►
From YouTube: IETF102-NMRG-20180719-1330
Description
NMRG meeting session at IETF102
2018/07/19 1330
https://datatracker.ietf.org/meeting/102/proceedings/
A
And
it's
time
to
start
so
welcome
everyone.
This
is
the
first
energy
session.
For
this
afternoon.
We
have
a
second
session
at
the
end
of
the
afternoon.
I
would
start
with
giving
a
short
introduction.
We
have
a
full
agenda
today,
so
I
will
ask
also
the
presenters
to
be
on
their
timing.
Total
maximum
discussion
for
today
can
somebody
also
close
the
door,
please
the
back
of
the
room.
Thank
you.
A
So,
starting
with
the
no
12,
so
the
IOT
RTF
follows
the
IETF
policy.
So
please
be
aware
of
the
different
recommendation
in
this
note
well
by
participating
in
the
IETF
meeting.
You
will
abide
by
these
recommendations
and
before
we
start
the
actual
discussion
technical
discussion,
we
need
at
least
two
person
to
volunteer
for
being
a
minute
taker
and
another
one
for
being.
The
Charter
scribe
meet
a
co-driver
scribe,
so
we
cannot
stopped
without
having
at
least
two
volunteers
for
that
somebody
for
the
minutes
and
somebody
for
the
scribe.
A
A
We
have
a
follow-up
session,
the
last
session
of
this
afternoon,
which
will
be
a
bit
beyond
the
energy
but
still
hosted
by
energy.
It's
a
OGF
special
session
on
AI
and
Emily
networking.
This
will
be
a
more
discussion
meeting
to
understand
a
bit
more,
the
relationship
between
the
detracting
part
and
the
AI
ml
part
and
what
we
could
derive
from
that
either
in
IETF
or
its
most
specially
in
RDF,
and
then
also
an
announcement
for
interim
meeting
of
energy.
So
on
Friday
tomorrow
we
have
a
one-day
meeting
workshop
dedicated
to
intern
base.
A
Networking
the
meeting
will
not
be
in
this
utterly.
It
will
be
hosted
by
ETS,
a
Cola
technology
super
oil,
which
is
two
blocks
away
from
from
here.
So
a
nine-minute
walk
all
the
information
on
the
details
on
the
meaning
list
or
if
you
are
just
it
please,
at
the
end
of
this
session,
you
can
come
to
discuss
with
me
and
pattern,
which
is
also
here
in
the
room,
which
will
be
our
local.
A
stand.
Helped
me
to
organize
this
meeting
with
a
great.
A
Patty,
quick,
a
look
at
our
agenda,
so
this
is
the
introduction
and
I
try
to
group
the
different
presentation
of
today
under
some
topics.
So
it's
sometimes
not
always
the
right
way
of
grouping,
but
the
first
topic
will
be
around
aspect
pertain
to
telemetry,
with
the
first
presentation
from
Pedro
on
exploiting,
if
external,
even
detectors
and
their
use
to
do
some
management
activities.
A
The
second
plantation
going
to
be
from
are
you
on
the
the
network
telemetry
framework
that
has
been
presented
also
in
other
groups
and
understand
what
we
can
do
in
the
scope
of
energy
around
this
telemetry
aspect
and
especially
the
more
forward-looking
activities?
The
second
topic
will
be
on
management
aspect
of
network
slicing.
There
will
be
two
presentation,
one
from
Alex
on
the
management
problem
or
the
problem
of
the
managing
slides
in
the
gap,
analysis
and
the
second
presentation
from
Kieran
which
will
be
blending
remotely
on
some
use
cases
of
the
truck
slicing.
A
And
finally,
the
last
topic
will
be
on
so-called
novel
approaches
to
network
management.
It's
a
way
of
grouping
different
presentation,
so
we
will
have
I
think
it
was
crimson
on
the
intelligent
network
management,
using
reinforcement,
learning
a
second
presentation
on
a
roadmap
to
a
network
class
world
from
being
new,
then
a
remote
presentation
from
Jefferson
on
autonomic,
v,
dot,
zero
and
finally
Alex
galley
Alex
Clare
Torry
will
give
an
update
on
ongoing
draft
distinguishing
intern
policy
and
service
models.
A
Thank
you.
So
a
couple
of
things
I
would
like
to
mention
on
the
stake
use
of
the
research
group.
The
meetings
been
for
this
year,
so
we
had
the
begin
of
1980
f-101
in
London,
because
usually
we
try
to
co-locate
at
least
one
of
the
early
meeting
with
on
one
of
the
scientific
conferences
in
network
management.
This
year
it
was
with
announced
2018
in
Taiwan
in
April.
So
this
is
past
and
we
are
currently
having
this
meeting
today
tomorrow.
A
It's
an
interim
still
in
morale
and
the
last
meeting
we
will
plan
to
have
this
year
will
be
here
with
ITF
Onofre
in
Bangkok
in
November,
and
in
parallel
to
that
you
can
come
discuss
with
the
chair
or
suggest
in
the
meaning
list,
a
doc
interim
or
topical
meeting.
We
used
to
add
this
like,
for
instance,
tomorrow
the
meeting
on
Ethan
base
networking.
So
if
you
have
any
suggestion
or
proposal
or
things
you
like
to
discuss,
we
are
here
to
help.
A
Status
on
some
documents
for
the
related
to
the
research
group,
so
recently
in
February,
we
had
new
RFC
published
under
the
energies
or
ever
see.
Eighty
three,
sixteen
on
autonomic
networking
use
case
for
distributed
detection
of
SLI
violations.
So
congratulations
to
the
author's
because
it
was
a
long-standing
effort
to
get
this
published.
A
For
those
that
are
not
fully
aware,
we
weld
a
what
we
call
the
research
agenda
for
the
research
group
in
order
to
organize
the
activities
or
the
the
document
in
there
in
the
group
around
a
set
of
films
currently
defined.
We
have
in
turn
based
network
management,
the
use
of
AI
techniques
for
network
management
and
another
thing
that
we
call
for
the
moment
on
sonic,
3.0
or
self
driving
network
or
z-rotation
working.
A
So
the
idea
is
to
understand
what
we
could
do
in
the
scope
of
autonomic
networking
and
new
approaches
to
approach
this
isn't
this
environment?
Of
course.
This
is
not
limited
to
these
three
themes.
Based
on
the
presentation
we
can
avalor
suggestion.
We
can
develop
new
topics
for
the
research
agenda,
so
it's
based
on
the
interest
and
its
contribution
difference.
So
if
you
have
proposal,
we
are
here
to
hear
you
and
give
you
a
fluorophore
for
some
suggestions.
A
A
lot
of
those
will
say
themes
of
the
research
agenda.
