►
From YouTube: NMRG Interim Meeting, 2021-05-17
Description
NMRG Interim Meeting, 2021-05-17
A
A
It's
important
that
you
have
a
look
that
research
group
from
the
irtf.
You
follow
the
ietf
intellectual
property
rights,
disclosure
rules
and
in
previously,
if
you
look
at
the
slide,
it
means
that
if
you
present
somewhere
related
to
a
patent
or
something
that
would
be
that
you
started
to
patent,
it's
important
to
disclose
that
and
before
the
meeting
and
please
look
on
this
slide.
A
A
Okay,
so
another
rule
is,
I
think
you
probably
if
it's
like
me,
you
you
this
pop-up
window
that
started
when,
as
pop-up
at
the
meeting
started,
that
it's
important
to
to
know
that,
in
the
in
the
case
of
online
meeting,
everything
that
you
will
say
or
if
your
camera
is
on
will
be
recorded
so
by
participating
the
this
interesting
to
some
to
to
agree.
You
agreed
to
to
this
rule,
and
so,
as
you
can
see
here,
the
meeting
is
starting
to
make.
A
Now,
although
important,
we
have
privacy
and
code
of
conduct
so
know
that
what
you
present,
what
you
will
say
will
be
will
be
public.
It's
important
to
know,
and
also
that
you
have
important
is
that
you
have
to
be
respectful
of
what
other
participants
will
present
and
say
it's
really
important
in
case
of
of
concern.
A
So,
although
just
remind
that
this
is
a
ayat
group,
so
of
course
we
will
detail
a
bit
what
is
iota
for
newcomers
in
the
next
presentation,
but
ayat
basically
conduct
research.
We
are
not
focusing
on
developing
standard,
so
we're
focusing
on
long-term
research-
and
this
is
the
standardization
is
done-
is
the
parallel
organization
that
is
ietf.
A
So
this
is
an
online
meeting.
Fortunately,
so
when
you
are
not
speaking,
please
mute
yourself
and
when
speaking
important,
to
clearly
state
your
name.
If
you
cannot
really
see
what
it's
always
good
to
restate
your
name,
you
have
all
the
link
for
the
for
the
meeting
that
you
can
consult
and
important.
Please
go
to
the
to
the
to
the
note
and
add
your
name
to
the
meeting
participant
list
that
can
keep
track
of
people
participate
and,
although,
as
I
said,
it
is
recorded,
so
this
meeting
can
be
can
be
replayed
replay
can
be.
A
You
can
see
this
replay
of
this
meeting
afterward
on
the
youtube
channel
of
the
iet
yeah.
A
So
after
this
quick
introduction,
we
will
give
you
more
in
detail,
introduction
about
what
is
an
energy,
and
what
is
an
irtf
in
particular
is
important,
because
we,
we
are
peace,
that
this
emergement.
A
I
is
to
be
presented
within
the
iem
conference
and
we
so
much
more
participant
to
the
conference
as
well.
So
it's
important
to
present
you.
What
is
the
energy
with
in
the
next
presentation?
And
then
you
will.
You
will
have
two
technical
tools
so
one
from
amin,
abu
bandhia
with
this
routine
and
zero
touch
management
of
network
slices
in
beyond
5g
and
the
other
one
from
jgo
lopez,
which
is
about
the
concept
of
digital
free
network.
B
A
My
screen
or
no
it's
no,
it's
like
this
since
the
beginning,
oh
okay,
so
we
just
stop
and
restart.
C
B
All
right,
it
says
starting
to
share,
but
it's
still
a
grey
screen.
B
D
A
A
So
basically
you
you
consent
to
that,
and
so
there
is
yes,
the
really
inter
intellectual
property.
So
here
you
have
all
the
references
and
that
I
said
that
with
the
disclosure
rule
in
case
of
particular
in
case
of
patents,
and
then,
if
you
go
to
the
next
slide,.
B
A
A
A
It's
on
okay,
thank
you,
so
yeah,
so
we
will
present
now
with.
I
think
we
prepare
the
slide
this
lower
because
together
culture
of
an
energy
just
to
present
you
what
is
irtf
and
what
is
enemy
and
although
what
is
somehow
is
in
relation
to
the
ietf
here
in
this,
in
this
short
introduction,
next
slide,
please
law.
A
Okay,
this,
I
think
it's
a
version
but
anyway,
so
the
questions
that
we
will
try
to
to
answer
today
are
some
of
the
forums
is
that
they
sound
like
a
very
basic,
but
it's
not
so
so
I'm
also
easy
to
to
to
answer
them.
A
It's
mostly
someone,
but
the
idea
is
that,
after
this
meeting
in
particular
for
people
who
are
not
very
familiar
with
irtf
with
an
mlg
to
understand
what
is
the
difference,
ietf,
that
is
more
related,
of
course,
that
is
related
to
standardization
and
irtf
also
to
understand
how
you
could
contribute
to
a
ietf
or
your
kiev
and
in
particular,
really
iot.
That
is
about
research.
What
is
the
difference
with
an
academic
conference
when
you
participate
to
research
group
as
the
iatf
and
mostly
today,
because
we
are
talking
about
network
management,
we
will
so.
A
Okay,
so,
as
you
know,
as
our,
we
could
think
that
the
research
and
center
are
two
isolated
worlds.
You
see
here
on
the
left,
you
have
some
overall,
you
have
the
let's
say
the
research
conference
academic
conference,
a
small
response
organized
by
using
kcm
or
tripoli
at
the
right
side.
You
have
all
these
standardization
organizations
like
iets,
that
you
talked
today,
but
there
are
others
like
3gpp,
xeiq
and
so
on
and,
of
course,
are
not
isolated.
We
we
need
some.
A
They
need
some
interconnection
between
the
two,
because
we
need,
of
course,
that
some
research
can
be
transferred
on
to
standardization
in
order
to
make
things
evolving
and
also
that
we
need
also
as
a
researcher.
We
also
need
some
input
from
more,
let's
say:
work
in
the
field:
practical
and
working,
so
in
particular
next
slide.
A
Okay.
So
the
first
question
is
that
why,
as
a
researcher,
you
should
contribute
to
standard
an
easy
answer
which
would
be
to
say
that
it
will
help
to
make
the
network
work
better.
A
So
it's
so
simple,
but
of
course,
if
you
just
click
three
times
long
and
make
everything
appear,
it's
really
easy.
When
you,
I
think,
as
a
researcher
like
I
mean
every
researcher
when
they
are
working
on
their,
of
course,
our
topic
at
some
point,
I
would
like
also
that
the
research
does
not
only
remain
as
political
research
or
apply
in
a
very
in
a
lab
environment,
but
understand
that
the
research
can
really
be
useful
can
be
reused
in
real
world.
A
That's
why
it's
it's
a
way
that
for
researchers,
it's
important,
it
can
be
important
to
contribute
to
standard.
Does
that
mean
that
you,
as
a
researcher
you
are
directly
to
right
standard
but
to
be
in
touch
with
people
working
and
stand
up,
and
this
is
important
because
my
contribution
to
standard
it's
also?
A
It
also
means
that
discussing
with
people
in
the
english
queries
that
work
on
standard
those
authors
that
feedback
regarding
the,
for
example,
the
proposal
research
proposal
of
the
technology
you
propose
and
see
if
it
could
be
applied
in
the
real
world.
What
other
problems
that
you
may
may
have
not
foreseen,
and
so
on
and
important.
I
think
the
last
point
is
very
important.
If
you
look
basically
on
standard,
you
will
realize
that
most
of
standards
are
have
written,
led
by
a
major
industry.
A
Of
course,
with
your
business
point
of
view
and
as
a
researcher,
I
think
we
also
less
and
we
have
less
driven
by
my
business,
and
so
it's
good
that
researcher
and
also
interconnection
installed
out
to
where,
let's
say
known
as
point
of
view
of
the
technology
as
well,
next
slide,
please
so
how
to
contribute
so
in
very
schematic
way.
A
Yes
looks
like
so
you
have
the
difference,
of
course,
as
I
said
before,
different
standardization
organization
here,
just
a
landscape
of
some
of
say
some
of
the
protocols
or
technology,
or
that
has
been
somehow
that
has
been
sorry
at
the
etf
you.
