►
From YouTube: IETF102-NFVRG-20180718-1520
Description
NFVRG meeting session at IETF102
2018/07/18 1520
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
A
B
B
B
B
B
B
Dissipation
remote
participation,
but
just
in
case
you
have
the
report
and
the
presentations
in
that
link
and
well,
there
is
a
proposal
that
will
discuss
to
specifically
consider
natural
slicing
among
the
group
work
items.
I
know
that
comes
discounts
were
after
some
bhavam
or
neutral
slicing
pieces.
B
B
D
As
charged
you're
really
looking
for
more
trafficking
in
the
list,
especially
for
the
model
driven
optimization,
it
was
a
lot
of
interest
in
discussions.
The
interim
meeting,
but
you're
hoping
there
are
more
discussions
of
the
list,
but
it
quieten
down.
So
we're
really
allowed
to
see
more.
You
know,
discussions
yeah,.
E
It
works
so
since
the
last
idea,
but
what
I
did
was
corrected
minor
typos
thanks
for
the
thanks
for
the
pointers
and
it
was
requested,
I
added
a
section
on
operational
experience,
which
I'm
going
to
be
elaborating
a
bit
of
right
now
in
this
talk,
so
there's
a
based
on
real-life
experience.
We
have
a
project
called
mommy
there.
You
see
the
log
of
that
project
in
the
upper
corner
and
on
that
we
have.
E
E
What
have
I
touched?
Okay,
so
we
have
two
V,
two
components
that
are
based
on
topic,
which
is
a
traffic
generator
and
we're
just
actually
wrapping
on.
On
top
of
our
three,
then
I
have
a
daughter
base
to
feed
data
from
the
measurements
and
that's
based
on
it,
looks
DB
and
then
I
have
a
traffic
sniffer
that
is
based
on
aunty
shark.
E
So
in
the
nice
thing
about
this
environment,
is
that,
depending
on
what
you
want
to
do,
if
you
have
the
Midianites
the
traffic
leontief's,
what
you're
going
to
have
is
add
whatever
you
need
for
you
particular
environment.
So
we
start
with
something
like
this.
You
have
a
client
and
a
server
and
the
network
you're
testing
there,
and
then
it
can
go
to
well,
of
course,
better.
If
you
also
have
the
database
to
collect
your
measurements,
because
afterwards,
really
the
data
from
the
influx
TV
match
very
well
into
anything.
E
Any
kind
of
reboot
you
want
today
can
be
even
better.
You
can
also
mix
in
a
t-shirt
a
Wireshark,
a
traffic
safer,
and
you
can,
instead
of
getting
the
results
from
the
measurements
or
you
get
from
traffic.
You
can
also
go
and
go
directly
into
the
packets
and
analyze
the
packets,
to
see
your
things
and
to
make
things
complete.
You
can
also
have
everything
you
have
for
possible
network
services,
which
are,
if
you
look
at
them,
they
are
incremental
based
on
the
on
the
network
service.
E
E
However,
the
experience
I'm
showing
here
is
that's
much
better
to
have
everything
written
in
with
Nemo.
It's
like
this
question
of
yeah
and
in
the
80s
of
the
last
century.
It
was
still
part
of
the
cool
stuff
to
write
assembler.
But
it's
a
this.
You
started
to
have
other
wonderful,
high
level
programs
to
do
real
programming.
So,
in
this
version
of
the
draft,
one
of
the
things
I'm
showing
is
how
to
use
Nemo
to
build
things
incrementally.
E
So
you
have
something
like
this.
You
know
in
this
case
that
is
the
universe
sniffing.
When
you
put
the
sniffer
and
we've
been
starting
to
see
that
non
infrastructure
support
things
like
are
starting
to
support
things
like
tabs
as
a
service,
for
example
in
OpenStack.
So
you
can,
you
could
actually
extend
even
the
link
model
we
currently
we
have
point-to-point
and
for
some
pay
and
Elon
or
villain.
E
You
can
go
and
add
a
third
link
model
that
implements
this
tapas
of
service
in
order
to
do
this,
instead
of
having
to
write
lots
of
things
and
putting
in
a
hopper
outside
the
network
to
do
the
sniffing.
So
this
is
something
that
is
already
available
in
OpenStack,
so
one
of
the
things
that's
currently
being
part
of
the
work
is
once
the
mono
we
are
using,
which
is
OSM,
has
a
notion
of
this
at
opposite
service,
plug-in
in
in
OpenStack.
We
are
going
to
be
having
the
link
below
the
specific
link
model
that
implements
sniffing.
E
So
conclusion,
it's
a
sixth
round
of
the
draft.
We
have
included
all
the
corrections
from
previous
critical
readings
and
I
want
to
thank
again
here.
All
the
people
that
took
the
time
read
the
previous
versions
of
the
draft
and
sent
comments
included
the
operational
experience
as
requested
in
London.
So
just
a
question
and
request
I
think
we
should
move
this
forward
and
have
it
part
of
adoption
aisle
so
I
would
ask
for
research
group
adoption
of
this
prop
as
it
is
now
and
that's
it.
D
This
is
very
interesting,
especially
the
modeling
aspect,
so
one
thing
you
may
want
to
take
a
look.
There
is
some
work
happening
around
networks
of
East
Nash.
You
know
it's
trying
to
abstract
sort
of
the
reference.
Different
physical
I
mean
basically
hardware
representations.
For
example,
you
have
InfiniBand
or
you
could.
You
know
different
types
of
transport
right
so
rather
than
right
now
the
approaches,
the
exposed
ethical
interference
network
functions
and
you
exactly
they're
all
synonymous
to
the
hardware
implementation.
Instead,
network
service
nash
is
looking
to
abstract
those.
So
here.
C
E
E
E
D
D
F
Okay,
thank
you.
My
name
is
Karla
Bernardo
from
uc3m
and
I'm
gonna
present
some
insights
about
discovery
mechanisms
in
the
HC
and
every
manner
stack.
So
they
are
here
to
basically
present
in
the
problem
and
then
I
will
basically
also
present
a
one
example
of
solution
that
we
have
been
working
on,
but
this
just
for
the
sake
of
presenting
an
example,
MN
ABS
to
present
a
problem
and
see
whether
there
is
interest
in
the
room
for
this
for
working
in
this
kind
of
of
issues.
F
So
this
is
the
outline
of
my
presentation,
so
first
very
brief.
