►
From YouTube: IETF99-NMRG-20170720-1810
Description
NMRG meeting session at IETF99
2017/07/20 1810
https://datatracker.ietf.org/meeting/99/proceedings/
A
B
B
B
Okay,
so
we
have
one
and
all
stretches
with
a
three,
so
this
is
the
usual
not
well.
Today,
we're
gonna
have
a
very
short
meeting.
This
is
actually
the
second
slot
that
we
have
on
the
week,
because
we
have
the
first
stock
on
Monday,
which
was
actually
a
workshop,
which
is
part
of
a
series
of
workshops
that
we
have
on
network
measurements
in
monitoring.
B
C
B
C
B
The
other
presentations
that
we
have
first
one
is
by
new
king
and
then
we
have
a
Klaus
talking
about
his
work
on
stn
and
then
Angela
will
talk
also
about
management
out
on
on
management
in
SDM,
and
then
that
we
follow
with
that
discussion.
I
just
mentioned
before,
okay,
so
let's
move
on
for
the
first
presentation
means
to
please.
E
Yeah
hi,
my
name
is:
music,
came
from
a
tree
I'm,
going
to
present
our
drafted
for
the
intelligent
management
using
the
collaborative
enforcement
multi-agent
system.
So
we
have
another
authors
whose
name
is
I
go
home
from
a
tree
and
I
only
am
from
the
career-tech.
E
This
is
our
depth
of
abuse,
so
if
you
wanna
I
have
only
ten
minutes.
So
if
you
wanna
check
the
more
details,
so
you
can
go
over
to
the
metal
management
IG
data
tracker,
so
you
can
check
everything
all
retaking
the
web.
The
machine
learning
algorithm
that
is,
reinforcement,
learning,
I,
think
reinforcement.
Learning
is
really
useful
to
you
to
manage
it
then,
and
the
mentoring
in
some
system,
intelligent,
aliso
reinforcement,
learning,
consists
of
a
three
components.
E
So
first
one
is
action
than
that
and
stick
and
we
can
get
some
reward
from
the
for
the
optimal
value
and
the
policies
all
the
important
component
for
the
repos
belonging.
So
in
the
post
we
can
evaluate
to
some
data
from
the
action
and
state
and
weird
value.
The
disease
page
column
was
mountain
diagram,
so
we
have
a
general
motivation
for
the
reinforcement.
Learning
so
we're
both
month
alone
is
system
capable
of
the
autonomous,
acquirement
and
incorporate
of
knowledge.
So
we
can
repost
month.
E
E
Architecture
overview,
so
we
can,
but
we
have
made
multi
agent
that
has
agent
in
the
dispute
to
note
that
what
the
agent
can
detect
can
take
action.
Then,
then,
the
agent
can
ten
can
make
some
resort
to
the
transport
to
the
global
environment.
The
grouper
umbrella
meant
means
centralized
load.
Then,
after
evaluation
in
the
global
environment,
the
we
can
get
reword
the
transport
to
the
multi
agent
for
the
next
available
step.
E
This
is
our
simulation
to
compare
to
compare
with
a
normal
process
versus
reverse
Monta
learning
process,
so
the
lattice
lattice
normal
process
has
so
really
fluctuate
and,
like
one
is,
we
can
get
the
more
optimizer
to
tell
you.
So
first
one
is
a
random
exploration,
so
after
land
arms
we
can
make
some
optimizer.
We
would
to
get
some
repose
multi
learning
system.
E
So
this
is
a
multi
agent
to
repose
multi
learning
techniques.
So
the
centralized
loss
has
a
different
note
that
each
node
have
some
node
train.
They
don't
train,
manage
it
and
control
holder
some
some
ages,
so
so
in
this
purple,
so
we
took
the
distance
and
the
frequency
schemer
support
the
policy
and
the
next
one
is
second
proposed
math
dot.
This
one
is
sharing
information,
so
the
globe
the
agent
can
take
some
actions.
Then
that
is
this.
One
is
transferred
to
the
I.
E
The
or
the
reward
I
mean
the
the
optimized
values,
the
broadcasts,
the
whole
everything
suit
for
the
next
available
step.
So
this
is
a
second
result,
photo
our
scenario.
