►
From YouTube: NMRG Interim Meeting, 2020-04-14
Description
NMRG Interim Meeting, 2020-04-14
A
A
First,
a
couple
of
reminding
slides
about
the
goal
of
the
Archaea,
the
IHF
conducts
research.
It
is
not
a
standards,
development
organization,
the
for
an
idea
for
cases
on
longer
term
research,
issues
related
to
the
Internet
and
our
affiliated
or
sister
organization.
The
internet,
Engineering
Task
Force
focuses
on
shorter
term
issues
of
engineering
and
standards
making.
We
would
like
to
make
it
clear
the
distinction
between
both
and
our
GF
can
publish
information
on
an
experimental
documents
in
the
RFC
series.
A
A
So
these
are
important
elements
we
invite
you
to
to
be
aware
of,
and
especially
as
a
reminder,
by
participating
to
follow
ITF
processes
and
policies.
You're
aware
that
any
IHF
contributions
is
an
application,
but
are
all
more
control
by
you
or
your
sponsors
must
be
disclosed
as
a
participants
or
at
indeed
to
any
ATF
activity.
You
acknowledge
that
written
audio,
video
and
photographic
records
of
meetings
may
be
made
public
and
for
information.
This
meeting
will
be
recorded.
A
Now
that
we
are
also
conducting
more
virtual
meetings,
and
especially
with
series
of
virtual
interims
that
have
been
following
the
IETF
107,
that
was
helpful
eerily,
there
are
some
kind
of
global
etiquette
across
research
group
and
working
group.
So
we
would
like
to
highlight
a
few
points.
Please
remember
that
all
sessions
are
being
recorded,
as
it
is
the
case
for
this
meeting.
Make
sure
that
your
video
is
off
to
save
man
with
keep
yourself
muted.
Unless
you
are
speaking,
we
will
end
on
the
cue
WebEx
chat.
A
You
want
to
say
something,
and
if
you
want
to
remove
from
the
queue
or
not
any
more
comments,
you
may
type
LH
and
the
cheers
will
end
the
queue
okay
and
when
you
are
speaking
please
for
the
sake
of
time,
please
stay
to
the
point,
be
concise
and
precise,
I
hope
everything
is
clear
until
now,
a
few
links
I
think
if
you
are
already
online,
it
means
that
you
know
a
bit
to
navigate,
but
this
may
always
be
useful.
The
agenda
materials
are
available
on
the
IETF
data
tracker.
A
You
have
the
etherpad,
which
is
a
shared
path.
You
may
know
its
past
events
etc
to
the
path.
This
is
also
what
we
will
use
as
a
basis
for
the
minutes.
The
recording
will
be
made
available
after
the
meeting,
and
please
aside,
please
add
your
name
at
the
bottom
of,
if
not
touch.
The
page
is
very
useful
to
track
attendance.
A
Overview
of
our
agenda,
for
today
we
have
a
packed
agenda,
a
lot
of
topics
we
would
like
to
cover,
and
especially
initially
we
had
planned
to
have
two
sessions
in
the
IETF
one
of
seven
meetings,
so
we
want
it
still
to
be
able
to
to
cover
most
of
those
topics.
So
this
will
be
a
bit
along
meeting
today,
but
we
hope
you
can
stay
with
us
and
Cydia.
So
this
is
the
current
first
load
meeting
resisting
an
agenda.
A
It
will
be
finished
in
1
minute
and
then
we
will
start
our
technical
presentations
with
Pedro
for
intelligent
reasoning
on
external
events
on
the
track
management
and
then
Sonia
will
give
us
an
overview
of
their
work
in
telefónica
on
metadata
based
aggregation
of
telemetry
flows.
We
will
have
then
new,
coming
from
from
me,
lad
on
fault.
Tolerance
for
cervix
function,
change
as
an
introduction
of
their
work,
and
we
see
also
connections
with
some
other
activities
in
IETF
on
SFC.
Then
a
new
draft
coming
to
the
energy
with
transpose
lies
in
tenth
and
final
notice.
A
Coming
from
based
on
some
work
in
a
European
project,
then
automate
the
design
of
network
services
from
network
service
requirement
is
a
research
article,
research
work
he's
also
in
the
scope
of
again
work
plan.
Then
we
have
some
draft
energy
and
ops
area
working
group
intent.
Classification
with
chu-young
will
give
us
an
update
on
the
current
document
and
some
next
steps
we
will
have
been.
You
are
giving
us
an
update
on
assurance
for
ibn
architecture,
which
is
working.
B
A
And
definition,
and
finally,
if
we
hope
we
have
still
time
to
give
you,
the
main
research
group
updates
at
the
end
of
this
meeting,
I
hope
the
agenda
is
fine
for
you.
If
you
have
any
questions
or
changes,
please
let
us
know,
because
it's
very
packed
and
we
would
like
to
start
right
away,
share
home
I,
don't
know
if
you
want
to
add
something:
oh,
if
we
can
move
on
no.
A
D
D
Only
certain
happens
for
network
management
and
slightly
I
have
been
working
on
this
together
with
Hamas
and
from
entity
to
mainly
particularize
the
requirements
from
Network
slicing
for
for
especially
from
this
AI
involvement
in
network
management.
So
first
quick
context
that
we
have
is
the
solutions
must
work
in
Symphony
with
all
the
network
management
solutions,
so
they
alone
it
works
to
grow
in
complexity,
to
be
faster
decisions,
etc.
So,
but
solutions
work
only
on
performance,
they
normally
the
other
AI
solutions
require
more
information
and
intelligent
reasoning.
D
Solutions
need
to
collaborate
with
network
to
retrieve
topology,
real-time
situation
etc,
and
they
need
some
special
efficient
semantic
representation,
annexation
or
data.
For
for
this
adoption
of
in
AI
promise
me
this
is
the
main
context.
We
have
been
working
on
research
challenges
and
topics
in
in
this
overall
integration
of
different,
not
just
machine
learning,
about
different
kinds
of
machine
of
artificial
intelligence
solutions
in
in
network
management
and
especially
again,
I
insist
that
we,
we
have
been
working
together
with
entity
to
get
requirements
from
the
West
lysine
in
order
to
fulfill
this.
D
So
it's
not
bad
thing,
but
what
I
mean
is
that
we
have
exploited
the
learning
face
of
intelligence
very
well,
but
now
we
have
to
move
forward
and
in
the
artificial,
intelligent
world
we
have
the
next
step,
and
it
was
also
commented
in
this
latest
conference.
The
next
step
is
rationing.
So
what
is
what
are
the
requirements
in
this
sense
of
from
reasoning
to
exploit
the
network?
Sorry
destroyed
artificial
intelligence
in
network
management.
D
We
aim
to
define
some
gaps
to
this
required
us
to
this
region
in
application
and
what
will
be
the
challenges
to
integrate
reasoning
into
next
slide?
Please
so
I
think
that
we
can
skip.
This
is
like
that.
Please
notice
in
this
one,
because
we
have
the
background.
I
will
just
comment
it,
so
we
have
different
kinds
of
network
systems
that
will
be
to
computers,
Sdn,
etc,
and
we
are
also
focusing
on
on
management
and
slice
slicing.
D
D
We
are
emphasizing
this
so
to
apply
the
item
and
romance
mean
we
need
to
move
beyond
machine
learning,
machine
learning,
it's
very
specific
tool,
and
it
has
very
specific
way
of
doing
things
like
it
requires
one
formalization
of
information
in
vectors,
etc,
and
some
very
efficient
or
even
though
it
can
be
computational
intensive,
is
very
efficient
calculations
to
get
some
kind
of
model
after
that
can
be
exploited.
Okay,
but
for
now
is.
D
It
would
be
good
for
freeing
the
machine.
Is
there
a
problem,
my
system,
but
now
we
have
more
complex
versions
like
well,
should
emigrate
what
this
built
machine
to
accomplish
my
goals.
Okay,
so
intelligence
is
needed
for
this
and
the
new
function,
for
example,
that
answer
this
question
hello
network
devices
to
be
more
autonomic
and
can
be
fed
to
the
animal
world
next
slide,
please.
So
one
key
that
we
are
working
on
on
these
is
to
extend
management
operation
to
achieve
some
intelligent
network
management
process,
and
this
in
this
sense,
we
have
four
to
make.
D
D
This
information
is
fed
to
current,
for
example,
machine
learning
tools
and
we
obtain
the
model.
After
that,
we
also
execute
the
rezoning
process.
Recently,
processes
will
infer
new
knowledge
and
rules
etc
from
this
information
and
models
both
together.
So
we
have
new
rules
that
will
be
also
fed
back
to
the
mobile,
so
it
will
be
constantly
building
a
better
view
of
the
network
management
from
the
management
point
of
view.
In
order
to
go
to
this
nexus
10,
the
system
will
be
to
solving
potential
problems.
D
This
view
will
find
problems
and
some
specific
operation
process
from
what
happened
Scott
would
be
executed
in
order
to
resolve
those
problems,
so
there
will
be
the
solving
part.
And
finally,
once
the
solution
is
phone,
we
have
to
plan
that
plan
how
to
implement
that
solution
into
the
network,
how
to
require,
in
the
network
to
adjust
the
amount
of
resources
that
are
involved,
or
things
like
that.
D
So
this
four
processes
will
be
executed,
and
one
key
point
is
that
they
executed
in
parallel,
so
they
will
be
inter
dependencies,
but
also
some
way
of
independence
from
them
among
them.
So
it
is
very
tricky
way.
So
there
are
challenges
there,
interface
definitions,
that
we
must
consider
in
order
to
to
make
all
these
processes
work
together.
Okay,
we
also
have
been
have
been
working
on
the
what
we
call
the
closed
control
loop
management
approach.
D
No,
so
what
we
do
is
that
the
close
this
control
loop
so
one
way
or
another,
this
processes
I
mentioned
before
they
will
be
required
to
connect
their
outputs
to
the
inputs,
but
in
different
kinds
of
a
different
way,
so
some
planning
would
connect
to
solving
etc.
So,
at
the
end,
you
Tavish
that
data
plane
elements
would
will
be
involved
also
in
this
reasoning,
because
they
must
provide
Sun,
even
small
information,
to
check
that
the
that
the
enforce
decision
is
taking
some
is
being
effective.
So
we
need
that
okay
Inessa
slide,
please.
So
what
we.
E
D
Data
and
we
want
to
achieve
wisdom,
so
all
our
solutions
become
like
being
able
to
take
this
kind
of
taking
decisions
by
getting
data
processing
the
data
getting
information,
what
we
is
called
or
the
data
formatted
in
some
kind
of
information,
then
from
the
information.
We
reason
we
obtain
knowledge
and
we,
but
we
are
familiar
the
knowledge
base,
but
with
that
knowledge,
well
after
the
application,
it's
at
the
exploitation
of
the
knowledge,
what
we
get
is
that
is
a
system
is
a
wise
system.
D
D
Also,
we
hold
the
knowledge,
storing
that
in
the
knowledge
base
in
order
to
achieve
a
system
that
is
more
wise.
So
when
I
see
this
wisdom,
okay,
so
that
that
would
be
the
key
of
this
exploitation
part.
So
next
slide,
please
that
will
be
the
last
slide.
I,
don't
know
how,
if
I
am
on
time
of
not
that
I
will
just
finish
so
for
now.
What
we
are
identifying
in
this
draft
is
that
the
challenges
we
are
now
facing
out
the
reason
or
network
behavior
from
perform
measurements
and
Stephanie
means
that's
our
main
challenge.
D
We
have
a
mainly
identify
right
now
in
the
sense
of
never
slicing
from
network
slicing
part,
so
the
gaps.
What
we
find
also
identify
there
would
be.
We
need
some
metal
for
frog
for
a
low
in
different,
provides
some
method
and
then
dos
be
able
to
coax
it
and
work
together.
That
means
that
they
can
exchange
information.
They
can
understand
to
each
other
in
the
terms
of
these
intelligence
data
and,
as
I
mentioned
before,
the
wisdom
of
that
they
obtained
from
all
these
processes.
So
one
key
point
also
what
we
have
identified
in
the
draft.
D
B
D
Am
more
focusing
on
assess
that
they
are
working
as
they
are
expected,
because
one
key
difference
is
that
today,
when
we
buy
or
when
we
integrate
some
component
physical
or
logical
component,
they
normally
we
we
expect.
No,
they
work.
Well,
no,
we
are
not.
Let's
say
we,
we
can
be
deterministic
no
in
the
sense
that
we
know
that
it
will
work
in
some
way.
But
when
we
are
introducing
some
AI
module,
we
cannot
be
deterministic.
We
cannot
know.
D
We
cannot
be
sure
that
the
module
will
work,
so
we
need
some
way
to
assess
the
quality
of
the
model.
That
means
that
the
model
will
work
most
of
the
time.
That's
it
things
like
MTV
eventhough,
something
like
that,
but
something
more
focused
on
the
intelligence
is
house
marked.
How
much
is
the
intelligent
of
this
module?
D
And,
finally,
what
we
have
identified
that
the
protocol
just
published
information
must
respond
to
some
concentrate
on
the
target
uses.
That
means
that,
for
example,
when
we
are
sharing
some
data,
some
information
of
knowledge
items
among
different
providers,
different
modules
or
different
modules
of
that
will
analyze
that
information.
A
F
I'm,
sorry,
when
you
go
into
WebEx
and
you
use
whatever
accounting
I'm,
not
here
as
the
ecology
chair
I'm,
he
was
a
person.
Okay,
I
have
two
questions
comments
when
I
think
it's
interesting
that
you
didn't
mention
datasets
hear
you
talk
about
ontology,
but
it's
more
than
that.
It's
the
data
set.
What
are
the
data?
Is
that
you're
going
to
use
for
what
you
call
network
management
and
they
should
be
problem
specific,
there's
a
ton
of
things
that
happen
in
a
network
that
are
completely
disconnected.
F
So
what
do
you
mean
by
having
this
very
generic
reasoning,
the
other
one?
You
seem
to
be
kind
of
confused
about
the
role
of
machine
learning.
Machine
learning
is
just
not
just
accumulate
data
for
in
comma
learning.
It's
also
to
infer
on
that
learning
what
you
call
the
AI
part.
So
again
you
have
a
good
data
set
and
you
actually
have
your
machine
learning.
