
very
high
level
talking
about
slos
and
sles,
and
connectivity,
matrices
and
endpoints.
B
We
could
put
more
in
along
the
lines
of
an
information
model
or
we
could
just
say,
let's
just
cut
to
writing
the
young
model
for
the
customer
service
model,
for
an
mba
for
a
network
slice
and
not
put
anything
more
in
this
document.
A
Not
to
myself,
hopefully
yeah
personally,
I
think
the
information
model
adding
that
to
this
framework
document
is
a
bit
of
an
overkill
like
you're
saying
I
think
you
do
have
most
of
the
components
already
listed,
maybe
a
bit
more
text,
and
I
think
we
should
be
good.
I
I
think
info
model
in
this
document
at
least
is
oracle.
B
E
Eco,
hey
guys,
hey
adrian,
just
a
quick
question:
if
you
have,
I
say
two
connectivity
mattresses
in
the
same
slice,
okay
and
so
two
pair
of
endpoints
belong
to
both
connectivity
matrices.
How
do
you
distinguish
which
connectivity
methods
needs
to
be
used,
in
particular
case.
B
So
yeah
that
that
question
is
is
real
and
needs
an
answer,
but
I
believe
that
the
answer
is
solution
dependent.
B
Let
me
give
you
an
example
of
a
way
you
might
do
it
the
the
way
you
might
do.
It
is
simply
to
assign
a
different
vlan
tag
to
each
connectivity
matrix.
E
B
I
so
I
I
I
really
desperately
don't
want
to
get
sucked
into
the
how
to
make
this
work,
using
particular
technology
solutions.
B
E
My
point
is
that
it's,
it's
probably
unnecessary,
so
why
not
to
use
separate
slices
and
use
one
connectivity
methods
per
slice
and
it's
very
clear.
B
So
you
need
to
go
back
and
look
at
that
definition
of
connectivity
matrix.
If
what
you
want
to
do
is
stick
with
that
definition,
then
all
you're
saying
is:
let's
have
lots
and
lots
of
slices
compared
to
right
compared
to
one
slice
with
multiple
matrices
lots
of
lots
of
them
yeah.
So
my
my
original
point
was:
if
an
implementer
wants
to
support
multiple
matrices
on
a
slice
and
they
can
do
it,
why
should
we
stop
them?.
E
B
E
So
are
you
talking
about
like
the
same
name,
space
across
all
slices
and
connectivity
matrixes?
No,
so
if
I'm
using
several
slices
and
within
each
slice,
I
have
several
connectivity
matrices
right.
So
how
do
I
distinguish
that.
E
B
B
G
We
can
steals
a
little
time
from
the
final
discussion,
but
we
probably
shouldn't
steal
too
much
so
I'm
going
to
let
you
manage
the
queue
as
you'd
like
and
john.
If
you
want
to
jump,
if
you
want
to
go
into
the
queue
you
should
enter
the
queue
we
have
joel
and
john
scudder
in
the
queue.
F
I
am
a
little
confused
by
igor's
assertions,
because
I'm
looking
at
the
email
that
kristoff
sent
to
the
list,
which
pointed
out
that
you
may
want
multiple
connectivity
matrices
to
represent
different
services
for
different
subsets
of
traffic
within
the
slice
and
separately.
You
may
have
properties,
for
example,
of
the
port
that
are
slice
wide.
F
Therefore,
it
is
not
sufficient
to
just
say
well,
just
have
to
use
a
separate
slice
for
each
connectivity
matrix,
because
there
are
properties
that
are
for
the
slice
and
there
are
properties
that
are
for
the
connectivity
matrix,
and
if
that
is
true-
and
I
believe
he
is
correct,
then
the
structure
you're
using
is
correct.
We
may
need
to
allow
more
cases
where
there
are
multiple
connectivity
matrices,
but
again
exactly
how
the
traffic
is
bound
is
not
a
problem.
We
need
to
deal
with
at
this
level.
B
Okay,
thanks
joel
what
I've
just
written
down
for
myself
is
that
I
need
to
draw
and
show
everybody
a
venn
diagram
that
I
believe
will
explain
and
at
least
allow
us
to
make
a
decision
and
john
has
gone
away.
B
Yeah
we
have
that
text
in
the
document
and
it's
just
a
question
of
whether
we
can
get
everybody
to
agree
that
it
is
at
at
least
harmless
okay
force
for
time.
Let's
move
on
to
the
next
issue
next
slide.
B
The
debate
was:
where
is
the
slice
service
delivered,
and
a
lot
of
this
discussion
has
happened
in
the
context
of
cs
and
pes,
but
we
should
recognize
that
this
terminology
of
cemp
is
not
universally
helpful
when
people
are
interested
in,
for
example,
network
functions
and
so
on.
But
it's
it's
a
useful
picture
and
I
identified
four
cases
of
where
the
service
endpoint
might
be,
as
shown
on
this
figure
and
the
text
goes
through
those.
B
Yes,
that's
it
that
one!
Thank
you.
What
I
want
to
do
is
cut
to
my
conclusions
here
and
say
that
we
should
embrace
all
four
possible
endpoint
locations.