So
that
there
is
the
surely
infamous
even
reassuring
information
can
get
a
more
efficient
result
rather
than
I'm,
not
sharing
information,
but
in
the
writer
result
we
can
lesson
widget
and
share
the
information,
but
this
one
is
still
had
a
nice
great
widget
photo
because
we
use
the
repost
belonging.
E
So
we
have
possible
Network
scenarios.
So
first
one
is
an
autonomous
driving
system.
The
autonomous
driving
system
is
a
set
automobile
driving
without
human
control,
so
we
can
set
automatic
restoration
policy.
We
also
consider
about
wireless
network
because,
while
sensor
networks
are
already
supplied,
many
wires
of
the
schemes
such
as
corporate
communication
routing
and
any
other
intelligent
management
and
monitoring
system,
so
last
one
is
routing
enhancement,
an
incan
we
have
minutes
so
routers
can
determine
the
disk,
are
optimal
outs
with
the
predictor
D
word,
so
routing
can
reduce
it
operate,
so
we
can
believe
so.
E
This
is
a
long
way
walk.
So
we
sat
environment,
the
environment
and
consists
of
the
routing
protocol.
Then
the
environment,
each
environment
has
controlled
by
the
emotion
on
your
platform.
So
we
created,
we
are
creating
the
machine
on
a
platform
and
this
one
provided
that
a
promotional
average,
so
we
assumed
an
agent,
is
routers
and
policies
routing
table.
So
we
will
not
matching
something
to
set
a
architecture
to
provide
optimal
test
and
reward
and
the
we
just
on
going
and
keep
going
to
make
some
architecture
photo
repos
mataloni.
E
So
next
step
is
so
we
created
or
the
keeper
intelligent
networking
management
model,
with
the
reinforcement
learning-
and
we
also
said
repost
motor
learning
scenario,
and
then
we
also
need
to
the
purpose
and
pre-processing
for
the
data
set.
So
we
is
a
really
hard
to
look
at
some
data
is
a
real
data
set
from
the
network
via,
so
we
just
collecting
as
mix
effort,
real
data
and
then
make
some
pre-processing.
So
my
study
just
s
it.
So
if
you
have
questions
and
comments,
let
me
know
some
questions
or.
F
F
E
We
consulted
things
but
first,
first
of
all,
we
consider
autonomic
network
system
using
the
reinforcement,
alladhi
and
if
we
got
some
real
result
from
the
Potomac
and
our
implementations,
we
also
adapted
this
though
whole
dental
field.
So
that's
why
we
just
keep
going
and
make
some
effort
to
the
first
I
think
we
need
to
take
some
first
step
to
the
intelligent
management,
so
so
I
think
your
witness
is
first
step.
So
we
also
consider
about
yeah.
F
Ultimate
system-
and
the
second
question
is
about
because,
if,
as
you
using
reinforcement,
learning
I
think
that
you're,
you
have
to
define
the
utility
function
and
also
you
have
to
specify,
which
are
the
data
models
that
you
use
in
to
to
call
to
to
feed
your
system.
So,
for
example,
in
the
second
and
third
use
case,
I'm
wondering
if
you're
using
some
IETF
data
model,
such
as
yang
or
something
you
know,
another
tool
to
define
how
your
your
network
is
organized
and
then
configured
for
the
for
the
the
performance
of
the
further
to
perform.
G
C
C
H
H
Good
evening
my
name
is
Klaus
I'll
talk
about
reference
models
for
a
representation
for
Sdn,
and
this
talk
is
related
to
the
draft
that
we
recently
uploaded
the
times.
What
I
have
is
short
about
10
minutes,
so
I
only
will
be
able
to
go
to
some
main
points
after
the
work.
If
anybody
has
an
interest
to
more
details,
I'd
careered
you
to
read
the
draft
and
you
after
that
contact
us
so
well
as
the
end
can
be,
can
become
very
complex.
H
H
H
We
have
two
papers,
the
the
first
one
deals
with
the
theoretical
part
of
this
work,
and
there
it's
set
of
theorems
that
prove
the
properties
were
claiming
they're,
the
subject
to
have,
and
the
second
one
deals
with
algorithm
or
representation
of
this
kind
of
graph
and
algorithms
to
to
work
with
them.
So
this
desire
the
full.
H
Papers
for
for
explaining
this,
this
object
and
what
we
are
really
proposing.
This
draft
is
a
special
case
of
this,
which
is
specifically
done
for
multi-layer
just
multi-layer.