You
can
see
a
lot
of
different
technologies
that
you
use
every
day,
probably,
and
but
there
is
no
detail,
then,
if
you
go
to
the
next
slide,.
A
Please
so
here
is
just
a
very
a
quick
summary
about
the
step:
how
to
make
standardization,
basically
ultra
etf.
So
you
start
from
maybe
from
research,
but
we
will
discuss
a
bit
that
it's
not
a
research
at
the
it's,
not
primary
research.
It's
research
that
you
already
know
that
can
fit
some
problem
for
the
standardization
of
the
aether
that
has
been
already
identified
by
the
atm
once
you're,
something
that
is
appropriate.
Then
you
have
to
build
your
community
around
what
you
propose
and
basically
to
convince
people.
A
So
you
will
present,
like
your
purpose,
will
commence
people
to
also
help
to
help
you
to
refine
the
proposal
in
order
to
push
the
proposal
and
at
some
point,
if
your
proposal
is
enough,
but
here
it
should
be
developed
within
within
a
working
within
a
working
group
again
with
different
type
of
refinement,
a
different
step
of
review
and
and
at
the
end
it
might,
it
may
lead
to
publication
at
the
rfc
as
a
standard.
A
A
A
So
when
you
propose
something
it
has
to
fix,
it
has
to
be
aligned
with
the
ietf
carol
is
a
architectural
symbol
of
the
internet,
and
only
if
it's
here
then
you
could,
you
could
propose
something,
but
it
has
to
be
enough
mature
in
terms
of
development.
Just
that
I
mean
a
theoretical
id.
You
need
some
real
development,
already
prototype
implementation,
probably
to
be
convincing
before
going
to
the
iep
next
slide.
Please.
A
And
so
then
is
a
question
about
research,
because
here
in
particular
today,
because
we
are
collocated
with,
we
have
a
lot
of
researchers
in
the
virtual
room
and
what
will
be
the
role
of
research
in
in
idea.
Actually
starting
with
the.
A
Inaudible,
and
as
you
can
see
on
this,
this
figure
here
is
that
it
could
be
seen
as
a
maybe
a
nice
character,
an
enabler
if
you
want
to
go,
maybe
for
standardization,
it's
not
only
the
way
to
go
for
sterilization,
but,
as
you
can
see
here
from,
let's
say,
academic
work.
You
can
go
through
iotf
to
discuss
more
and
to
work
more
research
problem
and
then,
if
it
comes
enough
matters
and
you
can
think
about
going
to
the
ietf
next
slide,
please.
A
A
Each
group
works
in
a
different
manner,
but
basically
the
main
principle
is
that
it's
an
open
forum
wherever
searcher
can
exchange
id,
can
present
ids,
even
primarily
ids
can
present
experience
can
present
prototype
and,
and
then
it's
some
place
where
the
more
let's
say,
academic
research
like
we
have
the
conference
meeting
today
in.
I
am
on
this
weekend,
where
you
have
more
opportunity
to
meet,
discuss
with
engineer
working
on
practical
analysis,
mainly
the
difference
when
you
participate
to
irtf
research
group
or
meetings
with
an
academic
conference.
A
You
have
also
more
interaction
with
engineers
at
work
in
the
field
and
can
also
give
feedback
with
problems
they
may
encounter
in
reality,
and
also
you
can
also
propose
them
some
solutions.
They
do
not
envision
for
their
own
program,
so
this
is
where
it
is
different.
I
think,
of
course,
even
in
iron
conference
we
have
a
mix
between
industry
and
academia,
but
I
think
in
iotf
you
may
find
a
better
me
because
also
we
have
collocated
meetings
with
iitf,
but
we
have
a
lot
of
people
working
on
engineering
aspects
next
slide.
Please.
A
Okay,
we,
I
will
go
quickly
through
the
the
next
slide.
There
are
different-
let's
say,
global
activity
at
the
iert
level,
to
support
applying
networking
research,
so
for
so
applying
research
in
networking
one
is
a
workshop
that
is
organized
and
this
collocative
ietf
meeting
in
july.
So
it's
processing.
So
if
you're
interesting,
you
can
change
the
same
next
slide.
Please.
A
And
another
one
is,
I
would
say,
apply
networking
with
search
price.
That
again,
is
a
rather
to
recognize
the
best
results
that
you
may
have,
or
any
any
other
can
have
in
applied
networking,
because
you
can
nominate
self
dominate
on
the
minute.
Also
and
interesting.
Of
course,
there
is
a
cash
prize,
but
also
you're
invited
to
the
ietf,
and
you
give
a
talk.
A
Okay,
so
these
two,
these
two
I
mean
the
workshop
and
the
applied
networking
research
prize-
are
two
way
also
to
some
or
make
good
publicity
of
your
work
and
to
the
iatf
community.
So
again,
if
you're
interested,
you
can
check
the
link
or
presentation.
Just
I
didn't
say
that,
because
it's
a
common
for
a
regular
attendant
or
presentation
are
available
on
the
website
afterwards,
so
you
will
find
all
the
links
next
slide.
Please.
A
So,
let's
start
now
with
the
nmrg
network
management
research
group
next
slide,
so
again,
just
to
repeat
again
that
this
is
really
a
forum
for
researchers
to
to
exchange
together
to
explore
new
technology
for
the
management
of
the
internet.
So
this
is
a
very
general
thing
like
that,
so
there
are
different
type
of
frogs
that
is
performed
some
there's
some
work
serving
some
some
approaches.
Some
work
on
the
new
architecture,
maybe
for
network
management,
specification
of
solution
and
so
on.
A
Here
is
a
main.
Let's
say
landing
page
for
energy.
If
you
want
to
learn
more
about
the
all
the
activities
that
have
been
done
in
the
book
and
knowing
that
it
started
in
19,
so
it's
quite
old
and
I
think
they're
all
best
still
active
now
and
there
have
been
successive
waves
of
some
thematic
waves,
the
first
one
about
management
technologies,
the
second
one
with
autonomous
autonomic
network
management,
and
no,
I
will
discuss
a
bit
more
in
the
next.
A
But
what
is
important
to
understand
here
is
that
the
some
all
the
thematic,
the
objective
of
the
group
are
not
fixed
and
can
be
an
are
regularly
revised,
released
thanks
to
the
participant
proposition
interest.
It's
not
only,
of
course,
the
chairs
that
we
say
what
we
will
do
in
the
group.
It's
really
a
community
effort
to
refine
me
virus
with
his
or
agenda
next
slide.
Please
now.
A
So
now
look
we
will
look
a
bit
more
at
the
current
activities
in
the
group,
so
the
overarching
theme
is
surveying
self-managed
network,
which
is
somewhat
available
of
automatic
networking,
but
with
maybe
a
more,
we
are
expecting
more
automation
in
network
management.
I
think
you're.
A
I
mean
I've
seen
a
lot
of
people
talking
about
that
and
you
will
see
a
lot.
I
think
this
week.
I
am
as
well,
of
course,
if
he,
even
if
we
think
that
we'll
have
several
driving
or
certain
library
we
also.
We
were
so
aware
that
we
listed
some
other
humanities
look
somewhere
and
we
need
infinite
interface
between
the
human
and
the
self-driving
system,
and
that's
why
we
will
we
particularly
promote
intent-based
networking
intel
group,
our
ibm,
that
I
will
describe
a
bit
later
and
also
for
having
self-driving
networks
like
any
severing
system.
A
A
So
so,
for
here
is
the
this:
is
some
of
the
research
item
a
work
item
of
the
ibm
within
the
group?
A
So
we
it's
a
somological
logical
flow,
so
we
start
with
product
statement
and
forget
them
at
that
concept
and
if
you
just
click
on
so
in
regard
to
that,
we
particularly
have
a
document
which
is
a
quite
natural
no,
for
which
we
target
an
informational
rfc,
where
we
give
our
own
definition
on
what
is
an
intent.
In
particular,
we
say
that
declaration
of
operation
goal,
so
you
declare
the
outcome
of
what
you
expect
to
have
in
the
network,
not
the
way
to
do
it.