Motivation
of
why
we
need
discovery
mechanisms
and
why
what
we
are
why
what
I
understand
by
discovery
mechanisms
in
the
HC
energy
mono
stack,
then
we
will
basically
try
to
identify
in
the
monastic
where
we
need
this
discovery
mechanisms.
F
There
are
like
three
four
reference
points
there
that
are
identified
by
HC.
That
require
is
kind
of
despair,
American
instance
and
then
I
am
just
publishing
one
of
them.
There
VI
were
exact
in
the
one
that
is
relevant
for
economists
and
I
thought
we
are
a
duty
three,
and
particularly
working
on
and
I
will
present
an
example
of
solution
based
on
ipv6.
But
this
yes,
an
example,
as
I
mentioned
before,
understand
conclusions
next
steps,
so
motivation
I
always
like
to
try
to
use
a
deal
worse
trip
for
this
kind
of
motivation.
F
B
F
Already,
but
it's
also
true
that
today
is
something
that
is
mostly
limited
to
data
center
environment,
so
on
by
data
center
environments.
In
the
scope
of
four
day
in
this
talk,
Bay
before
I
highlight
the
characteristics.
There
is
a
semi
static
environment
where
all
the
entities
that
are
there
are
pre-configured
and
all
the
entities
that
are
involved
already
know
the
other
entities
that
they
will
interact
with.
F
But
this
is
not
the
only
scenario
that
we
we
can
consider
one
before
I
go
into
other
scenarios.
This
is
an
example
of
what
I
mean
by
data
center
environments.
This
is
a
figure
taken
from
one
by
paper
and
is
basically
showing
the
unique
architecture,
a
unique
addition
which
is
from
telefónica,
but
this
would
apply
to
many
other
companies
where
basically
in
the
vision
that
they
have
in
how
to
you
know,
function
bit
recession
in
the
end-to-end
services.
F
C
F
F
But
this
again
is
not
the
only
potential
deployment
environment
for
network
function
virtualization
another
one
is
what
we
call
this
edge
for
dynamic
environment,
basically
is
where
we
put,
or
we
push
the
utilization
closer
to
the
users
in
a
microliter
center
that
are
in
the
premises
of
the
user
or
even
at
mobile
devices.
We
may
think
of
a
car,
or
we
may
even
think
of
a
mobile
phone
as
a
device
that
may
host
visualization
resources
and
even
and
it's
ENFP
kind
of
manual
stack
on
that
and
in
what
we
refer
to
the
phone.
F
And
then
we
have
this
continuum
from
the
folk
at
the
very
edge
of
the
network,
even
at
the
end-user
devices
to
the
centralized
cloud,
the
typical
glad
that
we
are
more
familiar
with.
So
this
is
scenario
we
have
these
resources
on
the
end
devices
these
devices
may
move,
may
appear
and
disappear.
So
it's
volatile
dynamic.
F
In
those
cases,
we
really
need
to
have
mechanisms
to
discover
or
to
advertise
what
are
the
resources
that
are
available,
so
they
involve
entities
can
know
that
they
are
there
that
they
can
interact,
that
they
can
engage
into
a
relationship
and
what
are
the
resources
that
they
offer?
What
others
was
it
a
dark
behind
those
entities?
So
there
is
something
that
has
been
already
identified
by
Etsy
for
some
point
in
in
some
of
the
specs
on
a
knife
at
when
in
effort
on
it,
for
example,
for
multi-domain
marine
I-55.
F
They
also
discover
that,
for
example,
between
an
MPO,
an
Orchestrator,
an
and
a
beam
on
the
resources
these
ratios,
it
is
now
assumed
to
be
pretty
known
pre-established,
but
there
may
be
environments.
Where
do
we
need
to
have
discovery
mechanisms
photos?
So
this
is
one
example
that
we
will
talk
about
more
later.
F
F
So
that's
kind
of
the
motivation
with
a
particular
example
which
is
kind
of
dynamic,
distributed
environments,
the
fog.
This
is
not
the
only
one
and
we'll
talk
a
bit
more
later
about
multi-domain,
which
is
another
environment
or
scenario.
That
means
this
need
for
discovery
functions,
but
then,
let's
take
a
look
at
the
Etsy
NFB
frame
or
architecture
for
an
MV
this
well,
no
one
is
not
the
only
one
that
we
can
think
of,
but
anyway,
what
I
am
gonna
present
here
will
apply
to
other
a
potential
architecture.
F
F
These
resources
may
not
be
always
there
where
the
AVO
is,
for
example,
because
of
this
mobility
stuff.
This
one
is
one
point
where
we
may
need
this
dysphoria
mechanisms
and
other
ponies
between
orchestrators.
If
we
think
of
multi
domain
environments
like
the
ones
address
in
haifa
28,
we
may
have
different
orchestrators
in
both
and
we
may
need
to
discovered
those
and
not
only
that
they
have
the
ability
or
the
presence
of
those
orchestrators.
But,
more
importantly,
what
is
available
behind
those
orchestrators.
So
what
are
they
function?
F
The
resources
that
those
of
pace
raters
manage
so
this
is
another
point:
will
women
need
discovery
functions?
Another
one
is
between
the
vnf
managers
and
their
beams,
so
you
may
have
a
via
net
manager
needs
to
interact
in
operating
in
what
is
called
Eric
mode.
Two
would
be
friendly
with
different
beams,
so
we
may
need
to
discover
what
are
the
beams
that
are
available
and
similarly
to
the
or
bi
to
the
interface
between
the
orchestrator
and
the
beams.
There
is
also
in
the
context
of
life.
F
A
to
Nate
in
the
context,
was
multi-domain
the
interface
between
different
beams
and
then
maybe
also
the
need
for
for
discovering
what
are
the
beams
that
are
available.
So
those
are
the
main
four
points
that
we
consider
as
potential
targets
for
these
qualifications.
So
I
mentioned
before
that,
then
a
misty
fog
environment
is
not
the
only
use
case
for
having
this
kind
of
need
for
the
scoring
mechanisms.
F
Another
one
is
multi
domain,
but
the
main
is
basically
the
idea
that
we
may
have
multiple
initiative
domains
where
we
may
make
them
collaborate
in
order
to
instantiate,
deploy
and
into
an
service.
In
order
to
do
so,
you
need
to
have
interaction
between
the
orchestrators,
this
aura
or
our
interface,
or
between
beams
or
a
combination
of
both
in
these
scenarios.