The
idea
is
to
derive
more
specific
research
items
identified
with
documents,
activities
of
the
research
group,
experimental
result
or
experimental
validation,
but
also
possibly
also
work
plan
and
somebody's
understand
that
it's
a
research
activity.
So
it's
not
like
a
working
group
where
you
have
a
more
strict
time
management,
but
still
if
we
can
derive
some
milestones
for
those
activity,
it's
important
to
see
the
progress.
A
And,
finally,
if
you
want
to
read
a
bit
more
but
the
content
of
each
of
those
themes
with
lisandro,
we
had
published
in
March
in
the
I
Triple
E
communication
standard
series.
There
is
some
stat
news
which
are
very
short
articles,
highlighting
progress
in
different
areas,
and
there
is
so
a
short
article
for
energy
in
this
issue.
I
think
it's
free
available.
A
Now
a
word
on
the
so-called
future
of
energy,
so
I
think
for
those
that
follows
the
meaning
list.
You
are
aware
that
listen
of
all
my
culture
decided
to
step
down
after
seven
years
of
service
as
coach
of
energy.
So
I
would
like
to
really
address
my
thanks
to
to
listen
role
and
for
she
cannot
be
here
for
budget
reason
for
all
what
she
has
done
in
the
energy
and
the
interaction
we
had.
The
good
thing
is
that
lisandro
will
still
be
an
active
participant
of
the
group,
not
as
a
chair
but
as
a
contributor.
A
So
we
will
not
lose
this
good
spirit
here,
but
there
are
also
two
consequences:
our
actions
that
we
take
after
this
announcement.
So
we
search
for
a
new
culture,
so
the
the
call
from
candidates
has
already
been
sent
to
the
meaning
list.
A
monster
goes
language.
We
already
have
some
can
yet,
but
still
exciting.
A
Volunteers
and
we
will
review
the
conjugates
with
the
IHF
chair.
The
second
aspect-
and
this
is
also
following
some
discussion-
I,
had
this
week
with
the
rhf
Jay
Allison.
We
come
to
the
conclusion
that
it
can
be
a
good
time
also
to
use
this
opportunity.
I
said
in
the
search
of
a
new
culture
to
rethink
what
is
energy
and
what
will
be
the
next
step,
energy
just
for
the
memory
energy
started
in
March
1999.
So
it's
a
I,
think
one
of
the
oldest
research
group
still
active
in
energy.
A
So
the
idea
will
be
not
to
erase
everything
but
to
reconsider,
maybe
to
be
a
bit
more
in
phase
with
the
evolution
of
network
management
and
the
managed
environment
so
to
capture
may
be
a
bit
better
in
the
Charter
in
the
activity
of
those
newer.
Those
new
things
and
also
to
use
this
effort
to
have
a
big
new
ambitions.
I
will
say,
trying
to
highlight
a
bit
more
hot
topics
in
the
area
of
network
management,
but
also
understand
a
bit
more,
some
orientation.
A
So
what
what
it
means
to
manage
or
to
operate
at
scale
and
at
speed
in
those
new
environments.
So
this
is
ongoing
work,
ongoing
discussions,
ongoing
thoughts,
so
we
will
propose
an
concert
with
the
research
group
for
a
new
name
for
the
scope,
the
Charter
over
the
coming
months.
We
think
that
by
the
next
meeting
in
November,
we
would
like
to
have
progressed
on
this
thing.
A
We
don't
put
ourselves
or
had
my
my
stone
to
to
conclude
by
the
next
idea,
but
still
we
would
like
to
rather
start
it
as
this
new
evolutions
and,
of
course,
you
are
really
really
welcome
to
to
send
your
thoughts
and
contribution
if
you
have
good
ideas
for
all
topics,
if
you
want
to
candidate
for
new
culture
or
any
elements,
you
would
like
to
see
in
the
future
of
this
energy
and
final
note
on
that,
for
me
doesn't
mean
that
we
get
read
about
the
existing
work.
This
work
will
continue.
They
are.
A
C
C
D
C
C
Well,
this
is
the
Avenue
I
presented
in
which
they
would
now
I
am
concentrating
how
to
get
this
information
from
the
external
event
detectors
to
the
system
and
I
started
this
in
this
update
by
introducing
the
the
information
model
and
in
Jongno
that
is
required
in
order
to
get
that
information
first,
it
is
quite
simple,
but
it
is
based
on
two
models.
Is
the
the
events
and
the
notification
they
having
one
is
based
on
pillows
and
the
pillows
I
have
defined
only
three
basic
payloads.
C
We
have
also
defined
the
information
model,
this
even
itself
with
all
the
necessary
fields
that
are
typically
used
in
order
to
communicate
those
events
and,
finally,
the
notification
that
is
basically
this.
Is
this
this
using
this
structure
in
order
to
to
perform
this
communication
as
a
notification
that
this
in
this
case,
coming
from
outside
of
the
system?
C
C
The
second
topic
I
wanted
to
discuss
was
about
the
essential
artifacts
that
we
have
will
require
in
order
to
achieve
what
we
finance
intelligent
driving
networks
well,
first,
to
clarify
some
concepts:
I
want
to
define
that
here
that
AI
is
not
ml.
Normally
many
people
consider
machine
learning
is
hi
Laura,
yes,
but
AI
has
a
broader
spectrum.
It
is
based
on
perception,
rationing,
planning
and
and-
and
it
is
not
fully
exploded,
but
it's
already
present
in
some
part.
C
So
we
have
not
to
be
a
scary
of
incorporating
AI,
especially
for
network
management,
as
I
have
I
will
present
the
family
no
later.
Also,
another
concept
is
that
intelligence
is
different
from
intelligent.
That
means
that
intelligence
is
more
also
more
broader
terms
that
emphasize
the
Catalan
of
the
information
and
how,
on
the
management
at
information
intelligent
emphasizes
the
smartness
of
one
device.
No,
they
don't
have
to
be
present
at
the
same
time
or
I
mean
so
why
applying
a.
C
E
C
Iii
I
cannot
give
you
all,
for
example,
any
planning
that
you
can
be
used.
Let's
say
in
a
rotor
Rooter
is
constantly
taking
the
the
best
route
from
point
A
to
point
B
if
it
is
taking
some
heuristic
in
order
to
determine
that,
for
example,
in
geographical
routine,
or
something
like
that
that
every
stick
is
part
of
a
I.
What
I
I
want
to
point
something?
Ok,
no
sorry
I
want
to
point
something.
Normally,
the
definition
of
AI
in
my
in
my
in
my
experience
according
to
years
is
everything
that
is
not
being
used
today.
B
C
Why
I
apply
this
deep
application
of
AI
into
the
network
management?
It's
because
Madison
management
decisions
are,
we
know,
are
becoming
more
and
more
complex.