B
I
think
this
came
out
are
in
a
thread
on
the
list,
probably
as
far
back
as
april,
where
reza
and
a
couple
of
others
were
exchanging
emails
about
where
the
end
points
might
be.
B
I
B
Brilliant.
Thank
you.
That's
that's
good
to
hear
because
you
were
the
main
person
in
that
discussion,
so
I
will
convert
this
into
text.
As
you
see,
I've
already
done
the
ascii
art.
B
Let's
the
management
architecture
is
a
bit
boring
and
it's
a
bit
ndi
document.
Let's
go
to
the
next
slide.
B
Right,
realization,
architecture,
the
the
the
desire
is
to
have
a
common
and
generic
architecture.
That's
independent
of
solution
approaches,
so
we
need
to
get
terminology
that
we
can
all
agree
on
and
then
technology
solutions
can
point
back.
B
The
challenge
is
that
the
solutions
work
is
quite
advanced,
and
so
we've
got
a
bit
stuck
with
terminology
and
models,
but
really
good
offline
discussion
between
the
the
authors
of
two
sets
of
drafts,
the
enhanced
vpn
and
the
best
bar
families
they're
making
good
progress.
So
next
slide.
B
What
we
can
agree
on
is
that
we
need
some
kind
of
scaling
mechanism
so
that
the
very
many
sizes
don't
need
to
be
individually
micromanaged
in
the
network.
B
There's
a
question
there
as
to
whether
we
need
to
be
able
to
break
out
individual
flows
from
services,
slice
services
and
map
them
onto
different
sets
of
network
resources.
Some
debate
on
that.
I
I
find
it
hard
to
justify,
but
anyway,
next
slide.
B
How
do
we
build
that
set
of
resources
and
what
is
the
mapping
of
services
to
a
set
of
resources
one
to
one
and
enter
one
seem
to
be
obvious
for
scaling
enter
into
m.
Is
this
is
a
bit
confusing
both
from
an
implementer's
point
of
view,
but
also
how
to
how
to
express
it.
B
This
pretty
picture
is
attempting
to
get
some
terminology
out,
and
so
the
way
to
read
this
picture
is
to
read
it
from
the
top
down
and
the
bottom
up.
At
the
same
time,
from
the
top
down
a
network
slice
is
this
set
of
potential
connections
between
end
points
and
that's
a
customer
view,
and
from
the
bottom
upwards
we
have
a
a
physical
network
that
has
to
be
programmed
to
forward
and
treat
the
traffic
correctly.
B
Then
continuing
from
the
bottom
upwards,
there's
the
potential
to
filter
a
topology
to
produce
a
set
of
nodes
links,
resources
that
have
specific
capabilities,
for
example,
they
might
be
links
that
support
maxsec
whatever
from
the
top
down.
We
have
this
thing
of
grouping
slices
into
a
bundler
set
that
we
operate
on
a
partition
of
the
network
resources,
and
then
we
have
this
thing
of
generating
that
set
of
resources
from
the
available
topology.
B
The
names
we
have
here-
ietf
network
slice
and
itf
network
size
service.
I
think
we're
happy
with
physical
network
we're
happy
with,
and
that
just
gives
us
two
names
to
deal
with
the
network
resource
partition
and
I
think
I'm
seeing
debate
now
that
we're
down
to
either
network
resource
partition
or
network
resource
group
with
slightly
different
overtones
for
group
or
partition,
and
then
we've
not
really
talked
about
the
filter
topology.
B
E
G
Adrian
does
the
research
is
the
resource
partition
defined
by
the
filter
topology
or
what
is
the
relationship
between
the
resource
partition
group
and
that
filter
technology.
G
So
a
group
inherently
has
topological
relevance
or
significant
scope.
I
guess
is
the
right
word
to
say
it.
B
G
Okay,
thank
you
for
for
me,
as
participant
partition
makes
sense
only
if
it's
related
to
a
specific
topology,
because
you
know
we
always
talk
about
partitions
as
separation
of
parts
of
the
network.
So
from
that
context
a
group
to
me
would
be
independent
or
could
be
independent
of
topology.
B
C
Yes,
I
I
think
this
is
a
very
useful
diagram.
Honestly,
thanks
for
putting
it
together,
it
clarifies
a
lot.
You
know
how
we,
at
the
end
of
the
day,
realize
the
service
in
our
on
on
the
underlay
transport
network,
and
you
know
the
steps
might
vary,
which
one
starts
before
the
next
one,
but
overall
at
least
the
solution
that
we
are
pushing
for
is
aligned
with
this.
So
I
would
like
to
see
this
in
our
network
slice
draft,
at
least
this
diagram.
If
you,
if
you
can.
B
Yes,
yeah,
I
you'll
notice
it's
pre-prepared
in
ascii
as
well
and
you're
right.
I
I
had
started
work
on
a
workflow
and
then
I
believe
that
which
order
you
build
these
things
in
is
very
much
implementation,
specific
and
probably
doesn't
matter
from
a
framework
point
of
view.
Thanks
chairs
over
to
you.
B
A
Okay,
we
will
now
shift
our
focus
to
the
network
slice
nbi
discussion.
We
did
initiate
that
option
for
draft
wd
on
august
27th.
