►
From YouTube: IETF115-NMRG-20221107-0930
Description
NMRG meeting session at IETF115
2022/11/07 0930
https://datatracker.ietf.org/meeting/115/proceedings/
A
A
A
A
So,
let's
start
the
meeting
and
with
the
first
of
all,
of
course,
the
not
very
intellectual
property.
So
you
must
know
that
the
area
follows
the
ITF
intellectual
property
right.
Let's
discuss
your
rules,
so
you
can
here
read
the
different
rules,
which
is
important
to
note
that,
if
some
contribution
you
would
like
to
share
that
are
that
are
covered
by
your
patent
or
application
now
patent,
you
should
disclose
it
before
in
a
timely
manner.
A
A
Of
course
you
can.
You
can
have
this
red
to
network
like
landyard.
So
if
you
do
not
wear
it,
you
can
send
to
our
prayer
in
Social
coding
and,
of
course,
if
you
speak
in
a
microphone
or
if
you
even
online
or
if
you
put
your
video
online
or
either
on
site
online,
you're
speaking
microphone,
you
are
concerned
to
be
recorded.
A
B
A
A
A
Why
the
artificial
informational
or
expansion,
experimental
documents
in
geography
series
is
primary
goal
is
to
promote
development
of
research,
collaboration
and
teamwork
and
spring
research,
security
to
internet
protocols,
application
architecture
and
Technology.
Another
area
of
the
RFC
74
are
18
to
that
refers
to
that.
A
Here
is
a
link
for
the
for
the
first
meeting.
What
is
important
and
what
is
I
think
written.
The
slide
is
that
even
if
you're
on
site
is
importance
that
you
connect
to
mythical,
because
this
will
be
necessary
to
collect
the
blue
sheet,
it
will
be
done
by
this
way,
so
very
important
that
you
collectors
automatico,
is
not
done
yet.
A
So,
let's
move
on
now
more
to
the
content
of
the
meeting
so
we'll
be.
There
will
be
a
brief
introduction
here
regarding
the
searches
of
the
group,
so
from
the
last
meeting
that
we
have
declared
are
the
two
I
receive
that
was
in
the
queue
regarding
IBM
that
have
been
published
so
the
first
one
which
is
RFC
9315
about
Indian
Bay,
networking
concept
and
definition
and
rrc
93
16,
with
intent,
classification.
A
Or
those
are
the
bunch
of
documents
which
are
run
in
the
group
for
let's
say
for
a
while,
quite
stable,
and
maybe
there
will
be
for
some
of
them
for
one
of
them.
There
has
been
a
call
for
adoption
before,
but
we
think,
as
also
as
as
the
chairs
of
the
group,
that
there
will
be
potential
upcoming
call
for
Redemption,
for
this
document
is
still
here,
the
first
one
about
Network
measurement
interns
around
about
IDF,
Network,
size
intent
and
the
last
one
about
interconnection.
A
Intents,
the
question
also
because
this
documents
these
documents
will
be
presented
today.
The
question
will
be
all
with
the
evolved
towards
are
having
this
document
and
adopted
and
maybe
and
then
to
to
propose
you
for
possible
publication
as
races
to
have
a
singular,
multiple
use
case
documents.
This
question
is
still
open
and
although
there
are
documents
that
are
possible
coming
references
will
be
about
the
research
challenges
in
artificial
intelligence
for
Network
management,
so
we'll
be
under
the
presented
today
as
well.
A
Although
we
would
like
to
also
like
to
to
share
with
you
the
maybe
the
to
recall
the
research
agenda
of
an
emergency
that
you
can
also
find
on
the
data
tracker,
one
major
Rock
item
of
the
group
was
related
to
IBM.
As
you
know,
know
we
have
as
a
touristy
that
has
been
published.
We
have
the
use
cases
and
the
questions
that
would
as
a
group
if
you
want
to
pursue
this
work
on
other,
let's
say:
work
items,
for
example,
architecture
on
that
it
was
initially
easy.
A
The
objectives,
but
after
submitting
there
was
no
so
much
content
about
that,
but
it's
still
open
and
so
we'd,
like
also
to
have
your
feedback
with
our
indict
in
the
discussion
of
course,
I
kind
of
rather
a
new
topic
in
the
group
like
the
AI,
Network
and
recently,
more
recently
green
networking.
So
the
it's
so
also
the
question
will
be
show
these
different
topic
continue
to
be
part
of
an
MLG
should
be
somewhere
else
for
them
the
right
place
for
this
topic
or
not
for
for
future
works.
A
A
C
C
So
as
a
reminder,
the
idea
would
be
to
leverage
Network
Technologies
for
requesting
IDF
number
slices,
so
the
use
case,
one
of
one
of
them-
could
be
the
the
sonar.
Persistent
is
requesting
and
slice,
and
it
will
the
IDF,
let's
say,
capabilities,
controller
and
on
this
very
immediate
use
case,
could
be
the
3dp
management
system.
For
instance,
we
will
recommend
later
on
the
idea
of
a
POC
on
this
respect,
so
essentially
through
the
intent,
we
will
have
a
declarative
intention
of
highness.
B
C
And
this
declarative
intention
will
be
translated
to
what
we
expect
to
be
translated
to
a
kind
of
an
MBI
that
could
be
supported
by
a
NSE
controller
and
so
on
so
far,
so
the
benefits
of
going
in
the
direction
of
Indian
base
will
be
the
the
portability.
So
we
could
have
a
solution
that
could
be
consumed
by
different
implementations
later
on
in
the
control
part.
C
Also
simple
way
of
expressing
transmission
size
needs,
so
the
clarity
wave
expressing
the
slices
and
so
and
as
any
kind
of
internal
system
to
focus
on
the
world
and
not
on
how
right
the
Hub
will
be
later
on
being
decided
by
the
ITF
number
or
less
control
in
this
case.
So
we
see
this
as
a
compliment
for
the
this
for
the
working
group.
Sorry
for
the
slice
work
in
this
working
group,
so
complementing
all
that
stuff,
all
those
ideas
of
nevertheless
controller
with
an
intent-based
system.
C
C
In
the
upper
part,
in
order
to
do
the
an
easy
translation
from
the
declarative,
slice
intent
towards
the
MBA
young
model
and
also
we
have
provided
details
on
the
Assurance
phase
of
the
for
the
internal
life
cycle.
So,
as
you
can
see
there,
we
were
well
in
in
the
previous
section.
We
propose
the
Fulfillment
part,
so
we
are
here
complementing
with
the
Assurance
part.
C
We
have
on
the
monitoring
side,
the
collection
of
monitoring
information
related
to
the
slice,
so
whatever
is
has
to
deal
with
the
slos
and
slice
slice
that
are
respected
for
the
for
their
slice
in
the
media.
In
the
middle
part,
we
will
have
the
aggregation
of
this
information,
this
monitoring
data
data
that
could
fit
the
validate
component.
That
will
help
us
to
to
realize
the
inner
closed
loop
for
controlling
the
the
intent
in
the
provider
side
and
also
on
on
your
left.
C
You
can
see
the
the
components
for
a
providing
the
report
to
the
to
the
customer,
so
so
how
extracting
the
monitoring
information,
the
monitoring
and
Telemetry
information
in
such
a
way
that
the
customer
can
receive
feedback
from
the
behavior
of
the
slides
and
then
the
customer
can
perform
the
outer
close
loop
for
yeah
for
fixing
whatever
problem
that
could
be
with
the
slides.
So
next,
please.
C
We
are
targeting
a
POC,
so
essentially
we
understand
we
have,
let's
say
like
five
main
components
to
address
the
first
one
will
be
the
translation
from
this
upper
system,
so
take
it
into
consideration.
Information
that
could
come
for
a
in
this
case
in
the
in
this
particular
example
from
the
3bp
system,
so
understand
the
kind
of
objects
that
VPP
system
can
propose
to
the
intent-based
system,
so
taking
them
and
translating
them.
This
would
be
the
component
one.
The
component
two
will
be
essentially
the
mapping
to
the
normal
interface
of
the
network,
size
controller.
C
So
for
that
we
are
also
running
some
parallel
work
in
this
working
group.
We
have
the
I
I
have
put
that
a
couple
of
drafts,
one
the
first
one
in
in
different
parties,
essentially
or
we
intend
to
describe
there-
the
mapping
of
the
information
from
three
Epp
to
what
would
be
the
mbir
model
and
the
wall
at
the
bottom
is
essentially
the
realization
with
current
VIP
and
TLS
Technologies.
So
the
idea
is
to
leverage
on
those
drafts
for
let's
say,
accomplishing
the
Fulfillment
part
for
the
Assurance
part
that
will
correspond
to
a
component
number
three.
C
We,
the
idea,
is
to
leverage
on
on
some
other
work
in
ATF,
maybe
the
the
service
Assurance
architecture
that
is
being
proposed
in
in
noksa
also,
and
then
we
will
have
the
component
4
and
the
component
five.