So
it's
a
bit
different
representation
where
we
have
a
vertex
set,
an
edge
set
in
a
layer
set,
and
the
edges
in
this
case
will
be
represented
as
I
ordered
quadruples
as
and
you.
D
H
Initial
vertex
initial
layer,
destination,
vertex
and
a
destination
layer.
With
this,
we
are
able
to
maintain
a
property-
that's
basic
4/4
graph,
which
is
it's
it's
property
of
resembling
binary
relation
to
do
that.
We
see
this
edge
actually
as
a
pair
of
pairs,
and
then
the
binary
relation
is
maintained
between
pairs.
This
makes
that
his
objects
has
the
same
structure
of
abnormal
directed
graph.
H
Further
from
this,
we
can
infer
that
there
different
kinds
of
edges
we
can
have
interlayer
edges
in
which
objects
of
different
layers
are
being
connected,
but
actually
this
is
the
same
object
in
different
layers.
We
can
have
inter
layer
edges
which
connects
objects
in
the
same
way
here
in
represented
in
blue
and
finally
Mixit
edges,
which
can
connect
different
objects
in
different
layers.
Here
they
are
represented
in
green
and
we
want
to
represent
undirected
edges.
We
do
this
by
using
two
edges
pretty
much
as
we
do.
H
Undirected
graph
by
a
matrix
Y,
we
have
two
entrance
with
once
to
represent
two
edges,
so
it's
a
pretty
much
the
same
case
here
from
this.
We
can
also
see
that
we
can
split
the
edge
in
two
pairs
and
treat
each
pair
as
being
vertex
of
what
we
have.
What
will
be
a
directed
graph
so
from
this
we
can
have
the
exactly
the
structure
of
a
normal
directed
graph.
However,
the
additional
information
that
is
brought
by
the
multi-layer
part
is
directly
encoded,
either
in
the
vertices
or
in
the
edges.
So,
even
though.
I
H
H
H
Layer,
then,
from
the
one
data
layer
to
the
three
data
layer
and
so
on.
So
this
is
a
path
fully
built
in
the
same
layer.
It's
an
internal
air
path.
We
can
have
other
forms
of
paths
like
this
in
green,
which
connects
the
controller
one,
a
controller
layer
to
the
d2
at
data
layer,
and
so
this
is
a
mix
it
edge
and
also
a
mix
it
paths.
H
Both
these
paths
will
reappear
later
and
the
caller
so
have
to
remember
here
that
green
is
for
for
mix
at
paths
and
blue
is
for
inter
layer
we
can
also
have
notions
of
shortest
paths.
Actually,
with
these,
we
can
reconstruct
all
the
notions
that
are
already
known
from
the
from
the
graph
theory,
which
is
a
very
extensive
body
of
knowledge,
and
we
can
also
have
solution
for
some
problems
that
we
have
a
multi-layer
regarding
with
projections
of
the
paths.
H
So
here
we
have
the
matrix
representation
of
exactly.
This
is
amol
example
that
we
presented
initially,
and
we
can
see
that
this
has
a
special
block
structure.
This
can
be
achieved
that
we
call
these
the
the
canonical
graph
representation
in
adjacency
matrix.
So
here
the
main
block
diagonal
represents
only
the
intro
layer
nets
and
the
other
kinds
of
nets
are
in
the
off
diagonal
blocks.
We
can
see
this
here.
H
H
Normal
graph,
however,
in
the
block
that
goes
controller,
2
data
layer
and
the
other
one
in
the
block
that
goes
data
layer,
2
controller.
In
this
way,
we
can
do
the
the
representation
in
matrix
form
and
they
take
away
of
mate.
Messages
of
this
is
that
we
can
use
these
to
represent
as
the
n.
Why
pertaining
the
full
information
of
the
layered
structure?
H
H
B
H
A
A
A
And
and
I
get
you're
aware
of
tools
that
would
help
us
in
doing
so,
I
mean
just
just
a
bit
because
in
the
draft
making
some
reference,
so
you
can,
you
can
use
these
or
that
or
that
I
mean
would
be
important
for
us.
We
have
been
working
in
the
past
in
in
in
an
environment
that
you
could
use
a
capes
environment
to
develop
and
to
validate
networks.
Probably
trying
to
integrate
these
two
things
would
be
really
interesting.