There
seems
to
be
consensus
on
the
need,
at
least
at
least
on
the
need
for
an
ietf
network
slice,
nba
data
model
to
be
furnished.
There
is
reasonable
support
for
adopting
this
document
as
the
starting
point
for
this
work,
but
we
do
have
a
reasonable
number
of
working
group
participants
who
have
raised
concerns
on
the
modeling
approach
used
in
this
document.
A
We
do
need
to
resolve
these
concerns,
and
that
would
be
the
objective
for
this
section
of
the
session.
At
a
high
level,
there
are
primarily
two
questions
that
we
would
like
to
get
discussed
and
resolved.
The
first
question
is
with
regards
to
the
scope
of
the
nbi:
is
this
a
service
model,
or
is
it
something
else?
There
seems
to
be
a
school
of
thought
positioning
this
as
a
network
model,
we
will
have
somebody
we'll
have
daniel
represent
that.
A
The
second
question
is:
what
should
the
nbi
be
based
on?
Should
we
define
a
new
independent
service
model,
as
is
defined
in
traffic
wd,
or
should
we
use
the
existing
vn
model,
or
should
we
define
a
new
model
by
augmenting
the
vn
model
or
any
other
existing
service
model
like
l3s
model
tsm,
or
do
we
leave
room
for
all
options,
so,
hopefully,
by
the
end
of
this
session,
we
will
have
some
clarity
on
the
way
forward
for
these
questions,
so
without
any
further
delay.
A
Let's
bring
on
bo
first
to
give
the
presentation,
on
behalf
of
the
authors
of
draft
wd,
oh
you're,
up
next,
give
me
just
a
minute
to
shift
lights.
A
J
Yeah,
okay
thanks
this
is
for,
and
I
will
be
present
on
behalf
of
all
these
courses.
J
Oh,
I
cannot
see
the
screen.
Okay,
do
anybody?
Do
anyone
can
see
this?
Oh
yeah?
Yes,
during
a
working
group,
adoption
call.
J
We
received
various
comments
on
this
model,
and
so
here
I
would
like
to
highlight
that
because
there's
some
comments
about
the
model's
purpose
and
some
about
the
modeling
approach
and
the
first
thing
that
I
would
like
to
highlight
that
this
model,
given
the
definition
on
8309
its
service
model,
explained
that
this
model
is
a
customer
service
model.
J
As
adrian
just
presented,
the
network
slice
framework
defines
ietf
networks
like
service
and
and
the
doc.
The
framework
also
defined
that
natural
slice,
northbound
interface
will
be
used
by
customer
to
request
for
the
network
slices
so
based
on
the
customer
service
model
definition.
J
This
mbi
model
is
class,
classified
as
a
customer
service
model,
so
and
since
this
is
a
customer
service
model
and
they
during
the
modeling
process,
we
investigate
the
current
existing
customer
service
modules.
J
So,
the
this
slice
service
model
is
used
when
service
provider
for
provides
like
slicing
as
service,
like
so
different
from
the
existing
one
like
layer,
three
vpns
service
or
ac
tmvn
as
service.
So
in
this
way,
customer
could
focus
on
the
service
requirements
rather
than
whether
it's
a
vpn
technology
or
whether
it's
a
ctn
technology.
J
So
this
is
the
model
position
that
next
slide.
Please.
G
I'm
not
seeing
the
same
delay.
I
can't
tell
for
other
people,
but
maybe
just
get
start
just
start.
J
Yes,
I
can
go
with
this
one,
but
actually.
J
H
J
Is
the
number
three
one,
but
I
can
go
with
this
because,
like
in
in
the
second
slide,
I
summarize
all
the
comments
regarding
this
modeling
approach,
like
someone
proposed
to
use
weight,
reuse
with
a
model,
and
someone
also
proposed
to
use
augmenting
wind
model.
J
Yes,
now
I
can
see
the
the
summary
about
this
modeling
options
for
network
slice,
service
model.
J
Coupling
with
t
topology
model
customer
has
to
use
a
whale
model
and
together
with
t
topology,
to
define
the
service,
and
apart
from
that,
real
model
also
use
a
definition
with
like
weigh-in
ap
access,
point
or
virtual
network
access
point,
which
these
terms
are
quite
different
with
the
the
slice
network
network,
slice
definition
so
and
also
this
model
means
a
lot
of
these
unless
features
defining
framework
like
slo
and
I
sell
e
and
or
they
they
have
some
sl
definition,
but
they
use
t
topology
past
constraints
to
represent
that.
J
So
it's
very
in
the
model
it's
hard
to
use,
so
we
are
thinking
reusing.
This
wheel
model
is,
is
pro,
is
quite
hard
and
the
other
approach
is
augmenting
with
model
and
then
the
same
issue
applied
here.
So
still
this
is
t
topology
coupling
and
there's
also
some
comments
proposed
that
they
want
to
remove
t
topology
banding,
but
because
this
model
doesn't
exist
and
and
right
now,
ietf
network
slice
definition
doesn't
see
that
a
network
slice
is
a
type
of
virtual
network
and
and
nse
is
also
defining
ap
or
vnap.
J
So
we,
the
authors,
anal,
gave
the
analysis
on
these
models
still
think
we.
We
still
suggest
that
an
independent
model
is
needed.