The
component
four
sincere,
will
be
to
report
to
the
CBB
management
system,
the
behavior
of
the
slides,
so
the
reporting
some
kpis
essentially
and
the
number
five
will
be
the
validation
of
those
kpis
for
performing
the
inner
close
loop,
Control
Plus
Loop.
So
next,
please
what
sorry,
the
other.
C
The
question
I
put
also
this
question,
maybe
for
discussion.
If
we
should
document
the
fully
the
use
case
in
the
in
the
in
this
draft
or
essentially
pointing
out
to
other
work
in
these,
and
so
so.
This
is
a
question
maybe
for
for
discussion
next,
please
so
key
questions
from
the
chairs.
The
chairs
propose
us
to
solve
to
answer
a
number
of
questions,
so
the
objective
of
the
work
was
one
of
them,
so
yeah.
This
is
clear.
The
Our
intention
is,
to
elaborate
intents,
that
could
facilitate
the
request
of
ITF
Network
slices
so
complementing.
C
with
that,
we
would
have
the
ultra
closed
loop,
more
or
less
or
some
kind
of
proof
of
concept
in
that
respect
and
yeah
trying
to
work
on
the
component
five.
This
is
for
further
analysis.
Probably
we
need
more
capabilities,
maybe
the
enormous
artificial
intelligence.
I,
don't
know.
We
need
more
time
for
for
analyzing
that
so
next,
please-
and
this
is
the
the
final
one
so
yeah
we.
A
C
A
lowest
level
be
part
of
the
translation
process
because
could
be
the
case.
I
mean
what
what
the
network
will
be
able
to
manage
our
slos
related
to
the
network,
but
we
are
not
so
sure
what
could
be
the
the
kind
of
I
mean
the
sellers
from
the
perspective
of
the
customer.
If
we
take
the
three
Epp
example
probably
could
be
similar.
C
A
B
D
So
I
think
this
is
a
very
nice
use.
It's
a
very,
very
nice
topic,
but
I
do
have
one
question
concerning
nmrg
versus
standardization,
so
the
question
is
busy.
How
much
do
you
think
is
a
research
type
of
agenda
versus
how
much
might
be
ready
for
civilization
you're
talking
about,
but
it
would
be
an
IDF
as
well?
Can
you
perhaps
common.
C
Thanks
for
for
the
question,
certainly,
there
will
be
aspect
more
related
to
research,
for
instance
all
the
control
loops.
This
is
probably
more
more
focused
on
research.
The
pure
translation
itself
is
maybe
it's
not.
There
is
not
so
much
research
in
that
respect.
Maybe
there
could
be
areas.
We
commenting
the
class
now
the
translation
from
services
and
laws
to
network
slos,
so
some
ideas
of
how
how
to
digest-
let's
say
the
service
Expressions,
to
relate
to
transform
to
network
Expressions.
Maybe
that
could
be
much
more
related
to
to
research.
D
C
How
is
to
offer
the
intent
in
a
general
manners,
the
relation
with
TM
Forum
run
and
some
other
apis
on
top
is
not
I
mean,
probably
could
be
even
on
top
of
the
upper
system
that
will
request
an
average
lies,
so
maybe
not
data.
It
could
be
that
there
is
no
direct
relation
between
the
TM
for
an
apis
and
the
idea
of
memorized
stuff.
A
C
Was
a
reminder,
so
the
idea
is
to
leverage
on
internet-based
Technologies
for
handle
this
enrich
interconnection
as
analysis
of
applicability.
We
listed
there
a
number
of
them
that
could
be
others
for
sure
could
be
the
interconnection
of
non-public
Network
to
public
networks
in
5G
or
multi-domain
level
as
a
service
request,
as
in
section
4.4
of
RFC
85688
or
multi-domain
network
virtualization,
like
in
the
data
I
reference
there,
there
could
be
others
as
well
for
sure
so
the
modes
of
usage
of
this
interconnection
intent.
We
can
just
simply
stay
with
a
IP
traffic
integration.
C
C
C
So
it
could
be
maybe
a
very
good
topic
for
having
interpretability
in
in
the
hackathon
in
iti118.
So
next,
please
yeah
for
moving
forward.
They
drive
ideas.
The
idea
will
be
to
define
a
more
detail,
the
structure
of
the
intent
that
the
template,
the
model,
the
the
to
model-
let's
say
the
audio,
the
objects
for
interchange
between
the
peers
and
to
identify
protocols
or
apis
or
the
lack
of
them
for
accomplishing
the
different
kind
of
interconnection
types
considered.
C
So
not
only
happy
Traffic,
Service
function,
resources
and
so
on
so
far,
so
the
feedback
is
more
than
welcome
for
sure,
and
there
is
just
one
reminder
in
this
life
Jerome,
because
this
is
a
question
for
the
research
group.
We
have
been
in
contact
with
the
authors
of
the
measurement
intent
with
colleagues
that
will
talk
after
me
and
and
we
have
common
interests
in
in
the
in
the
drugs
of
each
other.
So
we
have
two
options
for
progressing
the
work
either.
C
F
C
F
I
bring
this
up
is
during
the
ITF
Network,
slides
discussion.
There
were
this
talk.
Also
about
like
you
know,
I
might
need
to
create
more
logical
interfaces
as
I
move
things
and
I
would
like
to
do
it
in
a
much
simpler
way.
So
I
think
the
what
you're
proposing
here
could
be
used
even
for
doing
that,
so
it
might
be
worth
expanding
it
a
little
bit
and
including
that.
A
Oh
okay,
I'm,
sorry,
you
didn't
get
the
name
so
for
maybe
as
a
maybe
as
a
terrible
question.
I
think
are
the
two
options
that
you
are
mentioning
I
think
the
it's
hard
to
just
to
give
you
an
answer
like
that.
Maybe
it
should
be
good
that
he's
user
will
tell
you
sure
what
are
the
common
part,
maybe
in
another
meeting
what
should
be
merged
and
what
should
be
in
commands
to
to
figure
out
what
exactly?
What
would
be
the
mirror?
A
B
B
G
Hello,
everyone
I
am
currently
from
China
mobile
and
this
time
I'll
present
about
Ami,
Network,
environment
intent.
So
two
working
items
have
been
published
as
5Cs,
namely
the
concepts
and
definition
of
IBM
classification
of
IBM,
so
I
think
in
The
Next
Step.
There
is
a
main
focus
on
studying
the
use
case
of
IBN.
What
exactly
IBN
can
bring
in
what
kind
of
season
rails.
G
So
that's
basically
why
we
propose
a
typical
use
case
of
IBM
Network
management
intent,
which
we
think
is
a
very
important
and
practical
use
case
of
IBM
and
after
several
updates
of
the
draft,
we
have
received
very
valuable
suggestions
from
juices
and
chain
from
Huawei
and
Lewis
from
telefonka,
and
we
all
think
that
Airline
is
well
worthy
of
studying
in
Mig
and
we
are
keeping
for
looking
forward
to
more
suggestions.
So
next
slide.
Please.
G
So
here
is
a
very
brave
recap
of
the
draft
in
the
recent
update.
So
the
introduction
about
nmi
Ami
is
to
use
a
network
measurement
intent
to
achieve
a
network
performance
aggregation
and
from
user
or
network
administrator
intent
to
verify
whether
a
network
measurement
result
meet
the
environment
content
and
further
improve
the
equity
of
the
configuration
in
the
network.
So
and
this
time
we
have
make
a
update
of
the
relationship
with
two
offices,
that
is,
the
concepts
and
and
definition
of
IBN
and
the
classification
of
IBM.
G
G
So
it's
static,
Mi
is
about
the
the
measurement
purposes
remain
unchanged,
and
it's
independent
in
the
network
State
about
Dynamic
Mi.
Is
it
changes
dynamically
according
to
the
network,
State
environment?
So
these
These
Are
The
Brave
recap
of
the
draft
and
the
recent
updates.
So
next
time
at
least.
G
So,
after
the
last
time,
the
calling
for
adoption
procedure
we
have
received
about
total
of
nine
members
from
different
entities
to
support
the
draft.
We
have
solved
questions
raising
the
main
release
and
we
have
updated
the
references
of
other
offices
and
the
working
group
documents
and
the
next
slide
please
so
for
the
next
step.
Just
like
Luis
mentioned
in
his
draft
interconnection
intent,
so
different
personal
drafts
focused
on
IBM
use,
cases
have
been
proposed
to
nmrg.
We
can
either
merge
them
into
a
whole
working
group
document
of
IBM
use
case.
G
A
Okay,
so
so
maybe
for
again,
this
is
a
a
bit
similar
that
the
the
question
before
about
the
the
emerging
at
work,
if
I
think,
if
I
am,
if
I'm
interested
in
a
while
you
you
are,
there
is,
as
I
said
in
the
introduction,
this
option
for
emerging
type,
one
single
document
for
all
living
use
case
or
not
I
think
maybe
it's
a
Jefferson
you
can.