A
In
particular,
we
make
the
distinction
when
I
say
we
just
as
a
group,
because
no
it's
a
it's
a
group
document,
but
this
is
led
by
by
some
waterers,
and
so
we
make
the
difference
between
internet
and
policies.
That
is
very
important
to
send
with
details
of
what
are
the
expected
functionality,
living
system
and
life
cycle.
So
this
is
again.
This
is
the
the
vision
that
we
have
in
energy,
so
ibn
in
other.
A
In
other
places
that
can
be
with
different
vision,
but
it's
important
for
our
work,
maybe
that
we
set
up
what
are
the
basis
of
our
work
then
next
slide
along.
A
Please,
then,
we
have
also
one
work,
which
is
quite
natural
for,
for,
although
we
are
targeting
the
informational
access,
which
is
about
internal
classification,
where
it's
somehow
helped
to
in
the
first
graph
I
mentioned
it
was
a
general
definition
of
indentation
here,
it's
put
in,
let's
say
with
it's
illustrated
with
some
some
example
in
different
type
of
intense
relation,
carinator
dc
and
entrepreneur
network.
A
It's
important
to
see
that
this
draft
also
provide
a
way
for
anyone
who
would
like
to
classify
the
type
of
content.
You
would
like
to
use
a
methodology
by
identifying
the
user,
the
types,
cop
and
absorption,
and
so
now
what
what
is
the
status
of
the
group.
So
those
two
work
are
quite
are
quite
mature
now,
but
now
we
are
really
in
the
step
four
and
five.
A
A
So
this
is
really,
let's
say
one
of
the
other
topic
of
the
moment
in
the
group.
So
if
you
are
interested
to
participate,
if
you
work
on
ibm,
I
would
like
to
propose
or
use
cases.
This
is
really
a
good
timing,
because
this
effort
is
really
under
consideration
now,
but
then
next
step
will
be
also
implementation,
integrity
and
to
priority
and
next
slide.
Please
law-
and
we
have
also
some
some
work
but
from
here
because
we
participated
to
a
hackathon
with
a
multiple,
multi-level
approach,
ibm
implementation.
A
A
So
for
the
first
draft,
the
drive
is
in
my
hand,
you
know
that
it
has
been.
It
has
been
revised
according
to
the
to
the
review
it
is
received
and
as
a
shepherd,
I
have
to
check
if
no
it's
enough
mature
and
I
mean
if
it
has
to
go
again
for
a
last
call
or
if
it's
then
okay,
to
to
propose
and
for
the
sigmund
russ.
As
far
as
I
know,
lauren
is
a
shepherd
and
we'll
use.
We
do
some
summary
of
the
of
the
comments
we
received.
B
D
A
A
A
And
so
here
is
a
somehow
what
we
expect
is
either
white
paper
or
informational
fc,
but
basically
no
is
a
collaborative
document
where
you
try
to
involve
both
network
management
and
the
expert
to
list.
What
are
the
challenges
of
the
coupling
between
ai
and
m?
So
it's
not
the
list
of
use
cases
because
we
had
the
experience
before
we
start
from
use
cases,
but
some
more
people
come
with
use
cases
present
use
cases.
It's
nice.
A
We
have
a
lot
of
discussion,
but
then
from
that
just
it
was
not
easy
to
somehow
extract
some
common
challenges
that
we
face
in
when
using
ai
network
management.
A
G
A
Also,
very
in
our
domain,
because
we
are
targeting
the
driving
network,
we
need
all
the
ai
for
taking
decisions.
This
is
a
a
bit
different,
not
from
all
domain,
but
at
least
it's
it's
some
more,
not
only
just
analyze
the
data
and
give
some
results,
it
needs
to
take
some
decision
and
also.
We
also
think
that
it's
not
only
using
ai
for
network
management,
although
network
management
can
support
a
bait
area.
A
So
this
is
also
a
work
which
is
really
let's
say
under
consideration:
consolidation
now
so
really
open.
If
you
want
to
contribute,
you
can
you're
really
welcome
to
do
so.
We
will
also
discuss
this
topic
in
the
panel
wednesday
morning
at
the
conference.
That
is
also
related
to
the
research
challenge
of
using
or
of
ai
and
network
management.
A
There
are
tools,
our
point,
one
is
about
organizing
the
practical
competition
and
the
other
one
about
supporting
more
discussion
as
an
open
forum.
So
for
the
competition
is
a
bit
on
pause
because,
due
to
lack
of
time
so
excited,
is
our
volunteers
ready
to
also
to
support
these
activities?
So
this
kind
of
a
practical
competition,
like
you,
have
an
other
domain,
but
of
course
we
are
also
regular
talk
so
easily
from
people
that
presented
our
paper.
People
who
have
papers
and
conferences
can
present
against
their
work
into
the
easy
group.
A
So
I
want
to
highlight
that
here
if
you
never
participated
to
to
an
energy,
so
a
good,
good
and
easy
way
to
do.
Let's
say
first
participation
is
that
you
can
just
basically
represent
the
work
that
you're
already
presenting
in
the
conference.
So
if
you're
interesting
to
do
so,
you
can
really
contact
us.
A
We
also
envisioned
to
organize
something
more
regular,
like
a
ribena
series.
To
be
honest,
we
we
put
that
on
on
pause
as
well,
because
you
know
with
with
the
let's
say,
yeah
fundamental
timing.
We
are
all
on
online
event
in
web
conference
and
telco,
so
we
thought
that
it
was
not
a
good
timing
for
launching
a
new
online
event,
so
we
prefer
maybe
to
to
resume
a
bit
later.
A
Yes,
for
the
yeah,
exactly
thank
you
laura
yeah.
We
have
to
also
resume
a
bits
for
the
google
docs,
so
we'll
organize
a
dedicated
meeting
for
that
as
well
and
next
slide
law.
Please.
A
A
Residual
configuration
next
slide.
Please
don't
speed
up
a
bit
because
I
think
I'm
ready
late.
So
in
summary,
I
took
our
topic
ibn
and
ai
with
some
somehow
I
presented.
What
is
the
status
and
what
was
the
comment?
Let's
say
topic
in
this
topic,
so
it's
important
that
for
people
who
are
not
familiar
with
irtf
and
emerges
that
we
are
not,
as
usually
you
say
that
as
a
irtf
research
group.
Yes,
we
are
a
personalization
group.
You
know
we,
of
course
you
can
go
through
it.
A
You
have
to
do
some
standardization,
but
we
are
a
research
group,
not
a
pre-standardization
group,
so
we
do
a
lot
of
things
that
our
expectation
is
well
to
have
everything
published
as
a
standard
as
the
ifc,
but
also
to
support
really
a
research
collaboration.
His
technical
presentation
with
academic
joint
implementation
is
really
to
somehow
make
people
working,
collaborating
together
from
engineering
from
a
research
world
as
well.
A
So
in
practice
you
can
start
from
the
group
web
page
and
subscribe
to
mailings.
If
you
want
to
be
aware
of
the
different
meetings,
everything
is
announced
on
the
mailing
list,
usually
at
the
virtual
meeting,
almost
monthly,
not
exactly,
and
we
have
collocated
with
the
etf,
which
are
normally
a
physical,
but
of
course
the
last
one
were
also
nine.
You
can
participate
to
a
meeting
for
free
for
meetings
that
are
collected
with
ietf
when
it
is
physically.
A
Of
course
you
have
to
pay
for
the
etf
and
currently
also
for
ietf,
which
is
online
you're,
also
too
paid,
but
you
have.
You
can
also
apply
for
fee
waiver
to
participate
online
and
there
is
no,
I
think,
no
limitation
for
fee
weaver.
So
it's
really
open.
So
you
don't
need
to
pay.
You
don't
need
to
have
your
institution
affiliated
with
the
the
iatf
or
your
itf.
B
I
just
would
like
to
add
on
top
of
what
jerome
said
about
what
rtf
and
nmg
are.
Is
that
really
we
are
a
research
community
on
network
management?
So
that's
why
we're
also
participating
with
imm
and
noms.
We
want
to
really
lays
with
a
tripoli,
synonym
and
ifip
working
group
6.6
communities
on
network
management
network
operations.
B
So
we
think
this
is
really
complementary.