Okay,
you.
H
Can
do
that
with
a
little
problem?
The
the
basic
algorithms
you
have
for
for
directed
graphs
will
work
in
this
environment.
So
only
if
you
want
some
expressional
kind
of
results
where
you
only
want
to
see
one
layer
or
you
only
want
to
see
edges,
then
you
have
to
do
composition,
actually
projection
of
the
edges
and.
D
H
A
D
A
H
H
H
H
H
A
No,
no,
not
necessarily
what
I
was
asking
you
think
there
is,
could
be
some
links
about
which
are
the
truths
for
those
of
us
who
are
familiar
with
graph
theory
is
about
to
say
you
can
use
X,
Y
or
Z
for
these
days,
and
that
well
suggests
in
some
hints
on
how
a
language
could
use
as
input
for
building
the
matrix
would
be
helpful
as
well.
Okay,
thanks.
D
Okay,
hello,
everyone:
this
is
Angela
Rodriguez
from
University
of
Calca
in
Colombia,
a
I'm
going
to
present
my
my
PhD
proposal.
No
draft
not
stick
just
create
a
proposal
which
title,
which
is
titled
framework
for
autonomic
management,
so
far
define
networks,
so
here's
the
outline
and
I
will
skip
through
introduction.
So
in
my
context,
area
is
also
software-defined
networks,
so
just
just
enough
race.
So
far,
different
networks
is
the
Copeland
control
task
from
the
forwarding
devices.
Okay,
however,
a
when
Sdn
was
first
stated,
the
management
was
not
taken
into
account
a
some
project.
D
Some
projects
I
just
tried
to
put
some
patch,
but
here
in
the
application
plane
to
perform,
manage
management
tasks
or
there's
a
disputed
in
the
in
the
control,
trained
act
for
kind
of
mixing
of
controlling
and
management
functions.
A
recently
some
proposals,
proposals
from
oh
and
fe2
and
has
a
stated,
a
vertical
management
plane.
So
why
this
necessary?
A
vertical
management
Elena's
began
well,
just
because
each
plane
of
the
Sdn
needs
to
be
managed
it.
For
example,
here
we
have
a
forwarding
devices
the
initial
setting
up
of
these
devices.
D
D
So
a
monitoring
the
controller's
status
in
any
abnormal
behavior
is
a
management
tasks
at
the
application
planes
plane
a
have
to
manage
also
they,
for
example,
their
sales
level
agreement
is
last.
A
contracts
of
the
stuff
has
to
be
managed
it.
So
that's
why
a
vertical
plane
it's
necessary.
I
well.
We
know
where
that
is
the
energy
has
of
NAFC,
where
they
stated
management
plane
at
the
same
level
of
the
control
plane,
okay,
but
they
in
in
this
average
RFC
management
plane
has
now
a
direct
communication
with
the
control
plane.
D
So
that's
why
we
decided
to
use
this
one.
On
the
other
hand,
I
have
the
autonomic
network
management
paradigm.
So
a
what
that's
for
I
am
a.
The
main
goal
is
to
reduce
human
intervention
during
a
management
operations.
So
how
can
this
be
achieved?
Well,
when
we
need
that
the
network,
the
system
organ?
This
way,
the
network
can
be
able
to
be
a
way
of
context,
change
and
in
adapt
to
those
change.
So
we
have
a
map
model
like
you
see
here.
D
All
of
these,
the
idea
is
that
at
systemic
achieved
self
is
okay.
All
of
this
I.
Almost
all
these
issues
are
currently
being
working
at
a
ITF
animal
working
group
when
the
key
motivation
I
put
on
board
a
open
flow
specification.
So
what's
that
a
we
have
a
control
it
that
defines
a
or
that
feels
they
forwarding
tables
that
I
finally
implemented
in
at
a
network
devices
based
on
open
flow.
D
So,
let's
suppose
that
we
have
a
or
Sdn
network
with
some
a
open
flow
based
devices
and,
let's
suppose
one
of
those
devices
have
failed
and
of
course,
some
of
the
forwarding
paths
will
be
no
longer
available.
So,
as
the
N
has
two
methods
to
solve
this
situation,
first
one
is
restoration.
How
does
it
works
so
in
restoration
they
control?
It
will
refill
all
the
forwarding
tables
of
the
devices
that
have
been
affected
according
to
administrator
instructions.