Then
your
system
becomes
by
itself
autonomous
and
and
so
and
when
you
say
that
you
you,
you
need
to
know
if
the
information
is
right
or
not.
F
Well,
that's
actually
what
machine
learning
is
designed
to
do
throughout
the
machine
learning
process,
either
by
automatic
decision
making
or
by
annotating
by
experts?
So
you
know,
while
you
train
your
algorithm,
you
can
have
inputs
from
experts
who
say
well.
This
is
the
behavior
that
we
we
expect.
So
you
have
your
training
set,
so
I
think
you're.
What
I
see
is
that
yet
there's
a
lot
of
good
stuff
in
there,
but
there's
a
lot
of
things
that
seem
to
be
incomplete
or
mixed
up.
F
D
Yes,
that's
and
so
for
now
we
are
not
focusing
in
just
one
problem,
but
again,
I
insist
to
gather
this
kind
of
challenges.
Okay,
I
understand
your
your
points.
Well,
then,
I
am
more
in
this
sense.
Maybe
it's
not
so
clear.
The
difference
that
we
are
trying
to
clarify
here,
I
said
in
the
sense
of
this
particular
difference
between
the
laning
phase.
Not
just
I
am
NOT
saying
that
machine
learning
only
gathers
information.
Machine
learning
is
what
it
has.
It
builds
a
model
from
the
information
that
we
we
have.
D
It
will
say
model
of
knowledge,
it's
okay,
but
the
inference
that
machine
learning
has
is
just
for
the
output
of
this
information.
What
we
want
is
to
go
beyond
that
inference,
to
do
reasoning
on
that
inference.
Let's
say
not
just
learning
and
and
stop
there,
but
build
the
model.
Okay,
but
dynamically
change
the
model.
In
order
to
do
that,
we
must
establish
new
rules
and
new
ways
to
do
that.
So
the
idea
is
that
after
machine
learning
process
has
finished
okay,
we
can
take
decisions
with
the
output,
but
what
the
next
step
would
be.
D
Those
decisions,
also
output
from
massive
level,
would
be
reasoning
beyond
that
I
put
in
order
to
obtain
meta
information,
let's
say
information
about
the
machine
learning
tool
and,
let's
say
change
that
tool
with
new
rules
and
new
conditions,
etc,
that
the
reasoning
operation
is
doing
in
that
sense.
After
that,
we
again
I
insisted,
we
have
the
planning
and
set
robot
will
be
video,
but
the
main
point
where
now
the
deaf,
the
next
step
right
now
would
be
to
exploit
the
risk.
D
This
reasoning
part-
and
we
want
to
clarify
that
so
if
you
want,
for
example,
collaborating
this
draft
in
order
to
clarify
the
point-
the
involvement
of
machine
learning
in
all
this
process,
we
are
open
to
new
collaborative
work
with
our
target
right
now.
We
are
not
focusing
in
in
specific
problems,
but
in
order
we
have
more
wide
spectrum
to
define
the
challenges
and
gaps
and
find
collaborators
that
can
provide
this
kind
of
details
that
we
are
missing
in
the
draft.
Ok,
ok,.
A
So
thanks
Mary
Joseph
for
raising
your
points
and
thank
Pedro
trying
to
address
them.
We
are
a
bit
tight
on
schedule,
so
my
recommendation
and
because
I
think
this
is
good
discussion
and
we
is
to
follow
up
with
your
visual
draft,
maybe
with
a
dressings
on
the
mailing
list.
The
comments
raised
by
Rosie
trying
to
develop
further.
A
What
could
be
your
your
common
understanding
on
that?
As
you
will
see
later,
Jerome
will
present.
Also
we
have.
We
are
producing
a
new
documents
on
research
challenges
for
AI
in
network
management.
I
believe
that
some
of
your
inputs
could
really
go
to
these
documents
also
to
contribute
to
these
other
documents
and,
as
we
have
seen,
it
will
be
nice
also,
if
you
can
try
to
to
develop
a
bit
further.
The
next
step
for
your
documents
and
and
going
from
this
I
will
see
more
higher
level
views
but
more
concrete
proposals.
A
D
D
A
G
Good
afternoon
my
name
is
Sonia
and
I
work
in
the
transport,
an
IP
network
division
in
the
loyalty
tio
unit
in
telefónica
research
and
development.
So
a
bit
into
introversion,
myself
did
in
the
last
two
years,
I've
been
working
in
different
European
breeds
regarding
new
telemetry
mechanism
for
Network
Devices
in
15
that
works.
So
in
relation
with
these
now
we
are
exploring
about
a
metadata
base
obligation
of
telemetry
flows,
which
is
the
main
peak
of
this
presentation.
Next
slide.
Please.
G
So
for
this
presentation
we
are
going
through
a
short
introduction,
followed
by
the
description
of
the
standard
in
which
we
base
our
initial
work,
and
then
we
will
see
the
initial
report
of
our
work
and,
finally,
we
will
state
some
conclusions
and
a
request
for
this
research
group.
Next
next
one
please.
G
As
you
already
know,
over
the
years
telecommunication
network
have
evolved
in
order
to
others,
they
did
a
man
of
the
new
services,
so
the
increasing
number
of
nose
that
forms
the
the
networks
and
the
advanced
capabilities,
the
evolution
of
the
underlying
technologies
and
all
the
different
ways
to
access.
The
network
has
turned
a
telecommunication
networks
into
battery
genius
networks,
not
only
regarding
the
technologies,
but
also
traffic
requirements
from
the
provision
of
different
services.
G
Next,
please,
in
these
data,
oriented
networks
in
which
is
necessary
to
handle
the
rigidity
of
all
of
these
components
is
important
to
offer
data
in
a
consistent
way
to
all
of
the
different
applications
needed
for
the
closed
loop
automation
systems
such
as
some
application
about
artificial
intelligence
or
Big
Data
or
whatever.
So,
for
this
work,
we
propose
the
use
of
a
semantic
metadata
framework
for
aggregation
of
all
of
the
data
coming
from
the
network.
G
G
So
this
semantic
metadata
framework
is
based
on
on
the
idea
of
using
a
model
to
describe
the
different
network
elements
that
are
present
in
a
data
flow.
This
metadata
will
not
be
used
for
sending
the
data
itself,
but
it
models
the
data
the
device
is
capable
of
providing
so
using
the
semantics
metadata
framework.
The
different
data
sources
or
flows
will
be
described
by
means
of
this
metadata,
and
it
would
then
be
necessary
to
adapt
all
of
the
control
elements
one
by
one.
G
So,
apart
from
this,
the
application
of
this
semantic
model
will
imply
some
changes
in
the
current
definition
of
the
descriptors,
but
as
we
are
in
a
very
early
stage,
we
are
not
focusing
in
this
part
yet
so
for
the
starting
point
of
this
idea
of
defining
the
framework
is
based
on
and
the
T
standard
called
context,
information
management,
so
I'm
going
to
introduce
you
a
little
bit
of
this
concept.
This
is
standard,
defines
an
API
for
context.
G
Information
management
enable
access
to
information
coming
from
many
different
sources
so
for
implementing
this,
the
standard
provides
an
information
model
that
defines
the
structure
of
context,
information
that
must
be
supported
by
the
system.
Initially,
it
was
thought
for
IOT
data
sources,
but
the
scope
is
not
tight
for
these
data
sources,
so
this
standard
allows
users
to
provide,
consume
and
subscribe
to
context
information
in
different
scenarios.
G
So
besides,
the
sim
standard
does
not
find
a
specific
architecture
for
implement
in
this
system.
They
provide
some
recommendations
about
three
possible
architectures:
a
centralized,
a
distributed
and
a
federated
architecture
a
nevertheless.
The
three
architecture
have
the
same
key
components
that
are
the
ones
that
you
can
see
here,
which
are
the
contacts
consumer.
That
is
basically
the
application
that
needs
this
context:
information
a
of
the
data
sources
for
the
question
where
information
or
another
they're
ones,
that
fits
for
the
application.
G
Then
it
is
the
context
producer
or
context
sources
which
corresponds
with
all
the
data
sources
that
produce
some
data,
and
thus
they
need
the
definition
of
this
context
data.
Then
there
is
a
central
piece,
which
is
the
main.
The
main
one,
which
is
the
broker,
is
responsible
for
handling
these
requests
from
the
consumers,
application
and
asking
information
to
the
source
and
finally,
depending
on
the
on
the
architecture
that
you
choose.
G
There
is
another
component,
which
is
the
contacts
registry,
which
is
a
component
that
can
be
in
the
broker
or
can
be
in
a
separate
one
in
which
basically,
the
data
source
register
the
context
information
they
can
provide
and
is
available.
We
request
by
the
broker
for
the
contract,
corresponding
request
a
done
by
the
consumers
in
this
one.
G
So,
as
you
can
see
here,
this
is
the
proposal
that
is
in
the
standard
for
distributed
architecture.
As
you
can
see,
all
the
interactions
between
the
consumers
and
the
source
are
gone
through
a
broker
who
requests
the
context,
information
of
the
data
sources
to
the
contacts
registry
and
the
contacts
source
a
previously
a
were
registered
in
the
contact
registry
next
one.
G
So,
as
I
said
before,
this
standard
provides
an
information
model
that
defines
the
structure
of
the
contacts
information
that
should
be
supported
by
the
system
that
I
shown
in
a
previous
slide.
So
this
information
model
relies
on
the
idea
of
having
an
ontology
which
basically
provides
the
definition
of
a
set
of
concepts,
data
taxonomy
and
the
relationship
that
governs
this
concept.
So
this
figure
shown
a
UML
representation
of
the
information
model
defined
in
the
standard
which
is
composed
by
different
entities,
relationships
and
properties.
G
So
one
of
the
final
goals
we
have
for
implementing
this
work
is
trying
to
build
a
generic
patterns
for
metadata
definition.
However,
until
reaching
these
general
patterns,
we
start
first
collecting
multiple
context
serve
that
we
will
have
available
in
different
scenarios
and
an
initial
static
definition
of
sand
information
models
for
this
context
or
that
we
identify.
So
at
the
same
time,
we
are
defining
all
of
these
information
models.
G
We
would
like
to
start
building
a
semantic
aggregator
or
the
broker
at
desist
call
in
the
sim
stander,
by
making
use
of
tools
already
available
in
the
industry
for
making
processing
of
the
data
and
introduction
for
the
digital
models
of
this
architecture.
And,
finally,
all
of
this
work
is
trying
to
be
aligned
with
the
sim
standard
as
much
as
possible
as
we
take
it
as
a
reference.
Next,
one.
G
Apart
from
from
that,
we
start
to
think
how
can
we
integrate
this
semantic
framework
into
a
general
and
a
commune
architecture
using
the
current
projects
that
we
are
working
on?
So,
as
you
can
see
here,
we
had
the
semantic
elevator
plus
the
contest
registry
in
in
two
possible
domains:
the
network
service
provider
domain
and
the
local
one.
G
So
here
you
can
see
some
of
the
data
source
we
have
already
identified
in
the
place
that
we
are
working,
so
some
of
them
are
network
based
profs
such
as
an
ICMP
probe,
HTTP,
DNS,
IP
fix
and
so
on,
and
we
are
also
considering
a
devices
that
already
defines
the
information
through
a
general
model,
a
for
example
that
are
located
in
some
time
series
database.
Next,
one.
G
So,
starting
from
in
the
next
two
slides,
you
will
see
some
examples
of
the
information
model.
We
have
start
to
define
for
some
of
the
context
source
that
I
shown
before
so
in
this
case.
In
this
slide,
this
correspond
to
the
information
model
for
the
network
based
probe
such
as
an
ICMP
probe.
So,
as
you
can
see,
it
follows
the
sim
standard
in
which
the
information
model
is
composed
of
different
entities.
The
the
white
boxes
properties
some
relationships
between
these
entities.
G
So,
apart
from
the
specific
things
of
this
context,
source
the
ICMP
probe,
this
information
model
implements
two
important
characteristics:
the
first
one.
How
is
the
information
structure
of
this
ICMP
probe
by
the
definition
of
the
schema
property,
as
you
can
see,
with
the
German
dictionary,
and
the
second
property
will
be
how
the
semantics
of
relator
the
broker
can
reach
this
ICMP
probe
by
the
definition
of
the
entity
employ
characterized
by
the
entity
URI
next
one,
please
so
the
second.
G
The
second
example
will
be
for
John
based
device,
so
we
are
still
defining
this
information
model
and
it
could
be
sound
a
little
repetitive,
because
the
youngin
data
models
have
already
defined
the
structure
of
the
information
and
that
word
device
can
provide.
But
a
idea
of
the
semantic
aggregator
is
trying
to
unify
all
the
possible
data
source
needed
and
not
all
of
these
data
source
implements
yon
data
motives.
G
For
this
reason,
we
thought
that
it
would
be
good
to
take
advantage
of
the
already
defined
game
models
in
the
device
for
inferring
the
properties
for
this
entity
device
that
owns
multiple
entities
of
young
model
next
one
next
one,
please
is
the
last
last
slide.
So
as
summary,
we
have
been
seen
that
the
data
driven
network
management
requires
the
aggregation
of
the
original
sources
of
data
that
we
have
to
do
to
the
different
services
that
are
provided
and
all
of
the
devices
that
confirm
the
network's.
G
Also,
as
we
have
previously
comment,
we
aim
to
extend
existing
metadata
model
for
IOT
environment
to
the
network
management
environment.
And
finally,
we
would
like
to
kindly
ask
you
if
you
are
interested
in
this
work
and,
if
so,
trying
to
document
it
in
an
ID.
So
this
is
a
very
early
stage
of
the
work.
So
this
is
all
from
my
side.
We
are
happy
to
hear
from
you
anything
you
want
to
ask.
A
G
In
your
system
that
all
the
sources
that
you
have
available,
you
can
register
once
you
know
it,
and
then
this
model
can
sent
to
the
application
needed
by
your
system.
This
information,
for
example,
you
have
these
three
data
sources
and
information
that
they
can
provide
to.
You
is
these
information,
this
information
and
this
information.
So
by
doing
this,
the
application
can
select,
which
is
the
information
that
really
needs
for
working
properly.