As
you
all
mentioned,
we
are
not
doing
standardization
but
participating
to
the
rtf,
gives
you
also
experience
and
opportunities
to
discuss
with
operators,
engineers,
vendors
that
are
actually
implementing
and
developing
the
internet
standards.
So
this
is
an
invaluable
proximity
and
experience.
B
So
it's,
I
think
if
you
ask
different
colleagues
in
the
academia-
and
there
are
many
in
the
in
imm
that
have
already
participated
in
iitf-
it's
very
ietf
and
rtf-
it's
very
valuable
experience,
not
necessarily
for
your
academic
cv,
but
from
a
I
mean
from
work,
experience
and
and
diversity
of
what
you
can
can
can
meet
and
and
do
so.
I
think
it's
really.
We
try
to
it's
a
very
open
forum,
bring
your
your
problems,
bring
your
research
questions.
This
is
completely
open
and
you
can.
B
B
Maybe
we
can
continue
with
your
presentation.
A
Yeah.
Thank
you
lord.
I
think
if
there
is
no
question
yes,
we
can
move
to
the
explanation.
So
next
presentation
is
given
by
by
I
mean
switching
and
the
retouch
management
of
network
slices
in
bm
5g.
D
G
Thank
you
very
much.
Okay,
yeah,
so
hello,
everyone,
nice
to
see
you
again.
I
think
I
am
is
a
nice
opportunity,
but
also
an
emoji.
So
I
would
like
to
thank
lauren
and
jerome
for
the
opportunity
I
so
I'm
aminable,
I'm
a
researcher
and
project
manager
in
orange
labs
in
france,
and
I
would
be
speaking
on
the
name
of
the
consortium,
all
the
partners
of
mumbi
5g
project
dealing
with
management
and
orchestration
of
networks
for
beyond
5g.
So
it's
an
ict
20
project
as
it
deals
with
management.
G
That's
a
very
nice
just
to
join
what
jerome
was
saying
about
being
with
im
and
noms
conference
and
cnsm
as
well.
That's
a
very
management
focused
collaborative
projects.
Your
european
project
and
the
focus
as
we
I
will
be
presenting
is
the
distribution
of
management
and
the
automation
of
management
with
also
a
network
slicing
as
a
target
as
jerome
and
lauren
suggested.
G
I
I
kept
the
focus
more
on
the
architecture
rather
than
the
very
scientific
contributions,
but
as
I
said
that
that
would
be
an
overview
and
I'll
be
happy
to
to
answer
your
questions
if
you
have
any
by
email.
So
I
I
you
have
my
email
here,
so
5g
project
really
uses
as.
D
G
Some
some
some
keywords
here
are
distributed:
management,
zero
touch,
management
and
orchestration
network
slicing.
Massive
scales,
bn5g,
but
also
the
project
deals
a
lot
with
deep
reinforcement,
learning,
federated
learning,
and
I
have
here
a
teaser
m
s,
a
e
n
d
e,
which
are
very
specific
to
the
project.
G
So
an
overview
of
the
consortium,
so
mumbi
5g
project
is
composed
of
12
partners,
so
along
eight
countries
and
nine
from
industry
and
three
from
academia
and
has
a
budget
of
5.5
million
euros
and
is
up
for
three
36
months
and
now
we're
we're
exactly
at
the
mid
term
of
the
project.
G
And
so,
if
I'd
start
from
from
the
left,
we
have
ericsson
and
ireland
become
in
france,
orange
france,
uricom
france,
city
dc,
spain,
ecuador,
spain,
neck
germany,
alto
finland,
university
of
alto
in
finland,
orange
poland,
and
I
don't
see
the
rest
of
the
screen
citrix
in
greece,
ot
in
greece
as
well
and
eboz
in
cyprus.
G
So
it's
quite
well
balanced
consortium,
mainly
with
regards
to
management
and
machine
learning
expertise.
G
The
structure
of
the
project
is
into
seven
work
packages,
five
main
and
technical
work
packages
and
two
there
are
more
project
management
and
communication.
So
dealing
with
management
and
architecture.
G
We
have
work
package
2
and
distribute
more
focused
our
distributed
monitoring
and
analytics
engine
very
linked
to
the
distributed
ai
driven
decision
engine
for
slice
management
and
a
work
package
that
is
cross-functional,
which
is
security
and
energy
efficiency
and,
of
course,
a
main.
Another
main
work
package
is
everything
related
to
implementation
and
bugs
and
demos.
G
The
approach
that
the
project
is
following
as
a
methodology-
let's
say,
of
of
organizing
the
project,
is
to
well.
Some
of
them
has
already
finished,
but
working
on
everything
related
to
network
architecture
and
use
cases.
So
thinking
of
how
this
massive
scale
can
it
can
can
be
highlighted
through
5g
use
cases.
What
kpis
should
be
considered?
What
are
the
requirements?
Mainly
the
ones
related
to
the
analytical
engine
and
decision
engine
phase?
Two
is
the
data-driven
algorithmic
innovation
and
that's
the
deep
scientific,
let's
say
technical
work.
G
So
that's
the
design
of
the
algorithms
going
inside
the
analytic
engine,
the
design
of
the
decision
engine
and
all
the
mechanisms
related
to
security
and
energy
efficiency
and
phase
three
is
the
the
evaluation
and
testing
and
that's
the
the
design
of
let's
say
a
global
framework
platform
to
to
have
the
pieces
of
box
gathered
to
show
the
the
whole
vision.
G
The
evaluation
of
the
ae,
so
analytical
engineering
decision
engine
alone,
separate
separate
testing
and
also
the
same
mechanisms
of
security
and
energy
separate
and
the
project
is
targeting
two
main
parks.
First,
buck
on
zero
touch
service
management
to
show
it
end
to
end
with
some
scalability
use
cases,
and
also,
as
security
is
an
important
part
of
the
project.
G
G
The
main
concept
of
monbi
5g
is
to
focus
on
the
distribution
of
management
functions,
especially
especially
for
network
slicing,
so
it
defines
a
let's
say,
a
central
element
with
central
policies
defined
by
the
slice
owner
or
the
slice
provider.
It
can
be
network
operator
and
relies
on
two
centralized
analytical
engine
and
decision
engine
and
it
defines
several
technological
domains.
G
Here
I
have
only
three,
but
actually
it's
the
edge
domain
run
domain
cloud
domain
with
with
core
functions
so
and
actually
we
also
have
ones
related
to
devices.
But
this
is
my
the
what
we
call
extreme
edge,
so
each
technical
domain
relies
on
the
three
functions:
monitoring
system,
analytical
engine
and
decision
engine-
and
these
are-
should
cope
and
cope
with
the
technical
features
and
the
technical
data
of
each
technical
technological
domain.
G
So
this
is
what
we
are
insisting
on
in
terms
of
separation
of
concerns
and
separation,
abstraction
of
technical
details
from
a
domain
to
another,
and
also
we
assume
that
within
each
technological
domain
we
have,
we
can
have
a
slice
that
should
be
managed
within
the
domain
by
independent
network
functions
for
management.
Orchestration
functions,
I
mean.
G
So
of
course,
we
assume
that
we
are
supporting
services,
communication
services
from
different
providers
that
would
need
support
from
the
network
in
terms
of
network
slicing
and
the
answer
of
a
network
slice
coming
from
a
network
slice
provider
would
be
managed
by
the
same
provider
or
from
another
one,
but
the
same
provider
should
rely
on
different
infrastructures,
especially
if
it
spans
different
technological
technologies
or
generations.
G
And
of
course,
each
client
is
a
slice
tenant
in
terms
of
using
consuming
the
network.
The
network
slice
and
it
relies,
of
course,
on
on
different
providers
of
virtual
network
functions.
G
So
behind
each
each,
let's
say
a
break
here.
We
can
see
also
a
stakeholder
in
terms
of
of
service
service
provider
as
a
service
provider,
if
it,
if
it's
service
level,
application
level
or
network
or
infrastructure,
we
needed
to
do
some
work,
but
I'm
not
going
to
to
go
into
details.
Some
some
work
on
the
the
impact
and
relevance
with
regards
to
five
gpp
use
kpis.
G
So
I
assume
many
projects
need
to
do
that.