F
Let's
see
or
identify
that
there
are
two
potential
options:
either
you
have
everything
pre-configured
in
advance,
which
may
fit
in
several
scenarios,
but
certainly
nothing
in
all
of
them
or
you
may
need
to
have
some
Alto
discovery
functions
to
discovery.
What
are
these
orchestrators?
What
are
these
functions
that
are
available
and
what
are
the
resources
that
are
behind
those?
So
this
is
basically
what
we
will
also
focus
in
in
the
context
of
this
talk,
so
we
have
identified
four
different
kind
of
points
where
we
need
discovery
functions.
F
Not
all
of
them
have
the
same
requirements,
because
not
all
of
them
require
the
same
amount
of
information
or
the
same
type
of
information
or
assume
the
same
three
requirements.
So
here
we
list
the
full
of
them.
So
basically,
for
example,
the
you
are
who
are
between
orchestrators
theses,
basically
something
between
different
domains.
So
here
is
important
not
only
discovering
those,
but
mainly
once
you
have
those
discovery
discover.
So
they
know
that
there
is
an
Orchestrator.
There
are
50
total.
F
What
are
the
resources
and
the
functions
that
are
behind
those
orchestrators
are
controlled
by
those
Orchestrator.
So
if
I'ma
and
mr.
dip
domain
and
I
want
to
deploy
something
that
I
cannot
deploy
my
own
infrastructure
because,
for
example,
I,
don't
have
the
graphic,
a
footprint
that
is
required
for
that
service.
F
F
As
I
mentioned
I'm
a
fan,
Orchestrator
I
may
have
several
resources,
but
some
point
in
time.
Additional
resources
may
appear
or
disappear.
So
I
need
to
be
able
to
discover
that
that
happen,
and
also
if
new
resources
appear
I
may
need
to
remain
want
to
know
what
are
those
resources
that
appear
and
in
in
case
I'm
interested
in
inducing
them
in
registering
with
them.
To
do
so
so
I
mean
it's
not
only
about
the
scoring
that
they
are
there,
but
also
discovering
what
is
available
with
is
provided
by
those
bibs.
F
We
will
see
a
bit
more
in
the
next
slides
and
then
the
Vav
is
similar
to
the
RBI
and
basically
the
needs
for
that
is
in
the
context
of
multi-domain.
When
you
have
been
implementing
what
is
called
the
single
logical
point
of
contact.
Basically,
you
have
been
interacting
with
multiple
babies
and
traded,
a
single
interface
to
be
to
whatever
is
above
the
nav
you
do
you
need
that
beam
to
be
able
to
discover
the
other
beams
so
basically
similar
to
the
RBI
kind
of
requirements,
but
in
the
context
of
between
different
bibs.
F
So
this,
basically
generally
where
we,
where
do,
we
need
discovery
functions,
and
why
do
we
need?
Those
is
Cori
functions?
So
now,
let's
focus
on
one
particular
thing:
this
is
the
RBI
and
but
a
sample
of
solution.
Ok,
again,
is
just
an
example,
so
the
generic
case
scenario
potential
example
of
the
scenario
that
we
have
in
mind.
So
we
have
on
any
big
platform,
basically
like
a
kind
of
data
center
that
is
available.
F
There
is
what
we
have
we
are
used
to
and
in
the
risks
kind
of
a
static,
and
then
we
have
a
resource
that
is
mobile
or
a
device
that
is
mobile.
Maybe
a
car,
maybe
something
that
that,
since
its
mobile,
it
may
appear
disappear
in
the
same
network
where
we
have
the
cysteine
data,
centers,
cysteine
and
every
platform,
and
this
device
here
may
have
available
resources
for
other
to
use.
And
how
do
we
make
that
happen
today?
We,
we
cannot
do
this
with
the
system
standards,
because
this
interface
is
something
is
pre-configure.
F
The
this
reference
point
is
used
to
attach
it's
used
to
query
what
is
available,
but
before
doing
that,
you
need
to
have
that
relationship
in
place.
So
in
order
to
extend
this
for
a
more
dynamic
scenario,
we
need
mechanisms
that
allow
this
an
AVL
to
know
that
this
beam
is
available
here,
that
there
are
some
resources
that
it
can
provide
and
how
these
nabo
can
actually
get
in
touch
or
contact
with
this
beam
and
register
a
bit
more
detail
on
examples
of
use
of
this.
F
So
this
is
basically
what
I
mentioned
before
we
have
a
platform:
a
system
platform
here,
where
we
have
resource
services
deployed
that
are
consumed
by
somebody.
Our
service
at
the
end
is
consumed
by
by
a
user
identity,
and
this
platform
may
be
extended
by
using
resources
from
other
devices
that
are
male.
So
that's
one
particular
example
of
applicability
of
this.
F
So
basically,
what
we
have
here
is
just
an
example:
I
will
not
go
into
the
details,
but
is
basically
using
a
go
discovery
to
advertise.
The
availability
of
this
resource
so
in
case
that
the
nab
other
castrated,
a
system
platform
really
wants
to
use
those
resources.
It
can
then
talk
to
the
whim
and
register.
So
basically,
here
is
where
we
embed,
extend
and
nd
signaling
to
advertise
those
resources
and
also
the
identity
of
the
beam.
And
then
here
we
can
do
some
registration
of
the
between
the
Navy
and
the
wind.
F
F
This
is
an
example,
and
what
I
worry
more
is
more
interested
in
is
to
know
whether
you
think
that
this
kind
of
mechanisms
for
auto
discovery
functions
in
general
in
the
Navy
monastic
are
useful.
I
tried
to
to
justify
that
I
required,
not
only
because
of
fog,
dynamic
environments,
but
also
in
in
data
center
environments
because
of
the
multi
domain
aspects
at
Siena.
We
already
in
defy
this,
but
I'm,
not
a
worry.
G
B
F
By
some
rogue
one,
some
element
that's
going
to
try
and
find
out
what
kind
of
resources
exist
in
already
systems.
Okay.
So
thanks
for
the
question
the
first
one
at
this
point,
we
are
kind
of
trying
to
reduce
the
data
model
that
sits
in
its
energy.
So
we're
trying
to
use
the
same
thing
that
for
somebody
in
the
RPI
there
are
some
there
is
something
define
I
mean
if
I
remember,
but
there
is
for
this
reference
point
when
you
clear
query
the
resources
that
are
available,
there
are
some
models
available
there.