A
Okay,
thanks
for
the
explanation,
I've
seen
also
that
in
that
other,
have
been
several
comments
and
exchanged
on
different
aspects.
We
will
try
to
capture
that
also
in
the
minutes
of
the
meeting,
and
also
from
from
your
last
point
about
the
interest
from
the
research
group
and
relating
the
newest
comments.
Yes,
please
documented,
given
for
I
will
prefer
that
a
model
instead
of
an
information
model
and
data
models
provide
API,
is
a
new.
A
A
data
model
versus
I
am
so
I
think
at
least
personally,
not
as
the
chair
and
I
think
from
what
I
saw
on
the
chat
lady.
There
are
some
interest
today
to
move
further
with
this
proposal,
and
it
will
be
nice
to
get
a
bit
more.
I
would
say
much
of
yours
to
read
from
you.
I
think
it
could
really
benefit
to
order
items
we
have
in
the
research
group,
either
the
the
Ramon
I
related
or
even
maybe,
or
what
the
pillow
just
mentioned,
which
needs
for
ontologies
and
semantic
aspects
of
data.
A
A
G
A
D
Fact
what
I
just
wanted
to
point
that
see
the
CIA
model
can
be
a
good
way
to
let's
say,
implement
the
ontology.
That
I
was
mentioning
before
I
was
a
few
years
ago,
I'm
working
with
CIA
man.
It
is
quite
complete,
complete
for
this
task,
so
we
can
collaborate
in
these
two
topics
together
to
put
them
together
in
the
same
place,
let's
say
the
relation
between
the
AI
application
and
the
CI
M
model
for
the
for
this
information
retrieval
today:
okay,
yes,
let's,
let's
talk
in
the
future,
please,
okay,.
A
A
A
H
A
H
H
So
in
this
presentation,
I
will
talk
about
fault,
our
search
function,
training
and
the
idea
is
that
how
can
keep
a
service
function
chain
running
after
f-number
of
its
service
functions?
Failure
and
I
will
talk
about
how
we
can
integrate
FCC
fault
armed
chaining,
with
an
asset
to
provide
the
fault
tolerance
for
industry.
H
Functions
or
middleboxes
our
network
function
that
received
the
traffic
processor
traffic
and
forward
the
traffic.
Here
are
two
examples.
For
example,
on
left
side
you
see
network
address
translators
and
they
act
as
a
bridge
between
privates,
Network
and
public
network,
for
example.
In
in
this
figure,
Alice
and
Bob
are
citing
the
privates
network.
H
They
send
a
traffic
and
they
use
private
IP
addresses,
and
then
that
translates
these
private
IP
addresses
to
public
IP
addresses
that
are
accessible
in
in
the
internet
and
change
the
header
of
the
packet
that
it
receives
from
Alice
and
Bob
to
public
IP
addresses
and
send
a
packet
to
the
internet.
Indeterminant
receives
the
packets
back.
H
H
The
to
combat
this
issue,
there
are
fall,
turns
approaches,
and
the
main
idea
behind
alternate
approaches
is
that
we
only
need
to
replicate
the
state
of
that
particular
service
function.
For
example,
in
this
example
to
make
and
to
make
not
fall
turns
we
only
we
need
to
replicate
its
connection
state.
It
uses
this
connection
state
to
remember
who
is
talking
to
whom,
for
example,
in
here
it's
the
not
remembers
that
Alice
is
talking
to
Apple.
Bob
is
talking
to
Bing
and
based
on
these
connection
estates.
H
It
can
translates
the
packets,
like
an
forth
between
private
IP
addresses
and
public
IP
addresses,
so
to
make
that
fault-tolerant.
We
have
a
replica
and
naturally
that
exactly
replicates
that
connection
state
so
that
if
the
original
NAT
fails
and
if
you
lose
the
original
last
connection
state,
we
still
have
a
replica
of
that,
and
the
only
thing
that
we
need
to
do
is
redirect
the
traffic
through
the
secondary.
H
Not,
but
actually
the
devil
is
in
details
and
there
are
main
work
that
describe
how
we
can
achieve
these
replication
and
they
mainly
focus
on
single
middleboxes
single
for
service
functions
and
they
are
a
snapshot
based
system.
But
it's
not
should
I
mean
that
they
frequently
take
the
snapshot
of
the
service
phone
service
function,
state
and
replicate
it
into
some
a
separate
replicas
so
that,
if
the
primary
fails,
we
have
the
replica
a
snapshot
of
this
that,
in
some
other
place
in
the
network,.
H
H
From
making
individual
middleboxes
individual
service
functions
for
turns
it
turns
out
if
we
use
existing
system
existing
snapshot
based
system
for
a
chain
of
service
function.
The
throughput
drop
is
massive
on
left.
So
you
see
that
if
we
have
a
change
of
5
meter
box,
there's
five
service
functions.
We
almost
lose
half
of
their
true
put
because
of
the
snapshots.
The
reason
that
we
lose
that
much
of
a
true
put
is
that.
H
Then
one
of
service
function
is
taking
a
snapshot,
it
has
to
bus
its
operation,
but
by
pausing
its
operation
then
touch
and
stall
and
with
longer
chains.
We
have
more
passed
time
and
knows
more
no
service
time
and
that's
why
we
have
a
lot
of
throughput.
Overhead
approach
is
different.
We
don't
do
a
snapshot
based
and
we
make
the
entire
chain
fall
turned
the
key
here.
The
key
difference
is
that
we
lose
the
chain
itself
to
make
a
service
function
chain
faltered.
But
how
do
we
do
that.
H
If
you
don't
mind,
I
first
talked
about
in
China
application
and
then
I
come
back
to
state
to
get
back
in
so
as
I
as
I
mentioned
before
in
existing
approach
to
to
provide
all
turns
for
a
service
function.
What
we
need
is
that
we
replicate
the
state
of
service
function
in
some
separate
replicas
across
the
network.
We
wait
for
their
acknowledgment
and
then
at
that
time
we
can
release
any
packet
that
has
modified
the
state,
but
in
our
system
we
use
the
chain
itself
to
make
the
chain
alternate.
H
We
observe
that
already
traffic
goes
through
the
chain
from
service
function.
One
search
function
to
2003,
for
example.
So
why
not?
We
use
the
next.
The
subsequent
service
functions
in
the
chain,
as
the
replica
for
the,
for
example.
First
service
function.
In
this
example,
for
example,
we
can
use
sales
function
as
a
replica
of
service
function.
One.
H
Let
me
combine
when
we
combine
into
a
replication
with
state
piggybacking,
but
it's
a
backing
I
mean
that
existing
approach,
each
we
observe
two
traffic
streams
coming
out
of
each
service
function.
One
is
that
they
are
data
traffic,
the
normal
traffic
that
we
process
and
the
other
traffic
is
stream
is
to
disseminate,
is
to
disseminate
the
state
updates
to
wrap
the
cost
for
a
state
ratification
in
our
approach.
H
So,
let's,
let's
see,
let's
see
in
more
detail
our
system
work.
So
let's
say
that
we
want
to
provide
fall,
turns
for
two
service
functions
as
f1,
and
s
f2
and
they're
running
in
separate
servers
to
tolerate
one
failure.
We
instantiate
three
replicas
r1,
r2
and
r3,
and
they
are
they're
located
in
separate
servers
in
order
to
isolate
their
failures
so
that
if
one
of
them
fail,
its
failure
is
independent
of
the
others.
It
doesn't
cause
the
failure
in
other
replicas.
H
Also
intercepts
packets
and
they
provide
state
management
API
for
service
function
so
that
they
can
trace
and
they
can
track
whatever
it
is.
That
has
been
changed
by
the
service
function
so
that
they
can
replicate
it
and
send
it
to
them
through
next
replica,
as
I
mentioned
before,
whatever
state
has
been
changed
by
processing
a
packet
or
system
piggyback
that
onto
the
packet.
H
Now
part
two
receives
that
packet
replicates
the
state
detaches,
the
piggyback,
let's
say:
send
it
to
the
service
function,
track.
Whatever
state
could
be
changed
by
in
processing
that
packet
and
when
service
function
to
sends
the
packet
out,
it
receives
the
packet
piggyback
state
changes
into
the
packet
and
sorry
to
the
third
replica,
an
archer
even
which
receives
the
packet.
It
replicates
this
thing
piggyback
the
state
street,
that
the
piggyback
is
States
from
the
packet
and
then
finally
release
the
original
packet
to
the
outside
wall.
H
While
doing
so,
basically,
we
achieve
high
performance-
and
here
is
one
of
the
results
that
we
achieved
by
running
our
system.
We
compared
our
system
with
non
fault,
Ireland,
a
service
with
non
faltering
service
function
chain,
which
is
basically
the
baseline,
the
highest
performance
that
we
can
achieve,
and
we
compare
it
with
their
state
of
art,
FTM
B.
H
You
see
two
labels
for
FDM
B,
Afghan
people,
and
if
they
only
plus
a
snapshot,
can
ignore
FTM
B
everything
B
plus
a
snapshot
is
actually
the
actual
system
and,
as
you
can
see,
for
a
chain
link
are
four
different
chain
lengths.
We
achieve
almost
twice
as
the
state
of
the
state
of
the
art
and
for
the
chain
length
of
five
we
achieved
3.5
times
higher
throughput.
H
Believe
that
FTC
could
provide
a
good
platform
for
fault,
tolerance
of
service
functions
and
service
function
chains
and
combined
with
NS
edge.
We
believe
that
we
can
provide
a
great
platforms
that
is
supported
by
because
nsh
are,
which
is
a
protocol
designed
by
IETF,
is
already
supported
by
industry,
and
we
think
that,
while
combining
an
aesthetic
with
FDC,
we
can
provide
a
great
platform
for
fault
and
service
functions.
H
A
H
So
we
have,
and
a
message
has
three
main
component
service
function
for
warrior:
service
function,
which
is
as
before
and
in
a
such
packets
service
section
for
order
can
implement
a
replica,
and
the
task
lady
does
is
basically
I
cannot,
okay,
that's
working,
so
SFF
can
replicate,
can
implement
or
a
replica
currently
NIH
supports
packet
forwarding
to
chain.
But
our
contribution
and
our
ongoing
work
is
extending
an
SH
for
extending
a
service
function
for
order
to
provide
also
state
management,
API,
understates
replication
and
we
add
the
support
of
a
message
into
click
modular.
H
Rather
we
also
working
on
supporting
and
as
such,
using
any
such
header
to
a
piggyback
estate
and
I
searched
header.
So
here
we
see
the
a
message
header
and
there
are
lots
of
fields
here,
but
the
only
thing
that
I
want
to
focus
here
is
variable
length,
context
header,
which
is
the
variable
which
allows
us
to
carry
which
allows
us
to
carry
metadata
inside
service
function
chain,
and
we
can
use
that
to
piggyback
estate.
H
I
passed
this
one
interest
of
time
currently
in
a
search,
provides
packet,
encapsulation
and
variable
link
from
the
data.
But
our
contribution
is,
we
define
how
we
can
use
an
assessment
data
to
us
to
piggyback
a
state
and
because
they're
these
information
are
sensitive
information
about
the
network.
We
also
need
to
devise
our
secure
approaches
that
maintains
the
confidentiality
of
the
data.
Thank
you
very
much,
and
that
was
money.
Our
presentation.
C
Yes,
Thank
You
millet
for
your
presentation.
I
know
we
have
already
had
some
talk,
but
can
you
give
me
because
I
think
it's
very
interesting,
the
idea
of
piggybacking
of
the
states,
but
what?
What
is
a
because
you
want
to
put
in
some
a
Indiana
sage
what
we
can
expect
in
terms
of
size
that
we
need
here
to
piggyback
States
I
mean
for
certain
service?
If
you
have
some
example
in
mind,.
H
So
in
our
system
we
work.
We
implemented
that
for
not
usually
one
packet
and
may
cause
adding
some
information
to
connection
state
which
is
around
40
bytes,
so
load
balance
are
also
stateful
load
balancer.
They
are
around
the
same
size
around
like
you
know,
40
to
60
bytes,
so
lots
of
lots
of
middle
boxes
lots
of
service
function.
We
expect
that
one
packet
shouldn't
go
beyond
a
few
dozen
bytes
and
we
think
that
an
asset
should
be
enough
and
such
a
that
should
be
enough
to
carry
this
native
one.
H
So,
for
example,
let's
say
that
you
want
to
stop
port
5
failures,
so
we
need
to
pack.
It
needs
to
carry
the
state
of
6
middle
boxes.
6
service
functions
for
those
cases
could
be
challenging
because
it
could
go
pro
to
myself.
There
is
a
problem.
There
is
a
limitation
there
that
we
cannot
go
beyond
some
some
number
of
failures
or
beyond
some
size
of
service
option
chain.
Actually.
A
A
A
I
I
I
So
the
background
for
this
work
is
basically
we
started
to
explore
these
ideas
of
hidden
based
mechanisms,
but
applied
to
the
nebulous
Lyceum
concept.
It
was
already
presented
in
the
presentation
that
I
refer
there
in
the
slides
in
the
Singapore
meeting.
At
that
time,
the
world
was
basically
focused
on
the
Asian
and
is
the
instantiation
and
placement
of
different
billionaires
some
des
facilities,
and
so,
even
though
this
PNM's
are
finally
connected,
there
was
not
an
emphasis
on
the
transpolar
slice
in
itself
that
there
is
some
initiative
in
in
the
teens
were
working
group.
I
I
I
Regarding
this
initiative
in
entities
working
group,
there
are
actually
currently
two
outcomes.
One
transporters
lies
definition,
draft
and
another
described
in
the
framework
for
these
times.
Realizing.
If
we
look
at
the
first
page
and
a
transpose
lies
definition
is
measuring
their
respected
by
the
triangle,
potential
transpose
less
controller
would
be
a
way
in
the
novel.
In
the
face,
I
mean
a
way
of
expressing
the
need
of
connectivity
in
a
technology
agnostic
way.
I
Okay,
looking
at
this,
we
thought
that
the
team-based
approach
could
be
for
the
provision
of
this
transport
level
slices.
So,
as
far
as
we
could
provide
appropriate
level
of
abstraction,
we
could
manage
manage
those
requests
with
these
indicts
based
mechanisms.
This
is
basically,
we
are
sprouting
in
this
draft
next
slide.