You
can
pick
a
little
bit,
I
think
they're
easy.
A
This
is
still
open.
The
main,
the
main,
let's
say
concern
will
be
that
to
identify
command
structure
and
template.
I
think
Jefferson
started
to
do
so
and
then
a
bit
of
the
question
to
to
those
to
all
this
author
working
and
use
case,
the
community.
This
structure
is
okay,
you
are
who
they
would
like
to
to
change
it,
or
maybe
they
see
that
it
will
not
possible
to
have
a
common
use
case
like
a
common
document,
it
will
be
just
adjust.
A
H
You
know
what
I
think
that,
maybe,
if
because
we
we
had
this,
this
template
for
for
the
document.
So
if
the
authors
of
the
these
drafts
can
you
know
address
that
like
saying
that
it's
okay
or
you
can
maybe
change
the
template,
it
would
be
nice.
Otherwise,
in
the
the
next
meetings
we
can
have
presentations
pointing
out
the
the
similarities
you
know
like
to
to
show
if
it's
it's
possible
at
least
because
I
think
it
I
also
agree.
B
A
A
Sawyers
again
just
a
summary
of
have
the
outline
of
the
draft,
so
maybe
just
as
a
as
a
remaining
disease.
So
the
second
version,
but
of
course,
if
you
remember,
if
you're
participating
to
the
to
the
group,
you
find
the
discussion
we
have
this
share
document
that
you
work
on
before
it
was
in
in
the
in
the
format
of
the
draft.
A
A
A
So
from
from
last
time,
we
received
a
few
comments,
so
so
we
basically
change
the
document.
According
to
them,
we
agree.
We
complete
the
document
to
cover
some
issues
or
some
Concepts
that
were
not
covered.
Well.
In
particular,
we
refer
to
the
data
mesh
concept,
I
think
other
than
the
I
need.
The
concept
which
it
was
very
interesting
to
highlight
here
is
that,
in
the
previous
version
realized
that
there
are
a
different
granularity
of
datas
or
a
different
volumetry
different,
it's
a
dynamically
change.
A
The
data
that
come
at
different
time
and
so
on.
One
important
thing
that
we
did
not
really
highlight
is
that
okay,
there
are
different
type
of
database.
They
are
natural
by
the
way
they
are
provided,
but
when
you
have
to
aggregate
this,
data
as
I
might
not
be
aligned
in
terms
of
time
as
well,
and
this
can.
This
can
also
raise
some
issue
and
challenges.
So
it's
not
just
only
via
real-time
data,
or
we
have
data
that
arrives
very
fastly.
A
Some
can
be
very
fast
or
not,
and
so
sometimes
we
need
to
still
align
them
next
slide,
please.
So
for
the
next
step.
What
we
would
like
to
to
have,
of
course,
I
have
more
feedback
from
the
group
and
also
maybe
I
have
a
conversation
that
we'd
like
to
ask
for
this
in
the
in
the
next
and
the
next
step.
We
have
already
identified
possible
addition.
We
don't
know
if
we
need
to
include
already
then
or
if
you
need
to
include
at
all
regarding
lightweight
yeah
within
distributed.
A
Ai
is
not
well
integrated,
though
there
is
some
discussion
about
the
impact
of
level
versus
unlably
data.
Also,
some
discussion,
I
mean
internally
between
the
let's
say
the
the
group
of
water,
but
having
some
more
content
about
as
a
legal
or
regulatory
product
of
let's
say
using
AI
for
Network
management.
A
What
do
you
think
is
that
we
cannot
be
exhaustive
anyway,
so
we
try
to
figure
out
what
would
be
the
most
valuable
challenges
for
the
group
in
the
documents
that
we
put.
Of
course,
all
your
feedback
is
welcome
to
be
sure
that
not
that
we
are
not
validating
something,
because
I
think
it
would
be
very
hard
to
be
any
way
to
be
exhaustive
and
might
not
be
the
call
here
and
also
an
additional
Next
Step
will
be.
That
know
that
we
are
this
document.
That
is,
we
think
that
is
quite
stable.
A
A
Try,
maybe
maybe
another
possible
approach
will
be
try
to
try
to
identify-
and
maybe
some
kind
of,
please
what
has
been
proposed
in
nature
Returns
on
to
solve
these
challenges
at
least
partially
to
just
also
other
than
just
purely.
We
have
already
references
of
friends
and
around,
but
rather
than
purely
just
the
the
challenges,
maybe
some
kind
of
let's
see
living
document
where
we
say:
okay,
there
is
new
Solutions,
that's
been
proposed,
I,
don't
know
here
or
in
this
conference
or
by
this
tool
and
so
on.
A
So
that
can
fulfill
the
challenge
and
can
be
also
a
goal
of
of
the
group
to
maintain
this
as
well.
Of
course,
maybe
in
some
some
more
let's
say
Drive,
which
are
more
focused
and
certain
Challenge
and
I
suppose,
then
we
have
a
presentation
about
that,
and
that's
all
for
me.
So
if
you
have
any
question
or
comment.
C
I
A
I
Forgot
to
kick
the
to
click
the
the
button.
This
is
a
developer.
It's
when
you're
talking
about
the
data
challenges,
Etc
I
I
there
I
miss
a
couple
of
things.
One
is
the
suitability
of
unsupervised
environmenting
in
networks
because
of
the
stability
concerns.
Another
like
this
is
one
thing.
The
second
and
I
believe
is
something
that
is
critical
in
the
application
available
to
networks
is
precisely
distributed,
AI
or
not
only
the
Federated
models,
but
anything
that
is
implied
by
the
using
localized,
Ai
and
knowledge
sharing.
A
We
think
so
that
maybe
what
was
missing
at
least
maybe
in
the
group
of
photos,
somebody
that
is
very-
let's
say
this
treaty-
that
what
Jesus?
So
let's
say,
let's
say
2012
is
a
Federated
AI.
You
are
looking
for
some
help,
so
if
in
either
room
is
volunteer
to
help
us
for
this
part
in
the
document
will
be
very
happy
to
have
it.
J
J
J
J
We
support
divide
the
section
4
to
section
4.1,
section
4.2
and
the
section
4.3,
the
section
4.1
is
to
order
functional
characteristic
of
the
hardware
and
section
4.2
is
the
characteristic
of
the
air
model
and
sections
4.3
is
the
characteristic
of
the
communication
method
and
we
add
one
reference:
that
is
the
etsa
document.
The
title
is
overlay
Computing
Market,
acceleration
me
symmetrics
best
vertex
and
guideline
groups.
Specification
ETI
is,
let
me
see
iez006
plus
1.1.1,
and
we
invite
to
export
from
etli
next
page.
J
Please,
okay-
or
this
is
the
motivation
of
this,
the
previous
little
chair
or
present
or
on
challenge
document
so
based
on.
We
are
investigate
the
development
of
AI
Services,
because
now
the
focus
is
moving
to
training,
to
influence
from
learning
through
prediction.
So
you
know
for
influence
not
only
high
performance
server,
but
also
small
Hardware
microcontroller,
low
performance,
CPU
and
AI
chipset
are
the
optimal
Target
device.
Provision
is
to
you,
know
the
cost,
so
we
are
considered
the
computation
of
the
system
in
terms
of
AI
influence
services,
not
of
training.
J
If
you
think
about
Target
device,
we
can
choose
a
local
device,
edgy
device
or
Cloud
Server,
and
if
we
find
Watchers
the
objectives
objective
of
the
air
model,
it
can
be
diversed,
for
you
then
put
accuracy
latency
for
real-time
services
and
electrostatic
resource
utilization,
Etc,
so
I
found
some
consideration
for
the
system
for
AI
Implement
services
or
one
is
a
model
and
the
other
starting
grammar
communication
method,
device
capacity,
important
data
Etc.
So
the
last
sentence
is
the
main,
the
purpose
of
draft.
J
We
want
to
accelerate
the
study,
AI
issue
and
find
some
possible
standardization
items
in
the
MLG
that
space.
Please
yeah,
so
you
know
I
guess
you
are
so
familiar
with
the
generic
procedures
or
AI
Services.
The
first
one
is
data
collection
store,
and
the
next
stage
is
to
data
analysis,
property
process
and
the
other
next
stages:
AR,
model
training
and
Air
Model
deploy
and
implants,
and
the
final
stage
is
Monitor
and
the
maintain
accuracy.
J
J
Next,
please
also,
we
can
combine
between
the
ETI
device
was,
and
the
Cloud
Server
for
router
Theory
or
task
operating
issue
next
page
yeah.
This
is
the
main
part
of
our
draft
to
the
first.
The
Contrition
is
to
functional
characteristic
of
the
hardware,
so
indoor
or
it
has
a
document,
so
it
describes
various
metric
which
can
be
a
potentially
be
improved
through
deploying
or
service.