A
F
Mean
for
sample
something
that
we
need
is
this
thing
I
mean
this
kind
of
thing
is
basically
taken
from
a
cysteine
and
every
documents,
but
the
the
point
of
conveying
the
the
URI
to
get
to
the
beam
discontinue,
because
in
in
let's
see
this
is
assumed.
That
is
know
in
advance
and
then
on
the
second
point
on
the
security
aspects,
there's
a
definitely
very
valid
point
in
the
context
of
multi-domain.
We
have
already
some
work
done
in
different
projects
on
what
is
required
in
the
context
of
this,
but
but
in
this
multi
domain
stuff.
F
Typically,
you
have
some
pre
relationship
between
the
actors.
The
stake
holders,
so
you
can
rely
on
that
in
the
in
the
context
of
the
fog
environment
is
something
that
we
need
to
see
because,
of
course
you
need
something
you
cannot
do
whatever
you
cannot
expose
anything
to
anyone.
You
can
play
with
different
levels
of
granularity.
You
can
play
with
exposing
something
very
simple
and
then,
depending
on
the
relationship
you
have
with
the
other
one
provide
more
detail.
So
these
kind
of
things
are
details
that
need
to
be
work
out.
This
increments.
H
Neurons
have
a
Leon,
okay,
I
have
two
questions.
One
is
you
consider
discovery,
and
you
also
think
about
the
other
way
around
I
mean
to
have
self
advertisement
of
variable
resources
or
capabilities
so
that
you
can
other,
because
if
you
do
discover
it's
an
active
process,
you
want
to
discover
something.
So
it
can
work
in
some
situations,
but
you
can
also
have
the
reverse
situation
and
you
have
subsidizing.
So
you
know
that
some
resources
will
be
available
when
you
want
to
compose
them.
So
just
yeah.
F
That's
actually
probably
a
misuse.
The
term
discovery,
because
you
actually
in
the
in
the
draft
will
cover
both,
but
it's
basically
based
on
the
advertisement,
so
we
define
some
mechanisms
for
advertise
so
do
attach
to
the
network
and
your
birthday
resources,
okay,
and
then
we
also
support
somebody
big
question:
the
other
system
devices
to
advertise
that
will
be
like
kind
of
the
discovery
part
like
okay
and
there
you
have
something
to
tell
me
tell
me,
but
is
everything
based
on
the
advertisement?
Okay,.
H
Thank
you
and
second
point
is
I,
see
not
so
in
your
presentation
that
you
consider
different
requirements
of
discovery
for
different
interfaces
or
different
situation.
Again.
I
was
also
in
different
working
group
or
research
group
that
there
is
this.
We
still
come
back
to
this
need
of
discovering
new
capability
in
the
network,
and
you
have
a
proposal
here
to
piggyback
on
IP
visit,
but
II
also
considered
considering
it
kind
of
generic
approach
to
discovery
where
you
have
want
to
discover
capabilities,
and
then
you
have
a
model
to
say
that
kind
of
cut.
H
What
I'm
looking
for
what
the
network
or
the
company
can
advertisement
because
I
see
the
need
that
in
some
other
working
group
we
also
designing
new
discovery
mechanism
of
protocols.
So
if
you
have
one
for
nav
and
one
for
anima
and
one
for
others,
so
maybe
there
are
some
common
needs
and
having
a
more
generic
design,
we
can
I
mean
profile
some
of
the
capabilities
or
dynamically.
You
insert
what
you
are
looking
for.
Yeah.
F
That's
a
good
point.
Definitely
we
need
to
look
into
that
this.
This
specific
solution
came
from
one,
a
scenario
that
we
are
working
on,
where
it
fits
very
well,
but
it
is
I
fully
agree
that
it's
good
to
look
into
other
options
and
try
to
I
mean
have
one
single
solution
with
different
profiles
instead
of
having
like
multiple
approaches.
Thank
you
very
much.
D
Very
interesting
topic,
in
fact,
of
investigating
this
and
wound
up
in
the
context
of
edge.
So
what
we
realized
was,
as
you
walk
to
the
use
case,
especially
the
mobile
limbs.
I
mean
as
soon
as
critical
as
the
network
functions
are
the
applications,
the
one
which
actually
drive
it
so
sort
of
saying
that
hey
I
need
air.
We
are
like
the
low
latency
in
the
edge
and
then
I
need
this
need
to
capitalize
on
mobile
infrastructure
right.
So
that
said,
what
gets
interesting
is
so
how
do
you
manage
the
separation
of
concern?
D
So,
for
example,
especially,
you
know
you
have
a
constrained
liberal
source.
You
know
few
servers,
then
do
you
really
partition
the
resources
between
applications
and
network
functions?
Or
how
do
you
manage
them?
You
know
and
also
the
who's
going
to
manage
the
service
like?
Is
the
application
be
delivered
by
a
different
provider,
or
is
it
a
load
by
the
same
operator
who
is
also
dealing
in
the
network
savvy
some
interesting
aspects
to
think
about?
But
it's
a
very
interesting
topic.
Yeah
thanks.
F
I
am
NOT
here
we
just
focus
on
the
the
advertisement
and
discovery,
but
it's
this
is
something
that
is
to
be
I
mean
for
the
actual
services
to
be
employed.
You
need
to
have
that
gonna
be
the
one
actually
deploying
so
who
is
in
charge
so,
for
example,
what
is
the
entry
point
for
a
service
to
be
deployed,
in
this
case
the
different
actors
that
are
involved?
F
D
B
D
I
Also
just
remind
you
around
a
year
ago,
at
in
in
Chicago
at
the
ITF
I
made
a
big
presentation
and
also
a
lot
of
discussion,
vo
options.
What
are
how
you
can
fit
in
network
slicing
in
an
IDF
concept
since
then,
Etsy
had
issues
three
separate
reports,
always
saying
that
this
is
feasible
and
important,
but
also
currently,
all
the
other
SDL's
are
issuing
reports
on
on
this
topic.
I
Recently,
I
made
a
substantial
effort
to
bring
all
the
information
available
together
with
a
V
of
network
slicing
in
a
form
of
a
tutorial
which
was
presented
two
weeks
ago
in
Montreal
at
a
top
conference
on
networks,
authorization
and
the
topics
are
there
and
out
of
this
I
felt
that
maybe
it's
a
good
idea
to
present
to
you
a
good
shopping
list
of
potential
problems
to
be
solved,
not
the
ones
which
are
potentially
sorted
out.
That's
the
scope
of
my
presentation.