For
example,
one
mention
is
that
before
we
had
the
what
to
answer
the
question,
is
there
a
problem,
my
system,
but
today
the
question
is
more
broad
like:
where
should
I
meet
this
bit
or
machine
according
to
my
goals,
and
the
goals
can
be
very
complex
so
and
that
must
be
resolved
automatically
for
thousands
or
even
more
elements
at
the
same
time.
C
Well,
also,
the
operational
environments
where
the
systems
are
working
are
more
and
more
dynamic,
and
this
is
emphasized
by
supposition
and
probability
and
never
returns.
A
shoe
and
automation
are
required
for
that.
Also
we
have
that
network
devices
are
becoming
autonomic.
We
have
here
also
in
IDF
with
animals,
and
there
are
complex
decisions
that
had
been
taken
and
Sylla
touch
network,
for
example,
they
have
to
fully
exploit
programmable
elements
and,
finally,
when
I,
why
not?
Why
not
using
AI
is
just
a
method
on
resources,
not
solutions.
Solutions
are
made
after
that.
Well.
C
We
have
access
permit
to
that
technology
to
have
access
to
his
amount
of
intelligence
data
from
the
system
that
they
are
managing.
Okay,
the
knowledge
derivative
can
be
used
to
decide
the
stability
response
on
any
event
or
situation.
So,
finally,
we
have
a
constantly
evolving
model
that
we
can.
Why
not
say?
C
Today
are
gonna,
say,
dreams
are
quite
general
and
have
not
particularly
specific
specific
for
networking,
and
we
have
some
news
cases
like
predicting
traffic,
behavior,
iterative
network
communication
or
the
assessment
of
administrative
policies
that
is
quite
important
in
networking,
so
the
standardization
issues
that
are
behind
or
imagine
known
from
this
application
would
be
to
facilitate
in
order
to
facilitate
the
system
of
methods
from
different
vendors
would
be
that
the
methods
used
to
retrieve
the
information
must
be
the
quality
of
those.
The
methods
must
be
assured,
so
we
have
to
assess
the
quality
of
the
method.
C
So
when
you
should
know
customer
or
whatever
they
they
can
check,
if
one
method
is
better
than
other
in
some
in
some
terms
that
can
be
validated,
but
that
stakeholder
also,
the
types
and
quality
of
information
that
is
retrieved
from
a
system
or
object
must
be
consistent.
Consistent
in
terms
of
that,
if
we
are
receiving
one
mesh.
C
D
Now
yeah
yeah,
so
I
believe
for
the
for
the
different
plane.
We
may
have
a
different
situation
for
the
control
plane.
We
may
be
able
to
directly
monitor
the
interactions
of
the
routing
protocols
and
we
also
already
have
the
you
know
the
lot
of
database
for
the
management
management
train
data
and
also
for
to
monitor
user.
As
a
Ras
and
the
we
are,
you
know
the
operators
say
is
able
to
pin
down
some
user
flows
to
monitor
their
performance
experience.
So
I
think
we
as
a
winner.
D
E
Not
problem
damn
Bogdanovich,
it's
not
problem
in
the
possibility.
It's
a
problem
in
the
willingness-
and
my
other
question
is
which
is
my
that
was
a
comment
and
like
my
question
here
is:
what
are
you
missing
in
the
existing
technologies
that
we
have
today
that
you
cannot
develop
to
collect
the
data
from
the
network?
What's.
D
E
Cannot
unify
it?
Let
me
answer
this.
You
cannot
unify
different
types
of
data,
and
single
data
is
because
they
have
some.
They
have
different
characteristics
that
you
cannot
put
in
a
single
database.
You
have
to
have
multiple
different
databases,
one
has
to
be
chronological
and
the
other
one
will
be
no
transact.
Yeah.
D
Yeah
I
mean
that's
a
different
layer
right.
You
are
talking
about
how
the
data
are
organized
for
for
the
application,
but
there
are
also
interface
how
the
data
extracted
from
the
network
devices.
So
there
are
currently
are
multiple
different
proposals,
and
maybe
we
don't
try
to
unify
to
the
single
interface,
but
at
least
for
each
each
plane
is
preferred
to
have
a
sense
of
one
or
two
where
few
are
standard
interface.
So
they
don't
need
to
talk
to
run
it
for
battery
in.
E
My
opinion,
it
is
important
to
understand
the
data
in
which
interface
it
comes
over.
It
is
much
less
because
what
we
have
today
might
not
be
sufficient
tomorrow
and
the
date
the
data
has
to
be
consistent,
but
the
interface
is
more
or
less
than
it
can
be
adapted.
That
could
be
changed.
Okay.
Thank
you.
One.
G
I
think
there
are
a
couple
of
things
I'd
like
to
point
out:
telemetry
means
different
things
at
different
layers.
So
if
we
look
at
it
at
the
bottom,
we
are
looking
at,
you
know
bites,
and
packets
per
second
kind
of
a
thing
may
be
related
to
a
single
connection
like
a
per
connection
information,
but
very
quickly.
It
becomes
a
response
time,
calculation
for
HTTP
or
it
could
be
responsiveness
for
some
other
application.
So
at
different
layers.
That
means
different
things
and.
H
G
G
G
D
G
D
B
D
Those
issues
like
some
institution
right:
it's
try
to
provide
the
real-time
traffic
performance
measurement
for
force
for
user
flows.
So
that's
not
available
today,
mm-hmm
right,
that's
just
one
example,
and
there
may
be
some
other
other
works,
for
example,
for
the
we
are
doing
some
work
on
the
control
plane,
our
monitoring,
so
we
try
to
extend
the
BMP
protocol
to
support
more
thoroughly
control,
plane,
merriment
so
that
some
ongoing
works,
but
there's
not
existing
today,
so
I.
So
that's
why
this
work
we
want
to
identify.
I
D
There
are
a
lot
of
contour
resources
available,
but
but
for
example,
if
you
want
to
get
some
statistics
for
some
special,
specifically
user
flows,
it's
a
difficult
to
do
today,
because
there
are
limited
resource,
and
sometimes
you
may
need
to
dynamically
interactively
a
hit
that
resource
to
include
designate.
What
fellow
you
want
to
monitor
so
there's
something
we
also
see
hist.
You
thanks.
A
Okay,
just
commenting
as
myself
one
area
I
think
that
is
a
research
group
could
contribute
or
work
on
is
the
aspect
of
metadata,
because
I
think
we
already
have
quite
a
lot
of
data
and
measurement
protocol
and
whatever.
But
as
Dean
pointed
out
it's
to
understand
this
data,
usually
we
there
is
a
challenge
to
be
able
to
describe
metadata
so
that
this
can
be
used
by
other
application
or
analytics
in
the
pipeline
to
to
understand
the
data
that
is
coming.