J
So
in
this
way
the
network
slice
service
model
is
modeling,
a
user
modeling
approach
as
layer,
3,
sm
and
layer,
2
sm,
and
it
doesn't
use
other
t,
topology
model
or
other
top
topology
model
to
support
the
service
and
because
and
also
right
now,
in
teas
network.
There
are
other
approaches
to
implement
slice
services,
something
like
multi-topology
flexago.
J
These
are
not
t
techniques,
and
so
we
are
thinking
a
vm
model
can
be
used
to
realize
the
slide
service.
J
This
could
be
like
together,
used
together
with
other,
like
there's
three
assembler
2sm
and
also
other
network
model.
So
next
slide,
please.
J
Yes,
here's
we
want
to
give
a
more
detailed
modeling
approach,
comparison
on
like
the
what's
the
difference
with
accused
point,
defining
win
model
and
what's
and
the
nse
definition-
and
you
can
see
in
this
picture
that
I
show
a
t
topology
here
and
access
point.
Defining
wheel
model
refers
to
the
access
link
between
customer
node
and
also
the
provider
node.
So
with
each
access
link,
then
we'll
we'll
use
an
access
point
to
identify
this
access
link.
J
So
it's
a
one-to-one
mapping,
this
access
link
and
with
this
and
we'll
also
use
t
topology,
abstract
node,
to
define
the
weigh-in
connectivity
matrix.
So
they
are
using
this
modeling
approach.
But
given
the
difference
with
networks,
slice
endpoint,
actually
network
slice,
endpoint,
just
like
adrian,
just
give
the
definition
on
the
and
service
endpoint
this
modeling.
J
J
But
the
win
access
point
actually
is
the
are
the
two
options
that
the
access
link
is
equivalent
to
the
access
connectivity
and
that
two
options
is
wins
modeling
approach,
so
here
is
a
difference
about
this
modeling
of
endpoint
difference.
J
Oh,
I
think,
there's
one
yes
and
here's.
We
gave
the
summarization
of
these
two
different.
The
difference
between
these
two
young
models.
Young
structures
of
this
two
young
model
has
quite
different
sayings
like
in
weigh-in.
They
use
they.
They
maintain
a
ap
list
and
win
ap
association.
J
This
association
doesn't
defining
a
network
slice
framework,
and
you
can
also
see
that
we
show
here
the
young
model.
The
wind
will
referring
to
the
leaf
ref
to
t
topology
definition
like
abstract,
node
and
also
connect
connectivity
matrix.
J
J
It's
the
two
options
defining
in
this
in
those
use
cases,
and-
and
we
think
that
when
and
t
topology
young
model
are
just
one
of
the
slice,
selection,
slice,
realization
techniques.
J
So,
and
here
we
like
each
and
as
end
point,
we
we
modeling
here
as
we
can
associate
one
each
endpoint
can
associate
with
multiple
access
link
here,
so
we
can
cover
the
four
options
that
defining
network
slice
framework.
J
Yes
and
then
there
is
also
comments
about
that
is
network
slice
service
young.
This,
the
only
network
slide
service
model.
J
Our
answer
is
no
because
customer,
like
service
provider,
could
use
layer,
3,
sm
or
their
2sm
or
weigh
in,
can
use
for
slicing
and
also
realization
for
this
technology
agnostic
service
model,
and
we
think
that
both
all
these
models
can
be
coexist
and
can
be
provided
for
customers
for
different
usage,
especially
when,
when
a
service
provider
may
implement
multiple
technologies
in
their
network
and
to
support
different
slice
to
to
support
slicing
service,
and
this
independent
service
model
can
help
to
hide
the
implementation
difference.
J
So
our
consideration
on
this,
maybe
in
the
future
revision
of
this
draft.
We
can
add
more
text
to
clarify
this,
that
that
why
this
model
can
be
coexisted
with
other
customer
service
models.
J
Here
we
give
a
summary
on
the
on
the
concerns
raised
on
the
mailing
list,
and
here
we
want
to
like
to
clarify
most
of
concerns
so
peop,
so
so
people
can
get
get
more
understanding
about
this
model's
relationship
with
the
current
ac
tn
framework
and
also
other
models.
J
Our
answer
is:
this
model
can
be
part
of
a
cmi
interface
along
with
a
wheel
model
or
layer,
3,
sm
or
layer,
2
sm,
and
whether
this
model
is
a
replacement
wheel
model.
No,
we
are
thinking
about
real
model.
They
can
be
coexistent
and
can
be
used
to
realize.
J
And
I
think
I
will
not
read
out
all
these
texts-
and
these
are
just
for
information
for
the
working
group.
J
And
next
pack,
please
yeah.
So
our
conclusion
still
is
that
we
believe
a
new
independent
network
slice
service
module
is
needed
as
per
a
current
ietf
network
slice
definition
and
this
model
doesn't
have
some
can
coexist
with
other
existing
customer
service
model,
and
so
this
is
very
helpful
for
this
network
slide
service
work.
J
So
that's
all
for
the
presentation
thanks
is
there
are
some
questions
questions.
Maybe
it
makes
sense
to.