I
The
center
of
attention
is
a
way
of
networking,
and
in
that
sense
a
network
slice
has
to
be
like
any
type
of
network
self-contained
isolated,
because
I
talking
out
many
parallel
network
slicing,
it
has
to
be
programmed
in
a
sense
that
you
cannot
just
leave
behind
of
hewing
and
some
extent.
This
concept
is
orthogonal
to
the
infrastructure
architecture
which
exists
today
and
does
not
depend
on
a
particular
technology
to
they
wear
it
as
a
data
plane.
The
partitioning
which
is
only
one
part
of
the
network
slicing
this
issue
in
mind.
I
It's
obvious
that
there
is
no
point
to
think
in
terms
of
building
new
or
deploying
new
services
by
creating
an
infrastructure
all
the
time.
This
is
inefficient
and
extremely
expensive
and
is
going
to
change
substantially,
potentially
due
to
the
advent
of
metric
slicing
impact
in
the
industry.
In
my
view
now
there
are
many
ways
to
look
at
the
where
this
sort
of
concept
could
be
used,
and
this
is
just
a
shopping
list
of
potential
massive
user
scenarios.
I
I
will
point
out
only
three
of
them:
the
emerging
high
precision
networks,
which
need
to
be
created,
bearing
in
mind
that
IP
is
not
deterministic
and
also
does
not
guarantee
anything.
So
the
risk
Linnaean
need
to
understand
what
to
change
in
network
functions
and
grouping
them
and
how
to
bring
them
together
with
resource
partitioning
trust
services,
in
order
to
have
specific
guarantees
for
quality
of
service
or
KPIs,
and
you
cannot
date
for
all
the
KPIs
in
the
world
or
all
the
types
of
partition,
all
the
issue.
I
Ethic
quality
of
service,
but
needs
to
be
done
per
groups
ie
per
twelve
slice,
and
this
is
a
key
issue,
for
example,
in
defining
very
low
latency
type
of
network,
and
this
becoming
an
important
topic.
I,
don't
know
how
you
can
do
it
without
looking
at
slices
overall
means
that
to
redo
the
whole
internet
and
to
end.
But
there
are
many
other
opportunity
to
look
at
the
way
in
which
slice
it
could
be
adding
value
in
in
user
scenarios.
I
Now
dimension
is
already
Etsy
has
issued
different
three
documents,
actually
mainly
as
a
whitepaper
level
about
how
metric
slicing
can
fit
in,
and
this
is
a
diagram
which
basically
shows
that
this.
How
this
may
happen
is
a
way
of
grouping
network
functions
in
addition
to
resources
with
services
in
a
context
on
V,
and
this
is
an
opportunity
to
move
forward
in
a
big
way.
C
I
Be
dissolved,
eating
from
outside
includes
around
20,
some
are
listed
here
and
in
the
area
of
multi
domain
nav
network
slicing
and
there
are
different
granularity,
but
also
some
related
to
other
elements
which
need
to
be
built
and
in
order
to
ensure
that
the
capabilities
of
network
slicing
in
this
context
are
realized.
Some
related
to
the
the
winning
services
and
more
need
to
be
mapped
into
system
concept
are,
very
importantly,
look
at
the
issues
related
to
guarantees
slices
with
guarantees,
other
quality
of
service
of
API,
Gertie's
or
scalability.
We're
talking
about
paralyzing.
I
Not
in
two
or
three
but
many
tens
of
thousands
of
dynamic
quite
some
years
ago,
nav
VPN
was
great.
If
now
everybody's
using
it
it's
a
time
to
sink
in
systems
as
well,
so
that
the
question
put
forward
is:
are
these
problems?
Maybe
all
the
reporters
could
be
resolved
partially,
not
all
of
them,
but
some
in
this,
what
you
group
and
your
feedback
now,
oh
good,
either
one
viewpoint
which
says
that
people
are
moving
away
from
NAV.
I
Towards
nad
centered
around
network
slices,
which
is
at
the
center,
this
is
a
unit
of
networking
which
need
to
be
highlighted
for
different
type
of
services
are
many
around
talking
network
services
and
they
have
to
be
supported
by
some
sort
of
lightweight
smart
fabric
voilá
coordinator
yep,
because
this
diagram
shows
a
slightly
different
view
of
when
AV
architecture
shown
centered
around
slices,
which
are
becoming
the
unit
or
could
become
the
center
and
a
unit
of
any
visa.
For
maybe
this
will
be
thinner,
some
sorts.
I
C
J
Reza,
yes,
if
you
go
back
to
a
slide
number
four,
there
are
a
couple
of
thing
that
you
know
in
the
context
of
network
is
slicing.
First
thing
is
network
is
slicing
is
an
end-to-end
concept.
When
we
are
talking
about
end-to-end
everything
that
we
discussed,
it
should
be
starting
from
Yui
all
the
way
to
application
anything.
C
J
Is
not
really
into
it
and
in
that
context,
if
you
want
to
be
pragmatic
in
the
picture
that
you
put
here,
for
example,
is
very
simplistic.
We
have
to
consider
distributed,
run
crowd,
rants
the
on
the
EPC
side.
You
know
4G
5g,
and
if
you
see
this
one
from
end
to
end,
not
all
the
components
are
vnf.
There
is
for
sure
there
are
P
n
FS
v
NS
and
depends
how
you
want
deployed.
J
I
I
I
I
J
J
Know
I
understand,
but
the
one
thing
that
you
mentioned
and
I
really
agree
with
that
net
forget
so
I
think
5g
network
is
slicing
is
not
VPN.
It's
not
LSP
is
not
Hana,
isn't
much
more
than
that,
and
this
is
whatever
you
imagine,
and
this
like
and
one
last
thing:
I
don't
want
to
waste
time
here
and
it's
life.
I
have
you
mentioned
about
tenancy,
and
this
is
something
that
I
in
my
opinions
overlooked.
J
The
multi-tenancy
that
we
are
talking
about
5g
is
different
from
hierarchical,
tenancy
or
aka
nest,
the
tenancy
that
we
have
in
industrial.
This
is
something
that
I
the
you
know.
We
can,
you
know,
talk
pair
off
line
and
that
one,
but
this
is
something
that
I
want
to
make
sure
that
this
is
address
in
one
way
or
another,
because
some
in
even
some
context,
people
talk
about
network
a
slicing.
Whatever
they
talk
about
is
really
hierarchical,
tenancy.
They,
you
know
they
did
already
use
this
word.
You
know
the
the
in
the
different
context.