D
D
So
how
does
it
work
a
protection
is
that
they
control
it
at
the
moment
of
the
initial
setting
up
of
the
network
feels
not
only
one
forwarding
tables
parent
device,
but
another
one
that
has
a
bucket
pad
okay,
but
let's
suppose
that
the
net
world
has
suffered
a
big
damage
like
an
invasive
one,
that
has
a
cost
a
northern,
what
device
being
failed.
So
it's
possible
that
also
the
bucket
parts
don't
are
no
longer
available.
So
that's
a
problem
for
the
network,
the
network
operation,
so
a
state
that
is
definitely
needed.
D
A
management
system
that
provides
an
appropriate
solution
based
on
the
analysis
of
the
current
network
state
or
maybe
it
can
analyze.
Also,
it
also
passed
a
state
and
give
the
appropriate
solution
management
solution.
Each
situation
or
each
network
went
also
original
source.
He'll
have
stated
that
Sdn
and
enemy
shaved,
the
objective
that
you
can
read
it
in
it's
just
like
before,
since
other
cable
management,
all
self
management.
D
When
in
the
state
of
our,
we
have
a
list
of
quite
recent
works
that
have
worked
it
on
the
relation
between
as
the
nnnn
a
in
aim,
so
we
have
identified
identified
some
common
shortcomings
of
those
related
works.
All
of
them
ignore
recent
proposal
of
vertical
management
plane.
So
they
still
remind
that
holding
management
tasks
at
application
plane
that
control
thing.
They
don't
have
a
strong
emphasis
on
intelligent
algorithmic
components
and
up
to
us,
a
analysis
in
systems
need
strong
and
intelligent
components.
None
of
them
have
autonomy
city
at
low
level.
This
is
Sam.
D
This
is
being
excuse
me
and
my
colleagues
have
been
excused
whether
now
Sdn
devices
have
to
be
autonomic
with
discussing,
but
up
to
now
yet
consider
him
and
Willie
I'll
follow
increasing
prevalence
models
for
autonomic
network
management
that
for
us,
could
be
adapted
to
as
the
end
conditions
already
and
requirements.
We
have
genetic
economic,
Network,
a
detector
from
Ito
and
they
Fe
economic
future
internet
and
the
animal
reference
model
from
the
IDF
anima
working
group.
D
So
we
have
received
question
M
hypothesis,
as
you
can
read
how
to
efficiently
manage
and,
at
the
end,
with
vertical
management,
point
being
a
word
of
change
in
the
state
with
minute
human
intervention
under
which
so
the
hypothesis
is
like
your
decision
elements
in
and
at
the
end,
with
vertical
management,
plane
support
web
unintelligent
autonomic
control
groups
could
contribute
to
pay
for
itself
functionalities.
Achieving
efficient
management,
so
I
know
common
question
is
how
can
we
test
them
that
we
achieve
this?
D
We
still
don't
have
the
fall,
the
hole
and
sweat,
but
we
have
like
it
spected
contributions.
They.
We
are
filling
this
list,
so
maybe
reducing
configuration
time.
It
will
be
our
first
measure
that
we
have,
because
a
reconfiguration
is
one
of
the
most
a
common
management
test,
and
we
expect
to
achieve
this
didn't
ask
pudding.
Micro
looks
that
they
go
on
devices
and
normal
looks.
They
have
three
reaction.
They
go
in
the
management
plane.
D
D
Needed
information
for
the
whole
system,
like
the
policies
policies,
any
roles,
all
information
about
the
base
is
a
capability
so
that
stuff,
for
example,
we
can
have
a
set
of
plans,
action
plans,
all
that
stuff
can
be
here.
This
is
like
a
translation
service.
It's
a
traditional
translation
services,
rate
requests
and
responses.
Here
we
have
a
representation
model
can
be
junk.
For
example,
resource
representation
model
is
required
a
here.
We
have
the
network
elements,
notes
a
with
the
micro
loops
that
we
are
considering.
D
Well,
this
is
like
to
see
the
device
with
our
zoom
in
this
is
I
have
to
say
this
is
the
proposal
from
Ghana
detector
from
Ito
okay,
but
our
concern
is
how
to
achieve
this
layered
decision
elements.
What
decision
elements
in
an
open
flow
based
switch
for
the
someone?
That's
our
urine,
Carson
concern.