G
Jump
right
into
danielle's
presentation
and
then,
unless
someone
has
a
big
clarifying
question,
you
know
if
you're
talking
about
the
questions
that
have
come
up
on
this,
let's
go
to
danielle,
first,
so
hearing
no
one!
Why
don't
we
jump
there.
K
Okay,
so
questions
yeah,
perfect
good
thanks
a
lot.
So
this
is
a
summary
I
think
when,
when
the
agenda
was
put
together,
the
title
of
concerns
was
perfectly
identifying
the
issue.
So,
let's
start
from
the
very
beginning.
K
The
draft
is
presented
as
a
customer
service
model,
but
our
impression
was
that
it
was
used
a
little
bit
differently.
It
was
claimed
to
be
a
customer
service
model,
but
it
was
used
a
little
bit
differently,
so
we
took
rfc8309
and
we
started
to
analyze
the
various
models
here,
given
that
we
often
speak
about
customers.
K
K
We
started
to
analyze
a
little
bit
where
this
model
could
fit
and
let,
let's
start
from
from
the
right.
So
is
it
the
service
delivery
model,
so
something
that
sits
between
the
service
orchestrator
and
the
network
orchestrator
both
of
them
managed
by
the
operator?
K
Probably
the
answer
the
answer
is
not
is
no,
and
I
mean
the
previous
preventive
presentation
also
said
clearly
that
it's
not
something
that
sits
on
interface
b
and
by
the
way.
One
thing
that
we
seem
to
agree
on
is
the
fact
that
the
service
delivery
model
doesn't
really
exist
in
in
real
networks,
because
the
service
orchestrator
and
the
network
orchestrator
in
many
cases,
are
collapsed
into
into
a
single
entity,
which
is
what
we
call
here
for
simplicity,
oss,
slash
orchestration.
K
So
then
this
is,
if
you
look
on
the
left
side
of
the
slide,
this
is
option
one.
This
is
a
customer
service
model,
so
the
one
that
sits
against
interface,
a
so
between
the
customer
and
the
oss
orchestrator.
K
K
K
If
you
look
at
the
right
side
of
the
picture,
we
have
what
we
thought
would
happen,
but
didn't
happen,
which
is
the
the
left
side
branch,
where
the
cnc,
the
customer
application
controller.
Whatever
is
direct
access
through
customer
services,
service
models
to
the
mdsc
to
the
operator
network
in
general.
K
What
really
happens
in
the
industry
is
what
you
see
on
the
right
where
you
have
the
oss,
the
orchestrator,
what
we
call
the
before
oss
orchestration,
which
again
is
joined
the
service
orchestrator
and
the
network
orchestrator,
but
managed
and
operated
by
the
operator
which
have
access
to
the
mdsc.
So
there
is
where
we
have
the
network
models.
In
fact,
it's
where
we
have
the
l2
nm
model,
3,
nm,
etc,
we
etc.
K
We
hardly
ever,
if
not
ever
see
the
service
models
between
those
two
components.
The
customer
then
sits
on
top
of
this
oss
box.
There
can
be
other
boxes,
the
bss,
etc,
and
it
uses
the
customer
service
model
to
request
for
services,
connections,
etcetera
to
the
operator
from
a
network.
Slicing
point
of
view,
we
agree
that
this
customer
service
model
is
missing.
K
K
So
discussion
a
a
ctn
or
non-ctn.
Actually,
this
is
something
that
applies
to
both
sctn
and
known
scpn
frameworks
which,
by
the
way,
are
extremely
similar.
K
If
you
see
on
the
left,
we
have
an
ect
and
enable
the
network
where
we
have
the
mdsc,
the
pnc
is
etc.
On
the
right,
if
we
have
a
known,
ctn
type
of
network
with
or
without
traffic
engineering,
for
example,
unf
topi,
tip
master
method,
so
the
architecture
is
is,
is
always
the
same
with
the
same
type
of
models
being
used,
and
here
we
are
adding
now
the
nfc,
the
nsc
I
mean
is
not.
K
K
Let's
assume
that
the
network
slicing
nbi
is
a
customer
service
model,
so
it
seems
that
we
are
all
starting
to
a
degree.
This
is
the
case.
A
service
model
for
network
slicing
is
missing.
Yes,
yes,
this
is
this
is
true.
We
have
service
models
over
there
del2
sam
three
sm,
but
they
are
not
enough
for
for
nato
slicing,
it
could
be
possible
to
augment
them,
but
I
mean
no
one
is
objecting
the
fact
that
a
new
model
can
be
can
be
used
there.
K
Then
the
model
presented
in
the
draft
is
is
a
candidate
to
fulfill
it.
Yes,
it's
something
that
should
be
used
by
the
customer
to
request
a
slicer
to
the
operator.
K
It's
expected
to
be
done
to
against
the
uss.
Not
so
it's
not
an
nfc
ndi.
If
we
all
agree
that
the
nsc
is
something
sitting
at
the
mdsc
level,
slash
hierarchical,
sdn,
controller
level
and
this
model
shouldn't
be
used
by
the
operator
as
a
network
model.
So
it's
not
something
that
should
be
used
by
between
the
two
operator
components
so
between
the
oss
and
the
mdsc
or
hierarchical
sdn
model,
simply
because
existing
models
can
already
meet
the
requirements.