I
In
this
presentation
are
plenty
of
other
information
and
other
20
slides
which
are
covering,
but
at
least
anyway,
this
presentation
was
prepared
by
Leon
China
Mobile,
but
together
with
the
other
people,
myself
included,
but
he
couldn't
come
to
deliver
it,
and
this
focuses
this
presentation
focuses
on
the
issue
on
architectural
elements
of
network
slicing
management,
which
need
to
be
looked
at
in
particular.
Here.
I
Just
to
recap,
the
same
concept,
then,
is
that
the
network
slice
is
a
set
of
infrastructure
resources
and
service
functions,
very
importantly,
which
need
to
be
put
together
in
order
to
serve
needs
of
a
big
industry.
As
long
as
a
concept
is
quite
far
reaching
now,
slicing
is
a
mekinese
about
creating
such
entities
very
easily
across
a
data
plan,
control
plan
management
and
service
plan
together
and
in
that
sense,
we'll
have
other
players
need
many
slice
provider.
C
I
Tanner
ones,
which
could
be
another
operator
or
another
operator
or
another
company
which
takes
over
a
slice
or
slice
number
of
slices
and
add
their
own
facility
to
it.
So
it
could
be
managed
by
the
initial
providers
facility,
but
also
they
could
be
managed
if
you
want
and
operated
use,
additional
tools
and
techniques
of
the
tenant.
So
this
is
quite
an
important
demarcation.
Now.
I
Did
this
slide
always
saying
that
this
mechanism
is
quite
absolutely
essential
and,
to
some
extent,
is
more
important
than
the
way
in
which
has
a
data
plane?
You
activate
partitions,
because
the
end-to-end
issue,
as
it
was
mentioned
by
Raisa,
cannot
be
resolved
per
leg
of
the
component
of
the
network,
which
is
a
radio
access
access,
core
edge
clouds,
edge
clouds
on
their
own,
let
alone
many
technologies.
Many
think
there
has
to
be.
The
mekinese
has
to
be
looked
at
the
global
things
in
order
to
ensure
the
usability
of
this
concept
at
large.
I
So
together.
This
is
like
a
architecture,
the
option
to
political
points
which
need
to
be
resolved.
They
are
not
resolved
yet.
You
know
end
to
end,
but
it
could
be,
but
the
Reza
my
desk,
around
10
or
20,
separate
architectural
networks,
I
see
in
the
research
literature,
research
projects
and
as
the
other
fields
too
much
of
it
and
everybody's
overruling.
Clearly,
it's
a
time
to
look
at
the
granularity
which
will
have
that
in
the
context
of
being
pragmatic,
I
don't
see
that
a
long
time,
but
this
architecture
more.
I
I
I
J
Razor
away
from
Nokia,
in
my
opinion,
there
is
another
model
missing
here-
is
from
the
orchestrator
to
the
network
controller,
and
this
is
that
connectivity
API,
that
it
is
something
that
again
it's
missing
here.
We
can,
you
know,
talk
offline
as
well.
So
in
my
opinion,
that
is
very
important
aspect
from
NSO
that
you
put
here
to
the
network
controller
or
network
model.
That
interface
is
important.
Why
it's
important?
Because
it
is
not
there
already.
J
You
have
but
the
network
resources
today
everything
votes
based
on
service
based,
you
say,
I
want
to
create
an
VPN
from
here
to
here
and
all
the
say
that
connectivity
is
one
level
abstraction
that
then
interface
that
I
am
referring
to.
Is
a
connectivity
API
again
to
be
pragmatic?
That
connection
says
I
want
to
connect
a
node
B
to
ask
a
tree.
I
wanna
connect
a
node
B,
you
appear,
it
doesn't
tell
you
which
service
to
use
and
how
to
do
it.
That
layer.
J
A
D
When
requests
besides
offline,
they
could
bring
their
discussions
in
the
group
will
be
much
appreciated
and
also
another
thing
to
think
about.
I
think
the
presentation
and
mobile
bins
and
the
discovery
was
very
interesting
and
I
do
think
it's
very
relevant
to
network
slicing,
because
today,
if
you
look
at
slicing
still
looking
at
more
of
static
static
aspects,
it's
basically
assuming
it
closed
the
static,
we're
trying
to
connect
them
pray
but
versus
the
real
problem
is
then
this
was
a
more
dynamic
raid
and
then
then,
how
do
you
create
a
slice
in
the
dynamic
environment.
I
I
I
I
I
K
K
Okay,
well,
the
context.
The
context
is
quite
straightforward:
we
know
that
current
networks
are
evolving
to
towards
deeper
bit
to
essentially
schemes.
We
are
here
for
that
now
in
a
piece
directly
related
to
that.
However,
a
slice
in
physical
and
mental
resources
in
a
bit
is
essential,
as
you
know,
force
protein
exploiting
the
the
capabilities.
K
In
this
sense,
we
are
with
a
slice
in
that
plasticity
and
composability
provide
benefits
that
must
be
incorporated
also
to
any
system,
especially
in
the
future,
because
for
commercial
and
engineering
organizational
applications
of
energy
systems
processes,
the
end
composability
are
a
must
and
of
course,
we
have
to
consider
also
ubiquity
in
vnf
hosting
services
like
edge
of
fork,
computing
that
add
additional
flexibility
to
the
whole
system.
So
an
in
general,
apart
from
from
this
week,
we
can
ask
for
improvised
distillation
stability
under
civility,
as
we
have.
K
Okay,
so
I
will
be
brief
from
these.
The
technical
viewpoints
that
we
find
in
this
in
this
scheme
are
two
first
would
be
from
provided
to
consumer
that
as
debut
that
providers
have
about
the
slicing
mechanism
and
how
they
provide
this
slice
into
the
consumer.
So
in
this
sense,
for
sambal
NAB
providers
I'm
to
maximize
the
exploitation
of
their
resources,
and
so
they
is
lies,
no,
they
restructure
to
processes
to
get
the
largest
set
of
smaller
resources.
K
Okay,
in
this
way,
both
providers
at
consumer
are
concentrated
on
their
own
business
on
the
same
moment,
in
the
same
time,
free
from
consumer
to
provide
that
debut
is
similar,
but
in
this
case
then
a
big
slice
consumer
or
like
we
can
say
here,
then
that
of
coming
from
operator
I
will
talk
about
this
later
I'm
to
meet
the
requirements
or
the
network's
said
by
the
applications.