A
Usually
you
can
get
one
flow
data
I
mean
protocol
are
designed
to
get
some
one
source
of
data,
the
data
they
expect
to
have,
but
you,
if
you
are,
if
you
are
also
a
capability,
if
you
have
multiple
source,
you
can
collect
data
from
multiple
protocols.
Multiple
points
there
is
also
the
aspect
of
being
able
to
get
those
multiple
sources
and
correlate
them
to
get
a
different
understanding
of
what's
happening
from
multiple
point
of
view,
either
different
dimension
or
different
endpoints.
Very.
D
A
J
J
H
J
J
That
what
they
miss,
amazed
and
I
think
this
is
important
in
the
direction
would
be
basically
a
mechanism
for
not
trying
to
to
build
a
theory
of
a
unified
protocol
for
everything
mechanism
for
managing
data.
I
mean
the
for
managing
metadata,
how
you
can
a
certain
element
that
is
going
to
consume
the
data
that
is
coming
from
whatever
the
measurement
points
is
able
to
to
know
which
others
later
on,
how
to
process
them.
J
Yeah
I
think
I
think
that
if
we
can
move
forward
in
a
way
in
which
we
would
facilitate
the
measurement
points,
the
way
in
which
they
can
be
care
which
they
are
they
are
going
to
provide
and,
on
the
other
hand,
the
the
others
to
announce
which
which
data
they
are
trying
to
to
consume.
That
would
be
an.
J
D
B
D
Used,
but
maybe
here
in
ITF,
we
have
more
about
the
standard
in
for
other
components,
as
you
said,
for
the
data
export
right.
If
you
the
data
plane,
we
may
use
the
IP
fix
and
UDP
directly
from
the
device
of
holding
phoning
pass
and
maybe
for
the
management
clean.
We
use
a
gr
PC
and
some
other
protocol
sleep.
We
don't
try
to
you
want
news,
but
we
first
we
need
to
identify
what's
available
and
even
for
this,
existing
protocols
was
missing.
So
that's
a
first
task.
We
want
to
want
to
do.
J
Yanis,
that
precise
naming,
maybe
something
about
thinking
or
something
or
stop
playing
with
the
D
and
research
terms,
to
start
with
how
something
that
would
be
similar
to
what
John
has
become
for
management
would
be
basically
how
to
do.
Declare
the
scheme
of
data
that
you're
willing
to
consume
or
produce
yeah.
Just
as
a
suggestion
is
something
that
you.
A
A
F
F
By
now,
everybody
should
know
that
Network
slicing
is
now
concept
which
is
now
not
on
research,
but
it's
engineered
worldwide
by
all
the
players,
as
I
understand
in
some
form
of
shape.
It's
probably,
as
the
concept
is
older
than
me,
but
it
will
outlive
me
by
far
and
the
concept
itself
is
not
new,
namely
that
the
problem
of
network
services
is
partition
different
in
such
a
way
that
the
infrastructure
supporting
it
will
be
able
to
deliver
or
to
support
each
partition
account
separately.
F
F
This
sort
of
concept
appears
to
be
very
useful
in
a
and
a
number
of
important
scenarios
which
are
now
as
I
understand
it,
driven
from
key
requirements
of
the
industry.
There
are
many
listed
here,
but
one
of
the
special
one
which
I
think
that
the
ship
may
be
one
focus
point
is
about
the
high
precision
service
networks
which
are
appearing,
namely
the
possibility
to
create
parallel
slices
in
such
a
way
that
each
of
them
to
have
a
level
of
guarantees
either
for
the
quality
of
service
or
kpi's.
F
It's
unconceivable
to
have
something
new
based
on
IP,
because
either
is
not
guaranteeing
anything
to
guarantee,
let's
say,
low,
latency,
plus
many
other
things
for
all
the
type
of
services.
So
it's
a
good
idea
to
split
beyond
the
problem
space
and
to
understand
how
to
do
this
well
before
this
is
fundamental,
but
it's
something
practical
as
well.
In
addition,
there
is
another
practicality
of
this
splitting
is
that
some
of
these
IDF
protocols
could
be
deployed
in
a
slice.
F
F
In
addition,
that
the
problem
of
how
you
compose
decompose
on
demand
and
how
to
do
all
the
other
things
are
our
least
important.
In
addition
to
this
detailed
one,
I
would
like
to
add
two
more,
namely
about
the
way
in
which
networks
I
see
could
be
scaled
up,
in
particular
management
scale,
up
not
to
tens,
but
maybe
10,000,
maybe
millions
of
possibilities.
The
practicality
as
well
as
how
you
perform
it.
F
It's
probably
a
key
research
topic
now
in
the
future
as
well,
and
also
to
how
these
slices
could
be
used
for
precision,
service
or
network
services.
Now
I
point
out.
One
thing,
which
is,
it
might
be,
that
the
networking
element
of
the
future
will
be
based
on
slices,
there'll
be
many
types
and
they
are
supported
by
an
underlying
lightweight
mechanism
to
make
the
existing
network
fabric
a
bit
more
soft,
but
also,
if
you
aren't
controlled
by
the
management
which
is
in
slice
management
as
well
as
overall
end-to-end
orchestration,
which
neatly
done.
F
F
Clearly,
this
is
a
scope
or
problems.
These
problems
are
in
the
scope
of
NMR
G,
so
I'm
suggesting
that
maybe
you
need
to
look
at
it
and
your
feedback
will
be
highly
appreciated
and
once
again,
just
to
say
that
this
material
is
not
driven
by
one
RFC
or
many,
but
is
a
way
to
present
some
but
a
longer
list.
The
management
problems
which
need
to
be
resolved
here
as
well.
Another
part
some
some
engineering
problems
as
well.
F
That's
for
sure,
but
before
concluding
a
part
of
this
tutorial,
which
I
presented
two
weeks
ago
in
Montreal
at
the
academic
conference,
including
around
a
hundred
academic
references,
but
also
a
road
map
of
existing
references
of
work
in
in
slices,
which
is
resistances.
That
concludes
my
own
presentation.
Now,
it's
your
turn
to
lock
me
down.
Thank
you
or
not,.
E
They
are
Bogdanovich,
so
network
slicing
is
an
interesting
management
of
network.
Slicing
is
an
interesting
problem
and
I
spent
personally
and
within
my
work
time
thinking
about
management
of
the
network,
slices
of
the
control
and
management
pipe,
but
I'm
so
struggling
to
get
what
will
be
definition
of
the
network
slicing
in
a
data
plane
because
with
whenever
I
hear
the
airplane,
I'm
hearing
different
impressive
interpretation
and
they
all
start
dedicated
network
resources
are
too
expensive.
So
I'm
thinking
are
we
talking
about.