K
K
K
By
the
way.
Speaking
about
models
we
went
to,
we
tried
to
understand
a
little
bit
better
better
where
the
l2
sm
and
nm
and
three
sm
and
then
m
position
themselves
to
see.
If
there
was
alignment
with
the
sorry
I
I'm
already
jumping
into
the
next
slide.
I
can.
Can
you
go
to
that
thanks
thanks
a
lot,
so
we
we.
K
We
wanted
to
check
how
the
service
and
network
models
position
themselves
against
rfc
8309,
and
we
found
that
there
is
actually
a
lot,
a
lot
of
confusion,
because
the
l3
nm
says
it's
a
service
delivery
model.
The
l2
nm
says
it's,
it's
a
network
model,
so
I
mean
we
have
some
confusion
in
mind,
but
I
would
say
this
is
something
that
is
a
pretty
much
widespread.
K
K
So
again,
what
is
this
component
to
this
customer
level
operation
for
what
we
said
before?
This
is
what
we
call
the
dss
orchestration,
something
that
belongs
to
the
operator
that
doesn't
belong
to
the
customer.
In
fact,
if
you
see
the
mapping
on
the
right,
everything
is
red,
light
red
and
not
and
not
blue.
K
So
once
again,
the
nfc
is
a
functionality
of
the
hierarchical
sdn
controller
in
honesty
and
networks
or
of
the
mdsc
in
actin
networks.
You
see
it
sits
here
in
the
in
the
middle
of
this
hierarchy,
then
a
cnbi
is
towards
the
uss,
it's
not
towards
the
customer.
K
Next,
please,
so
a
very
quick
overview
of
the
existing
models
with
them.
Without
a
sctn
there
between
the
oss
and
the
hierarchical
sdn
controller,
we
have
a
wide
set
of
modules:
vpn
network
models,
l1
csm,
l2
and
m3,
and
m
network
models,
not
not
service,
not
to
the
service
version.
Topology.
We
have
the
network,
topology
t
topology
for
the
infrastructure.
We
have
t
tunnels,
virtual
network
lsps
policies
and
the
t
service
mapping.
How
is
it
possible
to
to
to
to
realize
a
an
ethos?
K
Slicer
today
in
a
non-te
type
of
network,
you
can
ask
for
a
plain
vpn
layer,
2
or
layer
3..
You
can
ask
for
plain
infrastructure
in
lsp,
a
simple
policy
that
can
be
implemented
via
flexalgo
segment,
routing,
etc.
If
yen,
with
or
without
tea.
Today,
the
vienna
doesn't
support
the
non-tea
scenarios,
but
the
augmentation
is
is
extremely
extremely
simple.
K
In
a
tea
type
of
environment,
you
can
do
something
a
little
bit
more
more
sophisticated.
You
can
create
a
vpn.
You
can
create
a
a
t
tunnel,
a
and
then
bind
the
vpn
to
the
t
timer
using
the
t
service
mapping,
which
is
the
reason
why
the
t
service
mapping
was
created.
You
can
do
it
with
a
t
tunnel.
You
can
do
it
with
a
vm,
I
mean
both.
Both
of
them
are
perfectly
perfectly
work,
as
I
said
before.
K
K
Proposed
augmentations
so
first
one
supporter
for
technology,
agnostic,
topology,
the
vienna
abstract
node,
to
be
used
to
reference,
a
topology
which
today
is
a
t
topology
using
a
profile
of
the
t.
Topology
for
non-t
scenarios
is
perfectly
possible.
It's
as
it's
explained
today,
in
an
existing
draft.
In
the
end
that
is
working
group,
so
it's
possible
to
apply
the
the
abstract
node,
also
to
non-t
topologies.
K
Second,
the
usage
of
draft
liu
is
the
the
draft
that
the
shuffang
and
his
code
also
put
together,
which
augments
the
network
topology.
K
So
here
we
just
need
to
change
the
reference
of
abstract
node
into
node
id
so
again,
something
very,
very
simple:
exposing
the
path
constraints
of
the
vn
and
debian
member
as
slo
and
sle,
instead
of
having
them
hidden
in
the
connectivity
matrix
like
it
is
today,
but
again
I
mean
it's
taking
one
parameter
and
removing
it
exposing
it
outside,
instead
of
keeping
it
into
the
connectivity
matrix
support
of
technology
agnostic
connectivity
matrix.
K
This
is
something
that
needs
to
be
a
little
bit
discussed
and
then
a
simple
mapping
of
terminologies
like
like
the
following
one
net
slice:
virtual
network,
slo,
sle
policy,
part
constraints,
nsn,
point
access,
point,
mdn,
access,
point,
ns
connection
is
a
vn
member.
The
all
of
these
things
that
perfectly
map
one
to
one
and
they
meet
all
all
the
constraints
of
the
nether's
lies.
K
Basically,
this
comes
from
the
people
in
c
camp
working
on
on
otn
slicing.
They
would
prefer
to
have
a
model
convergence
in
the
sense
that
tn
slicing
is
something
that
is
a
technology
specific.
K
And
then
work
on
a
technology
independent
augmentation
of
of
this
model.
I
just
wanted
to
bring
this
up
as
a
further
consideration.