On
our
MHC
platform,
MSC
means
the
mobile
Edge
Computing,
so
it
can
be
identified
in
order
to
highlight
the
benefit
of
deploying
a
machine
for
better
service
and
application.
J
So
the
first
is
the
functional
metric.
It
includes
latency
and
its
efficiency
through
food.
Good
food
lose
weight
Etc,
so
in
the
other,
Metro
is
non-functional
metrics.
It
includes
service,
drives
car
life
cycle
service,
update
Etc,
so
I
guess
that,
if
possible
in
the
MLG,
we
can
follow
his
the
approach
to
find
the
relevant
metric
for
the
AI
development
services.
J
J
Model
so
in
the
distant
Cloud
Server
it
provides
high
accuracy
short
different
time,
but
it
has
a
long
delay
to
transmit
dual
device
service.
It
provides
a
medium
accuracy,
medium
influence
time
and
the
medium
delay
to
transmit
in
a
local
machine.
It
provides
low
accuracy
long
interest
time,
but
it
has
the
short
trade
to
transmit
the
experience
yeah.
So
we
can
sum
the
period
from
the
Google
accounts
Pro.
It
shows
the
trade-off
between
water
size
accuracy
as
a
latency.
As
you
can
see,
we
can
paint
the
optimal
model.
J
So
it
is
the
characteristic
of
the
air
model.
So
in
usually
in
the
Cloud
Server,
we
deploy
the
heavy
Air
Model.
It
provides
high
accuracy,
but
it
has
a
big
file
size
and
the
report,
and
it
has
a
long
important
time
in
their
HD
device.
It
provides
video
utilized,
medium,
Air,
Model
and
the
medium
accuracy,
medium
size
and
medium
Implement
time
and
in
local
machine,
usually
utilize
right
to
create
Air
Model
and
it
provides
the
low
accuracy,
but
it
has
small
size
and
it
has
the
short
imprint
time.
J
J
J
So
it
is
some
kind
of
the
results
of
those
object,
detection
services
and
local
machine
HD
device
and
Cloud
Server
next
page.
Please
so.
I
also
request
to
the
church
to
or
share
our
opinion
for
the
relationship
to
challenge
document.
So
I
believe
that
the
challenge
document
is
the
main
document
for
handling
AI
issue.
So
our
ref
is
also
relates
to
the
chilling
document,
and
some
tests
can
be
added.
J
One
boarded,
for
example,
distributed
AI,
Services,
lightweight
AI
service
and
deployment
AI
services,
or
we
just
our
trip-
can
be
deployed
as
a
different
document
to
focus
on
AI
influence.
The
development
AI
Services,
because
the
existing
challenge
document
include
many
items,
so
we
can
focus
on
some
of
the
specific
area.
Okay
next
place.
So
that's
all
thanks.
B
D
Actually,
how
much
your
draft
is
really
more
about
system
management
versus
Network
management.
A
lot
of
the
challenges
that
you've
described
seem
to
apply
more
okay.
What
do
I
have
on
the
server
the
less
spots
on
the
network,
for
instance,
also
some
related
challenges
to
deploy
points
and
things
like
fed
rated
learning
or
transfer
learning
across
a
network.
It's
about
I
think
these
are
somewhere
where
some
of
the
real
challenges
also
lie
in
that
in
that
space,
I
don't
see
the
mentioned
in
there
in
their
draft,
so
I'm
wondering.
D
J
Yes,
you
know
right
thanks
for
your
application
command.
Yes,
you're
right
in
the
in
the
current
document
is
focusing
on
the
System
point
of
view,
but
in
the
animals
and
in
item.
If
here
we
are
finding
some
of
the
network
point
of
view,
so
this
is
our
homework,
so
we
will
find
the
two
revised
in
the
next
person
to
include
your
point.
K
J
K
J
You
are
right,
yes,
so
in
our
experiment
we
have
those
similar
approach.
We
have
the
multiple
active
device
so
take
the
test
off-roading
and
find
some
some
the
optimal
way.
How
to
do
so.
Yes,
you're
right,
we
can
not
only
vertical
area,
but
we
also
find
the
horizontal
area
for
the
distributed
approach.
Would.
K
H
I
have
a
comment:
Jefferson
I
think
it
would
be
nice
to
have
to
have
parted
out
the
the
same
dieties
that
you
see
among
the
documents.
You
know
like
the
one
that
you
presented
and
the
research
challenge.
In
order
to
see
you
know
how
big
are
the
difference
between
and
what
can
be
merged.
It's
more
like
a
comment.
H
L
L
Generalizes
the
concept
of
a
performance
digital
twin,
so
we
introduced
the
term
performance,
oriented
digital
twins.
It
reframes
the
originally
presented
NDT
as
something
specific
to
packet
Networks
and
introduces
an
optical
performance.
Digital
twin
opt
that's
something
specific
to
Optical
transmission
networks
as
a
second
example
of
podt.
L
L
Okay,
so
again,
going
back
to
the
first
draft,
the
performance
oriented
digital
twin.
The
idea
is
to
predict
or
replicate
pick
your
verb
performance,
behaviors
of
a
physical
network.
Of
course,
the
particular
behaviors
of
Interest
are
determined
by
use
case
requirements,
but
they
must
be
rendered
by
available
and
appropriate
functional
models.
L
We
introduce
here
the
notion
of
a
so-called
Network
information
interface
nii.
This
is
optional
or
or
present
case
specific,
and
it's
really
a
twinning
interface.
It's
where
we
can
absorb
data
from
the
physical
Network
and
or
management
plane
about
the
Network
Services
traffic
environmental
conditions.
L
What
have
you
really
all
the
information
we
would
require
to
construct
on
an
ongoing
basis,
say
a
faithful
replica
of
of
a
physical
Network,
the
digital
twin
interface,
as
previously
defined
as
a
runtime
interface,
and
we
can
conceive
of
the
inputs
as
defining
the
scenario
for
behavioral
assessment,
and
so
in
that
sense
they
would
select,
replace
or
complement
data
present
at
the
nii.
The
outputs,
of
course,
are
modeled
behavioral
parameters
and
in
the
interior
of
the
podt
congruent
with
the
Notions
described
in
the
architecture
document.
L
L
So
the
DTI,
the
inputs
as
they'll,
show
they
depend
on
the
use
case.
The
outputs
for
the
opt
really
are
Optical
Channel
parameters
so
path,
terminating
Channel
powers
and
particularly
things
like
generalized
signal
to
noise
ratios
margins,
things
of
that
nature.
At
the
nii
from
the
management
plane,
we
would
get
network
and
service
topology.
L
These
are
provisioned
Networks
and,
of
course,
ITF
defines
a
host
of
relevant
models
and
Frameworks
actn
abstraction
and
control
of
Transport
networks
from
teas
and
related
Yang
models
from
other
working
groups,
Being
Fit
for
the
purpose
from
the
physical
Network
inventory,
device
parameters
and
Status
instrumentation
data
things.
We
measure
on
the
network,
alarms
Etc
and
we
would
conceive
of
these
normally
as
coming
through
the
management
plane.
But
that's
an
implementation
matter,
behavioral
models.
L
So
a
quick
look
at
use
cases
and,
of
course,
planning
was
a
major
category
of
use
case
considered
in
the
first
draft
for
packet
networks
and
we
have
corresponding
planning
manifestations
in
the
optical
Network
case.
One
of
them
is
Optical
Services
planning.
So
if
we're
looking
at
introducing
or
provisioning
a
new
service
or
some
batch
of
services,
we
need
to
go
through
a
planning
exercise
and
figure
out
the
routes
and
spectral
allocation,
but
potentially
also
transponder
parameters
in
the
case
that
we
have
tunable
transponders.
L
L
So
obviously,
those
are
solutions
that
would
be
prospective
solutions
that
would
be
discarded
and
basically
the
findings
are
used
to
guide
a
search
for
Server
solutions
that
possibly
are
also
efficient
in
some
sense.
So
the
possible
embedding
of
optimization,
as
described
in
the
original
draft,
important
to
note
that
the
the
opt
doesn't
operate
that
search
it
merely
provides
information
to
it.
L
Optical
Network
planning,
basically
the
same
thing,
only
we're
looking
for
conjoint
service
and
Network
Solutions
that
are
both
functional
and
now
certainly
efficient,
for
example,
in
the
sense
of
some
lowest
cost
of
of
network,
and
in
this
case
the
DTI
net
inputs
would
have
to
Encompass
both
service,
topologies
and
channel
parameters,
and
also
the
newer
modified
network
topologies
being
postulated
next
slide.
Please.
L
Optical
Services
reprovisioning,
so
this
is
in
support
of
active
operations,
add
drop
reroute,
for
example.