You
know
the
lambdas
in
the
optical
layer.
Are
we
talking
about?
You
know?
E
E
F
Thank
you.
The
question
is
extremely
valid,
but
my
suggested
answer
is
that
the
current
important
topics
about
how
to
manage
an
orchestrate
this
new,
relatively
new
concept
now
in
this
forum
lies,
was
a
management
and
control,
plane
and
service.
Plane,
obviously,
was
an
interaction
between
the
three
in
the
footprint
in
the
data
play.
F
Don't
want
to
name
it
you
and
they
organize
it
was
the
issue
is
or
the
current
problematic
area
is
in
the
manage
management,
plane
and
interaction,
was
a
control,
trainer
service
way
and
to
act
as
one
and
bring
it
down
later,
who
some
ways
to
link?
It
was
one
of
the
many
or
all
the
way
with
all
the
things
related
to
data
plane.
F
Instantiation,
if
you
start
from
the
data
plane,
for
example,
form
the
segment
routine
you'll
end
up
with
not
much,
because
segmenting
will
not
be
applied
to
all
the
parts
of
the
network
as
well
as
other
things.
Therefore,
it's
I
would
suggest
at
the
priorities
to
use
a
management
plane
problem
first
and
then
map
to
a
number
of
options
for
the
data
plan.
Ok,.
A
A
B
K
So
let
me
get
started.
My
presentation
is
slightly
different
from
what
alex
talked
about.
It
is
initially
I
thought
we
will
talk
about
the
use
cases
in
network
management,
network
slice
management,
but
it
turned
out
to
be
a
very
high-level
overview
of
the
work
we
have
done
in
previous
couple
of
box
related
to
network
slices.
K
It
should
support
multi-tenancy,
and
one
of
the
differentiating
factor
is
that
these
resources
should
be
managed
by
the
tenants
themselves,
not
by
the
operator
of
the
infrastructure,
and
then
slices
should
also
have
some
sort
of
dynamic
capabilities.
If
you
want
to
augment
the
resources
within
a
particular
slice.
So
this
is
just
an
informal
definition,
because
if
you
try
to
put
it
in
one
sentence,
you
can't
finish
it.
So
let's
go
to
the
next
slide,
please,
and
so
why
do
we
want
to
work
on
networked
slices?
K
First
of
all,
we
have
to
differentiate
it
from
traditional
Network,
traditional
VPNs,
because
that
was
one
of
the
question
that
was
raised
in
one
of
in
first
boss
of
networked
slices,
and
we
said
that
the
kind
of
resource
guarantees
we
want
to
achieve
are
at
per
flow
level
and
they
are
different
from
Howrah.
Pians
are
built.
On
top
of
that,
we
look
at
slices
in
terms
of
service
verticals.
K
And
finally,
for
me,
the
most
important
aspect
is
to
look
at
management
of
network
slicing
as
a
way
of
understanding
slices
from
end
to
end
perspective
management
plane
is
the
only
is
the
only
level
from
which
we
can
look
at
end-to-end
slices,
how
resources
are
allocated,
abstracted
or
mapped
in
the
network.
So
if
you
talk
in
different
working
groups,
they
are
looking
some
functionality
related
to
the
slices,
but
it
is
very
difficult
to
put
it
all
together
and
if
we
take
up
this
work
here
in
NMR,
see
those
are
that's
exactly
what
we
can
achieve.
K
So
when
we
were
initially
working
on
a
network
slicing
buff,
we
first
started
with
the
functional
aspects
of
the
slices
and
we
classified
into
three
categories.
First,
one
is
pretty
simple
related
to
customization
that
an
infrastructure
provider
will
provide
Network
slice
as
a
service
to
enterprise
customers
and
using
word
exploiting
the
virtualization
capabilities.
You
could
allow
enterprise
customers
to
build
their
own
VC,
PE
or
Sdn
solutions
by
placing
their
network
functions
wherever
they
want
to
in
their
networks.
K
The
second
aspect
was
the
resource
assurance
which
most
of
the
people
always
talk
about
generally
related
to
low
latency
broadband
and
mntc
type
of
slices,
and
then
we
also
identified
another
category,
which
was
how
can
you
evolve
new
protocols
in
a
given
partitioned
network?
So
this
will
be
more
like
a
low-level
partitioning
of
the
slices
of
the
resources.
So
you
could
choose
to
run
your
own
application
protocols,
or
maybe
other
network
or
control
protocols,
given
a
particular
set
of
network
resources,
but
that
work
was
considered
pretty
broad.
K
So
we
just
started
focusing
on
the
network
management
aspects
and
from
from
the
management
of
slices,
we
identified
that
we
would
need
some
kind
of
end
to
an
orchestration
capability.
We
would
need
ability
for
tenants
to
customize
the
slices.
We
need
operations
and
management
of
the
slices,
F
gaps,
functionality
and,
more
importantly,
how
they
will
interoperate
across
multiple
domains
that
are
under
the
control
of
different
administrators.
And
how
would
you
interconnect
parts
of
the
slices
that
belong
to
different
Network
next
slide?
Please
you.
K
So
this
is
the
reference
architecture
where
we
talked
about
that
an
NS
provider
which
is
Network
slice
provider
will
have
a
very
we'll
have
a
kind
of
northbound
interface
to
the
tenant,
and
it
will
have
Saltburn
interface
to
network
sliced
orchestrators.
These
orchestras
talk
to
their
own
infrastructure
and
at
at
this
point,
when
we
are
talking
about
management,
we
don't
touch
the
control
plane
or
the
data
plane
functionality
next
slide.
K
We
want
to
aggregate
information
from
different
subnets
of
the
slices
and
put
them
together,
map
them
to
the
tenant
slices.
Then
in
services
next
slide,
please.
So,
let's
skip
this
part
since
I'm
running
out
of
time
and
next
slide,
so
the
previous
slide
was
just
running
through
the
use
case.
How
SDI
and
CSI
interfaces
are
useful
in
terms
of
entire
flow
of
instantiate
in
your
slice,
so
from
an
MRG
perspective.
K
For
me,
the
most
important
aspect
is:
how
do
you
see
end-to-end
slices
from
management
plane
and
how
do
we
integrate
the
operations
on
the
slices
and
the
sources
together?
Some
of
the
challenges
with
from
management
part
is
that
we
will
always
be
challenged
that
there
are
so
many
yang
models
available
which
are
service,
specific
or
infrastructure
specific.
How
are
you
going
to
use
them
or
exploit
them,
or
are
you
going
to
write
new
models?
K
E
A
E
E
E
You
are
talking
about
the
network
slicing
service,
but
where
does
your
service
model
sits
on
this?
Are
you
talking
to
this?