D
There's
a
theft
about
autonomic
control
groups,
but
it
works
more
focused
on
the
situation
amines,
so
I
don't
know
still
if
a
decision
element
is
a
component
of
the
loop
or
they
just
have
a
kind
of
relation
we
are
working
on
that
we
will
consider
an
OL.
Explain
that
sound
reasonable
has
stated
a
the
knowledge
train
is
to
to
have
the
information
model
and
the
data
model
that
could
be
junk
from
IDF
dam
formation
model.
D
We
know
a
song
so
about
if
both
are
important
or
just
the
data
model,
we
we
do
either
where
that
sets
exist,
an
error
see
from
IDF
a
an
old
one
that
talks
about
a
this
relation.
And,
finally,
we
want
to
provide
intelligent
capabilities
to
the
development
which
it's
been
kind
of
orchid
and
IDing
it
from
IDF
just
to
be
defined
a
policy
treatment.
If
we're
gonna
use
intense,
a
super
working
group
works
on
that.
Also,
how
will
be
the
communication
between
decision
elements
or
an
intergroup
communication
may
be
grasped.
We
will
use
grabs.
D
That
is
the
protocol
that
animal
is
working
on.
We
will
use
purity
of
events.
Reporting
is
also
an
important
part
of
the
look
we
did
looking
for
a
name.
This
is
the
day
for
this
presentation,
but
we're
looking
for
an
official
name
on
the
framework,
so
I
have
some
references.
Thank
you.
If
questions
please
slow,
spelling
thank.
B
You
questions
/
feedback.
I.
Have
one
comment
about
the
question
that
you
raised:
whether
a
network
node
would
be
autonomic
autonomous
or
not,
and
I
want
to
remember
that
the
definition
of
an
autonomous
element
may
not
be
an
advice.
It
could
be
something
side,
device
could
be,
the
device
itself
could
be.
A
bunch
of
those
devices
could
be
the
Sdn
network
itself.
So
it's.
D
A
The
what
are
the
individual
devices
individual
boxes,
autonomic
or
not,
is
something
that,
if
things
go
as
we
expect
them
to
go
in,
the
future
would
be
completing
material
I
mean
it
would
be
the
same,
because
the
devices
will
be
composed
and
their
services
will
run
will
will
overflow
devices
or
will
have
several
devices
in
several
services
inside
the
device,
so
it
will
become
a
different
I
mentioned
will
be
a
physical
dimension
that
would
be
not
relevant
for
the
services.
Ideally,
this
is.
A
Second
thing
is:
that
is
a
recommendation.
You
mention
gana
mana
man
is
not
because
we
are
in
the
ITF
use
any
map.
Just
you
know,
you're
not
gonna
compete
I
mean.
Is
it
maybe
you
something
that
it
was
overly
complicated
and
it
reflects
very
much
and
an
approach
that
is
sort
of
let's
say
old-fashioned.
I
would
say
this
from
the
old
times.
This
is
something
that
made
sense
some
time
ago.
A
A
F
This
is
Jefferson
table
just
to
mention
that
right
now,
I
think
that
that
your
work-
and
it
can
be
more
like
something
that
can
use
tools
from
aneema
because
of
the
the
current
charter
of
the
working
group,
it's
more
like
basically
infrastructure
stuff,
but
at
some
point
in
the
future,
maybe
I
can
be
more
open.
To
can
I
say
that
to
to
how
to
define
the
autonomic
way
to
perform
things
in
the
network,
or
something
like
this,
so
maybe
in
the
future,
your
work
can
be,
you
know,
can
can
can
fit
to
a
20-month.
F
B
Thanks
thanks,
we
need
some
additional
minutes.
Just
let
me
try
to
report
very
briefly
what
we
have
in
our
last
meeting
in
Lisbon
during
I
am
this
year.
In
that
meeting
we
were
discussing
about
topics
that
will
be
important
in
the
context
of
the
his
search
group,
and
one
of
them
was
try
to
come
up
with
some
say
clarifications
about
intent-based
networking
and
policy
based
network
management.
This
is
something
that
has
been
around.