Please.
I
So
the
for
first
polling
that
we
look
at
the
case
of
the
fib
services,
which
are
basically
the
ones
that
will
firstly
require
require
from
the
network
potatoes,
the
capability
of
instance
18
and
were
slices
in
the
3gpp
or
the
three
VP
defines
a
way
of
instantiating
and
requesting
these
slices
into
n
in
the
figure.
Basically,
it
is
expected
to
have
a
VP,
be
management
system
that
could
manage
control
seated.
I
I
The
network
management
system
will
be
the
one
receiving
they
come
the
request
or
deploy
in
the
transport
slice
in
the
underlying
support
interest
ability,
the
optics
IPO
we
be
here
basically
trying
to
address
the
interface
between
this
EP
management
system
and
the
transport
network
management
system
or
an
interface
I
mean
how
to
apply
the
internet-based
mechanisms
on
this
relationship
between
both
management
system
nexus.
Right,
please.
I
Even
just
may,
basically,
they
are
defining
a
generic
Islamist
and
plate
in
order
to
allow
users
of
the
network,
in
this
case
vertical
customers
who
specify
different
kind
of
attributes.
In
order
to
deploy
this
slice
into
an
and
for
those
attributes,
you
can
consider
things
like,
for
instance,
the
speed
of
the
healthy
terminals
or
the
location
or
whatever
pad.
Also,
there
are
also
other
attributes
that
are
related
to
the
transport
part
deserts
like
the
bandwidth,
lady
latency,
and
so
on.
So
far,
basically,
what
we
were
looking
at
is
that
probably
this
generic
is
lice.
I
I
Then,
looking
at
the
internal
life
cycle
that
is
been
defined
in
this
concept
definition
draft
on
energy
for
the
intent
based
systems.
Basically,
we
see
that
it
could
be
applicability
of
this
unique,
a
slice
template
concept,
please
in
the
slide,
because
this
is
a
limited
just
to
okay,
we
first,
we
do
foresee
that
at
this
point
there
would
be
some
processing
of
the
Jenica
sliced
and
plated.
Listen
in
the
transport
concerns
in
such
a
way
that
when
the
customer
requests.
I
Configuration
of
the
network,
so
we
will
help
somehow
to
processes
here,
so
they,
the
processing
and
understanding
of
the
day
and
slice
template
attributes
come
from
the
general,
be
concise,
template
and
another
process.
It's
very
fun
instructing
the
dead-ball
in
this
way,
the
transporter
slice
controller.
I
I
I
I
I
React
ok
as
nesic
steps,
so
we
will
keep
working
on
this
basically
with
these
ideas
of
the
natural
language
processing,
and
so
we
will
continue
working
on
analyzing
the
attributes
that
are
coming
from
the
generic
sliced
time,
listen
coming
from
the
industry
and
also
on
other
these
cases.
That
could
be
the
interconnection
of
any
be
environments
or
network
sharing
situations
or
evolution
of
wholesale
services,
and
so
we
want
to
keep
this
work
aligned
with
the
progress
in
these
neighbors,
like
in
design
team
and
basically
to
complement
the
work
being
done
there.
I
Here.
We
would
like
to
basically
request
comments
from
on
the
audience
from
the
group
from
the
village
group
and
input
for
new
versions.
We
are
in
the
process
of
preparing.
A
new
version
will
be
a
new
version
for
Madrid
meeting,
but
probably
before,
even
before
whether
they
were
based
on
other
questions
and
just
to
reflect
the
fact
that
gift
Angela
as
your
innocence
quarrel,
we
are
open
from
contributions
and
we
expect
that
this
could
have
interest
for
the
for
the
research
group
and
that's
all
from
my
side.
Thank
you.
J
I
just
have
one
question:
if
I
may
I'm
just
wondering
in
terms
of
your
attributes
at
the
intent
for
the
transport,
a
slice
layer,
do
you
envisage
that
the
attributes
would
be
delay,
jitter,
etc?
Or
would
the
attribute
be
more
high-level
decorative
in
terms
of
who
needs
this
kind
of
slice,
for
example?
Is
it
EMB
B?
Is
it
slice
for
EMB
B?
Is
it
slice
for
some
of
the
specific
types
for
3gpp
service
types?
Thank
you.
I
There
could
be
probably
a
combination
of
both
of
them,
so
there
would
be
something
that
could
be
more
or
less
straightforward,
translated
to
transport
requirements
like
throughput,
which
basically
can
be
translated
to
bandwidth
or
latency,
and
so
and
there
will
be
other
attributes
that
are
not
directly
are
integrally
impacting
the
transport
network.
For
instance,
mobility
reliability
is
maybe
he
could
make
the
the
transport
network
to
create
some
protecting
paths
in
order
to
accomplish
given
reliability.
I
Even
if
the
nice
template
declares
the
need
of
deployment
service
in
a
given
location,
so
indirectly,
you
can
infer
the
capabilities
in
that
area
that
the
geographical
areas
hope,
so
there
could
be
probably
a
mix
of
of
them
or
both
of
them
in
one
of
the
words
that
I
prefer
there
they
is
MPI
is
trying
to
digest
how
these
attributes
could
be
impact
directly
or
indirectly
in
the
transport
part.
So
there
is
a
parallel
ongoing
work
on
on
this
direction,
so
I
invite
you
to
check
and
document
it.
J
I
I
Generic
and
I'm
required
to
be
in
the
example
I
mentioned
before,
about
the
reliability.
So
if
99.99
percent
of
a
service
availability,
then
probably
I
need
to
instantiate
different
parts
in
the
transport
part
in
order
to
guarantee
such
level
of
availability
or
a
mix
of
them
kora
starting
point:
let
me
remark:
this
is
the
general
concise
template
from
that
going.
A
higher
level
like
mbb
or
you
ll,
see,
and
so
on
is
not
never.
J
B
The
the
intent
in
the
work
that's
going
on
right
now
is
that
the
transport
network
doesn't
understand
the
actual
application
that
are
applying
applying
for
us
transport
slice.
So
the
intent
would
not
include
any
information
about
exactly
what
this
slice
is
used
for.
It's
just
information
about
how
what
what
the
characterizations
are
attributes
as
we'd
says
that
are
to
be
communicated
to
the
transport
system.
A
So
I
don't
know
if
Eric
was
ok
for
you,
but
since
we're
also
running
a
bit
late
with
respect
to
the
initial
agenda,
I
would
like
to
close
the
discussion
now.
It's
you
another
draft
that
we
have
for
the
group.
There
are
interesting
points
to
be
further
develop
and
some
interaction,
as
we
have
seen
with
other
members
of
the
group,
so
I
hope
that
this
will
continue
on
the
meaning
list
or
M
before
the
next
virtual
or
in-person
meetings,
and
so
will
having
more
time
to
investigate
some
of
those
points.
I
A
E
Thank
you,
I'm,
not
saying
yes,
here
it
is
okay,
hello,
everyone,
my
name
is
Nabi,
then,
like
presentation
today
is
on
the
project,
automated
design
of
services
funded
for
service
requirements.
This
project
is
a
recent
project,
Concordia
University
in
Montreal,
and
it
is
in
the
collaboration
with
Ericsson
and
it's
supervised
by
dr.
fair,
attendant
dr.
Mario
tutorial
from
Ericsson
next
slide.
Please.
E
So
when,
outside
a
quick
background
during
the
motivation
of
the
goal
of
the
project,
an
overall
picture
for
method
and
after
that
talking
about
the
modeling
framework
that
we
have
and
then
discussing
the
the
method,
steps
and
prototype
children,
we
are
designed
developed
and
finally
fiction
next
slide.
Please.
E
Network
function
is
the
functional
block
in
the
network,
define
functionality,
external
interfaces
so,
as
I
hope
you
all
are
familiar
in
a
few
system
or
network
function,
virtualization
there's
more
functions
are
there
for
services
came
to
be
anything
system
and
one
of
the
main
management
management
locks
consisting
is
anything
management.
Orchestration
of
manner
requires
a
deployment
template
for
an
effort
service
in
order
to
deploy
in
emphasis
to
the
elephant
system,
and
that
template
was
called
little
service
descriptor,
which
describes
that
of
observers
from
all
virtualization
prospect
mix
like
this.
E
So
NEC
generation
corners
that
design
the
desired
network
sales
begin
and
the
design
process
is
the
process
of
selecting
proper
natural
functions
and
arranging
them
together
to
achieve
the
compass
of
functionality.
Behavior.
This
is
a
lovely
school
board.
Anything
and
it
is
done
manually
by
experts
today,
the
issue
that
is
anestis
on
processes,
knowledge
of
many
different
tenants
might
want
different
types
of
network
services
for
different
domains
and
that's
really
knowledge
excessively
design
and
many
other
NSE
generation
is
also
there
because
of
mainly
details
and
in
general
doesn't
fit
until
it
comes
into
touch
vision.
E
So
the
goal
of
our
project
is
to
devise
a
method
for
automatic
network
service
design
that
fulfills
the
tenants
intents
that
are
high
level
or
low
level.
Those
are
the
things
for
basically,
we
therefore
service
requirements.
We
call
them
industry
generated,
so
the
main
goal
is
to
generate
and
then
actually
design
the
Ennis
is
to
generate
the
innocence.
According
to
the
ED,
sees
any
from
standards
to
be
used
by
anything
at
all.
For
illustration,
illustration
and
part
because
of
the
direction
of
its
leverage,
their
models
and
excellence.
E
So
the
overall
picture
of
our
metal
starts
from
the
tenant
go
by
those
details,
and
the
output,
though,
is
the
NSD
at
standard
innocent,
but
there
is
a
huge
gap
in
its
because
the
intense
of
high-level
innocence,
so
we
feel
this
yet
with
our
method.
So
we
take
the
intents
or
the
inner
straight
from
the
tenants
and
the
method
technologies
which
he
called
little
function.
Ontology
it
has.
E
The
knowledge
of
function
of
the
composition
is
an
architectural
information
for
lift
or
sales
design,
samples,
our
industry
information
and
then
based
on
that
new
Civic
from
kettle,
and
then
we
generate
for
autographs
according
to
those
knowledge
and
then
finally,
the
jury
is
loose.
That's
going
to
use
by
anything
that'll
mix
like
things
so
as
I
said,
our
approaches
model-driven.
E
So
we
have
a
Model
T
framework
and
the
basis
of
the
start
on
this
picture.
Every
little
function
from
two
different
aspects:
one
is
functional
and
the
other
is
like
a
picture.
I
found
functional
perspective.
A
network
function
is
a
functionality
always
functional
and
from
an
architectural
perspective,
from
a
more
concrete
point
of
view,
that
functionality
can
be
realized
by
more
concrete
and
more
concrete
elements
like
an
architectural
block
and,
for
instance,
an
architecture.
What
defines
the
interfaces
and
the
protocols
used
to
realize
that
function,
and
our
assumption
is
based
on
our
definition.
E
So
keeping
that
exist
in
mind
I
start
from
introducing
in
astral
model
I
start
from
an
example.
Let's
consider
the
tenant
requires
the
vaults
and
syllabus
with
voice
for
our
registration
functionalities
and
6
megabits
per
second
throughput
for
small
and
50
registrations.
Persons
and
the
constraint
is
using
finest
occupation,
so
the
baby
model
is
shown
in
this
picture.
E
You
can
see
the
vaulting
decomposed
to
most
part
the
registration,
and
you
can
see
the
non-functional
requirements
attached
to
this.
It's
little
bit
personal,
the
treatment
end
of
the
crystals
and
the
architectural
constant.
So
that's
the
baby
model
and
if
you
go
to
the
looks
like
I,
can
show
the
overall
picture
of
the
little
model.
So
here
is
an
overview
of
the
industry
as
a
model
in
history.
Metal
model
consists
of
functional
requirements.
Architectures
and
non-functional,
as
you
can
see,
functional
requirements
can
be
composed
a
little
functional
units
and
other
blocks
attached
to
each
function.
B
E
E
So
next
up
is
that
the
ontology,
the
knowledge
base
that
you
capture
knowledge
into
the
ontology
structure,
is
quite
similar
to
the
model,
but
it
consists
of
more
details
in
the
world
information.
So
the
first
model
consists
of
functionalities
and
decomposition
to
architectural
blocks,
and
there
are
different,
decompositions
and
easy
compositions.
They
became
captures
from
standards
and
different
research
papers
or
any
knowledge
of
an
expert
and
in
which
this
ontology
and
here
at
the
bottom,
you
can
see
a
picture
of
area
an
instance
of
this
ontology.
E
So
you
can
see
a
volte
with
its
complete
decomposition
to
a
lower-level
functionalities
like
messaging
worst
reputation,
registration,
usually
for
short,
and
on
the
left
side,
you
can
see
what
the
picture
or
the
ILS
system-
and
you
can
see
the
relation
and
how
those
functionalities
can
be
realized.
Wait
is
iris
blocks
and
that's
how
to
capture
the
knowledge
of
functional
decomposition
in
general
facilities.
E
Here
quickly,
I
discuss
the
vnf,
neutral
method,
functions,
information
models
as
well,
so
the
energies
are
standardized
by
Etsy
and
they
provide
they
provide
on
the
metal
model
for
those
information
elements,
including
the
universe
which
in
some
other
party,
but
those
are
all
virtualization
for
right
information
from
the
aspect
of
virtualization
perspective.
So
what
for
in
its
design
in
the
application
perspective
information?
So
you
can
propose
an
extension
to
this
unit
information
and
nearly
40
picture.
E
This
group
of
architecture,
descriptor
provenza,
see
the
information
about
the
functionality
of
components,
functionalities
and
vnf
application
into
a
person's,
and
we
assume
that
this
unit
is
all
this
unit
architecture.
This
is
also
provided
from
the
vendor.
So
the
capture,
all
these
information
and
in
the
catalog.
E
So
I
talked
about
those
three
main
information
areas
that
we
use
as
our
inputs,
and
here
that's
how
we
process
them.
So
the
first
step
we
generate
a
solution
model
initialize
assumption
on
mobile.
It's
an
intermediary
model
that
captures
all
the
information
throughout
us
new
generation
costs
in
step.
Two
we
have
to
begin
with.