Before
executing
such
a
change,
we
would
want
to
assure
that
the
post-change
state
is
entirely
functional,
so
we
would
conceive
of
some
algorithms
looking
for
batching
and
sequencing
of
operations,
power,
optimizations
Etc
that
will
lead
to
a
successful
reprovisioning
and,
of
course,
in
that
search,
the
opt
can
function
can
assist
by
looking
at
the
functional
Integrity
of
all
of
the
postulated
service
state
steps.
L
There's
the
possibility
of
risk
mapping
to
look
at
any
prospective
event
scenarios
and
identify
potential
transmission
performance
issues.
For
example,
we
can
look
at
prospective
fiber
Cuts
examine
our
planned
restoration
responses
in
those
cases,
identify
possible
performance
risks
and
where
such
risks
are
identified,
we
can
construct
new,
planned
Dynamic
restoration
responses
and,
of
course,
that
whole
process
can
be
automated
as
a
closed
loot.
L
And
I'll,
just
close
with
one
sort
of
more
abstract
thought
this
digital
twin
we've
described,
is
what
you
might
call
an
information
digital
twin.
It
provides
information
given
data,
but
the
use
of
that
information
is
left
to
you
know
it's
a
service
and
whatever
the
consumers
of
that
service
happen
to
be.
L
You
can
conceive
of,
and
some
do,
a
larger
sort
of
sphere
of
action
of
a
digital
twin,
imagine,
ingesting
intent
and
and
issuing
Network
control.
Outputs.
Such
a
conception,
though,
obviously
encompasses
full
closed
Loops,
not
just
data
to
information,
but
information
to
decisions
and
decisions
to
action
right.
So
this
is
a
collection
of
functions,
not
a
single
function,
but
necessarily
including
the
information
digital
twin
described
as
one
and
such
a
larger
scope
of
function
frankly
starts
to
become
difficult
to
distinguish
from
a
management
plane
featuring
automation
as
a
whole.
L
A
B
M
Yes,
I.
Thank
you
for
the
presentation
very
interesting
I
wanted
to
ask.
Did
you
consider
making
digital
twins
of
the
traffic
patterns
on
a
granular
flow
basis,
meaning
a
digital
twin
of
a
subnet
or
a
group
of
users
to
detect
Anonymous
anomalies
potential,
lateral
movements?
Basically
ipfix
based
twins.
L
Yeah
so
I
mean
that
kind
of
thing
was
considered
in
the
first
draft
in
the
NDT
dealing
with
packet
networks,
it's
not
relevant
to
the
Optical
performance.
Digital
twin
right,
I
mean
there
we're
dealing
with
provisioned
Optical
paths,
but
I.
Think
if
you
go
to
the
go
to
the
draft
and
and
look
for
the
content
specific
to
the
packet
Network
performance,
digital
twin,
you'll,
see
you'll,
see
thoughts
related
to
that.
L
A
There
is
a
question
also
in
the
chat
from
long.
Maybe
you
take
take
this
first
as
it
was
first.
So
do
you
consider
the
configuration
type
of
interaction
of
on
the
interface,
for
example,
to
confuse
your
configure
collection
measurement,
monitoring
information
flows
to
the
podt,
all
right,
this
configuration
indeed
at
another
interfaces
by
other
processes.
E
L
L
I'm,
not
sure
if
that
was
the
sense
of
your
question,
but
to
be
very
clear
on
on
that
point,
and
that's
really
what
I'm
getting
at
in
this
last
slide
week.
You
know
the
conceptions
of
digital
twin
that
float
around
I
mean.
If
we
look
at
the
architecture
draft
right,
we
def
you
know
defined
as
interfaces
models
and
data
and
then
something
to
manage
those
together.
What
those
provide
is
information
right
to
provide
action,
outputs
or
to
ingest
intent.
L
You
need
more
functions
right
and,
ultimately,
to
issue
action
on
the
network.
You
need
full
closed
Loops.
So
in
our
conception,
that's
a
larger
set
of
functions
and
again
it
starts
to
become
pretty
much
what
a
management
plan
as
a
whole
does.
So
it
gets
a
little
bit
tricky,
intellectually
and
and
my
view
for
the
interfaces
themselves
again.
The
idea
is
that
we
have
a
so-called
twinning
interface.
L
That
you
know
is
the
steady
receiver
of
information
about
the
physical
twin
and,
of
course,
in
some
scenarios
there
may
not
be
a
physical
twin
right
if
we're
looking
at
Greenfield
planning,
but
in
general
the
idea
is
that
there's
a
physical
Network
that
we're
trying
to
replicate
as
closely
as
possible.
So
the
idea
is
that
the
nni
is
is
the
source
of
of
that
real
Network
information,
of
course,
on
a
provisioned
network,
the
service
topology,
for
example.
L
D
Xanax
term
yeah
I
just
have
a
comment
on
what
you
put
here
on
the
last
slide.
I
would
assume
actually
that
the
second
one
is
the
more
General
conception.
The
one
is
the
first
one
is
more
restricted
right,
because
I
mean
there
you're
committing
to
who
your
golden
friend
is
so
to
speak.
So
I
think
definitely
I.
Think
it's
more
interesting
to
look
at
also
number
two,
which
includes
number
one.
L
L
There
are
these
two
sort
of
conceptions
and
certainly
that
one
is
more
restrictive,
simply
pointing
out
that
two
is
have
to
be
a
little
bit
careful
because
it's
it's
really
a
collection
of
of
functions
and
it's
not
a
single
thing
and,
and
one
could
in
fact
represent
them.
As
you
know,
the
new
thing
here
is
an
information
digital
twin
management
planes
that
are
automated.
We
can
construct
those,
as
you
know,
the
Assembly
of
known
things.
L
B
N
N
Our
initial
scope
was
rated
on
performance,
so
targeting
lateral
attack
movements.
It's
not
in
our
scope
right
now,
Sharon,
it's
we
were
more
thinking
about
how
is
Network
performing.
So
are
you
having
some
delay
at
some
link?
Is
there
some
Jitter?
That
is
why
you
would
expect
and
similar
questions.
However,
it's
an
interesting
aspect
to
consider.
Also
I
I'll
try
to
drop
your
line
later
offline
about
this
topic
and
sorry
just
another
Point.
N
I
acknowledge
and
I
agree
that
it's
interesting
but
I
think
it's
broadening
too
much
the
scope
so
because,
if
you
get
a
management
plan,
well,
I
have
never
operated
the
network,
except
from
my
home
network
and
I.
Think
it's
too
simple,
but
I
mean
everyone
manages
networks
differently
and
there
are
like
tons
of
different
ways
to
manage
networks
so
more
perspective.
It's
a
scope
that
needs
to
to
Broad.
A
A
B
E
E
Digital
Team,
Network
Concepts
and
the
reference
architecture,
the
interfaces
of
digital
Network
can
be
divided
as
the
thing
Northbound
interfaces
in
southbound
interfaces
and
the
internal
interfaces.
So
in
order
to
build
this
routine
Network
and
realize
its
many
advantage
on
different
interfaces
should
be
able
to
meet
different
requirements,
and
this
draft
is
to
introduce
the
requirements
for
the
interfaces
of
dtn.
So
the
positions
of
different
interfaces
are
shown
in
the
figure
on
the
right.
E
So
next
right
piece
so
for
the
requirements
for
thin
Northbound
interfaces
and
the
thing
not
spawn
interfaces
is
the
interface
between
the
network
application
layer
and
the
team
Network
layer.
So
the
network
application
requirements
are
input
from
the
two
Northbound
interface
to
the
team,
Network
layer.
E
The
team
Northbound
interfaces
should
be
able
to
expand
the
in-time
to
meet
the
needs
of
new
network
applications,
and
the
third
layer
is
that
it
should
be
portability
far.
There
are
things
with
different
sizes
and
functions
in
the
team.
Network
layer
on
the
same
or
similar
requirements
of
various
applications
in
the
network.
Application
layer
may
be
deployed
on
different
things,
so
the
thing
not
responding
interface
should
be
easily
transplanted
and
deployed
on
different
things
and
last
one
is
it
should
that
it
should
be
flexible
deployments
that
to
reduce
deployment
time
and
cost.
E
Firstly,
it
should
be
United
that
each
subsystem
in
the
two
Network
layer
should
be
able
to
provide
the
same
data
fromage
and
the
data
service
for
other
subsystems
through
the
internal
interface
of
the
thing
layer.
That
is,
the
interface
should
have
United
and,
secondly,
is
the
adapted.
Adaptability
for
the
net
twin
Network
layer
must
interact
with
the
network,
application
layer
and
the
physical
Network
layer
and
should
be
well
adapted
to
various
Network
advances
and
interfaces.
It
should
be
adaptability,
so
suddenly
is
that
it
should
be
Port.
E
Building
the
data
model
instances
provided
by
the
service
mapping
model
subsystem
for
different
applications
may
have
a
high
degree
of
similarities.