So,
with
regards
to
some
use
cases,
many
of
the
performance
kpis
appear
to
be
to
be
to
have
a
high
high
impact
because
it,
of
course
the
project
is
with
automation,
but
also
in
terms
of
social,
social
and
social
kpis.
Let's
say
everything
related
to
security,
for
example,
goes
in
line
with
the
importance
that
the
the
eu
and
5gpp
gives
to
to
to
these
kpis
and
in
terms
of
business
kpis.
Well,
all
all
the
stakeholders,
as
as
we
saw.
G
Well,
it
goes
with
with
the
the
this
openness
and
this
duplication
of
roles
that
virtualization
and
automation
gives
to
to
the
network.
Let's
say
the
network
ecosystem
or
telecommunication
ecosystem.
G
What's
most
important
here
in
the
infrastructure
layer,
for
example,
we
define
as
many
infrastructure
domain
managers
as
as
info
as
infrastructure
domains,
so
we
should
be
able
to
include
any
a
technological
or
any
technology
as
an
infrastructure
domain
on
itself,
but
also
add
to
that
that
the
fact
that
it
comes
from
a
provider
a
of
provider
b.
So
this
structure
structures,
somehow
the
infrastructure
level
and
here
we're
focusing
on
to
show
the
management.
G
So
the
idm
here
is
the
manager
of
the
infrastructure
domain
and
it
can
be
independent,
of
course,
and
very
specific
to
the
technology.
At
the
management
and
orchestration
layer.
We
find
what
we
call
the
domain
management
and
orchestration.
So
this
has
a
let's
say,
a
slice
view
rather
than
an
unsub
slice
or
a
technological
slice,
part
in
the
infrastructure.
So
here
we
have
the
slice
level
and
we
assume
that
we
can
have
as
many
dmos
as
as
slices
as
we
target.
G
I
can
say
it
now,
so
there's
a
separate
slide
on
that.
The
the
project
targets
the
in
slice,
management
and
the
inter
slice
management,
so
the
dmo
level
deals
with
this.
The
intra
slice
management
and
the
the
opel
layer,
which
is
the
idmo
for
in
into
slice
into
domain
management
and
construction,
is
the
the
function
that
would
deal
with
the
end-to-end
view
it
so
the
inter-slice
view-
and
we
assume
that
we
need
an
exposure.
G
G
G
And
if
we
zoom
a
little
bit
on
the
mombivig
portal,
there
are
some
very
specific
functions
like
the
the
access
management
or
the
databases
for
the
tenant
subscribers
or
users.
Everything
related
to
what
the
monitoring
is
is
needs
to
show
at
this
portal.
In
terms
of
let's
say,
stability
or
sla,
sla
compliance,
there
are
some
apis
linked
to
the
life
cycle
management,
so
that
would
be.
There
are
links
to
the
idmo
and
the
dmo
and,
of
course,
the
validation
of
the
templates.
G
So
when,
when
we
have
a
request,
we
need
to
to
go
through
the
validation,
that's
quite
classic,
but
of
of
the
the
the
composition
of
the
of
the
slice
itself.
G
Although
it's
its
description
and
just
to
summarize
it
this
before
to
go
before
going
to
a
more
complex
figure,
is
that
the
fact
that
we
target
the
separation
of
technological
domains?
We
assume
that
we
also
separate
concerns.
I
know
not
everyone
likes
this
term,
but
it's
it's
somehow
a
separation
of
of
technical
features
and
very
specific
features
that
there
is
no
need
to
add
it
to
a
main
or
centralized
orchestrator.
G
So
this
is
what's
meant
here,
but
by
this
feature
the
distribution
of
the
functions
because
it
goes
down
to
distributed
infrastructure,
domains,
the
support
of
management
as
a
service.
So
each
actually
there
are
brokers
that
expose
that
allows
to
expose
management
functions
themselves.
So
the
the
the
idm
at
the
at
the
infrastructure
domain,
but
also
the
dmo
and
idm
idmo
in
this
slice
in
the
slice
level,
I'm
not
directly
involved
in
security,
but
there
are
very
good
works.
G
I
mean
a
lot
of
work
on
on
the
slide
security,
so
each
function
is
including
the
the
s
of
the
ff
cup
at
f
cups
for
security,
and
this
is
in
the
intra
and
into
slice
the
hierarchy
of
course.
So
we
have
the
the
inter
slice
level
that
allows
to
have
this
end-to-end
view
and
along
with
security,
the
energy
where
infrastructure
management
is
is
using
programmability
in
order
to
see
how
energy
related
policies
can
be
can
be
enforced
and
also
the
the
in-slice
management.
G
Usually
we
speak
about
sub-slices,
but
of
course
we
can
assume
that
a
tenant
is
requesting,
just
let's
say
a
run
slice,
and
that
would
be
for
him
and
in
just
the
complete
slice,
and
we
should
have
separate
management
functions
for
that,
and
that
would
be
the
in-slice
management
capabilities
and,
of
course,
scalability
for
the
slice
management.
Is
that
actually
we
we
see
and
when
we
look
closely
to
etsy
zsm,
so
they
retouch
service
and
network
and
service
management
work
and
we're
aligned
with
with
many
of
the
other
features?
G
And
we
try
to
go
further
on
that.
For
example,
how
we
can
enhance
scalability
using
the
the
closed
loop
automation,
but
also
the
massive
distribution,
and
how
this
helps
in
in
the
scalability,
and
also
just
considering
infrastructure
domains
that
that
are
large
scale.
G
So
this
is
the
let's
say,
a
complex
figure
to
show,
for
example,
so
many
of
the
notions
I
have
presented
before,
but
here
at
least
we
can
see
how
the
functions
can
can
be
exposed
and
how
we
can
assume
the
use
of
machine
learning,
of
course,
can
be
applied
to
the
same
functions
but
at
different
levels.
So
we
can,
we
can
think
of
a
management
function,
an
ml
based
management
function
applied
to,
let's
say,
a
network
slice
from
one
tenant,
but
also
to
another
tenant.
But
this
stays
completely
of
course
separate.
G
So,
if
I
go
beyond
that,
in
addition
to
the
the
machine
learning
algorithms
that
that
would
be
integrated
within
the
functions,
we
assume
the
use
of
control
loops
at
also
multiple
levels,
so
the
the
oss
basis
level,
with
ideally
well
for
for
the
uss,
mainly
and,
of
course,
the
technological
domains
themselves,
at
the
infrastructure
level,
at
the
slice
level
and
at
the
at
the
nodes
and,
let's
say
composed
composed
services
level.
G
And
if
we
look
at
actually,
I
have
many
slides
on
the
internal
structure
of
each
of
each
main
component
and
if
we
look
at
the
inter
domain
orchestrator,
the
one
that
gives
the
end-to-end
view,
so
it's
separated
into
functional
layer
and
non
non-functional
layer.
So
the
functional
layer
is
somehow
the
other
functions
the
management
functions
in
in
in
terms
of
of
apps
caps
and
orchestration.
So
it
it
also
have
the
the
all
the
databases
and
and
templates
for
the
the
the
main
management
functions.
G
But
the
the
non-functional
layer
related
to
to
the
inter-domain
orchestrator
is
based
on
these.
Ms
monitoring
system,
analytical
engine
and
decision
engine,
and
it's
it
stays
with
within
the
management,
the
management
layer.
G
But
it
can
go
to
very
specific
actions,
for
example,
mainly
if,
if,
if
the
for
sure
that
the
the
the
slice
is
distributed
or
have
different
technological
domains,
so
you
can
imagine
that
this
inter
domain
orchestration,
can
be
very
specific
at
the
technological
domains
and
that
would
fill
let's
say
the
the
sub
layers,
the
msa
e
and
d
sub
layers,
with
very
specific,
with
regards
to
data,
at
least
very
specific
solutions.
G
Okay,
I
can
I
can.
I
can
skip
the
the
details
here,
maybe
just
to
I
will
skip
the
the
internal
structures
just
to
maybe
show
very
quickly
the
monitoring
system
supplier
here,
for
example,
the
all
all
the
functions
of
the
monitoring
system
are,
can
be
exposed
to
the
elements
manager
and
have,
for
example,
deal
with
everything
related
to
data
collection,
information,
connect,
collection
and
aggregation
processing,
calculation,
etc.