E
We
have
the
information
from
the
industry
into
the
solution
so
instead
to
the
map
to
the
ontology
and
eat,
reverse
ontology
to
capture
the
subtrees
under
function
also
that
architectural
side
that
matches
those
into
those
requirements
and
he
captured
them
into
our
solution
map.
In
this
third
step,
we
take
those
captured
architectural
blocks
and
we
go
look
to
our
nyan
F
catalog
and
we
select
the
proper
nearest
matching
those
architectural
works
and
we
bring
them
and
put
in
your
solution
organization.
E
Then,
after
that
now
we
have
all
the
information
that
means
from
the
ontology,
but
you
need
to
separate
alternative
alternative
solutions,
because
you
can
see
that
we
have
multiple
levels
and
layers
of
the
compositions,
so
in
three
sub
steps,
this
different
layers
alternative
service.
First,
we
look
at
the
function
of
the
composition
and
we
generate
a
functional
form
are
the
worst.
Those
are
sequences
of
functionalities
that
each
of
them
all
of
its
functionalities
together,
can
compose
a
dual
functionality
that
is
desired
for
delivering
service.
E
So
after
the
opposite
from
each
of
the
regenerate
architectural
elements,
which
is
a
sequence
of
architecture.
What's
that
no
one's
of
those
functional
and
we
the
same
token,
you
generate
premia
50,
which
is
basically
Vienna
for
London
graphs,
but
without
with
40
elements
like
muchas
leads,
then
the
detail
flows
so
premium
atrophy
is
basically
a
combination
of
the
combination
of
units
that
can
all
together
compose
the
designer
for
sales.
E
As
I
said
each
per
unit
that
she
is
a
potential
solution
so
for
each
pre-university,
regenerate
regenerate
that
for
service
descriptive
because
it
has
founded
in
of
stats,
can
compose
the
little
sadist
and
I
call
it
January,
because
they
are
just
introspective
functional
of
the
working
picture,
just
composition
of
what
the
picture
boards,
not
the
networking
part
and
not
the
visual.
So
in
this
part
in
this
that
the
design
of
the
children's
he
designed
the
series
expose
descriptors
and
we
generate
Vienna
for
garden
graph.
E
So
this
is
the
final
step
of
our
leopard
snow.
In
this
step,
you
take
the
non-functional
requirements
into
account,
if
you
remember
in
my
example,
ratio
that
cheated
or
m/s,
it
was
kind
of
requirement.
So,
but
first
we
designed
the
flows
because
of
the
fields
according
to
the
tenants
requirements,
whether
he
or
she
has
required
Christopher,
specifically
on
some
functionality.
So,
according
to
the
design,
different
floors,
packing
flows
and
he
used
dependencies
between
the
mutual
meeting,
the
others
and
using
the
internal
phase
of
units
which
does
the
foundation
has
been
captured
so
for
design
flows.
E
And
then
you
take
using
non-functional
requirements
that
he
propagated
throughout
his
floors
to
anticipate
them
are
supposed
to
be
exposed
on
each
field.
Therefore,
we
are
able
to
dimension
html5,
that's
meeting
that
load
so
by
considering
that
and
deployment
fingers
on
each
the
dimensional
appearance
and.
E
E
So
so
far
we
have
proposed
and
it's
to
automatically
design
that
four
sentences
based
on
the
food
service
requirements
for
the
infants
for
benefit
services.
That
tenants
hope
by
decomposing
those
intense
using
a
lot
of
toys
and
then
select
a
suitable
units
from
the
catalog
really
and
then
designer
for
writing
drafts
and
based
on
that
design
generated.
E
A
E
The
inner
part
in
temperature,
the
scripture
is
different
from
yes,
so
well
in
Vienna
is
a
neutral
Network
function.
Network
service
is
a
composition
of
via
notes,
so
first
this
VN
of
architecture
district.
There
is
a
descriptor
for
that
year,
but
the
NSD
is
a
descriptor
for
the
network
sales,
which
is
a
higher
level
or
composites
element
other
than
that
also
lien
of
architecture.
E
Descriptor
is
our
extension
or
it
consists
of
information
regarding
application
aspect
of
that
mean
you
know
course,
two
descriptors,
each
Etsy
provides
those
are
just
neutralization
aspect
elements,
meaning
they
just
provide
information.
We
don't
know.
What's
the
structure
of
this,
let's
say:
vnf
has
visualized
element,
for
instance,
whether
its
components
is.
How
do
you
figure
to
get
in
how
they
are
connected
together
to
visual
is
over
there
external
connection
pools,
but
they
don't
talk
about
their
functionalities.
They
don't
talk
about
their,
for
instance,
8
yards
each
regarding
the
functionalities,
so
this
be
an
important
picture.
C
Easier
you
in
your
approach,
when
you
work,
is
there
a
way
to
have
requirements
that
have
some
dependency
in
a
way
that
I
mean
not
all
requirement
must
need
to
be
fulfilled.
At
the
same
time,
it
can
be
combination
of
requirement
and
some
exception,
or,
let's
say,
yeah,
alternative
requirements.
Is
there
a
way
to
do
25
such
kind
of
things
or
not?.
E
E
E
So
the
requirements
that
we
get
we
matched
to
the
ontology
and
from
the
normals
and
travels
every
time,
each
one
of
those
who
access
requirement
in
certain
extant
ontology,
because
we
rely
on
so
any
requirement.
That
is
in
contradiction
to
the
ontology.
For
instance.
Let's
say
there
is
a
composition
or
there
is
a
dependency
that.
E
A
E
K
K
A
K
Okay,
the
goal
of
this
draft
is
to
paren
clarity
to
what
an
intent
represents
for
different
state
colors
by
means
of
classification
of
various
their
missions,
such
as
solutions,
users
and
the
intent.
Hence,
this
classification
would
ensure
a
common
understanding
across
all
participates
and
it
can
be
used
to
identify
the
scope
and
the
priorities
of
individual
projects,
fuses
research
or
open
source
of
projects.
K
This
is
achieved
proposing
initial
classification
tables
and,
as
a
mythology
used
for
generating
zone,
this
mythology
can
be
used
to
update
the
table
space,
adding
or
removing
different
solutions,
users
or
incentives
in
order
to
cater
for
futures
and
Orioles
applications
or
domains.
This
drafted
together
with
drop
to
era
TF
and
MRG
ibn
concepts,
definitions
and
to
aims
to
become
the
dolphin
Dacian
for
future
intent,
related
topic
discussions
where
all
participants
have
the
same
common
understanding.
K
Since
the
update
of
the
crafter
variant
2
in
November
2019,
we
have
received
the
support
of
experts
and
suggestions
for
improvements
to
the
draft.
The
following
is
a
summary
of
all
the
comments
we
received
offline
and
online
and
address
the
corresponding
solutions,
one
by
one.
Let
the
comment
expressed
by
a
gherkin
that
alignment
with
the
draft,
the
Colombians
and
amarjot
it's
the
intent.
K
We
addressed
that
by
updating
the
draft
and
the
referring
in
the
introduction
time
and
explaining
what
the
scope
of
the
each
draft
the
wrong
and
Laurent
commented
that
the
value
of
the
document
and
the
tables
not
clearly
explained
as
well.
Randy
salami
Isaiah,
committed
that
intended
obstruction
should
be
formal.
Therefore,
we
address
the
by
updating
the
introduction
section
and
the
discussions
on
the
mailing
list,
as
well
as
explaining
in
the
introduction
the
purpose
of
this
draft,
and
that
is
not
planning
to
be
a
formal
definition
of
intent
here.
K
Al-Khatib
said
under
ministers
that
it
is
not
clear
what
is
a
difference
between
intent
and
the
policy
in
your
draft,
and
we
addressed
by
updating
the
introduction
and
the
referencing
drafted
that
describes
the
differences
between
service
policy
and
intent
as
well.
We
updated
the
section
four
and
the
subsections
of
five
four
points:
three,
four
five
and
six
by
making
them
intent
and
in
those
policy
focused.
Similarly,
we
also
have
received
some
suggestions,
such
as
your
ROM
and
the
law.
K
Rental
suggested
that
or
draft
shoe
the
presenter
intent
on
methodology
and
they
were
adopted
by
ending
a
new
subsection
5.1
and
a
5.2
rule
also,
otherwise
that
it
would
be
better
if
we
could
have
some
concrete
examples
of
intents
from
different
perspectives.
In
response
to
the
subject
to
this
suggestion,
we
have
added
examples
to
the
tables
in
Section
5.1.
K
K
First
Eclair,
we
first
her
classifier
in
turns
into
it
in
tabs
and
the
describe
each
tab
based
on
the
solution
is
a
blast
tool
and
the
word
intern
user.
It
is
a
form
with
them
present
different
categories
that
this
intent
can
belong
to,
based
on
intense
scope,
network
scope
into
the
obstruction
and
the
life
cycle
is
about
mythology.
K
This
should
we
have
either
added
an
Internet
examples
for
our
intern
tabs
in
all
the
tables
in
section
5.2,
3.1
and
the
four
point
one
and
this
5.1
the
picture.
Otherwise
in
snippet
the
snippet
from
high-resolution
table
instruction.
Five
point
three
point:
one
for
example,
for
a
network
operator
with
the
intent
AB
is
network,
our
service
intent,
the
operating
the
operations
that
may
be
desired
is
request
to
network
service
with
delay
guarantee
for
access
custom.
A
ok
next
slide.
Please.
K
The
intent
are
used
to
manager
the
behavior
of
the
network's
they
are
applied
to
and
all
intents
are
applied
within
a
specific
scope
such
as
connectivity,
security,
application
and
ACK
us,
and
the
in
Subsection
four
point:
four:
regardless
on
the
intended
user
type,
their
internal
request
is
affecting
the
network
or
network
components
which
are
responsible
representing
the
intent
pockets.
Thus,
it
and
blood
work
scope
can
represented,
we
haves
or
PDFs
physical
network
elements,
compressed
networks
as
d1
networks
and
the
cloud
edge,
clock
or
branch,
and
so
on.
K
We
updated
our
Phi
upon
me
by
renamed
as
interned,
obstruction
to
focus
on
characteristics
of
faith,
Bank
of
intent
and
the
Internet
can
be
classified
by
whether
it
is
necessary
to
fit
back
her
techno
technical
networker
information
or
non
technical
information
to
the
intended
proponent
after
the
intent
is
executed
as
well.
It
an
abstraction,
covers
the
level
of
technical
details
in
the
tank
in
turn
itself,
and
we
also
updated
4.6
internal
F
circle,
renamed
the
live
internal
app
cycle
to
classify
in
turn
into
transiency,
transcending
and
persistent.
K
If
it
was
the
internal
transient,
it
has
no
lifecycle
management.
As
soon
as
the
specific
specify
the
operation
is
successfully
carried
out,
the
intent
is
finished
and
they
can
no
longer
affect
as
a
target
object.
In
turn,
a
is
persistent.
It
has
left
cycle
management
once
the
intent
is
successfully
and
activated
and
deployed.
The
system
will
keep
all
relevant
and
intense
active
until
they
are
deactivated
or
removed.
The
next
slide,
please.
K
We
have
updated
in
the
introduction
to
better
explain
the
need
of
intent,
a
classification
and
acknowledged
contracts
for
different
different
additional
concepts
and
terminology
lab
is
currently
leading
these
efforts
by
defining
it
and
at
higher
level
the
corrupt
declarative,
positive
that
operates
at
the
level
of
network
and
a
services
it
provides.
However,
even
this
propose
the
internal
concepts
and
terminology
and
agreement
on
corner
internal
characteristics
and
instantly
still
be
billed
in
different
ways
by
different
stakeholders
for
different
use
cases
and
the
solutions.
K
Thus,
the
goal
of
this
draft
is
to
bring
clarity
to
what
an
incentive
represents
for
different
stakeholders
by
means
of
classification
of
various
dimensions.
The
classification
would
ensure
a
common
understanding
across
all
participates
and
it
can
be
used
to
identify
the
scope
and
the
priorities
of
individual
projects.
Theosis
research
or
open
source
projects.
This
drafted
together
must
climb
aims
to
become
the
foundation
for
future.
In
turn,
related
topic
discussions,
where
all
participants
have
the
same
common
understanding.
K
Concluded
the
corporate
vision
has
addressed
Oklahoma's
received
online
and
offline
and,
furthermore,
we
want
to
ask
for
adoption
as
a
research
group
practice,
especially
the
next
step
of
this
craft.
The
summary
is
a
following
of
work
plans.
First,
add
a
section
briefly
describing
intent
works
that
has
related
to
intended
classification
and
tax,
a
tax
on
autonomy
in
other,
a
skills
such
as
the
3gpp
and
ETS
a
and
mes
victims.
K
The
drops
are
aligned
with
the
trapped
air,
ta
mr.kia,
beyond
concepts,
definitions,
what's
more
work
with
the
community
to
derive
unify
the
intent
definition
that
encompasses
all
internet
apps
for
all
solutions
and
the
intent
users
as
last
improves
or
drop
to
folder,
based
on
any
new
comments
or
introductions
to
this
draft
update.
That's
all
thank
you
for
listening.
C
Okay,
maybe,
as
I
was
speaking
as
a
chair
regarding
your
your
last
night,
so
first
of
all,
thank
you
for
you
very
hard
work
to
improve
the
document
and
insert
different
question
regarding
the
process.
One
of
the
suggestion
I
would
like
to
give
to.
You
may
be
its
first
to
publish
the
release
version
as
as
the
version
feels
individual
draft
I,
think
that
will
be
the
first
step
and
then
before
ring
for
asking
for
the
adaption.
C
My
advice
will
be
first
to
go
for
a
very
short
round
of
comments
regarding
this
version
free
on
the
mainland
leagues
like
two
weeks,
but
we
could
are
from
people
living
comment
on
the
previous
version
to
see
if
they
are
satisfied
with
how
you
address
a
comment
and
that
after
this
check,
if
everything
is
fine,
maybe
there
could
be
we
can
we
last
for
adoption,
officially
fuzzy
fuzzy
document.
That
will
be
my
advice,
the
char
two
to
be
sure
that
before
you
as
reduction,
which
would
be
somehow
enter
into,
let's
say,
official
process.