B
C
C
At
the
same
time,
the
question
is:
what
is
intent
really,
and
particularly,
how
does
it
differ
actually
from
mother
efforts
before
it,
because
there's
also
the
policy
there's
also
the
term
service
models
and
the
vols
and
all
of
those
things
actually,
when
you
look
at
it
closer,
are
actually
fairly
closely
related
and
and
really
actually,
and
you
can
also
look
when
you
look
at
the
even
at
the
definition
of
intent
funds
in
the
things
that
basic
intent,
this
policy
so
basic.
That
is
the
case.
C
One
question
is,
certainly:
are
they
synonymous
or
to
really
mean
different
things
by
them,
and
it
would
be
good
actually
to
have
a
position
or
they'll
develop
a
position
on
that.
This
is
also
my
opinion
is
actually
that
the
way
they
are
used
today
is
is
pretty
much
interchangeably
and
in
in
many
contexts.
So
really,
basically,
it's
a
it's
a
newer
term
for
no
concept
at
the
same
time,
I
do
think.
Actually
there
is.
There
is
certainly
an
opportunity
in
having
and
having
different
terminology,
and
we
should
really
actually
work
out.
C
Basically,
the
distinction,
basically
between
between
them.
What
we
mean
with
one
foreign
policy
could
be
related
to
okay,
you
have
in
constant
just
the
the
event
condition
action
type
of
paradigm
that
goes
with
policy.
Basically,
they
indicate
that
versus
intent,
perhaps
being
something
high-level
declarative,
but
you
don't
know
actually
how
this
is
achieved.
So
buddy
policy
is
more
something
that
is
programmed
even
if
they
Clara
tively
versus
something
that
the
convergence
of,
but
actually
regardless,
actually
what
the
precise
distinction
is.
C
J
B
F
Jefferson
Wahby,
but
just
to
mention
that
the
drafting
in
a
NEMA
is
more
like
I,
don't
know
how
it's
say.
Something
like
intense
is
a
high-level
policy,
so
there's
nothing
more.
They
find
them
that,
and
also
probably
it's
just
it
just
takes
the
sideline.
It's
not
a
real
definition
of
intent
or
or
at
least
a
taxonomy
or
something.
This
will
be
helpful.
C
I
mean
my
opinion
is
basically
if
it
says,
if
the
definition
really
is
intent
is
a
policy,
then
we
should
actually
really
talk
about
policy.
I
mean
I,
don't
see
a
reason
to
talk
about
the
10,
so
if
intent
is,
there
is
something
that
distinguishes
intent.
That
is
something
that
we
that
would
be
worthwhile
actually
flashing,
flashing
out.
B
Well,
the
current
definition
for
examination
by
Jefferson
is
that
intent
is
a
special
case
of
a
policy
it's
up
because
it's
a
high
level
one.
So
if
we
can
say
that
I'm
talking
about
policies,
so
it
means
that
I'm
talking
about
every
kind
of
policy.
If
we
talk
about
intent,
then
we
are
talking
about
on
one
specific
kind
of
policy,
but
that's
not
clear.
In
that
sense,
there.
B
So
the
I
mentioned
before
we
have
this
discussion
started
in
I
am
in
Lisbon
the
audience.
There
was
much
smaller
than
this
one,
but
we
wanted
to
check
whether,
as
a
research
group,
we
agree
that
this
is
something
that
we
need
to
exploit
in
order
to
come
up
with
the
proper,
not
say
definitions
but
proper
clarifications
about
these
aspects
about
intend
policies
and
other
things
that
Alex
mentioned
too.
C
It
basically
came
like
I
think
what
one
historical
precedent
if
you
will
was
fought
since
the
the
confusion
in
the
past
concerning
data
models
and
information
models,
it's
sort
of
a
similar
type
of
thing,
but
was
it
in
the
early
2000s
something
then?
Basically,
the
draft
was
written
to
establish
basic
once
once
and
for
all
articulated
distinction,
and
something
like
that.
C
B
Maybe
we
can
organize
something
see
NSM,
but
probably
the
next
one
will
be
in
ITF
100.
So
until
there
we
may
use
the
mailing
list
to
collect
the
feedback
and
try
to
come
up
with
self
another
aspect,
but
I
think
Amon
is
not
on
remote
site.
But
another
aspect
was
that,
considering
that
we
have
a
more
clear
view
about
these
issues,
also,
they
imply
some
research
challenges
that
could
be
interesting
to
Belize
that
as
the
outcome
of
this
his
search
group,
so
that
we
can
say
to
people
that
look.