So,
in
order
to
improve
efficiency,
the
data
model
instances
must
be
able
to
be
provided
and
deployed
through
different
internal
interfaces
of
three
layers,
and
the
last
one
is
the
flexible
and
extensible.
E
E
E
Also,
control
updates
are
delivered
from
the
team,
suspend
interface
to
the
physical
and
99
work
and
the
wireless
network
elements
in
the
physical
entitled,
Network
exchange
exchange
network
data
and
the
next
Network
control
information
with
the
network
layer
through
the
thin
southbound
interface.
So
the
southbound
chain
interface
should
also
have
the
characteristics
as
follows.
E
First,
one
is
the
information
interaction
capability.
The
10
Southbound
interface
should
be
able
to
collect
the
information
of
different
physical
Network
elements
or
network
devices,
and
since
the
configuration
information
of
the
team
Network
to
the
physical
Network
for
execution,
each
can
realize
the
realize
the
information,
interaction
between
the
three
Network
layer
and
the
fiscal
and
testing
Network
layer
and
the
second
is
the
real
time
requirements.
E
The
realization
of
Team
network
configuration,
verification
and
other
functions
must
have
certain
real
time,
so
the
information
collected
and
uploaded
from
the
physical
entity
Network
and
the
configuration
information
sent
from
the
Q
Network
to
the
physical
Network
must
have
certain
real
time
in
order
to
meet
the
real-time
requirements
of
the
whole
digital
thing
Network
and
the
next.
The
third
layer
is
the
compatibility,
so
network
devices
and
network
elements
from
different
manufacturers
use
different
interfaces
and
protocols.
E
A
Okay,
I
think
Daniel,
as
you
recently
proposed
this,
this
new
contribution
I
think
people
might
need
to
have
some
time
to
read
the
document
and
provide
you
some
feedback,
maybe
on
the
mailing
list,
and
thank
you
again
for
the
presentation.
I
propose
that
we
continue.
There
is
no
question
now
and
maybe
on
the
chat.
There
is
one
just.
Let
me
check.
A
D
D
D
However,
the
question
is:
is
this
enough
and
also
networks
themselves
contribute
yeah,
but
through
the
covered
footprint,
so
hence
the
question
or
hence,
basically,
they
I
think.
The
fact
is
that
the
Net
Zero
mandates
will
also
apply
to
network
providers
as
well,
and
so
the
question
is:
what
can
we
do?
D
D
So
yeah,
so
so
there's
this
this
draft
draft
CX
green
ps01.
The
purpose
is
basically
to
analyze
the
challenges
and
opportunities
in
green
networking,
Network
management.
We
did
present
a
zero
zero
in
itf114,
but
there
are
a
number
of
updates
here
for
one.
Basically,
there
are
editorial
refinements
throughout
also.
The
original
draft
was
mostly
about
Energy,
Efficiency
or
energy
usage
efficiency.
D
This
has
been
updated
through
also
emphasize
energy
sources
and
deployment
aspects
in
general,
not
just
the
energy
usage
efficiency,
as
well
as
to
include
considerations
for
the
manufacturing
life
cycle
of
the
equipment
that
is
used
to
to
implement
these
there's.
Also,
there
are
some
changes
in
section
5.2,
which
grows
about
traffic
adaptation.
It's
actually
generalized,
also
expanded
through
it
will
be
about
protocol
optimization.
D
D
D
We're
focusing
them
mainly
on
the
ability
to
provide
visibility
as
a
foundational
problem
that
will
essentially
enable
a
lot
of
the
other
schemes
and
support
that
building
on
top
of
them
and
let's
basically,
to
assess
the
usage,
assess
and
validate
the
Effectiveness
and
efficiency
yeah,
essentially
of
the
energy
usage
and
so
forth.
And
this
is
then
also
the
basis
of
course,
to
enable
later
on
control
routes
for
various
optimization
schemes,
where
you'll
take
various
actions,
for
instance,
to
reduce
the
amount
of
carbon
footprint.
If
we
will
all,
of
course
validated.
D
But
you
need
to
see
the
feedback
of
that.
This
requires
also
instrumentation
for
corresponding
energy
metrics.
There
is
actually
a
companion
draft
on
re-networking
metrics,
which
basically
relates
to
the
metrics
not
just
to
equipment,
but
also
to
things
such
as
flows
and
the
passwords
of
all,
which
are
the
types
of
things
that
we
would
want
to
optimize
a
network
for
and
what
would
want
to
optimize
services
for
next.
One.
D
So
the
second
aspect
concern
is
the
network
as
a
whole,
so
basically
This
concerns
aspect
such
as
challenges
related
to
patrol
protocol
extensions.
What
would
be
needed
to
consider
energy
sustainability
and
support?
It
is
a
cost
factor
in
yeah
in
network
Patrol
decisions,
energy.
If
we're
routing
and
path
configuration
where
we
would
have
for
this
one
of
the
channel.
How
do
you?
How
can
you
assess
the
carbon
intensity
of
pass
if
you
have
the
to
choose
between
multiple
ones?
How
do
you
optimize
the
network
overall
to
minimize
the
overall
footprint?
D
D
Protocols
of
this
basically
challenges
related
to
enabling
Network
energy
saving
mechanisms.
This
blurs
to
some
degrees,
the
boundaries
between
membrane
and
control
plane,
because
parts
of
the
aspects
are
related
to
the
resource
management.
How
can
you
find
this
commercial
decommission
resources
on
very
short
time
scales?
This
requires
mechanism
for
fast
Discovery.
What's
going
on
the
network,
fast,
reconversions
and
so
forth,
here,
there's
a
role
for
autonomics
to
play.
There
are
also
for
intent-based
networking
where
we
had
various
use
cases
earlier.
D
Other
aspects
concern
Network,
addressing
instrumentation
traffic
adaptation.
How
can
manage
perhaps
the
burstiness
of
network
versus
smooth
transitions
to
to
affect
the
energy
transmission,
Z
and
so
forth?
Next
one
and
then
finally,
basically
at
the
architecture
level,
there's
a
question:
how
will
you
how
you
would
facilitate
the
organization
of
categories,
networking
applications
that
use
the
network
to
minimize
energy
consumption?
D
This
is
the
sponsors,
but
there
are
examples
such
as,
for
instance,
CDN,
ICN
and
coin,
which
do
some
of
those
things.
But
basically
the
question
is
here:
how
how
is
it
from
an
energy
perspective?
So,
for
instance,
when
would
you
deploy?
We
have
this
deploy
cache
information?
What
does
that
to
the
overall
balance
and
so
forth?
Next
one?
D
So
then,
basically,
the
the
question
we
discuss
here
is
basically
how
to
move
forward
with
this
draft,
and
we
believe
actually
that
nmg
is
a
good
landing
spot
for
this.
Many
of
the
challenges
not
all
of
them,
but
I,
think
the
majority
of
them
are
clearly
management
related.
This
includes
points
in
the
starting
point
of
the
visibility
and
the
insultation
the
metrics
and
support
all
of
those
things,
and
also
on
top
of
that.
D
Basically
many
opportunities,
then
involve
management,
control
groups
and
deployment
optimizations,
and
these
are
fundamentally
management,
related
problems
and
challenges,
and
again
some
of
those
examples
there.
How
do
you
plan
routes
segment
files?
How
do
you
moderate
the
various
trade-offs
that
you
have
Etc?
K
Oh
thanks,
Alex,
the
sculptor
of
the
draft
and
I
haven't
updated
it,
but
I
would
like
to
provide
perspective
from
very
different
point
of
view.
Urgency
of
this
work.
So
traditionally
this
kind
of
work
was
done
as
wishful
thinking
or
you
know
we
want
to
reduce
outputs
and
maybe
make
World
better
place.
We
are
getting
to
the
point
that
we
cannot
deploy
anymore.
If
you
look
at
how
Sardis
and
silicon
technology
evolves,
we
double
the
speed
every
two
years,
so
Marlo
is
still
applicable
while
power
consumption
goes
down
by
half,
so
I
wish
each
generation.
K
So
every
two
years
we
double
power
consumption
on
each
networking
device
we
deployed
from
size
perspective.
We
are
getting
to
1.2
gigawatt
data
centers,
so
there's
no
infrastructure
in
the
world
that
can
actually
power.
This
kind
of
data
center
cools
them
down,
provide
you
some
comparison
for
city
of
Dublin
concerns
to
gigabots
Microsoft
data
center
1.2,
so
they're
getting
close.
So
it's
not
only
nice
to
have.
We
cannot
really
continue
at
the
rate
we
go
today.
We
really
need
to
start
looking
into
how
to
optimize.
We
need
to
reduce
carbon
emission.
K
I
Sorry,
it
took
time
because
my
my
knee
is
killing
me
in
general.
I
believe
that
this
keeping
this
as
a
as
a
whole
study
of
the
of
the
concern
is
something
that
he
said
is
a
it's
a
bright
way.