C
The
commands
you
received
from
previews
from
people
who
have
previously
made
this
comment,
our
okay
with
and
the
manner
you
modify
the
document,
but,
of
course,
do
not
forget
to
first
published
version
fee
of
the
document.
I
know
you
sentence
a
mailing
list,
but
please
upload
on
the
data
tracker
that
is
is
fully
shared
and
into
Z
into
the
document
of.
J
And
I
just
give
a
comment.
Tara
like
I,
think.
Thank
you
very
much
for
your
suggestion.
Also,
but
I
just
wanted
to
say
in
regards
to
comments
to
engage
the
comment.
Authors
I
think
just
reminds
that
we
went
through
the
extensive
phase
of
emails
and
powerpoints
with
all
the
comments
and
asking
the
comment:
orders
for
the
for
the
conclusions
and
and
for
their
feedbacks
in
regards
to
how
we
wanted
to
address
the
comments.
J
Just
wanted
to
remind
you
about
that
meant
probably
I
would
say
a
month
of
exchanges
and
discussions,
but
I
do
believe
it
is
better
to
do
it
with
the
Version
three.
So
thank
you
very
much.
We
will
repeat
the
process,
but
I
just
want
to
remind
that.
We
tried
to
do
that
already
and
we
had
powerpoints
with
all
the
comments
and
asked
for
people
to
reply,
but
we
didn't
get
replies
from
everybody.
C
Okay,
I
understand
your
point
so
as
in
a
world
of
culture,
if
you
publish
it
as
version
free,
we
will
also
push
a
bit
of
the
mailing
list
said
that
people,
including
me
as
a
personal
as
well,
replied
to
the
to
the
new
document.
Oh
yeah
address
the
comments
to
give
you
the
feedback
in
a
relative
short
period
of
time
like
as
I
said
two
or
three
weeks
maximum.
C
C
It
will
be
to
be
sure
that
at
issue
more
chance
to
have
a
document,
then
I'll
be
the
first
in
the
group
you
just
just
kind
of
for,
for
the
sake
of
checking
that
the
way
you
address
the
document
is
aligned
with
the
document
you
receive.
Zara
Zara
has
no
misunderstanding,
but
of
course,
as
chair
we
can
push
rivets
on
the
mailing
list
to
receive
comments
and
as
well
that
okay.
A
B
A
Different,
so
maybe
we
could
clarify
by
the
least
the
discussion
today.
What
we
have
seen
in
the
slide
presented
today,
our
recommendation
as
chairs
for
publication
of
d03
to
be
the
new
baseline,
ask
for
comments
for
a
period
of
two
weeks
and
the
conclusion
of
this
period
to
be
able
to
restart
a
new
call
for
adoption.
If
you
can
clarify
this,
my
email
I
think
it
wouldn't
make
a
good
point
on
them
in
English.
With
respect
to
the
previous
column,.
A
L
All
right
so
in
this
15
minutes,
I
want
to
present
two
draft
that
have
been
presented
to
the
absolute
value
G
but,
as
you
were
mentioning
lo-har,
the
beginning,
they're
highly
relevant
to
you
anymore
G.
So
the
two
draft
are
mentioned
they're.
The
things
that
I
want
to
mention
first,
is
that
we
believe
we
address
mostly
feedback
received
in
NMR
G
last
time,
verbally
or
many
verbally
and
during
the
research
group
session,
and
also
the
feedback
received
from
the
AuSable
ug.
L
The
second
thing
is
that
you
would
see
from
the
list
of
authors:
we've
been
working
with
many
operators
and
I
see
Paulo
and
Diego
on
the
list
here.
So
much
feedback
address
there.
So
next
slide,
please
the
the
issue
were.
Having
is
that,
yes,
we've
got
yang.
Yes,
we
could
configure
networks,
but
the
fact
is
that
we
all
know
it.
We've
got
like
a
service
configure,
but
it
doesn't
mean
that
it's
operating
correctly
there
could
be
many
issues
there.
Now
we're
telling
okay
we're
giving
to
send
telemetry,
but
we
see
it
coming.
L
We've
got
like
way
too
much
telemetry.
This
is
the
issue
of
the
needle
in
the
haystack
right
now,
whatever
service
degrades,
the
the
the
thing
is
any
knock
wants
to
discover
world
fault.
Yes,
and
by
this
I
mean
there
are
easy
ones
like
okay,
there
is
the
Witco
interfaces
down,
for
example
great.
We
know
it
almost
directly,
but
whenever
we've
got
like
symptoms,
it's
get
like
most
difficult
lives,
a
and
B
and
sometimes
C.
Then
we
have
the
symptoms.
You
know
what's
happening.
The
other
issue
would
ring
to
solve.
L
L
We
have
to
understand
as
well.
The
first
step
is
like
close
to
perturbation.
This
is
what
we
trying
to
solve
here.
The
next
slide.
Please.
However,
we
know
that
the
end
goal,
which
is
like
intend
based
not
working.
It's
like
a
pyramid
of
needs
right,
you're
able
to
do
configuration
and
you've
got
operation
later
I
wish
telemetry.
Then
you
close
the
loop,
then
one
of
you
close
the
loops.
L
You
could
potentially
go
to
self
star
two
in
ten
base
right,
and
this
is
what
we're
trying
to
do
trying
to
decompose
this
from
bottom
at
and
twenty
decomposes
into
smaller
components.
What
we
call
is
sip
service,
then
we've
got
an
assurance
graph
that
is
going
to
link
those
services
and
we're
going
to
have
we're
going
to
map
it
to
service
in
tens.
L
I'm
going
to
show
you
how
we'll
do
it,
then
the
idea
would
be
to
assure
all
these
sub
services
independently
I
mean
it's
a
fact
that
we
all
collect
like
the
interface
counters,
but
okay.
So
what
you
have
a
nice
graph,
it
would
be
way
better
to
understand.
If
the
interface
behaves
correctly,
then
we
would
have
like
a
health
score
for
the
interface
health.
L
Now,
because
you've
got
this
runs
graph,
we
could
bubble
up
the
infirmed
score
of
the
service
and
I
want
to
stress
inferred
right,
because
this
this
architecture
complements
the
end-to-end
active
probing
right,
we're
going
to
get
like
the
the
real
packet
loss
and
then
delay
and
and
and
jitter,
and
all
this.
So
whatever
you've
got
those
two
solution
understand
it.
We
start
to
have
a
good
architecture
next
slide.
Please.
L
If
we
look
at
the
dependency
three
the
assurance
graph,
we
want
to
have
returners
to
be
healthy.
We
want
to
have
on
the
left-hand
side
the
two
interfaces
Pier
one
being
healthy,
as
well
as
the
pure
to
being
a
destination,
and
those
two
have
rely
on
the
house
of
a
physical
interface
right
and
and
then
that
you've
relied
on
the
house
device.
You
know
it
would
be
ways
year
if
you
would
have
like
a
device
and
the
ability
to
say
this
device
is
healthy.
L
So
if
you
look
for
a
problem,
this
is
not
where
the
problem
is
on
the
right
side.
The
Turner
service
instance
also
depends
on
the
IP
connectivity
between
A
to
B
or
p1
p2.
That
also
depends
on
on
a
control.
Plane,
in
this
case,
is
is
so
we've
got
to
search
the
top
and
the
SIP
services.
The
sub
service
instances
are
at
the
bottom
right
next.
L
Oh
yeah,
that's
one!
Yes,
so
then
what
you
could
be
doing
there
as
as
I
mentioned,
if
we
know
it
interface
is
like
unhealthy,
an
optic
is
bad,
is
degraded
right.
We
have
a
the
house
of
that
interface,
so
that
uptake
sorry
and
then
we
could
just
say
well,
obviously,
because
you
got
the
eshton's
graph.
L
The
specific
Turner
service
instance
is
affected,
so
I
could
infer
the
health
of
a
tunnel
source
instance
by
looking
at
the
health
of
the
components
into
three
the
assurance
graph
and
on
the
next
slide
I
want
you
sure
that
there
are
two
different
types
of
dependencies
right.
Obviously,
if
I
got
like
an
interface
being
down,
that's
obvious
that
my
tunnel
will
be
down,
certainly
call
an
impacting
dependency
right.
L
This
is
the
root
cause
we
know
it
and
on
the
next
slide,
we've
got
like
typically
an
informational
dependency,
meaning
I've
got
this
dinner
service
when
it
goes
over
for
ecmp
pause
right,
one
of
them
is
down.
Is
my
serious
impacted?
Yes
or
no?
Maybe
yes,
maybe
no
right,
but
the
point
is
that
I
want
to
be
aware
of
that
symptom,
and
this
is
what
we
mentioned
by
informal
informational
dependency
here,
because
that's
some
some
combined
with
a
different
symptoms.
L
Trying
to
summarize
what
we've
seen
so
far
when
a
service
degrade,
we
could
discover
what
a
fault
is
whether
it
is
a
symptom
or
root
cause,
but
most
of
the
time
the
most
interesting
ones
are
the
symptoms,
because
the
root
cause
are
easy
and
then
we
know
so
that
whenever
a
component
fails,
it
was
mentioning
the
object.
Then
we
know
which
service
instances
are
impacted
all
right.
Next,
please
I
want
to
show
here
that
so
nice,
something
here,
it's
like
in
architecture,
sorry-
and
we
try
to
do
it
for
multi-vendor.
L
It
has
to
be
because
they
are
not
many
networks
in
the
world
that
are
only
a
single
vendor
and
we
did
it
with
the
yang
module.
It
was
somehow
my
my
feedback
to
one
of
the
previous
draft
that
was
yang.
We
could
use,
we
could
generate
api's,
and
this
architecture
is
flexible
in
terms
of
components.
I
will
show
you
that
in
terms
of
service
and
sub
services,
so
we
could
be
adding
new
sub
services
as
we
see
them
right.
If
someone
needs
multicast,
we
could
add
a
multicast
service,
an
expander
shion's
graph.
L
L
So
next,
please
so
architecture.
So
we
get
the
intent
from
the
service
configuration
Orchestrator.
We
get
like
the
service
ID
the
service
config,
and
then
we
extract
the
intent
of
that
service.
We
create
the
assurance
graph
that
goes
into
some
sane
agent,
and
then
we
collect
the
require
metric
where
telemetry
and
we
send
the
house
core
and
the
symptoms
of
all
of
those
sub
services.
L
L
The
next
slides
I
want
to
show
that,
basically,
it's
flexible
in
terms
of
components,
we
could
have
like
the
same
agent
and
in
router
it
switches
and
we
could
have
like
the
orchestrator
and
the
scene
or
castrator
both
into
a
single
box.
The
point
in
next
slide
is
that
what
we
want
to
do
is
this
open
architecture
monitor.
So
you
want
to
specify
the
interfaces.
L
L
L
L
If
you
are
sharing
Sunanda,
shirring
interface,
you're
assuring
you
need
to
parameters
the
device
ID
interface
ID,
and
we
showed
how
we
could
extend
this
even
with
some
vendor
specific
sub
services
right
in
my
first
slide.
For
example,
if
I
got
like
a
Nokia
one
way
or
juniper
device,
there
should
be
the
one
having
the
device,
health
all
right
and
I
think
we'll
be
happy
lahar.
My
next
slide
is
the
last
one,
something
that
needs
to
be
improved.
L
This
is
I
need
to
align
the
the
draft
better
with
the
the
MRG
concept
and
definition
feedback
I've
been
asking
is
like.
Is
this
a
valid
problem
to
solve
at
the
ITF
and
now
we're
going
in
the
right
direction?
And
then
I
asked
that
question
to
the
observable
ug
it
was
for
adoption.
However,
the
feedback
of
this
group
is
that
it's
I
would
like
to
get
the
feedback
to
make
sure
that
all
the
concept
definition
are
going.
The
right
discussion
according
to
what
discussing
the
research.
So
thank
you.
I
I
L
A
very
good
question
and
I
believe
the
answer
is
yes,
and
no
yes,
because
there
are
some
of
them.
That
should
be
solid
eyes.
I'm
thinking,
you
know,
as
I
mentioned,
we
all
monitoring
like
the
interface
counters
and
whether
whatever
router
vendor
you
have
trying
to
get
like
the
Houseman
interface
should
be,
should
be
under
standard.
This
is
not
what
the
value
is
right.
The
value
is
higher
in
the
inner
in
the
stack,
so
some
of
them
yes,
protocols,
I,
was
mentioning
I,
Esiason
purpose
right
or
segment
rotting
or
spring.
What?
M
M
One
question
that
I
would
have
regarding
the
practicality
of
this,
because
presumably
you're
going
to
have
a
lot
of
in
the
dependencies
between
a
lot
of
objects.
So
do
you
have
any
comments
on
that?
How
these
relationships
a
lot
of
dependencies,
a
lot
of
which
maybe
may
be
highly
dynamic
and
yeah
and
so
forth?
That.
L
That's
right,
I
mean
you
you,
you
are
fully
right,
so
this
would
be
a
complex
graph,
so
we
could
set
different
levels.
You
could
just
say
say
at
what
was
configured
as
you
know,
getting
like
the
house
of
the
IP
connectivity
that
was
mentioning
in
arms
like
whatever
just
trying
to
get
that
and
potentially
trying
to
see
where
it
goes.
It
goes
from
A
to
Z,
going
from
BCD,
etc
or
ecmp.
Ecmp
is
dynamic
right,
poor
definition,
right
I
mean
if
we
could
just
solve
the
EMP
issue.
It
would
be
great.
L
N
I
think
the
one
of
the
interesting
question
to
me
is,
you
know,
to
run
through
some
example
of
workflows,
right,
I.
Think,
that's,
that's
that's!
You
know
some
some
amount
of
work.
That
would
be
good
understand
what
could
be
gotten
out
of
this.
So,
for
example,
the
way
I
kind
of
imagine
how
this
this
ultimately
works
is
that
I
have
you
know
different
levels
of
abstraction
right.
N
What
what
comes
from
the
service
configuration
orchestrators,
you
know
more
high-level
and
then,
when
it
goes
further
down,
then
then
I've
got
some
some
lower
level
like
let's
say
existing
yang
based
configuration
of
individual
Network
elements
and
the
way
I
see
it.
What
what
what
would
be
lovely
is
to
have
a
way
of
attaching
one
level
to
the
other
through
references
so
that
you
know
when
I
look
at
configuration
of
a
router
I'll
be
able
to
see
which
of
the
you
know.