G
You're
a
bit
over
you're
over
by
a
few
minutes,
and
we
should
get
the
conversation
going.
I
was
thinking
we
could
go
back
to
the
question
slide
if
that's
okay
with
pavon,
unless
he
wants
to
do
something
different
and
we
thank
you.
We
have
adrian
beau
reza.
So
when
we
get
started
adrian.
B
Yeah
I'll
start
talking,
while
you
find
the
connection
slide,
I
have
three
points
from
for
danielle.
The
first
one
is
to
be
really
careful
on
your
use
of
customer
versus
consumer.
B
The
next
small
point
is
that
when
I
look
at
bose
draft,
the
only
time
it
mentions
nbi
is
actually
in
the
file
name,
the
whole
of
the
rest
of
the
time.
It's
talking
about
network
slice,
customer
service
model,
and
my
belief
is
that
we
need
a
network
slice
customer
service
model
in
the
working
group
and
provided
we
make
sure
that
we
all
know
that.
That
is
what
we're
talking
about.
We
should
just
get
on
and
deal
with
it
and
then.
B
J
Yeah,
I
think
adrian
helps
us
to
clarify
the
fellow
name
yeah.
We
are
sorry
about
the
file
name.
It
may
be
misleading
that
that
seems
more
like
a
network
model,
but
actually
we
are
doing
a
customer
service
model
and-
and
the
other
thing
is
that
dan's
talking
about
the
net
slice
network
model
and
we
we
think
the
these
are
quite
useful.
But
this
is
not
our
model
and
we
hope
we
can
get
consensus
that
on
defining
this
customer
service
model.
J
First,
following
the
modeling
approach
like
their
3sm
and
their
2sm,
then
we
can
working
on
the
network
model
thanks.
L
I
Yeah,
I
guess
one
clarification
that
adrian
also
mentioned,
and
I
think
it's
very
important-
it's
the
concept
of
the
consumer
or
customers.
So
I
don't
want
to
repeat
what
adrian
mentioned,
but
the
important
aspect
here
is
that
depends
on
the
use
case.
You
know
we
put
a
group
of
use
cases
in
the
framework
document
and
definition
document
that
what
are
the
use
case
of
ietf
network
depends
on
the
use
case.
This
consumer
could
be
an
orchestrator,
could
be
end
user.
I
Another
thing
that
it
was
in
the
presentation
of
daniele
was
the
he
mentioned
about
network
slice
or
sorry.
Net
forecast
slice,
and
that
is
basically
from
our
terminology,
is
a
network
slice
realization.
Any
technique
that
you
mentioned
has
some
sort
of
the
technology
specific
referring
to
the
te
or
referring
to
l2sm
or
l3sm.
I
All
those
are
values,
including
hctn
itself,
are
the
realization
techniques
they
can
well
live
with
the
model
that
we
have
our
model
sitting
at
the
top
and
all
those
realization
models
can
be
used,
and
from
that
aspect
you
mentioned
that
mdsc
and
nsc
are
the
same
level
so
from
that
as
well.
You
will
see
that,
if
consider
that
realization
versus
technology
agnostic,
the
view
is
a
bit
different.
C
Yeah
thanks
danielle
for
the
great
presentation
I
do
want
to
mention
the
the
mappings
that
you
suggested
on
one
of
the
slides.
They
are
not
for
someone
working
on
network
slicing.
They
are
not
obvious
to
come
up
with
these
mappings
and
you
know,
although
it
is,
I
totally
agree,
it's
possible
to
you
know
augment
the
vn
model
to
achieve
the
you
know,
connectivity,
matrix
and
slos,
and
realize
the
network
slice
service,
possibly
those
mappings.
C
You
know
I
don't
know
where
we're
gonna,
maintain
them
where
someone
working
on
network
slice
service
will
find
them.
So
is
it
going
to
be
in
the
network
slice
definitions,
draft
that
a
v
nap
is
maybe
I'm
the
wrong
term,
but
a
virtual
yeah,
an
an
endpoint
or
a
vn
member
is
a
connection.
C
I
don't
know
how
many
know
that
you
can
map
a
vn
member
to
an
ns
network,
size
connection.
If
we
have
this
mapping
somewhere,
then-
and
I
don't
know
if
you
want
to
do
that-
it
would
be
more
a
little
bit
easier
to
do
this
extension
yeah.
That's
what
I
wanted
to
mention
thanks.
M
Go
ahead.
Okay,
I
just
I
was
a
little
bit
confused
by
the
discussion,
so
I
understand
the
difference
between
customer
and
as
an
end
user
and
and
and
a
consumer,
but
I
have
not
understood
whether
this
year
model,
the
wd
via
model,
is
a
customer
service
model
used
by
the
customer,
not
by
the
consumer
of
the
neto
slice,
to
request
a
service
or
it
is
a
consumer
service
model
or
to
be
used
by
the
consumer
of
the
neto
slicing.
I
have
been
a
little
bit
lost
by
that
discussion.
L
Hi
my
comment
to
danielle
one
was
danny
in
the
in
the
rfc
8453.
We
do
split
the
mdsc
functions
across
boxes
and
we
say
that
mdsc
can
easily
be
split
at
the
service
orchestrator
and
at
the
network
orchestrator.