G
If
I
try
to
go
very
quickly,
so
the
analytics
engine
deals
with
the
f
caps,
so
full
security
performance
accounting
and
also
exposes
these
capabilities
and
the
decision
engine
same
thing,
but
has
the
ability,
because
it
it
it
integrates
ai
based
algorithms
that
can
take
the
decision
and
give
it
to
the
to
the
dmo
and
idmo
have
the
same
management
functions,
but
have
the
decision
on
board,
and
so
it
it
doesn't
do
the
analytics,
because
this
is
in
the
in
the
ae
sub
layer.
G
D
G
That
yeah,
that's
very
short.
Okay,
maybe
just
this
slide.
If
we,
when
you
have
the
slides,
if
if
you
would
like,
if
you're
curious,
to
see
how
this
can
be
instantiated
within
within,
for
example,
different
technological
domains,
then
here
we
have
edge,
run
and
core,
and
we
have
tried
to
illustrate
one
single
tenant
over
three
technological
domains.
So
you
can
you
can
really
zoom
to
see
how
this
is
concretely.
G
I
can
be
concretely
instantiated.
Well,
maybe
I
can
rather
take
questions
then
continue
just
to
to
to
move
on
very
quickly
on
on
the
slides.
I
don't
know
so.
Two
two
main
use
cases
and
a
very
complete
platform
in
barcelona
that
is
very
integrative
so
to
host
the
box,
and
you
can
find
here
links
to
all
that
is
getting
out
from
from
from
from
the
the
project
in
terms
of
deliverables,
publications,
newsletters,
but,
and
also,
I
really
suggest
the
video
presentation
that
has
been
published
and
yeah.
Thank
you.
G
Sorry.
Sorry,
for
maybe
with
the
timing
I
see
I
I
think
we
were
quite
behind
schedule.
A
B
Yeah,
I
think
we
have
phil,
which
is
I
wanted
to
ask
something:
okay,
hello,.
C
It's
it's
yeah
phil
early
here,
thanks
very
much
amina.
That
was
some
good
overview,
interesting
stuff.
What
you're
up
to!
I
wonder
if
you
could
just
say
a
little
bit
about
in
terms
of
the
of
the
machine,
learning
that
the
project's
doing
what
I
didn't
quite
pick
up,
what
it's,
what
it's
learning
you
know,
what
it's
is
it
making?
It
is
this
about
just
how
you
know
what
choices
is
it
making
it
so
something
to
do
with
the
slicing,
but
I
didn't
quite
gather
what
it
was
actually
doing.
G
Yeah
yeah,
I
actually
at
least
from
from
the
the
orange
contribution
we're
dealing
with
with
resource
allocation.
So
it's
placement.
Sorry,
it's
placement
somehow!
So
when
one
of
the
functions
so
actually
the
the
the
main
proposal
was
to
so
it
was
drl.
So
deep
reinforcement
learning
algorithm
to
see
the
so
that
separates
the
domains
and
see
how
each
domain
can
learn
to
allocate
its
resources
separately.
G
So
this
is
part
of
the
orchestration,
the
the
the
infrastructure
domain
orchestration,
and
so
it
yeah
actually
it
it
learns
to
to
do
the
placement,
but
it's
not
that
independent
and
we
added
actually
a
control.
So
what
the
what
we
have
called
controlled
by
a
heuristic,
so
we
called
it
heuristically,
assisted
or
aided
deep
reinforcement,
learning
to
optimize
the
the
the
orchestration
of
of
network
slices.
G
So
the
same
learning
agents
are
within
the
separate
domains
and
we
have
a
centralized
learning
agent
centralized
trl
that
allows
to
yeah
to
have
the
the
end-to-end
view,
so
it
actually
what
what
it
learns.
So
we
have
a
a
feature:
extraction
automated
feature
extraction
from
the
the
physical
layer
so
from
from
the
infrastructure
layer,
and
it
learns
actually
to
to
orchestrate.
G
B
B
To
end
all
things,
diego,
are
you
ready.
E
I
was
in
the
process
of
muting
another
like
hello,
and
you.
E
E
The
idea
is
to
introduce
you
well,
this
is
a
this
is
a
sort
of
a
continuation
of
what
rohan
was
representing
about
how
the
the
whole
nmrg
works
and
the
and
the
process,
because
it's
precisely
the
introduction
of
an
ongoing
work
on
a
draft
on
a
particular
draft,
as
is
associated
with
that,
with
the
concept
that
this
list
of
people
you
see
there
are
working
on.
We
say
it's
a
document,
we
started,
I
don't
know.
Probably
it
was
presented
three
itf's
meetings
ago
and
right
now
we
are
about.
E
The
last
version
is
version.
Four
we
are
for
the
fourth
iteration.
I
am
preparing
personally
and
preparing
a
a
version
five,
and
this
is
how
it
evolves
by
by
the
interaction
with
the
rest
of
the
community,
the
feedback
we
get
and
additional
ideas
and
well
I'm
the
interaction
among
among
the
authors.
E
The
idea
is,
I
mean
the
idea
of
this
work
is
to
explore
the
the
application
of
digital
training
principles
to
to
network
management
and
and
to
network
experimentation
in
general.
E
We
we
started
from
the
this
idea
that
this,
I
guess
that
you
are
familiar
with-
is
digital
replicas
of
physical
entities
that
are
called
the
digital
trainings
that
are
rely
on
on
a
continuous
flow
of
data
from
the
from
the
physical
environment
and
that
use
and
rely
as
well
on
artificial
intelligence
to
simulate
to
to
emulate
or
make
a
synthetic
environment
that
behaves
like
the
the
physical,
the
physical
system
that
is,
or
let's
call
it
the
physical
twin.
E
So
it
is
possible
to
make
a
to
have
a
better
understanding
to
reduce
costs
in
in
evaluating
choices
and,
to
being,
let's
say,
on
the
safe
side,
when
trying
to
experiment
with
systems
that
are
extremely
complex
or
even
dangerous.
E
That-
and
this
is
precisely
this-
I
mean
when
we're
talking
about
digital
trains-
we're
talking
about
the
lemonade
digital
lemonade
that
we
can
use-
and
it's
not
so
so
dangerous
when
you're
in
in
the
case
of
experimenting
with
something
that
can
be
disruptive
in
terms
of
cost
or
even
or
even
in
terms
of
serious
damage
to
to
lives
or
nature
or
whatever.
E
The
connections
you
have
with
the
with
the
physical
system,
with
the
physical
training
in
terms
of
sensors
and
actuators
the
use
of
ai
to
to
precisely
to
simplify
and
to
somehow
to
reduce
the
amount
of
of
computing
or
simulation
power
that
you
need
for
for
reflecting
really
really
really
reflecting
the
behavior
the
use
of
communication
that
has
to
be
as
real-time
as
possible,
especially
in
in
cases
like
the
they
like
the
network
mechanisms
for
representation.
E
So
you
can
reason
I
mean
humans
can
reason
and
work
with
the
with
the
with
the
whole
system
and
understand
what
is
what's
going
on.
Essentially,
there
is
a
trust
fabric
that
the
the
real
train
to
trust
the
digital
train
with
some
decision
and
be
a
digital
trained
to
trust,
the
physical
train
when
collecting
data
and
the
and
the
mechanisms
around
privacy
and
security
that
are
especially
relevant
in
environments
like
the
like,
like
a
network,
because,
among
other
things,
it's
very
likely.
E
Well,
you
can't
simply
not
use
personal
data
for
any
kind
of
processing
without
the
express
authority,
authorization
of
the
owner
of
the
data,
and
that
poses
interesting
regulatory
issues.
If
you
go
that
way
in
the
in
the
case
of
networks
in
particular,
the
idea
is
that
well
the
idea
that,
as
I
said,
is
about
finding
a
way
in
which
an
efficient
and
we
can
rely
on
these
techniques
to
improve
the
the
life,
the
life
cycle
of
network
innovations
and
the
unshortened
the
time
to
market
in
in.
E
Oh
sorry,
I
have
a
problem
with
my
window.