We
provided
that
the
final
users
reduce
the
cost
of
ownership,
a
be
able
to
adapt
the
network's,
etc
and
everything
as
fast
as
possible,
of
course.
K
So
in
this,
in
this
sense,
what
they
do
is
to
obtain
every
slice
of
resources,
so
they
have
their
main
business
here.
The
main
objective
here
is
concentrating
only
on
those
slice
of
Russia's
is
not
how
they
are
a
split
or
how
they
are
built
on
their
underline
that
sparked
from
the
provider.
So
these
two
viewpoints
have
to
be
in
some
sense
considered
when
slicing
NAB
systems
well
and
here
gives
place.
The
rolls
I
have
mentioned
that
I
did
Utley
in
the
divot
from
these
viewpoints.
So
I
have
I
had
talked
about
the
nippy
provider.
K
It
is
the
owner
of
the
physical
orbital
resources,
I
include
also
built,
or
because
we
can't
have
some
sense
of
relativity
here,
some
provided
that
that
leases
one
physical
verses
and
provides
physical,
built
or
resources
to
other
operators.
So,
as
I
have
mentioned,
this
provided
slice
is
the
infrastructure
to
get
a
smaller
portions.
K
First,
topology
changes
and
dynamic
dynamical
specific
parameters;
okay,
so
it
in
interacts
with
the
consumers
that
I
have
mentioned
that
enemy
operators
to
execute
they
request
mainly
create,
read,
update
and
delete,
so
the
main
concepts
are
managing
the
physical
network
and
addressing
his
life
requirement
and
improving
ratios
efficiency
and
efficacy
in
the
underlying
system.
In
parallel,
the
energy
operator
Manas
the
slices
of
BNF
that
he
has
requested
on
obtaining
from
one
or
D
for
several
being
BNF
providers,
an
AP
provider
in.
K
Implemented
so
we
can
have
a
he
here,
our
CEO
of
NFB
o
provided
all
providers
so
well.
At
the
end,
we
have
the
application
series
provided,
that
is
which
establish
requirements
to
the
nabo,
and
then
we
we
can.
As
I
have
mentioned,
we
can
have
different
variation
of
this
instantly
theorem
the
network
program,
every
all
roles
can
be
played
by
the
same
entity
or
a
separate
entity
as
many
combination
of
them.
So
at
the
end
we
have
these
and
those
these
are
the
roles
and
these
in
wing.
We
see
the
interfaces
that
are.
K
Are
identified
once
the
roles
are
interconnected
one
to
each
over,
as
we
can
see
here
on
the
right,
we
have
some
that
they
the
way
passivity
I
have
mentioned.
We
can
see
that
in
this.
In
this
part,
sorry
in
this
part
and
then
none
more
more
classic
approach
in
this
side
with
a
provider
operator
and
the
under
plication
service
provider
will
be
their
final
consumer
fees,
so
the
interfaces
are
listed
here
and
that's
all
from
my
side.
So
if
you
have
any
questions,
I
hope
so.
A
B
K
L
My
name
is
Homer
from
entity
laboratories
in
this
presentation.
I
will
talk
about
the
concept
of
getaway
function,
format
for
pricing,
so,
firstly,
I
would
talk
about
the
motivation
problem
that
was
rising.
The
network's
rights
are
trying
to
get
away.
The
stress
get
away
is
a
different
function
to
handle
traffic,
for
that
was
rising.
So
after
you
know,
in
networks,
rising
operation
and
manage
orchestration
is
built
importance.
Meanwhile,
our
little
brain
functionality
for
the
Box
pricing
is
also
very
important
and
thrust
gateway.
Internet
draft
a
will.
L
L
So
this
show's
over
be
open.
It
was
rising.
That
was
rice
is
a
logical
network
deployed
based
on
service
requirements
for
examples
of
how
connected
people's
rule
agency
network
is
created
and
such
because
connected
to
the
network,
and
we
expect
that
was
rice
would
enable
subpart,
is
to
use
network
at
work
associate
as
a
part
of
it.
Their
own
subchief.
L
Server,
resources
and
network
choices
will
be
created
dynamically,
and
so
the
processes
in
lead
box
will
be
very
complex
and
we
would
liked
to
automate
these
processes.
So
this
figure
shows
frame,
love
framework
of
networks,
pricing
considered
in
cons
companies
both
held
in
the
last
IDF,
so
network
drives
Orchestrator
received.
L
Subsea
requirement
from
tenant
via
CSI
and
SDI
and
combat
the
requirement
to
the
complete
conflation
and
indicate
to
the
underlay
properly.
So
if
a
via
each
controller,
internal
network
trustees
will
be
deployed
dynamically
and
thus
networks
prices
would
be,
we
arrived
by
using
Sdn
and
NFV
technologists,
so
this
figure
shows
over
the
all
networks,
rising
winds
in
every
architecture.
So,
if
and
if
we
are
ticked,
we
use
in
every
architectures,
so
some
introduction
between
rice,
Orchestrator
and
a
network
anyway,
ok,
ok,
splitter,
will
be
required.
So.
L
L
So
in
and
2
and
networks
rising,
there
are
several
challenges,
so,
for
example,
which
information
will
be
required
to
provide
controllability
or
for
to
the
tenant
or
how
to
synchronize
the
domains
and
how
to
operate
in
each
subnet
and
its
data
frame
there.
There
are
also
several
challenges
so,
for
example,
how
to
detect
traffic
and
follow
the
traffic
to
the
appropriate
network
license
or
how
to
interconnect
between
the
subnet
and
how
to
keep
isolation,
doing
amongst
choices
to
keep
something
starts
40
so,
for
we
saw
these
Oh
being
such
a
challenges
on
data
plane.
L
L
L
H
I
have
a
bit
of
a
provocative
question
and
on
spitting
on
tomorrow,
there
will
be
also
presentation
on
network
slicing
in
the
network
management,
research,
whoo
yeah,
so
I
think
I,
like
the
presentation
that
we're
done
today
to
give
a
different
of
every
different
aspect
that
need
to
be
researched
in
an
aviary
or
elsewhere
in
ITF
rhf.
Why
they
don't
really
get
is,
if
you
want
to
have
really
a
discussion
on
this
topic
is:
do
we
want
to
have
research
aspect
investigated
in
energy
or
energy
on
networks,
Dyson.
C
H
Is
the
topic
of
today
and
tomorrow,
or
is
there
also
the
attempt
to
still
create
one
working
group
in
IETF
for
this
aspect?