I
will
go,
I
will
not
separate
it
in
different
matters
and
and
though
I
believe
that,
basically,
what
we're
talking
is
a
little
bit
similar
to
the
questions
that
Chris
was
assuming
before
about
this
idea
of
the
informational
and
action
digital
training
that
are
a
general
at
the
ends.
I
If
you
have
an
action,
digital
train,
you're,
talking
about
a
general
management
tool
like
AI
or
whatever,
is
how
you
put
these
in
in
the
general
management
flow
here.
The
question
is
precisely
how
we
put
Energy
Information
and
how
we
prioritize
energy
data
and
energy
related
decisions
with
respect
to
the
other
goals,
the
network.
I
So
that's
I'm,
making
emphasis
on
this
on
how
these
this
would
affect
policies
and
how
policies
have
to
be
expressed
and
and
enforced.
I
think
that
this
is
the
main,
the
main
question
here,
because
nobody
is
going
to
deny
the
importance
of
these
or
the
relevance
for
the
for
the
future
of
networking
and
even
the
future
of
humankind.
But
the
the
point
here
of
what
we
should
focus
is
precisely
in
this
new
dimension.
D
Right
so
for
those
types
of
implications,
if
you
will
on
also
existing
existing
lessons,
this
has
been
part
of
the
challenges
and
opportunities
that
we
would
want
to
precisely
actually
point
out
here,
and
some
of
them
may
be
straightforward
engineering
type
of
things.
Then.
Basically,
okay,
then
we
don't
need
to
talk
about
much
from
a
research
perspective,
but
some
other
things
will
be
more
closer
to
research.
I
O
About
Jody
from
Qualcomm,
so
yeah
for
I
guess
forever.
We've
been
sort
of
under
the
Inspire,
my
Paradigm
of
over
provisioning,
the
networks.
You
know,
science
right,
that's
sort
of
been
the
philosophy
like
and
we
as
Engineers
have
often
or
many
times
or
a
hard
time
being
told.
Don't
worry
about
optimizing
it's
over
provision,
so
you
have
you
know
bandwidth
and
so
on.
O
So
maybe
there's
an
element
about
changing
the
mindset
and
saying
it
really.
It's
really
important
to
design
elements
that
are
optimized
and
that
actually
try
to
to
do
better
than
what
we
do.
You
know
designing
to
some
extent
you
know
the
internet
is
designed
to.
You
know,
run
30
utilization.
So
is
that
really
the
the
when
we
talk
about
carbon
footprint?
Is
that
really
a
physical
to
start
front,
and
should
we
sort
of
also
talk
about
the
mindset
of
philosophy
of
how
we
design
networks?
D
Sustainability
also
because
then,
okay,
the
optimization,
is
just
don't
do
networking
right
just
turn
the
especially
whole
thing
I've
called
bother
with
it
clearly
so
clearly,
this
is
not
the
solution
either.
So
clearly,
you
basically
need
to
mitigate
and
moderate
between
different
goals,
which
are
conflict
so
basically
it's.
It
is
kind
of
like
a
constraint
based
problem,
because
the
constraint
is
still
that
you
need
to
as
far
as
resilience
your
cause
and
service
level
requirements.
D
As
far
as
all
of
those
things
are
concerned,
of
course
you
need
to
retain
those,
but
while
also
making
sure
that
you
do
what
you
can
there
and
the
opposite,
how
well,
as
always,
between
mitigating
goals-
and
you
have
also
the
challenge
of
of
policies,
for
instance,
how?
Where
are
you
willing
to
make
trade-offs?
How
do
you
mitigate
between
the
trade-offs?
How
do
you
manage
them
and
so
forth?
So
yeah?
This
is
about
part
of
the
challenges
exactly.
P
A
David
Summers
Harris
with
Rockton
I,
was
wondering
if
it'd
be
useful,
to
clarify
the
scope
of
the
research,
whether
this
is
for
a
single
operator
providing
them
with
ways
to
optimize
the
energy
consumption
or
sustainability
of
their
own
network
or
if
it's
also
including,
like
interoperability
between
so
one
operator
tests
I,
have
great
sustainability
in
my
network.
So
if
you
route
through
me,
then
you'll
be
green.
Then
how
do
you
prove
that,
and
it
reminds
me
of
you,
know
carbon
credits
and
governments
and
companies
have
bought
them.
P
D
That's
a
great
comment
actually
so
clearly
well
in
in
this
draft.
We
are
not
restricted
to
one
provider,
but
actually
some
of
the
challenges
are
related
to
specific
providers.
Some
are
related
to
vendor
somehow
operating
for
the
holistic
picture,
which
takes
all
of
them
into
account,
and
obviously
some
of
the
things
that
actually
the
draft
also
makes
mention
of
that.
D
You
could
say
well
as
a
network
provider
I
optimize
just
by
the
way,
I
do
it
in
my
network
and
it
doesn't
matter
what
happens
in
the
data
center,
for
instance,
you
could
do
that,
but,
of
course
that
well
then
you
shift
perhaps
the
the
Virgin
burden
somewhere
else
which
helps
you,
which
helps
your
energy
cost
us,
but
which
does
not
help
the
holistic
picture.
It's
appear
and
clearly
there
are
both
of
these,
because
that's
another,
of
course,
of
the
challenge
is:
how
do
we
moderate
between
that
and
yeah?
D
Q
On
mobile
here,
the
disease,
so
far,
if
you
look
at
the
green
topic,
a
lot
of
providers
is
not
just
facility
provider.
There
are
some
like
otd
companies
talking
about
the
to
recycle
the
waste,
the
heat,
the
heat
waste
and
like
by
using
like
a
liquid
cooling
and
then
from
the
hardware.
It's
going
to
heat
up
the
water,
that's
going
to
be
recycled
used
in
the
community.
Q
So
that
means,
if
you
look
at
your
topic,
so
I
I
agree.
This
is
very
interesting
topic
and
then
I
I
support
it.
The
thing
is
here:
if
you
consider
those
type
of
scenario,
since
you
know
this,
one
is
not
a
standard
alone,
not
just
like
right
now,
working
part,
it's
going
to
be
impacted
either
a
one
minute
or
the
other,
so
I
don't
know
whether
you
have
already
considered
suppose
you
can
feel
something
if
not
green,
since
it
consumes
a
lot
of
energy.
B
D
Recycling
the
whole
manufacturing
life
cycle
is
it
so.
From
that
perspective,
it's
certainly
good
if
you
can
extend
the
life
time
for,
for
instance,
of
your
Hardware
by
perhaps
making
more
software
base
if
the
addressing
tool
software
upgrades
versus
relying
on
Hardware
upgrades
and
all
the
basically
how
to
how
to
again
manage
it.
B
K
Going
back
to
networking
I
think
we're
working
in
metric
normalization,
really
if
you're
IPOs
Network
today
the
number
of
components
that
will
fit
into
PC
computation
and
provide
you
either
short
or
slow
slat
and
see
our
login
with
maximum
protection
path.
Right,
we
don't
include
power,
consumptions,
computation,
parameter
or
constraint,
job
by
having
normalized
metrics
that
could
potentially
come
out
of
this
work
towards
ATF.
We
could
use
it
in
normalized
PC,
computation
to
compute
the
plants
that
made
some
power
constraints.
K
A
R
I
was
in
danger
time
wise
there,
when
that's
such
an
important
presentation
was
only
for
me,
so
I
need
to
be
what's
the
question
so
I'm.
Looking
at
this
presentations
on
modeling
boundaries
and
its
boundaries
in
modeling
languages,
it's
actually
a
draft
into
netmod,
but
time
one
is
Brandon
here,
because
I
think
there's
some
forward-looking
stuff.
You
can
do
the
next
slides.
R
R
So
the
draft
actually
has
a
number
of
observations
of
challenges
in
modeling
and
it
focuses
a
lot
on
modeling
of
representations
that
involve
dealing
with
constraints
and
I've
picked
out
a
number
of
things
here,
all
of
which
you've
got
very
similar
requirements
in
10,
modeling
and
intent
of
capability
expressing
restrictions.
I
picked
out
one
their
partial
visibility,
which
is
one
that
doesn't
get
much
attention.
R
But
this
is
where
I
can't
I
can't
fully
see
the
the
the
monitoring
or
the
detectives
in
the
network
and
out
and
partial
view
of
that
which
looks
somewhat
like
constraints,
the
process
in
negotiation,
where
I
move
through
from
a
fairly
abstract
and
nebulous
view
of
what
I
want
through
to
a
much
more
specific
thing
that
again,
a
set
of
constraints.
These
two
in
technicals
and
all
of
these
require
some
view
of
uncertainty.
R
The
representation
of
these
things
progresses
through
a
set
of
gradual
narrowing,
so
especially
in
negotiation,
I
started
with
something
fairly
fabulous
and
I,
gradually
narrow
down
to
a
precise
level.