I
need
to
fix
it,
okay,
so
the
the
way
in
which
in
which
we
experiment
with
the
network,
and
we,
we
experiment
with
new
revolutions
and
sorry
new
new
innovations
and
how
we
can
make
them
available
in
the
networks
sooner
so
this
graph.
This
document
starts
by
defining
and
what
I,
what
comes
in
the
in
the
in
the
following
slides,
is
precisely
about
the
the
contents
of
the
documents
and,
and
the
current
state
starts
by
defining.
E
What
is
a
digit
between
network
is,
is
a
a
visual
representation
of
a
real.
We
call
it
a
physical
network.
This
is
something
that
this
can
be
a
little
bit.
E
We
should
go
into
something
more
talking
about
real,
but
is
that
whatever
the
real
network
is
and
is
used
for
for
the
goals
that
a
digital
twin
is
used
and
is
basically
based
on
flows
of
on
the
flow
of
data
on
network
models?
That
is
something
that
in
which
the
itf
is
extremely
active
and
for
sure
on
opening
interfaces
in
which
as
well,
there
is
a
a
clear
work
inside
the
xf.
So
these
are.
These
are
the
the
metro.
E
Digital
training
is
in
charge
of
analyzing
diagnosis
and
diagnosing,
emulating
and
controlling
the
eventually
controlling
the
the
real
network
and
is
composed
of
four
essential
elements.
E
The
first
one
going
from
top
top
left
left
in
the
and
the
clock.
Clockwise.
E
The
first
one
is
about
the
the
model:
the
mechanisms
for
mapping
so
supporting
how
a
real
network
maps
on
on
a
particular
digital
twin
and,
eventually,
how
two
virtual
two
virtual
trains
can
engage
as
well,
because
we
will
see
that
one
of
the
goals
of
the
of
this
environment
is
precisely
to
have
several
choices
at
the
same
time
running
and
derive
some
conditions
for
it,
the
the
interfaces
and,
basically
the
interfaces
down
to
the
down
to
the
or
down
the
interfaces
towards
and
from
the
the
real
network
and
the
interfaces
with
any
other
applications
that
can
be
used
to
exercise
the
the
digital
thing.
E
Basically,
the
data
that
goes
through
those
interfaces
and
data
in
this
sense
is
that
we
have
to
to
think
in
in
the
in
terms
of
this
data.
That
flows
in
both
directions
is
telling
me
telemetry
data,
but
as
well
is
about
action
data
that
is
taken.
I
mean
the
the
information
of
the
actions
that
are
taken
by
or
decided
by,
the
network
digital
twin
and
how
they
they
flow
and,
finally,
the
models
and
how
how
we
perform.
E
On
the
one
hand,
the
mapping
between
the
real
network
and
the
virtual,
the
virtual
train
and
the
mobiles
that
we're
using
for
emulating
the
behavior
of
the
real
network
inside
the
digital
game
from
the
from
the
function,
the
functional
point
of
view
is
analyzing,
diagnosing,
emulating
and
controlling
the
the
goal
is,
as
we
were
discussing
before,
is
about
the
possibility
of
of
optimizing
the
decision-making
optimizing.
E
The
the
network
layout
optimizing,
the
way
in
which
we
deal
with
natural
footprints,
lowering
the
costs,
making
a
much
safer
or
having
a
much
safer
procedures
for
performing
the
assessments
of
of
which
are
the
the
innovations
that
can
be
applied
to
the
network
being
doing
these
with
the
regulatory
compliance
and
I'm
respecting
the
the
privacy
of
our
users
and
something
something
that
is
equally
important,
which
is
something
that
as
well,
is
really
expensive
and
complicated,
is
to
customize
training,
training
of
all
kind
of
of
intelligence,
natural
and
artificial
humans
and
machines.
E
What
is
important
as
well
is
that
the
digital,
tooling,
being
a
com,
composition,
software
modules,
basically
something
that
allows
us
to
orchestrate
and
derive
and
change
the
the
system
behavior.
So
we
can
adapt
and
we
have.
We
can
evaluate
how
the
network
would
behave
with
different
configurations,
and
we
have
we
have
the
possibility
of
doing
it
in
a
reputable
way.
That
implies
that
we
can
replicate
different
network
conditions
on
demand,
and
this
is,
for
example,
think
about
training,
a
machine
learning
which
you're
interested
in
a
particular
pattern
of
traffic.
E
That
is
quite
quite
unusual,
but
this
is
interesting.
The
possibility
of
precisely
reproducing
this
sorry
repeating
this
or
reproducing
the
making
it
reproducible
that
implies
that
we
can
apply.
E
We
can
reply
different
combination
of
events
and
under
control
conditions,
so
we
can
derive
a
knowledge
of
which
are
the
the
parameters
and
the
factors
that
influence
the
natural
behavior
and
the
certain
and
the
particular
circumstances.
E
This
is
so
far
the
the
architectural
framework
we
have
which,
for
those
of
you
familiar
with
the
with
the
sdn
architecture,
but
somehow
resemble
it,
because
we
have
a
physical
network
that
provides
data
and
receives
control
actions.
We
have
the
network,
digital
stream,
and
here
we
are
playing
with
the
idea
that
we
have
a
digital
team
network
and
a
network
digital
twin.
E
The
physical
network
is
is,
as
I
said,
is
the
real
network
is
that
all
the
elements
of
data
from
the
the
network
element
is
fed
into
the
network
digital
technology
through
the
cell
phone
interfaces.
E
What
is
important
is
that,
again,
is
that
we
very
likely-
and
this
is
something
that
we
are
elaborating
right
now-
is
how
we
support
with
the
data
infrastructure
infrastructure,
how
the
data
is
segregated,
normalized,
anonymized,
translated
into
into
an
ontology
making
an
assessment
of
the
provenance
because
of
the
of
the
trust
you
can
put
or
not,
on
the
on
the
different
data
sources.
E
This
is
this
is
essential,
so
what
we
have
is
that
we
can
dynamically
select
different
data
sources
and
put
more
focus
on
particular
aspects
of
the
of
the
of
the
network
for
for
the
different
tasks
of
the
of
the
twin
or
for
the
different
twins
that
are
associated
with
different
segments
or
different
kind
of
a
different
kind
of
network.
E
The
important
thing
is
that
we
we
have
not
circumscribed
to
a
full
network
to
end,
or
we
are
nothing
conscribed
to
purely
in
the
the
internet
case.
We
are
considering
access
as
well,
and
the
idea
is
that,
in
principle,
the
real
network
is
something
that
which
are
the
limits
of
the
of
the
real
network
is
something
that
depends
precisely
on
the
people
running
the
the
the
network,
digital.
The
sorry,
the
dtm,
the
digital
network,
to
decide
where
how
many
domains
it
would
cover
and
to
decide
if
the
integration
can
happen
at
the
twin
layer.
E
The
network,
digital
training
itself,
basically
is
a
data
repository.
That
is
the
apparent
so
to
say
of
these
data
infrastructure,
we
were
talking
about
a
set
of
mechanisms
or
providing
the
mappings.
E
I'm
doing
this
into
a
set
of
basic
models
related
to
the
network
topology
and
the
elements
and
a
set
of
functional
models
regarding,
I
don't
know
security
or
regarding
the
the
goals
of
the
execution
or
whatever,
and
the
entity
management.
That
is
precisely
charge
of
these
orchestration
aspects
that
we
were
mentioning
before
about
this
siding,
which
is
a
topology
which
are
the
elements
which
are
the
models
that
are
active
in
a
certain
moment
and
for
sure
the
security
requirements
and
the
general
orchestration
orchestration
of
everything
that
is
happening.
E
When
talking
about
the
potential
applications,
we
have
a
number
of
them
on
top
of
the
so
we
we
can
have
applications
that
are
focused
basically
on
management
applications
that
are
based
on
on
intent,
based
networking
applying
conventional
technologies
simply
because
we
want
to
to.
We
want
to
have
a
indirect
control
of
the
of
the
of
the
real
network
or
more
innovative.
In
the
case.
In
the
case,
we
are
experimenting
or
making
a
first
assessment
of
a
particular
technology.