N
Services
are,
you
know
requiring
a
particular
configuration,
or
vice
versa,
when
I
basically
want
to
do
some
inventory
management
or
something
like
that
that
at
the
service
level
I
can
see.
You
know
what
network
components
or
you
know
configurations
are
there.
So
that
that
mutual
referencing-
you
know
I'm
not
seeing
this
expressed,
but
that
would
ultimately
I
guess
be
something
that
you
know
the
orchestrator
might
might
need
to
do.
What's.
What's
your
thought
on
this
right.
L
So
you're
right
in
a
sense
right,
so
basically
we
take
a
service
model
on
top
of
registrate
ER.
Then
we
configure
something
device.
There
are
two
different
choices:
either
it's
a
CLI
or
this
yang
Yang.
Then
we
have
a
starter
solution,
because
you
already
have
the
yang
models
that
we
have,
that
we
need
for
config
and
then
based
on
that
we
could
try
to
find
a
mapping
and
it's
easier
if
you
start
from
yang
models.
L
It's
number
two
is
that
you
start
from
CLI,
and
then
it
is
it's
harder
to
do
and
I
agree
with
you
with
the
lower
level,
and
that
may
be
your
so
to
answer.
Alex
is
that,
okay,
there
thing
that
we
know
from
the
config
of
the
service
there
are
that
we
know
from
the
network
right
I
mention
examples
is:
is
your
service
model
will
not
configure
eyes?
L
Is
the
IGP
is
there
by
default
right
so
as
a
consequence
need
to
learn
it
from
the
network,
and
there
are
extra
things
that
we
won't
want
to
have
like
I
mentioned
multicast
right.
We
we
might
want
to
see
a
specific
service
use,
multicast
or
not,
and
there
we
might
need
to
have
the
operator
input
to
constrict
the
first
graph
because
we
could
graph
the
first
time
we
got.
A
new
service
type,
make
sense.
Yeah.
N
In
the
most
simple
case,
we
have
two
levels,
let's
say
the
service
level
and
then
the
per
device
level
configuration
forget
about
the
CLI
stuff
agreed
makes
it
more
difficult,
so
device
config
is
also
yang,
but
purely
you
know
the
abstraction
models
by
which
I
can
say:
oh
okay,
here
I'm
looking
now
at
the
device
level
configuration
but
attached
to
each
I,
don't
know
level
or
something
else.
In
the
yang
device
level
configuration
are
pointers
pointing
back
to
the
services
that
are
requiring
this
so
that
I'm.
Basically,
this
edge
version
I
said.
L
N
L
Mention
that
I
mentioned
distributed,
the
router
would
know
about
part
of
his
serve
services,
the
device
house
interface,
health,
telemetry
house
whatever,
but
not
the
entire
tree.
However,
because
it's
yang,
you
could
just
have
a
leaf,
ref
right
and
just
and
okay,
so,
let's
a
give
cough
line.
I
guess
large
is
what
you
want
right
and.
N
A
Document
I
know
it's
progressed
more
in
ability,
but
I
think
again
links
to
to
the
activity
on
again
the
research
group
and
has
been
no
mention
with
flashlights,
expect
further
reviews
and
feedback
on
Fitness
of
the
concept
that
he's
considering
and
I.
Think
any
review
and
feedback
are
always
valuable.
B
B
A
So
next
in
the
agenda
will
be
a
very
quick
updates.
I
apologize
to
everyone
for
the
sport
agenda.
We
have
to
make
too
much
stuff
well
really
much
overtime.
We
will
try
to
go
very
quick
on
the
last
technical
presentation
and
give
save
a
few
minutes
to
give
some
research
group
updates
to
the
people
that
ken
has
still
be
connected
at
this
time.
So
the
next
presentation
will
be
on
the
14
concept
and
definition
draft.
Let
me
grab
the
slide.
This
should
be
very
quick
of
me.
A
A
I
will
present
as
a
cohort
dates,
so
this
draft
is
the
dinner.
So
this
is
a
person,
one
of
wrought
iron,
chef
and
energy
I'd
be
any
concept
and
definitions
very
recent
history,
so
this
document
was
adopted
as
we
serve
to
document
in
December.
Last
year
we
made
some
changes
in
run
essentially
completed
each
other
video
of
the
document.
We
have
also
revised
and
aligned
definitions
for,
in
terms
after
the
document,
which
is
basis
actually
on
the
introduction
mailing
list
and
with
the
draft
on
intern
classification
which
triggered
some.
A
Let
us
know
about
some
inconsistency
in
the
definition,
so
this
was
a
good
catch.
We
have
also
reworked
and
improved
refined
text
for
the
sections
five
imprinted.
Some
of
the
principles
were
so
big.
Slightly
worked,
we
swapped
section
6,
section
7.
So
now
the
new
section
6
is
functionality
and
section
7
is
a
life
cycle.
A
This
allows
us
to
explain
a
bit
before
going
into
the
details
of
the
diagram
a
bit
more
in
a
higher
level,
the
main
functionality
and
the
composition
of
the
infant
life
cycle
from
a
textual
point
of
view,
description,
point
of
view,
and
then
we
have
in
section
new
section,
7
the
life
cycle
diagram
and
some
quick
explanation
of
the
diagram
in
this
life
cycle
section.
Initially,
we
had
two
diagrams
two
kind
of
competing
diagram,
so
we
decided
to
keep
that
was
comprehensive.
A
Cover
all
the
different
elements
of
the
life
cycle,
we
thought
were
essential
for
the
job.
We
still
have
to
consolidate
a
bit
further
and
stealing
from
a
bit
on
the
description
of
this
life
cycle.
But
now
it's
a
bit
more
clear.
Also
for
the
external
readers
about
this
single
life
cycle,
diagram
and
I.
B
A
A
A
This
has
not
changed
mainly
some.
We
architect
clearing
and
improvement
of
the
take
up
the
next
steps,
so
we
are
working
on
the
Russian
0
2,
which
should
be
available
in
time
soon.
We
will
continue
to
improve
the
which
say
the
text
of
the
draft.
We
still
have
a
few
principles
that
we
wanted
to
add.
There
were
to-do
list,
so
we
will
provide
input
for
that.
We
will
also
work
in
the
concept
and
definition
draft
slightly
improve
all
sort
of
sectional
acronym
stands
and
definitions
to
have
proper
entries
into
this
section
of
the
document.
A
With
respect
to
the
concept
and
definitions
define
throughout
the
text,
as
I
said
complication
of
the
lifecycle
diagram,
we
will
also
collect
any
feedback
on
if
this
is
complete
and
angry
enough
to
proceed,
and
we
have
some
training
comments
and
reviews
that
we
received
some
months
ago
and
more
recently,
that
we
need
to
integrate
in
the
different
sections
of
the
talking.
So
this
is
pretty
straight
forward
in
terms
of
evolution,
knowing
that
this
document
we
believe,
is
pretty
stable,
it's
not
finished,
but
we
don't
expect
that
merger
reactivating
of
the
structure
of
the
document.
A
A
Not
necessarily
only
this
one
IRS
G
review
by
November
2020,
so
this
is
still
in
our
plan,
but
it
will
depend
how
we
move
on
with
the
next
version
and
the
comments
we
may
receive
for
that.
But
in
our
perspective,
as
a
group
of
co-authors,
we
think
it's
becoming
more
and
more
stable
and
it
would
be
nice
to
reach
these
targets
by
the
end
of
the
year.
A
It's
not
ok
thanks,
so
we
are
not
going
to
continue
with
the
chair,
slides
and
conclude
this
meeting
with
some
updates
on
the
research
group
so
to
broach
quickly
as
ever.
More
so
don't
hesitate.
Add
things
on
my
flow,
the
general
information
we
had.
Our
new
charter
approved
end
of
March,
so
this
is
still
quite
recent,
so
congratulations
and
thanks
to
everyone
that
participated
in
providing
inputs,
comments
and
support,
or
the
revised
Charter,
the
inputs
in
terms
of
technical
content,
review,
etc,
and
also
to
our
ITF
chair
for
bringing
this
forward.
A
It's
not
usual
for
wishes
group
milestones,
but
we
believe
that
some
of
our
work
items
deserve
it
and
it
will
help
us
also
to
try
to
put
energy
and
a
bit
of
pace
in
our
in
our
research
there,
because
I
think
the
industry
is
expecting
also
some
some
answers.
What
could
move
towards
also
ITN
and
other
groups?
A
We
have
also
initiated
a
few
months
ago
the
a
bit
up
organization
for
energy,
so
you
can
find
the
links
here.
We
invite
you
to
join
as
contributors
and
we
will
plan
to
host
repositories
for,
of
course,
research
group
documents,
but
we
can
also
offer
repositories
for
individual
documents
if
you
want
to
get
there.
Also
for
the
code,
we
have
a
few
implementation
and
be
an
ID
and
proof
of
concept
that
we
are
going
to
port
there,
and
also
we
are
looking
forward.
B
C
C
So
this
is
something
we
we
talked
a
bit
in
particle
in
the
in
Singapore,
where
we
wanted
to
document
in
order
to
actually
have
detailed
description
of
the
different
challenges
of
artificial
intelligence,
applied
to
network
management
use
case
I'm,
not
sure
we
have
time
to
go
through
the
document
itself.
Law
or
maybe
I
should
just
summarize
that
what
do
you
think
we
still
have
a
bit
of
time?
Okay,
so
maybe
you
can
open
it.
I
will.
C
A
C
To
okay,
so
nobody
have
the
document
so
that
you'll
appear
with
the
idea
that
right
because
I
know
that
we
know
that
a
lot
of
people
were
willing
to
contribute.
We
are
submitting
and
son,
but
we
need
a
little
bit
to
have
some
structure
to
12c
Dickey
Matt
written.
So
we
started
to
write
drafting
from
a
table
of
content
that
we
present
hint
in
previous
virtual
meeting.
C
So
the
idea
is
this
document.
Of
course
the
core
part
of
the
documented
big
challenge
is
ready
to
in
network
management.
We
have
identified
some
of
them
during
some
sign
meetings
in
previous
IETF.
We
also
be
doing
some
real
chair
meetings,
so
here
in
the
basically
in
the
in
the.
If
you
go
bit
below
Aloha,
is
you
have
the
you?
Have
those
a
talk
and
the
table
of
content?
That
is
ok
a
bit
expanded
so
that
to
two
main
sections.
C
So
challenges
as
well
as
the
goals
of
a
network
management
has
ones
are
the
two
main
sections
that
we
would
like
to
work
at
this
time
and
of
course,
maybe
on
the
section
and
research
challenges
for
each
section.
We
provide
some
kind
of
a
guidelines,
a
question
to
answer
to
our
people
to
know
what
type
of
contents
you
are
expecting
here,
yeah
as
long
I,
like
idea
for
challenges
and
so
for
the
different
challenge
we
have
identified.
But
of
course
you
can
add
some
challenges
that
you
have
in
mind.
C
You
should
try
to
write
the
text
in
a
way
that
you
answer
the
different
questions
and,
of
course,
shall
identify
yourself
and
interest.
Contribute
to
this
document.
Here
is
a
draft
example:
I've
worked
on
lightwei
I
for
network
management
of
course
base
you
have
different
sections
so
and
all
school.
It
needs
to
be
documented,
with
reference
and
so
on,
to
have
already
something
quite
concrete
and,
of
course,
an
attempt
to
free
font
files
are
that
are
other
challenges
we
have.
C
We
have
in
mind
based
on
the
discussion
where
the
previous
meetings,
so
basically,
if
people
are
willing
to
contribute
this
document,
of
course,
we'd
share
again
the
mailing
list.
Please
clearly
have
a
look
on
this
table.
Look
as
judge
challenger
interesting
into
participating,
identify
yourself
as
a
contributor
here
in
the
document
he
in
the
beginning
of
the
clinic
on
and
the
table
and
also
propose
some
contact.
So
you
have
detailed
instructions
at
the
beginning
of
the
document,
so
you
will
have
some
kind
of
clear
restriction.
C
To
be
honest,
we
we
have
a
use
case
section
it's
the
document,
but
we
do
not.
We
request
at
that
time
content
for
this
section,
because
we
know
from
previous
experience
that
if
you
go
directly
into
use
case,
it
would
be
it
would
be
completely
on,
let's
say
separate
use
cases
without
really
some
kind
of
some
kind
of
or
to
say
some
kind
of
commonalities
between
use
cases.
So
here's
the
goal-
you'll
really
tomorrow,
look
into
the
challenges.
C
Of
course,
in
the
goals
section,
you
can
provide
all
those
some
kind
of
overview
of
use
case,
but
not
detail,
use
case
at
that
time,
which
will
come
later
in
order
to
have
a
moth
entities
view
on
the
challenges.
It's
really
work,
yeah
on
purpose
that
we
did
that
and
so
yeah.
That's
really
key
is
a
document
document
document
version
and
if
you
go
back
to
the
slide,
please.
C
C
No
but
I
think
it
may
be
some
some
kind
of
delay,
but
so
then
it
just
only
on
the
other
part
of
the
stages
so
area
with
just
a
summary
of
the
process,
as
I
said
before,
if
you
are
willing
to
contribute
just
yeah,
you
identify
yourself
in
the
document
also
give
feedback.
If
you
disagree
with
what
you
proposed
in
terms
of
table
of
contents,
of
course,
it's
not
fixed
future.
It
may
evolve.
I.
C
It
should
be
yes
publishes
white
paper
or
Journal
magazine,
but
it
will
be
long
term
long
term
goals,
but
as
many
people
says
that
it's
interesting
to
participate
to
such
an
initiative,
but
we
will
need
visibilities
will
be
good
to
have
this
document
chest
as
a
draft
as
well.
So
here's
a
proposition
regarding.
Yes,
the
type
of
half
way
to
to
collaborate.
You
start
with
Google
Docs
it
just
to
facilitate
this.
The
starting
of
the
collaboration.
C
Does
that
mean
that
you
will
continue
with
Google
Docs
personally
phew
if
we
want
to
use
the
github
that
Aloha
mentioned
previously?
Is
that
yeah?
Yes,
we
will
probably
use
it
but
to
be
as
most
possible
as
open
at
the
beginning.