But
throughout
the
analysis,
I
think
you
kind
of
showed
mdsc
as
a
single
box,
and
I
was
kind
of
thinking
that
why
we
always
thought
mdsc,
not
as
a
physical
box
but
more
as
a
functional
elements
that
needs
to
do
that
and
even
when
we
map
itf
nfc.
L
D
E
I
have
a
question
actually
to
derek
and
paul
so
if,
for
example,
a
network
slice
has
several
connectivity,
mattresses,
okay,
what
would
it
make
to
the
definition
of
network
slicing
corrugator?
What
does
it
mean
in
this
case
this
one,
this
question
to
you
and
another
question
to
danielle:
how
do
you
say
that,
when
you're
using
a
particular
vm
service,
you
need
to
use
a
particular
connectivity
mattress
within
vm,
okay-
and
I
I,
as
far
as
I
remember,
it,
is
not
possible
today
to
have
more
than
one
connectivity
matters
per
gram.
G
So
danielle
says
that
he
has
to
go.
He
got
the
the.
J
Can
I
have
a
quick
response
to
ego
yeah
our
network
slice
service
model
currently
doesn't
support
a
multiple
connectivity
metrics,
but
the
authors
are
discussing
on
how
to
model
that
yeah.
That's
my
answer.
E
Okay-
let's
discuss
this
later
but,
for
example,
from
the
point
of
view
of
network
slice,
aggregation.
E
B
Yeah,
so
I
wanted
to
answer
atello's
question
about
customer
and
consumer
and.
B
I
think
it's
important
to
not
even
start
to
think
about
customers
as
end
users
or
paying
people
and
and
so
on.
When
we
wrote
rsc
8309,
which
talked
about
the
customer
service
model,
we
were
thinking
very
much
in
terms
of
customer
and
consumer
being
synonyms.
B
So,
for
example,
a
layer
3
vpn
customer
is
the
person
or
organization
that
uses
the
layer,
3
vpn
and
it
may
be
an
internal
structure
within
an
operator
and
it
may
be
cascaded
hierarchically.
It
doesn't
really
matter
and
that's
why
we
moved
to
the
term
consumer
for
for
slicing
and
probably
if
we
were
rewriting
the
8309
we
would.
We
would
call
it
a
consumer
service
model
rather
than
a
customer
service
model.
B
N
I
have
a
one
question
about
the
technology
agnostic
terminology.
N
My
question
is:
is
packet
switching
technology,
agnostic
or
technology
specific,
because
if
we
are
talk
when
I,
when
we
are
dealing
with
say
other
type
of
slicing
like
otn
slicing,
some
parameters
does
not
apply
right.
N
So
when,
when
we
talk
about
the
packet
agnostic
technology
agnostic,
we
need
to
define
the
model
and
the
parameters
on
the
slo
to
be
technology,
really
technology
agnostic.
C
Yeah
I
want
to
go
back
to
igor's
question
about
the
multiple
connectivity,
matrices
per
slice
and
assuming
we
want
to
carry
multiple
traffic
types
on
the
same
network
slice
one
connection
can
carry
multiple
traffic
types
or
we
can
create
multiple
connections,
one
per
traffic
type.
C
C
If
the
idea
of
multiple
connectivity
matrices
is
to
allow
for
load,
balancing
or
redundancy
or
something
of
the
like,
then
I
think
it
should
not.
It
should
be
hidden
from
the
service
request.
If
that's
what
my?
What
I'm
seeing
is
you
know
what
the
underlay
can
create
multiple
lsps
to
satisfy
the
same
connection,
that's
possible,
but
it's
still
one
connection.
At
the
end
of
the
end
of
the
day,.
H
G
Consumer
or
customer
whatever
we
want
to
call
that
consumer
facing,
I
think,
while
daniel
was
here,
he
I
think
he
agreed
with
that
point.
Unfortunately,
we
don't
have
him
here
right
now
to
say
he
yes
on
that,
so
you
know,
but
I
do
think
we
in
this
group
at
least
have
agreement
on
that.
That's
that's
an
important
distinction
of
scope
and
but
we
still
clearly
have
a
number
of
topics
that
we
have
to
flush
out
on
the
list.
G
Hopefully,
this
discussion
has
helped
move
that,
along
somewhat,
we
will
take
an
action
to
we,
as
chairs,
will
take
an
action
to
look
back
on
the
discussion
and
send
out
some
summary
points
but,
and
more
importantly,
also
put
it
forward
a
statement
on
how
we
we
think
we
should
take
the
next
step
on
the
adoption
poll,
and
I
think
those
are
the
main
summary
points.
Pavan
did
they
miss
anything.
A
G
But
also
don't
wait
for
us
if
you
have
a
topic
that
you'd
like
to
continue
discussing,
for
example,
igors.
Please
send
that
to
the
list
and
try
to
get
that
discussion
also
point
out
in
the
first
half
we
did
reach,
I
think
some
notable
agreements,
at
least
in
this
forum,
on
on
adrian's
questions.
So
we
will
also
send
that
to
the
list
for
general
agreement
as
pavan
started
us
out
with
we
reach
agreement
on
the
list,
not
in
these
meetings.