Your
infrastructure
configuration
model
or
where's
your
service
model
on
that
part,
and
then
the
last
part
saying
that
data
playing
the
control
point
will
resolve
somewhere
else
without
correlating
you
know
with
that
work,
I,
don't
think
we.
You
know
it
can
be
very
successful.
Yeah.
K
So
let
me
answer
your
questions
in
the
reverse
order.
It's
simpler
that
way,
so
we
have
to
start
somewhere
I.
What
you
have
in
mind
is
a
bottom-up
approach,
which
is
perfectly
fine,
there's
nothing
wrong
with
that.
But
what
we
are
trying
to
do
is
from
an
emoji.
We
are
taking
a
top-down
approach,
define
the
management
aspects
of
it
and
let's
tackle
the
control
and
data
plane
issues
those
issues
we
have
already
identified
in
our
initial
gap.
Analysis
document-
it's
just
that
we
haven't
discussed
it
over
here.
K
If
you
look
at
the
original
gap,
analysis
documents
lot
of
things.
What
we
wanted
to
do
about
data
plane
was
highlighted
there.
It's
just
that
no
working
group
is
willing
to
take
that
work
right
now
and
in
terms
of
service
model.
My
perception
is
that
my
service
model
will
be
inserted
from
the
CSI
interface
and
that's
from
where
my
service
description
will
come
in
two
ns
provider,
the
network
slice
provider,
which
is
actually
going
to
do
the
most
of
the
heavy
lifting
for
the
network,
slice
management.
E
E
Reserve
we
nokia
service
model,
the
one
thing
that
I
mentioned
also
yesterday
in
this
picture.
In
my
opinion,
the
most
important
interface
which
you
not
shown
here,
is
from
Orchestrator
to
the
sir.
So
basically,
the
model
that
we
are
referring
here
is
not
the
service
model
is
a
connectivity
model.
This
is
a
brand
new
model
that
I
mentioned
it
to
Alex
yesterday,
so
I'm
working
on
it
now
that
basically,
the
interface
between
Orchestrator
and
infrastructure
is
a
connectivity
model
and
from
there
is
a
service
model.
E
Basically,
you
receive
a
request
that
I
want
to
connect
a
BB,
u
2s
gateway,
and
we
translate
that
onto
a
service
model.
So
if
I
want
to
start
with
this
architecture,
I
start
with
that
interface,
in
my
opinion,
is
the
most
important
aspect.
The
interface
from
to
the
tenant
is
important,
but
it's
not
that
important.
I
agree.
K
D
K
E
Not
service
model
is
a
connectivity
model,
I'm,
just
a
stressing
that
one
David,
so
the
second
one
is
and
I
see
this
one
in
various
presentation
under
network.
As
I
say
there
are
all
the
time
they
talk
about
end
to
end,
but
I
never
see
end
to
end
concept.
Whatever
they
talk
about
is
not
really
an
network.
A
slice
is
a
transference
list.
We
just
doing
the
connectivity
at
the
transport
layer
if.
E
K
A
Sorry,
but
we
have
this
power
outage
and
we
are
really
missing
time,
so
we
can
not
solve
all
the
discussion
on
network
slicing
and
the
management
aspect
of
metal
slicing
now,
and
we
are
really
short
on
time.
Just
my
my
comment
as
a
chair
is
currently
I:
don't
really
see
clearly
what
the
research
challenges
are,
so
this
can
be
worked
out
offline,
the
mailing
list
with
presentation,
if
you
Tod
still
opportunity
to
express
what
really
the
pain
energy
could
investigate.
F
E
If
you
identify
a
problem,
the
point
of
the
research
group
and
working
groups
is
to
find
other
people
who
understand
that
problem
and
help
you
rehash
the
solution
of
that
problem.
So
if
nobody
understands
it,
we
want
to
learn
and
understand
what
is
the
problem
to
help
you
solve
that
problem,
because
many
of
us
are
working
on
the
same
or
similar
problems.
E
E
A
Are
no
problems
and
I
think
the
way
to
present
them
and
make
people
understand
and
adhere
to
the
vision?
It's
still
not
there.
So
we
have
we
had
this
first
opportunity
to
discuss,
but
I
think
we
need.
We
did
more
sorry
qur'anic
give
you
the
flower,
we
really
all
the
time
and
we
still
have
presenters
and
with
respect
to
their
also
their
preparation.
We
need
to
give
them
the
flow.
Thank
you.
A
Thank
you.
This
is
interesting
that
amendment
using
the
important
running
my
name
is
boom
gone
home
I'm
looking
for
ETL
I,
both
this
rep
and
his
flight
prey
attack
the
University
and
the
pure
telecom
help
us
to
prepare.
So
thank
you,
and
this
is
the
third
Bojan.
Also
we
have
a
one
time
to
good
each
end
in
last
99
meeting.
So
it
is
the
second
time
to
pretend.
A
Yeah,
so
in
this
rep
we
want
to
utilize
the
improved
learning.
I'll
show
you
what's
the
region,
what's
the
motivation,
and
nowadays
the
team
running
are
integrated
to
other,
so
we
are
also
of
allotted
approach
in
our
researching
yeah.
Why
we
want
to
use
all
the
important
running
in
the
tornado
measurement.
There
are
two
reason:
the
person
region
is
the
decision
problem.
A
As
you
know,
the
supervised
learning
in
suitable
for
single
decision,
for
example,
image
electrician,
but
in
real
world
in
network
it
is
not
composed
of
the
single
decision,
for
example,
body
for
Latin
all
were
multiple
links
and
the
Botsford
or
conjuration
are
existing
in
the
real
world
in
real
network.
So
in
the
sequential
decision
we've
lifted,
the
important
learning
is
water,
chute
world
or
through
sharp
the
problem,
and
the
other
reason
is
to
Dara
it.
You
know
in
supervised
learning
it
based
on
tour,
lots
of
doTERRA
so
and
is
lon
I
am
it
is
Vera.
A
So
in
the
immediate
recognition
we
can
study
and
we
can
teach
what
is
a
poor.
What
is
orange,
but
in
the
real
world
is
Pedro
set
and
a
song
says
that
it
is
not
possible
gathering
Terra.
So
if
we
deploy
and
any
preserved
the
problem
or
we
cannot
learn
by
example,
so
another
alternative
is
run
by
experience.
It
is
the
infection,
so
tasks
are
implicitly
given
through
the
reward.
A
A
On
Terrace,
so
it
is
the
a
cheap
paste,
intelligent,
ready,
control
scheme
using
parallel,
so
I,
don't
know,
you're
familiar
with
rotary
inverted
pendulum
system,
so
our
core
is
to
raise
the
pendulum
top
or
in
this
way,
so
to
do
this.
There
are
two
input
to
state.
It
is
a
pendulum
angled
era
and
am
endowed
era,
and
this
data
are
delivered
to
this.