If
I
look
at
intent,
I
I
may
have
an
intent
that
starts
off
as
relatively
vague
and
then
becomes
more
refined
as
the
as
the
implementation
takes
place
and
there's
a
number
of
levels
through
this.
It
also
applies
as
I
look
at,
especially
as
I
look
at
specification.
R
I've
got
an
example:
the
later
of
capability,
where
I
move
from
a
standard
specification
through
a
number
of
vendor
restrictions,
and
so
on,
I'll,
go
through
a
an
example
of
that
these
narrowings,
relating
what
we've
had
to
classify
with
a
new
term.
There's
a
big
terminology
problem.
This
is
proved
to
be
difficult
to
actually
discuss
this
space
because
of
lacking
terms.
We
have
classes
instances
and
then
we
run
out.
R
The
the
Assembly
of
a
system
you
get
many
similar,
but
not
identical
occurrences,
where
you're
placing
constraints
on
the
behavior
of
a
particular
thing
that
narrows
it
from
the
class,
but
doesn't
make
it
into
an
instance,
and
this
was
the
the
general
problem.
So
the
draft
goes
through
these,
these
these
General
sort
of
principles
and
so
on,
and
it
makes
a
number
of
observations
in
those
areas
against
the
next
slide.
Please
thanks!
R
R
Even
if
it's
a
single
value,
that's
still
a
narrow
constraint,
so
even
a
single,
even
even
an
expression,
of
a
an
instance
value
is
a
constraint
so
the
and
that
leads
us
to
potentially
a
single
way
of
expressing
everything
from
classes
all
the
way
through
to
to
instances
the
conventional
modeling
techniques
we
use
tend
to
break
it
into
those
two
things:
a
class
or
an
instance.
We
have
those
two
sort
of
levels:
we
don't
I
mean
all
meta
classes,
but
we
don't
have
this.
R
Sometimes
I
can't
see
it
at
all,
so
these
sorts
of
things
tend
to
push
us
towards
a
distinct
way
of
representing
compared
to
what
we
would
traditionally
do,
and
we
found
as
I've,
been
through
as
I
mentioned
a
long
time
challenging
with
this
as
we
go
through
the
challenge.
We've
through
the
challenge
of
modeling
particular
problems.
We
found
we
had
to
put
special
attributes
in
to
deal
with
these
funny
cases
and
so
on.
R
You
keep
finding
them
so
I
put
this
into
net
mod,
because
I
felt
that
or
we
felt
that
there
was
an
opportunity
in
Yang
itself
to
to
extend
the
language
to
deal
with
these
challenges,
and
we've
got
a
proposal
in
the
back
of
the
or
a
sketch
of
a
proposal
in
the
back
of
the
draft.
That
shows
some
of
the
things
you
could
do
potentially
with
Yan,
and
it
would
seem
to
be
possible
to
extend
it
within
its
current
definition.
So
it's
compatible
with
existing
things,
which
I
think
is
critical.
R
R
So,
just
to
get
some
the
things
here,
just
to
emphasize
so
so
intent
in
the
in
the
draft
I
use
expectation
and
tension
rather
than
intent,
really
because
there's
two
sides
of
that
there's
the
client
side
expect
stuff
and
the
provider
size
and
tends
to
deal
with
the
thing.
But
that's
desired
outcome
in
terms
of
constraints
are
to
emphasize
that
capability
is
about
opportunity
for
Behavior
to
be
exhibited,
and
that,
of
course,
is
again.
R
A
range
of
possible
restrictions
and
constraints,
and
the
partial
visibility,
as
I
said,
is,
is
imprecise,
measurements
in
a
lossy
or
environment,
where
I've
got
a
a
challenge
with
noise
or
something
of
this
sort,
where
I
can't
actually
see
it
fully.
What's
going
on,
she
goes
to
the
next
slide.
Please
thanks.
So
this
is
just
to
emphasize
this
gradual
progression
on
narrowing.
It's
a
it's
obviously,
in
our
arbitrary
rough
example,
but
I
may
start
with
a
a
standard
definition
of
something
where
I've
got
a
an
integer.
It's
boundless.
R
Essentially,
it's
full
range
of
the
institutes
specified,
inherits
no
constraints
whatsoever.
I
find
I
have
a
two
particular
implementations
that
a
vendor
has
where
and
why
it's
got
arbitrarily
chosen
here,
of
course,
10
gigs
to
100,
gigs
or
something
another
one's
got
five
to
fifty,
where
there's
some
constraints
in
the
particular
technology.
So
I've
got
two
different
realizations
in
my
particular
solution.
Here.
I
want
to
be
able
to
express
those
two
different
capabilities,
then,
in
an
application,
I
only
need
10
to
50
gigs
I
want
to
express
that
constraint.
R
I
have
a
further
point,
maybe
in
the
application,
where
it's
at
a
particular
point
in
the
solution.
There's
a
a
specific
restriction
on
the
Range,
so
it's
only
30
to
50
gigs
a
particular
place
same
implementation,
Hardware
we're
in
some
of
the
particular
place
and
in
a
proof
of
concept.
Sorry,
if
I've
got
some
busy
hour
thing,
I'll
say
it's
always
50
gigs,
so
I've
got
these
different
levels
of
setting
for
this
property
and
I'm
going
to
set
ranges
and
single
values
in
the
property
and
again
other
examples,
proof
of
concept
different
narrowing.
R
Maybe
it's
turned
into
an
enumeration,
because
I've
only
got
three
improved
points,
I'm
going
for
and
when
I
look
at
the
instance
at
the
bottom,
there
I've
got
an
intent
instance,
one
the
intent
is
set
to
30
or
40
another
one.
That's
particularly
40.,
so
I've
got
different
ways
of
of
constraining
this
very
simple
property
and
I've
got
various
different
occurrences
of
these
constraints
in
the
in
the
solution
spaces.
R
You
just
try
to
illustrate
the
the
sort
of
range
of
narrowing
from
a
standard
through
a
number
of
different
cases,
whether
each
expressions
are
the
same
property
they're
each
expressions
of
the
integer,
but
their
expression
to
the
integer
in
in
terms
of
different
nouns.
If
you
go
to
the
next
slide
this,
so
nothing
goes
through
this
in
detail,
but
the
this
this
the
draft
does.
R
But
the
the
key
thing
here
is
I've
assembled
the
system
out
of
the
same
pieces,
the
brown
pieces
here
with
the
same
letters
all
the
same
pieces,
but
each
has
been
used
in
a
certainly
different
role
and
each
is
being
used
in
a
a
way
where
it's
constrained
to
perform
a
different
function.
It's
the
same.
Underlying
thing,
but
the
specification
for
it
is
constrained
and
each
of
those
things
in
that
figure
are
essentially
occurrences
because
they're
not
instances
critical
happening
instantiated
when
they're
just
in
the
pan.
R
So
the
occurrence
here
is
a
is
a
pattern
tool
as
well
here
essential
that
lets
me
lay
out
different
cases
of
use
of
a
particular
single
thing
in
a
system
design,
and
the
this
this
actually
is
is
a
a
sketch
of
a
management
hierarchy.
Abstracting
signals
from
the
rnf
work
abstracting
from
a
network
View
to
a
a
sorry,
a
noble
virtual
Network.
You
do
the
next
slide.
Please
I
think
I'm
going
to
skip
the
next
one.
Yeah
I'll
skip
this
one.
R
Actually,
because
it's
a
detail
on
the
slide
and
the
the
key
thing
here
for
the
the
first
thing
is:
we've
had
terminology
problems
trying
to
try
to
discuss
this,
and
so
on
so
getting
a
better
view
of
the
terminology,
the
labels
we
use
and
the
problem
in
the
problem
spaces.
We
chose
the
term
occurrence
because
it
appeared
to
be
something
that
wasn't
being
used
and
it
did
appear
to
sort
of
emphasize
the
instance-like
nature
of
this
thing.
R
But
not
an
instance,
there's
not
a
good
language
for
this
again
so
representing
or
there
doesn't
appear
to
be,
we.
We
can't
find
a
language
that
seems
so.
Uml
has
classes
and
instances.
Yang
has
the
same
sort
of
two
positions
and
so
on
there
don't
seem
to
be
any
languages
that
give
us
this
gradual
slide
from
a
vague
notion
through
to
a
a
very
specific
definition,
so
we're
looking
continuing
to
explore.
R
We
haven't
found
anything
looking
again
into
the
group
to
see
if
anybody's
aware
of
anything
that
does
this
sort
of
represents
this
sort
of
thing
well
and,
as
I
said
in
the
internet,
mod,
because
we
feel
that
Yang
could
be
extended
in
this
way
then
into
this
direction,
and
we
felt
that
this
was
a
good
influence
in
that
mod.
So
that's
all
I
have
I've
got
some
backup
slices
with
some
stuff
in
the
in
the
draft
which,
hopefully,
you
could
read
and
maybe
give
some
feedback
on.