E
E
The
the
idea
is
that
the
applications
come
through
an
arphon
interface
of
the
of
the
natural
digital
train
and
the
important
thing-
and
this
is
something
that
the
applications
can
be
eventually
running
in
parallel,
several
instances
of
the
of
of
a
network
digital
train,
to
make
a
precisely
make
some
choices
or
decisions,
something
similar
to
the
av
testing
installer.
E
These
are
a
few
sample
application
scenarios.
We
are
considering
in
terms
of
training
at
both
at
both
levels.
As
I
said,
the
humans
and
machines.
The
idea
that
a
certain
certification
be
a
a
digital
tree
network
can
provide
an
ideal
environment
for
testing.
You
know
in
a
devops
loop,
apply
applied
to
the
network
or
running
specific
experiments
in
something
that
well
people
term
us
as
networks
as
fasting,
in
which
you
can
use
it
for
see
what,
if
in
some
cases
and
make
make
decisions.
E
Finally,
and
just
for
you
to
know,
if
you're
interested
in
working
with
this,
it's
about
the
challenges
we
we
we
see
right
now,
do
we
foresee
in
the
btn
technology
one
is
regarding
the
larger
scale
issues
I
have
associated
with
something.
That
is,
I
mean
I,
I
hope
you
know
it
about.
The
burgess
is
a
paradox
about
this.
This
idea
that,
if
you
want
a
complete
map,
a
map
that
completes
reflects
reflects
a
particular
territory.
E
E
The
idea
of
a
heterogeneity
at
the
rigidity
is
here
at
the
rigid
network
is
here.
One
of
the
successes
of
the
internet
comes
precisely
because
it
found
a
common
denominator
for
a
huge
amount
of
of
heterogeneity
is
the
idea
that
is
about
identifying,
where,
like
the
the
correct
balance
between
so
over,
specifying
something
in
terms
of
architectural
interfaces
and
making
it
flexible
enough,
so
it
is
possible
to
to
to
interpret
the
data
modeling.
E
Let
me
assess
we
are
very
much
focused
right
now,
monitoring,
because
it's
the
let's
say
it's
the
easier
part
of
the
part
that
is
better
understood,
but
in
the
in
the
in
the
control
action
is
equally
important,
the
the
need
for
real-time
requirements,
so
that
implies
that
the
the
our
control
army
is
stupid,
so
supply
lines
are
longer
and
more
difficult,
and
that
implies
that
the
that
would
require
a
quite
agile
processing
inside
the
digital
twin
network.
E
This
is
one
of
the
reasons
why,
for
example,
a
data
infrastructure
that
is
able
to
combine
and
put
together
infra
data
into
something
closer
to
knowledge
is
is
important,
and
these
well,
these
requirements
will
in
impact
as
well
the
large
I
mean
they
are
translated
as
well
to
to
the
or
translated
into
the
same
kind
of
issues
as
we
have
with
the
scale,
the
natural
scale
and
finally,
security
risk.
I
mean
the
same
time.
E
The
same
way
we
have
a
big
brother,
many
seeing
us
behind
the
all
the
network
facilities
we
have.
This
big
twin
issue
as
well
is
is
a
control
point
that
is
extremely
appealing
for
attack.
E
In
the
moment,
you
have
a
digital
twin
network,
running
in
parallel
with
the
real
network
and,
as
I
said
before,
privacy
preservation
in
all
sense
privacy
and
commercial
secrecy,
etc,
is
something
that
can
become
critical
here,
because
it's
a
it's
an
enormous
consumer
of
data
of
really
relevant
data
that
has
to
be
preserved,
and
with
this
I
mean
with
this
list
of
challenges
that
I
hope
will
encourage
you
to
to
collaborate
with
us
on
this
think
I'm
finished.
Thank
you.
A
B
E
Well,
this
is,
I
mean
this
list
of
challenges,
probably
not
exactly
expressed
as
this,
because
this
is
a
little
bit.
My
view
are
included
currently
on
the
draft.
Some
of
them
will
they
will
have
to
address
at
least
well
to
identify
which
certain
research
directions.
E
I
don't,
I
don't
think
that
we
will
be
in
the
position
of
of
solving
all
of
them
within
the
draft.
This
is
something
that
we
would
like:
love
to
see,
collaborations
and
probably
other
documents
more
detail,
or
probably
considering
particular
aspects
that
are
of
special
relevance,
so
they
are
considering
the
draft.
A
In
your
vision
of
the
digital
twin,
I
have
the
impression
that
mostly
it's
relative
to
the
data
that
you
can
take
as
input
of
the
digital
twin
or
maybe
you
can
also
put
given
input
events
and
so
on.
A
Do
you
think
that
digital
twin,
if
you
have
a
model,
so
can
serve
also
as
and
knowing
what
are
the
boundaries
somehow
the
data
related
to
a
particular
particular
function
of
the
system,
so,
rather
than
just
waiting
for
data
and
check
if
the
system
is
working
according
to
a
certain
function,
you
can
already
know
in
advance
what
would
be
acceptable
data.
E
No,
no,
for
sure
I
mean,
if
you're
talking
about
making
predictions
for
sure.
This
is
what
this
should
be
one
of
the
goals
of
of
a
digital
training
and
in
parallel,
something
that
we
are
experimenting
as
well
and
in
particular
what
I
mean
with
the
same.
We
is
my
mia
myself,
and
my
team
is
precisely
with
the
idea
of
injecting
having
the
something
that
is
close
as
close
as
possible
to
a
digital
trim.
E
Probably
we
are
not
that
we
are
not
there
yet,
but
something
that
is
very
close
to
a
digital
train
and
our
idea.
What
we
are
doing
is
injecting
synthetic
traffic
under
control
conditions
that
are
generated.
For
example,
there
are
a
group
from
a
university.
Madrid
is
working
with
us
in
running
guns.
You
know
these
adversarial
networks
that
make
a
real
the
behavior
of
a
real
group
of
users
of
real
segments
of
the
network,
injecting
it
in
the
twin
and
trying
to,
in
some
cases,
get
a
response
evaluate
a
particular
strategy
address.
E
A
Okay,
I
think,
can
be
nice
to
discuss.
I
was
not
thinking
about
not
pre,
somehow
not
really
predicting,
but
more
somewhat
evaluate
what
will
be
the
acceptable
data,
so
not
testing
data
in
evidence
but
know
by
the
model.
What
would
be
the
somehow?
Maybe
the
values
that
that's
higher
acceptance
input?
I
mean
maximum
minimum.
E
Values
that
make
sense
or
not
yeah,
yeah,
yeah
sure
sure
as
well.
I
mean
this
is
something
that
another
module
could
be
used
for
I
mean
once
you
have
something
that
is
able
to
collect,
able
to
mimic
natural
behavior,
then
that
allows
you
to
precisely
experiment
and
try
to
to
identify
patterns,
clusters
of
data
or
whatever.
Yes,
yes,
why
not?
E
H
Just
here
is
franco
from
bologna.
I
work
with
walter
that
I'm
sure
you
you
know,
because
it's
part
of
the
group
we
have
a
national
project
in
which
we
did
something
on
a
very
similar
concept
related
to
industrial
applications.
Since
our
area
is
very
manufacturer
oriented,
if
you
don't
mind,
I
will
send
you
some.
I
I
to
be
honest.
I
don't
understand
if
the
approach
we
followed
can
be
somewhat
the
framed
in
the
very
good
presentation
you
just
give.
Thank
you.
Thank
you
very
much.
By
the
way
it
was.
G
H
Would
be
very
interesting
to
see
whether
we
can
frame
that
approach
into
this
one,
because
we
already
have
a
sort
of
written
example
deployed
in
which
the
basic
concept
looks
the
same.
But
I
think
it's
probably
nice
to
to
talk
about
it.
A
little
bit.
E
That
would
be
lovely,
because
I
mean
we
are
here
exploring
something
that,
for
us
is
a
completely
uncharted
territory
but
well,
it
would
be.
It
would
be
quite
interesting
to
to
get
some
some
inputs
from
people
that
have
been
experiencing
this
with
these
ideas,
even
even
in
the
case,
this
has
not
directly
connected
with
network
management,
but
this
some
of
the
ideas
on
the
results.