A
Each
component,
though,
so
in
this
machine
we
run
the
tiara
module
and
the
features
the
action
is
to
both
power
power
during
the
era
so
and
then
this
motor
power
try
to
raise
the
poor
in
this
way.
So
yes,
so
this
for
connecting
this
system,
we
use
the
wire
connection
the
internet.
So
there
are
no
problem
so,
but
or
in
real
world.
It
is
not
the
same
easy
so
make
or
similar
story
award.
We
try
to
make
the
poor
sleepy
and
then
we
try
to
stop
the
apostrophe
without
ripping
measurement.
A
So
nowadays
we
are
industrial
alte
service.
There
is
the
load,
reimburse
see
dependencies
then
and
edge
computing,
a
intelligent
control
system.
We
are
succeed
to
integrate
it,
for
our
remaining
work
is
to
integrate
to
other
tripping
management.
Yes,
so
we
shows
flatly
to
our
simulated
environment.
We
use
the
open
air
GLM,
so
the
early
definition
is
racket
is,
and
we
use
three
data
model
to
every
jump,
the
tip
unit
or
a
through
C
and
ACC.
A
So,
as
I
said
that
there
are
two
anger
and
the
world
the
motor
power
and
the
reward
formula
is
like
this,
so
you
can
find
the
detailed
information
about
this
open
source
I
skip,
so
it's
the
reader
to
of
our
simulation.
So,
as
you
can
see
nowadays
the
better
than
typical
Network
a
she
is
somewhat
successful,
resulting
eqn.
So
our
remaining
words
to
integrate
a
threesome
order
in
our
head.
A
So
it
is
our
final
phase.
So
we
you
noted
a
we
spend
much
time
in
ITF
ir
tape
to
deploy
much
money
but,
for
example,
ID
net
net
emotional
named
RG,
but
nowadays
the
animal
is
the
best
play
to
handle
the
issue.
So
we
want
to
share
and
discuss
and
the
clerk
and
the
lon
how
to
deploy
on
much
learning
in
the
network
and
network
management.
So
to
do
this,
the
first
thing
is
yes,
so
many
person
asks
me
what's
the
reason
to
deploy
much
money
in
network
is
the
it
only
research.
A
K
A
Where
is
the
best
place
to
deploy
interests,
the
function
in
a
two
area
in
every
NATO
node?
Who
are
one
word
to
NATO
note
it
is
we
teddy
stretchy
and
all
the
reason
why
we
ask
that
is
if
the
environment
and
the
problem
is
too
complex
and
we
cannot
cuff
labrum,
so
simple
is
the
best.
So
we
believe
that
if
we
deploy
in
a
tumble
machine
learning,
a
top
function
tomorrow
is
from
Barbara,
and
the
final
is
that
we
show
the
Queen
a
to
see
and
a-three
see.
A
A
A
A
H
H
So,
as
we
know,
the
network
effect
is
always
a
challenge
in
the
industry.
Now
there
are
always
new
technologies
emerging
to
deal
with
the
problem,
for
example
the
STM
for
the
service,
autonomic
service
delivery
and
the
the
new
emerging
intent
driven
Network,
which
we
will
have
a
dedicated
discussion
on
Friday
in
emerging,
and
also
we
had
many
effort
on
the
how
to
utilize
artificial
intelligence,
machine
learning
technologies
for
automatic
control
loop.
H
H
H
So
so
far
we
figured
we
at
least.
We
need
two
dimensions
to
this
very
comprehensive
issue.
The
one
dimension
is
from
the
operators
perspective.
What's
a
final
effect.
For
example,
there
are
several
aspect:
the
self-organization
which
basically
related
to
a
various
kinds
of
a
park
and
play
technologies,
how
to
get
connect
with
a
network
menu
system
and
how
to
build
identities,
autonomically,
etc
and
self
configuration
how
to
make
the
parameters
automatically
be
configured
self
optimizing.
We
have
various
technology
to
make
the
traffic
more
in
a
more
efficient
way.
H
Our
self
diagnostic
and
self-healing
is
about
the
fourth
management.
Another
dimension
is
the
network
system.
Myself
needs
some
key
capabilities
to
support
the
final
effect.
They
are
so
in
a
mainly
three
areas
with
the
first
one
is
the
about
the
merriment.
We
call
it
a
never
perception
and
also
that
telemetry
and
never
visit.
H
So
we
come
up
with
this
table.
I'm,
sorry
for
the
readability
and
I,
don't
have
time
to
go
much
in
details,
I
just
peek,
maybe
one
a
couple
of
examples,
for
example,
photo
self-organization
the
level
sorry
I
only
include
from
the
level
2
to
level
5,
not
in
the
level
1,
because
never
1
all
about
something
already
widely
used
in
the
industry
and
and
from
level
3
to
level
5
does
something
we
just
imagined
whether
the
technology
will
go
in
this
approach,
for
example,
for
the
self-organization.
H
Maybe
the
next
step
should
be
the
the
network
can
figure
out
how
to
make
domains
by
itself
or
without
human
interview.
That's
only
a
very
simple
example
and
further
self
configuration
may
be
the
next
next
step
is.
The
network
element
can
can
compel
some
detailed
configuration
by
itself
rather
than
every
single
detail
will
be
calculated
in
the
centralized,
a
controller
or
never
menu
system.
H
So
we
we
really
want
to
hear
from
you
whether
this
kind
of
work
is
valuable
to
have
such
kind
of
levels
of
the
odd
numbers
Network,
and
we
already
have
two
dimensions
from
the
prototype
perspective
of
operators
and
the
key
capabilities
of
the
network
itself.
So
is
this
reasonable
or
something
missing
and
for
the
self
organizations
of
configuration
etc?
We
we
want
to
know
whether
there's
also
something
very
important
missing
in
the
in
the
table.
H
J
Developers,
I
think
that
in
general,
I
have
not
had
timing
to
go
in
detail
in
the
document
or
whatever
it's
been.
The
first
news
I
have
written
I
must
confess,
but
I
think
that
there
is
a
clear
interest
possessory
in
the
what
this
group
Conda
can
do
as
a
researcher
in
providing
precisely
you
know,
sort
of
a
theoretical
foundation
of
what
natural
management
should
become
in
the
new
era
and
I
think
a
finer.
This
is
available
as
a
research
activity.
J
The
message
I
don't
think
that
any
kind
of
pure
real
standardization
could
come
out
of
this,
but
on
the
other
hand
it
would
be
able
to
provide
for
nation's
forefathers
in
the
residual.
So
I
think
is
interesting
in
the
in
despite
land
visits,
I,
don't
know
exactly
what
the
to
DM
dimensions
at
the
trace
and
in
that
I
think
we
would
need
some
ma
father
thinking
that
as
a
as
a
first
move
in
providing
there's
a
new
theory
of
management,
if
you
like
is
something
that
I
think
it's
interesting.