Do
we
want
to
have
both
and
I
just
want
to
be
a
little
bit
more
anticipate
a
bit
on
the
discussion
and
what
we
could
conclude
on
those
things,
because
tomorrow
there
will
be
a
session
I,
don't
know
if
you
have
time
to
discuss
that
sort
of
to
know
bit
more
what
people
are
willing?
What
are
we,
what
can
really
go
into
the
research
side?
What
can
really
go
into
the
engineering
side?
L
I
Just
what
some
comment
about
this
question?
There
are
a
number
of
research
elements
which
probably
will
take
ten
years
to
resolve
related
to
metal
slicing,
and
these
research
questions
the
best
ways
to
handle
it
in
RTF,
some
specific
to
management,
some
specific
to
other
aspects
in
in
particular,
Navy,
which,
in
my
view,.
I
Out
of
network
functions
in
tools
into
slice
functions.
That's
my
reading
of
the
trend.
On
the
other
hand,
as
you
can
mention,
and
as
mentioned
as
okay,
this
concept
of
slicing
has
all
the
elements
related
to
the
next
level
of
networking,
which
means
that
there
is
an
opportunity
also
in
the
IDF,
to
dedicate
a
set
of
specific
questions
to
be
addressed
in
a
group
rather
than
to
spread
everything
into
pieces.
As
it
is
happened
now
and,
in
my
view,
that's
a
waste
of
energy
and
will
not
converge
ever
just.
I
It
but
the
industry
doesn't
but
clearly
networks
license
doesn't
depend
on
segmenting.
It's
about
the
use
of
services
of
the
infrastructure
underneath
in
the
combination
so
to
some
extent,
to
answer
the
question
in
a
deal
world,
we
should
have
a
research
set
of
question
in
our
ia
RTF,
but
also
a
dedicated
impossible
group.
I
Nowhere
already
I
have
problems
about
well,
what
are
these
drugs
appearing
to
do?
Thousands
of
drops
which
are
not
put
in
the
industry
there
it's
routine!
Well,
it's
a
big
question
in
my
mind
anyway,
I'm
just
expressing
that
in
idea
what
this
is
I
will
do,
but
we
have
to
be
pragmatic
in
our
next
period
how
to
do
to
move
forward.
D
Better
I
mean
respective
resource
challenge.
This
is
it
sort
of
solving
building
drop
problems,
for
example,
you
have
input
from
different
sources.
Now
you
have
you
optimized
rate,
for
example,
placement
of
resources
said
because
that's
one
of
the
key
problems
you're
trying
to
solve
saying
it's
a
guaranteed
slice.
That
means
you
shall
be
able
to
deliver
the
test
layer
right
so
is
that
more
of
a
resource
challenge
versus
the
interfaces
then
sense
that
perhaps
you
know
being
standardized
elsewhere,
I
mean
not
as
part
of
research.
D
J
Makia
one
thing
that
to
be
a
game
pragmatic
in
my
opinion,
the
first
thing
that
we
have
to
do:
we
have
to
define
the
API
API
between
the
API
that
was
missing
in
that
picture.
The
system
is
very
important.
You
know,
as
we
speak,
we
are
trying
to
implement
that
when
we
can
talk
about
that
as
well.
This
is
something
that,
in
my
opinion,
why
the
previous
part
wasn't
that
successful.
J
This
is
my
personal
view
is
because
we
didn't
define
exactly
what
are
missing
parts
and
what
we
want
to
do,
for
example,
to
define
that
API,
which
was
missing.
You
need
a
young
model
for
the
connection
connectivity
API.
We
define
that
and
that
one
becomes
specifically
in
5g
in
5g.
You
know
the
connectivity
is
a
bit
different
from
normal
point-to-point.
Multi
point-to-point
is
related
to
that.
So
this
is
one
practical
aspect
that
we
have
to
address:
that
the
second
one
is
the
multi-tenancy
versus
hierarchy
or
Nestor
Chauncey.
J
This
is
something
that
we
have
to
do
in
my
opinion,
again,
that
to
model
that
you
put
there
that
from
10
and
to
Orchestrator
and
from
the
orchestrator
to
the
controller
that
to
model
they
are
important
by
the.
In
my
opinion,
they
are
not
that
important,
especially
the
one
from
Orchestra
from
the
portal
from
customer
came
and
in
this
context,
is
a
customer
to
BMW.
You
know
the
City
of
Ottawa
being
owned
or
whatever.
J
I
C
I
Remark
is
that
the
world
of
networking
is
moving
towards
centering
on
slices
as
a
way
to
deliver
network
services.
If
that's
the
case,
then
you
have
to
readjust
a
bit
in
our
thinking
too,
and
rather
than
to
look
at
some
piecemeal
silo
issues
related
to
this
topic,
which
is
much
broader
in
terms
of
usage,
I
suspect
that,
40
years
ago,
when,
when
PPM's
were
invented,
there
were
similar
arguments.
I
Do
you
need
it
or
not,
but
now
it
nobody
cares
it's
using
it
before
we're
at
the
stage
were
to
use
it
and
I
also
add
two
key
research
aspects
of
this
topic,
which
is
scalability
large
number
and
also
slices
with
guarantees
which
are
going
to
be
the
norm
in
the
future.
These
are
two
subjects
which
may
refurbish
a
number
of
architecture:
substantial,
in
addition
to
all
the
others
which
are
I.
B
Don't
want
to
enter
in
an
argument
right
now
here,
but
let
me
remark:
these
groupies
focus
on
anything
if
I
mean
if
we
are
going
to
change
the
scope
please-
and
this
is
a
plea
for
for
opening
this
discussion
on
the
list-
please
think
about
what
the
means
said
and
discuss
here
and
if
you
can
identify
concrete
research
challenges
or
concrete
research
activities
to
be
reported
here
that
are
related
with
any
of
your
cost
ratio.
Network
attacks
are
derived
from
the
application
of
anything
performance
of
beautiful
eyes
or
mixed
environments.
B
I,
don't
care
whether
this
is
a
mix.
That's
our
our
matter
as
well,
when
you're
talking
about
a
mix
or
whatever
genius
deployment
of
physical
and
virtual
functions.
All
of
this
is
in
our
remit
and
whatever
you
feel
like
coming
with,
it
would
be
welcome
and
we
are
open
to
discuss
it.
This
said,
just
let
me
close
the
meeting.
Thank
you
so
much
see
you
next
time.