We
will
start
with
Google
Docs,
but
it's
not
does
not
mean
that
it
should
be.
The
final
final
version,
the
final
type
of
tools
that
we
will
use.
C
So
here
we
are
really
willing
for
contributors
and
feedback
on
the
process
and
the
document
content
that
we
would
like
to
have.
So
if
you
have
yeah
and
if
it
by
just
tell
us
now
or
in
mailing
this,
because
I
know
we
have
a
bit
short
of
time
now-
maybe
I
just
just
presented
the
full
slide,
and
then
you
can.
Somebody
asked
question:
can
just
raise
the
hand
regarding
a
hi
we
have
in
my
episode
forgot
that
is
more
regular.
C
That's
a
series
of
talk
regarding
regarding
this
topic
in
order
to
have
more
technical
talk,
as
it
was
suggested
to
us
as
well.
So
this
is
really
something
which
is
in
the
panimalai
step.
It
should
not
be
actually
we.
We
do
not
especially
see
that
as
only
enemy
activities,
but
we
would
like
something
more
global
with
other
group
and
community.
C
So
it
was
supposed
to
to
be
open
soon,
but
you
know
with
the
situation:
it's
not
clear
when
it
will
be
open
and
and
don't
think
we
don't
have
all
been
here
regarding
an
emoji
as
far
as
I
know,
all
the
unfortunately
all
the
project
a
bit
but
stop
now,
but
of
course
it's
still
open.
If
an
emoji
would
like
to
propose
a
technical
challenges,
I
mean
participant
of
an
emoji,
so
it's
still
open,
but
as
far
as
I
know,
there
is
no
project
that
is
continuing
for
this
last
data.
A
M
M
B
A
Of
subgroup,
so,
as
you
can
see,
this
is
the
table
of
my
two
students.
We
have
corresponding
to
what
you
have
in
the
Charter
I
try
to
map
everything
in
terms
of
work
items
with
respect
to
basic
milestones
when
the
adopted
and
the
rlg
review
the
dates
and
with
the
stages
just
wanted
to
highlight.
Okay,
it's
in
red,
but
it's
not
critical.
Okay,
we
are
not
a
working
group.
There
is
no
pgp
periods
or
whatever
it's
just
I
like
a
bit
of
a
critical
situation.
A
In
terms
of
where
we
stand
with
respect
to
our
initial
goals.
For
these
three
three
items,
the
other
ones
are
a
bit
more
on
track,
so
just
focusing
on
the
first
one,
the
first
Brigade
and
we
had
was
documented
problem
statement,
design,
goals
and
challenges
for
in
ten
days
of
working,
we
initially
a
discrete
first
work
items
together
as
a
batch
we
wanted
to
which
research
group
adoption
by
the
end
of
last
year.
Unfortunately,
we
have
no
corresponding
documents.
Currently,
on
this
specific
work,
I
tell
the
initial
goal
was
to
describe
the
problem
and
solution.
A
Space
identified,
the
limits
of
current
technologies
and
methods
and
variety
associated
research
challenges
needs
to
be
a
very
long
document,
but
at
least
I
believe
gentleman
I.
We
believe
that
this
is
of
utmost
importance
that
this
work
happens
in
the
research
group
becomes
help
us
to
define
clearly
what
we
are
trying
to
solve.
We
already
engage
in
two
other
documents,
but
I
think
this.
This
aspects
of
the
proto
statement
goals
are
important.
I
see
Alex
in
the
queue,
but
let
me
just
finish
with
the
slide
and
I
take
as
possible.
Next
steps.
A
This
is
just
open
proposal.
We
mainly
find
them,
of
course,
to
come
up
quickly
with
a
new
document
to
address
specifically
this
point
or
to
move
parts
or
all
this
work
item
to
existing
documents,
for
instance,
draft
IBM
concept
and
definition,
or
we
can
at
least
you
last
resort
drop
this
or
Qaeda,
which
I
don't
encourage
us
to
do
where
we
try
to
find
ultimately
Alex.
You
want
to
comment
on
this.
M
A
Is
no
I
was
in
developments
that
have
been
proposed
to
watch
the
energy
I
mean
personally
as
an
individual
I'm
working
on
that
for
Nokia.
So
we
have
a
draft,
not
a
draft,
but
a
draft
document
not
yet
ready
to
be
sure.
So
we
if
we
could
contribute
that
to
energy,
we
we
have
to
find
the
right
window
to
do
that
and
yes,
it
would
be
nice
from
a
research
group
perspective
to
have
something
coming
in
as
new
documents
or
based
on
something
else.
This
would
be
appreciated.
Okay,
thank.
E
A
M
A
Something
we
can
share
so
on
the
on
the
mailing
list
or
point-to-point
with
the
people
at
I'm
at
least
raise
the
call
for
for
contributions
to
the
meaningless
and
see
if
we
have
a
group
of
person
interested
to
move
forward
with
this
one
I
don't
believe
it's
so
complex,
so
difficult.
It's
just
that.
We
are
this
in
our
work
item
and
big
milestones.
So
if
we
can
achieve
that,
it
would
be
better
for
to
be
consistent
with
what
we
have
defined
in
the
chart.
A
M
A
Okay,
the
second
one
is
still
in
the
workmen
cohort,
then
there's
networking,
but
it's
addressing
a
different
part,
which
I
think
is
the
most
important
one.
We
should
work
on
in
the
research
group,
so
this
is
about
so
called
architecture.
So
the
goal
was
to
determine
the
elementary
functional
blocks
of
an
IBM
system,
their
interactions,
input
and
outputs
and
propose
different
techniques
applicable
for
the
different
functions.
A
A
To
have
something
as
an
early
result
of
adoption
for
July
this
year.
Currently
we
we
had
previously
some
individual
document
that
were
addressing
this
architecture
discussion
that
are
expired
now
there
have
been
a
few
meetings
where
we
have
have
technical
proposal
towards
architecture
discussion,
but
currently
we
don't
have
active
documents
and
new
proposal
for
that,
so
I
believe
it's
a
very
central
piece
of
the
Ibn
workman
will
be
able
to
come
with
this
common
architectural
framework,
especially
in
continuation
with
what
we
saw
currently
with
the
intent
classification
and
the
concept
and
definition
possible.
A
Next
steps
again
to
come
quickly
with
new
documents
or
activate
expired
documents
to
really
risk
in
this
activity
and
I
think
really
not
I.
Think
beyond
this
starting
the
restarting
of
this
activity,
we
should
really
increase
considerably
the
level
of
activity
and
research
group
on
this
item.
This
is
a
big
piece
of
work.
We
don't
go
into
formal
specification,
or
at
least
this
reference
model
will
be
very
useful
in
order
to
have
different
pieces
of
the
solutions
to
come
together.
A
We
have
seen
that
friends
but
know
with
each
party
ops
now
other
works
also
in
ops.
That
are
of
interest
in
any
model.
Also,
some
calls
in
the
past
I
think
there
are
many
groups
and
expectation,
and
if
we
can
contribute
with
something
that
apps
to
reach
this
common
architecture
of
view,
it
will
be
very
hard,
the
next
one.
A
A
We
have
a
some
proof-of-concept
demonstration
and
in
October
last
year,
but
from
it's
important
from
a
validation
point
of
view
and
so
potential
next
steps
to
come
again
with
documents
like
use
case
documents,
experimental
or
implementation
description
that
could
show
how
to
complete
application
of
our
ITN
concepts
in
real
settings,
also
improve
the
links
with
implementation
and
proof
of
concept.
This
is
the
item
six
in
the
work
plan.
A
So,
as
I
say,
we
have
some
proof
of
concept,
demos
and
two
that
are
being
the
block,
but
we
need
to
have
the
corresponding
description
of
those
of
those
experiments
to
be
able
to
progress
on
on
our
work
plan
and
again,
the
level
of
activity
in
that
space,
which
will
also
increase
Louis.
You
had
comments
notice
on
this
one
of
the
previous
one.
I
A
Yeah,
this
is
a
very
good
point,
because
when
I
was
reviewing
a
bit,
the
content
for
this
meeting
I
thought
that
the
work
that
you
are
proposing
with
the
transport
slice
intent
and
also
the
think
you
proposed
before
that
actually
could
be
really
related
to
that.
We
can
have
use
case
around
transport
slice
or
a
kind
of
leasing,
network
provider
plus
particle
interactions
for
for
intense
instantiation.
A
This
could
be
a
way
of
to
see
a
kind
of
experiments
or
use
case
is
concrete
boundaries
to
that
and
I
know
also
that
you're
also
on
this
number
five,
also
working
on
corresponding
I
mean
testbed
and
proof
of
concept,
so
I
believe
if
you
have
willingness
to
contribute,
this
could
be
a
good
starting
point.
Yes,.
I
A
Work
then,
we
also
had
in
mind
who
we
are
this
Akutan
project.
We
wanted
to
contribute
with
more
practical
aspect
of
IDL.
What
we
are
going
to
do
is
to
consult
with
the
different
in
more
teams
that
we
managed
to
contact
that
we
know
have
claims
in
that
space
interact
with
them
to
define
a
proper
scope
and
objectives
of
what
we
would
like
to
achieve
with
such
kind
of
academic
project
whole
overall
planning,
when
we
we
can
be
ready,
respected,
objective.
B
A
To
a
target
participation
in
the
next
IITs
meeting
in
July
in
Madrid,
there
are
still
some
uncertainty
from
the
meeting
point
of
view
if
it
will
be
in
person
and
what
will
be
the
ability
of
people
to
travel
or
participate
on
the
site
or
remotely.
So
this
is
very
open
and
then
shunt,
and
so
we
have
to
reassess
the
practicality
of
targeting
something
concrete
from
July
doesn't
mean
that
we
will
not
work,
but
we
have
to
understand
it.
It
will
be
a
bit
more
different
theme,
different
setting.
A
You
have
to
see
what
will
be
what
will
make
more
sense
and
also
for
the
future,
maybe
more
at
the
end
of
the
year.
So
we
will
communicate
towards
the
research
group
with
this
framework,
but
we
would
like
to
know
for
sure
so
that
everyone
can
be
aware
and
express
interest
and
willingness
to
contribute.
And
of
course,
if
you
have
any
willingness
to
support
and
help
us
in
this,
we're
really
welcoming
your
name
and
brain.
A
Last
point:
in
the
technical
update
for
the
research
group,
we
have
referred
for
research
activity
which
is
more
towards
the
self-driving
self
management
frameworks.
So
our
idea,
so
this
is
just
to
host
a
forum
of
discussion,
see
a
bit
what's
going
on
and
insurance
for
the
developments
of
internal
management
and
operations.
A
What
we
observe
a
bit
was
that
we
could
organize
a
kind
of
dedicated
session
based
on
what
we
have
seen
recently
inputs
coming
from
different
groups,
especially
the
ops
area
working
group,
or
we
have
seen
this.
We
see
a
framework
for
self
management
or
work
from
automating
service,
a
network
management
framework,
so
this
assurance
void
in
architecture
that
was
presented
at
this
meeting
by
Renoir.
So
there
are
already
knobs
several
works
that
are
touching
architecture
close.
E
A
Automation
policy
based
management,
automation,
all
of
correlated
items
that
I've
been
progress
by
different
organization,
different
groups
of
person
in
animal
working
group.
They
also
this
reference
model,
an
ACP
that
finally
will
be
able
to
move
forward.
So
this
can
be
also
a
way
to
reconnect
with
some
architectural
work.
That
was
a
pro
should
be
anima.
They
may
be
also
other
working.
This
is
not
exhaustive.
A
A
Okay,
so
you
just
because
we
already
quite
over
time
finally
just
a
very
quick
view
on
the
future
meetings.
So
next
week
we
should
have
had
a
in-person
intermitting
in
Budapest
for
the
ultra
polynomials
conference,
but
the
nomes
conference
will
be
fully
virtual
this
year,
so
we
decided
to
cancel
the
special
session
on
energy
because
the
idea
was
more
to
meet
people
and
gather
with
the
different
communities.
A
But
nonce
is
maintained.
It
will
be
a
virtual
compound,
so
we
can
still
register
and
attend
it's
very
good
program
that
will
be
next
week
all
the
keynotes
all
the
technical
speeches.
We
had
also
planned
special
session
dedicated
meeting
with
the
IPOP
conference
in
Tokyo.
But
again
this
conference
has
been
moved
to
September
this
year,
so
we
currently
have
to
cancel
our
kind
of
that
and
September
was
not
a
good
slot
to
an
energy
participation
due
to
absence
of
the
chairs.
A
So
we
decided
to
conserve
our
position
for
the
moment
we
desiccate
possibility,
makes
your
game
we
are
making
in
July
Madrid,
which
is
currently
maintained,
given
there
are
plans
to
consider
face
to
face
of
virtual
meetings,
but
we
have
there.
If
you
make
some
targets
in
terms
of
I
Catherine
and
Sarah,
we
shall
put
my
stone.
So
this
will
be
an
important
time
of
the
year
for
us
make
for
the
future
meetings
here
is
that
we
want
to
say
until
further
notice.
The
default
mode
of
operation
will
increasingly
rely
on
meetings.
A
We
want
to
use
more
the
meaning
list,
what
cross
our
technical
session
and
also
reliance
on
collaborative
platform
that
you
have
seen
pointment
with
the
document
at
Jerome
I
liked
it
for
these
Google
Docs,
if
it's
accessible
for
people
or
for
from
the
drafts
and
activities
and,
of
course,
not
educate
to
share
your
question
ideas
or
needs
with
the
entire
group
or
the
chairs,
and
we
know
that
maybe
it's
difficult
for
some
of
us
to
be
organized.
Since
we
are
not
meeting
face-to-face
for
Alex.
M
A
A
A
It's
always
a
challenge
from
our
perspective
to
find
the
right
slot,
but
thank
you
for
your
very
good
technical
presentation,
very
good
interactions,
also
for
the
chat,
I
hope
we
will
get
very
good
minutes
and
able
to
continue
actually
I've
seen
a
lot
of
you
good,
very
good,
technical
discussion
that
are
a
lot
of
potential
progress
and
again
contribution
driven
as
we
say.
So.
Please
continue
working
like
this.
Thank
you
very
much
analyze.
Whatever
part
of
the
day,
you
have
left.