J
H
I
A
M
M
The
network
can
must
be
automatically
adaptive
in
our,
so
this
is
translated
into
different
definitions
such
as
have
sharp
meta,
key
and
others,
and
we
use
this
kind
of
research
is
usually
addressed
by
a
network
manager
area
so,
for
example,
in
the
conference
such
as
I
am
norms
and
seen
the
same
and
the
in
in
IT
actually
and
I'm
going
to
talk
about.
This
may
be
the
correct
forum
for
that
is
DN
emerging.
So
next.
D
M
So
as
an
a
perspective,
this
is
the
main,
the
First
Nations
in
autonomic
network
in
the
Accord
2005
to
2006.
So
there
are
several
intestines
for
that,
for
example,
the
artificial
intelligence
in
network
management,
what
which
there
are
several
research
activities
in
the
90s,
then
self-organization
networks
and
declarative,
partisan
and
so
forth,
and
there
are
also
several
papers
and
some
of
these
papers
were
written
by
contributors,
often
MRG
such
as
John
and
other
people
next
slide.
M
So
there
is,
we
can
talk
about
in
research,
it's
called
for
the
autonomic
networking,
considering
different
kinds
of
technologies
and
the
perspective
and
approach
for
that.
So,
for
example,
our
chief
intelligence
policies,
machine
learning
and
some
of
this
of
these
topics
were
addressed
by
by
by
the
previous
presenters
in
this
meeting,
so
I
believe
this
kind
of
something
that
we
can
cannot
become
that
and
this,
the
the
research,
but
it's
balanced
in
the
whole
HD
spectrum,
so
this
can
be,
can
be
applied
for
violet
and
wireless
network
and
fixed
and
mobile
and
so
forth.
M
They
didn't
use
the
name
out
of
network
next
slide.
So
the
reason
and
adequacy
interest
in
autonomic
words
when
considering
academia,
research
perspectives
around
2011
thousand
thirteen
and
the
there
are
possibly
there
are
some
reasons
that
we
can.
We
can
cite
for
that,
for
example,
there
are
at
that
point
other
technology
that
were
gaining
momentum,
such
as
Sdn
and
NFV.
The
researchers
started
to
use
different
terms
for
at
our
network
and
such
as
cognitive
network
intelligence
briefing
network.
M
So,
for
example,
there
are
some
some
a
lot
of
frameworks
and
tools
and
the
and
the
researchers
parks
on
our
network,
but
it's
difficult
to
find
out,
especially
in
that
point
successful
deployment
case,
and
at
this
point
also,
we
started
some
some
work
on
the
standardization
for
Japan
and
et
assign,
and
also
in
the
IDF
with
the
animal
work
group.
Next
slide.
M
M
So
at
least
at
one
point:
there's
the
animal
working
group
and
at
the
the
organization
or
group
you
when
the
chairs
were
rising
the
charter
for
the
worker
group.
There
is
this
definition
that
which
the
the
data
points
are.
The
topics
that
are
outside
the
the
charter
can
be
encouraged
as
an
individual
submission,
but
most
more
quarter
as
an
emergency
submissions.
A
next
slide.
M
M
On
5g
and
IOT
and
smart
X
is
my
selfie
also
it's
important
to
have
a
better
link
with
real
world
operations,
for
example,
right
and
OGG
and
other
stuff,
and
also
it's
important
to
understand.
If
there
is
some
effective
deployment
in
the
reasons
for
the
lack
of
whites
gave
the
plant
all-na
dominant
work
next
slide
and
I'm
trying
to
be
very
fast,
so
I
want
that.
So
the
question
is
there
is
a
need
interest
or
inertia
for
automatics
3.0,
Indiana
Marty.
M
Also,
it's
important
to
have
an
integration
with
other
Network
automation
approach
such
as
zero
touch
network
and
service
management,
and
also
anything
to
make
me.
We
are
in
a
point
that
maybe
it
was
seem
to
revisit
the
75
75
s
and
five
and
six
we
shall
which
are
the
the
artsy's
that
puts
track
animal
working
group.
So
maybe
we
can
revisit
these
services,
for
example,
thinking
what
sort
of
anomie
scope
in
a
long
term
the
the
ruby
are.
We
charted
on
anima,
I
believe
in
in
november,
so
maybe
we
can.
M
A
E
A
big
proponent
of
autonomic
networking
I
disagree
that
we
have
to
go
ready
to
autonomic
3.0.
We
have
a
lot
of
work
to
do
in
the
current
framework
and
my
main
issue
is
these
days.
We
still
have
one
problem:
all
the
exception.
Handling
in
networks
is
still
done
by
the
human
operator.
We
have
to
figure
out
how
to
automate
the
exception,
handling
from
a
service
level
perspective
from
a
network
service
level
perspective
instead
of
just
focusing
on
the
device,
and
if
we
can
display
some
foundation
on
which
we
can
build
the
next
levels.
L
John
Strauss
nur,
so
I
broadly
agree
agree
with
Dean,
as
you
know,
I'm
a
big
fan
of
intent,
networking
I'm
working
on
it
actively
I
think
the
biggest
problem
with
going
to
autonomics,
3.0
or
X
dot
Y
is
that
everyone
forgets
about
knowledge
representation,
and
this
is
a
fundamental
requirement
for
any
serious
AI
practitioner
or
researcher
also
called
knowledge
engineering.
What
Dean
was
just
talking
about
has
been
called
the
qualification
problem
so
for
any
common-sense
rules
that
you
want
to
encode,
there's
a
huge
number
of
exceptions
right.
L
L
Most
IETF
errs,
don't
even
know
how
to
spell
the
word,
let
alone
what
it
means.
So
we
need
to
make
a
focused
effort
in
knowledge,
representation
and
as
a
general
thing.
Yes,
deep
learning
is
important
and
machine
learning
and
blah
blah
blah,
but
they're.
So
generic
that
we're
just
gonna
spin
our
wheels
and
not
do
anything
I
think.
A
L
A
We
are
over
time.
We
need
to
close,
because
there
will
be
another
working
group
after
that.
I
talked
with
Alex,
so
he's
not
it's.
Okay,
that
we
skip
his
presentation.
Thank
you
for
your
attention.
Thank
you
for
staying
with
us,
even
with
the
power
outage.
Thank
you
for
the
remote
presenters.
It
was
not
easy
to
to
understand
what's
happening,
so
we
have
still
ization
and
after
the
next
after
this
one,
which
will
be
a
more
general
discussion
about
networks
in
AI,
and
we
also
have
this
dedicated
meeting
tomorrow
on
Ethan
base
networking.