►
From YouTube: IETF102-TEAS-20180718-0930
Description
TEAS meeting session at IETF102
2018/07/18 0930
https://datatracker.ietf.org/meeting/102/proceedings/
A
Good
morning
welcome
to
T's
in
Montreal
I'm
Lou
burger.
This
is
the
bond
biram
chairs.
We
also
have
that,
partly
here
our
secretary,
who
usually
is
remote,
so
we
appreciate
them
being
in
the
room
for
those
who
are
willing.
Please
join
him
and
a
few
of
us
on
etherpad
and
help
out
with
the
joint
minute
taking
and
even
if
you
just
fix
a
comment
or
capture
one
comment
that
said
at
the
mic.
That
would
be
awesome.
We
don't
have
to
capture
everything.
That
said,
but
only
really,
the
gist
of
any
conversation.
A
That's
that
happens
as
well
as,
if
you
speak
at
the
mic,
it's
really
good
to
go
in
and
make
sure
that
your
name
is
accurately
reflected
in
the
notes
and
that's
the
the
URLs
right
here
on
the
bottom.
The
tools
that
IDF
door
working
group,
ease,
minutes
and
you'll
see
that
leave
from
there.
You
can
open
up
the
etherpad
102
we're
midweek,
so
everyone
should
be
familiar
now
with
our
how
we
operate
if
you're
not.
Please
take
a
look
at
the
note.
A
Well,
obviously,
this
is
going
to
go
by
pretty
quickly
so
there's
a
URL
on
the
bottom.
About
note
well
and
I'm
sure,
if
you,
google,
ITF
no
well
you'll,
find
this
message
and
basically
anything
that
you
say
here
in
session
is
part
of
becomes
part
of
our
record
and
that's
obviously
true
for
the
email
lists
as
well.
A
We're
screaming.
We
have
our
video
and
audio.
Please
use
the
microphones
state,
your
name
when
you
come
to
the
mic
and
I've
already
talked
about
helping
us
out
with
etherpad
I'm
on
Jabbar.
There
are
a
few
people
removed.
If
anyone
else
wants
to
jump
on
Jabbar,
that's
always
helpful,
and
for
those
of
you
who
are
remote,
please
feel
free
to
join
the
neat
echo
queue
when
it
comes
time
to
talk
or
talk
to
each
other.
That's
completely
up
to
you.
A
So
we've
had,
since
the
last
meeting
we've
had
three
RFC's
come
out.
That's
always
nice
to
see.
We
have
a
number
of
documents
in
the
editor
cute.
It's
a
good
number
I'm,
not
sure,
there's
anything
interesting
to
really
talk
about
there.
I
think
most
of
these
are
in
ref
weight
or
some
minor
things.
We
had
a
little
one,
but
one
document
delayed
a
little
bit
with
some
confusion
over
I
Anna.
But
that's
that's
all
sorted.
A
We
have
a
document,
that's
in
an
unusual
state,
or
at
least
for
us
it
would
this
ad
watch
list
on
hold
with
the
requirements.
There
was
some
discussion
about
that
I
think
everyone
should
have
seen
the
message
from
Deborah
where
she
liked
us
to
wait
and
see
how
the
other
documents
develop
and
then
decide
what
we
want
to
do
with
it.
B
Stepper,
it's
a
little
bit
different,
I'd
phrase
it
because
what
happened
was
the
framework.
I
think,
as
often
happens,
surpass
the
requirements
that
included
much
more
detailed
than
with
the
requirements.
So
what
what
we
had
said
on
the
with
authors
and
on
the
list
was
that
do
you
still
feel
need
to
publish
the
requirements,
and
so
the
on
hold
state
is
just
a
way
of
the
process
to
to
let
the
document
expire.
B
A
Some
of
us
are
a
little
dyslexic,
including
myself,
so
sometimes
that
happens,
although
I
can't
claim
this
one,
we
have
no
documents
in
working
group
last
call,
but
we
do
have
a
few
new
documents,
some
of
which
are
on
the
agenda,
actually
I.
Think
all
of
them
are
on
the
agenda.
So
we'll
keep
your
hearing
more
about
these
later.
A
A
A
See
camp
has
led
responses
to
sg-15
I'm,
not
sure
if
we
need
to
send
a
response
here,
even
though
ones
was
indicated
as
being
requested.
So
if
you
think
that
this
is
an
important
thing
to
respond
to,
please
send
a
message
to
this
list
or
send
a
message
to
C
camp
and
we'll
we'll
work
it
out.
But
right
now
we're
not
planning
on
taking
any
action
on
it
in
this
group.
A
C
Morning
before
we
get
to
the
working
of
document
status,
we
wanted
to
point
your
attention
to
some
work
that
we
are
planning
to
do
in
T's.
There
have
been
a
few
requests
in
recent
times
for
a
document
that
talks
about
IP,
MPLS
traffic
engineering
architecture,
landscape
with
an
IETF.
Basically,
the
ask
is
for
something
that
can
define
T
normatively,
we
usually
mean
in
the
past.
We
have
pointed
to
RFC
3272,
which
is
AP
principles
document.
This
is
about
16
years
old
or
in
recent
times
we
also
used
Adrienne's
RC
79
26.
C
That's
the
interface,
interconnected,
teen,
Network
stock
meant
as
a
useful
reference,
but
they
still
seem
to
be
some
gaps.
Some
polls
that
need
to
be
plugged
in
Lou
and
I
have
been
chatting.
We
are
still
looking
at
how
to
go
about
getting
this
work
done
if
anybody
is
interested
in
contributing
or
discussing
more
CEA's
offline,
and
they
can
take
it
from
there.
C
Next
up
is
working
group
document
status
of
the
24
working
group.
Documents
that
we
have
there
are.
There
are
eight
that
are
in
various
stages
of
iesg
processing.
We
have
11
documents
today
on
agenda.
That's
a
good
number,
so
that
leaves
us
with
these
file
and
I
will
walk
through
the
status
of
each
of
those
in
the
next
few.
Slides
first
up
is
the
a
CT
and
yang
document.
They
haven't
been
any
changes
to
this
since
early
this
month,
the
authors
have
promised
to
publish
an
update
fairly
soon.
C
C
D
E
F
A
So,
typically,
we
have
a
lot
of
different
ways
of
doing
archival
records.
Certainly
graphs
are
there,
but
if
we
want
an
archival
record,
that's
more
permanent.
We
usually
do
RFC's
if
they
want
to
do
the
ahrefs
to
fight
and
soothing
complete
this
work
and
move
it
through
the
process
and
the
working
group
doesn't
feel
it
objectionable.
Why
would
we
stop
that
so.
E
A
And
we
accepted
this
as
a
working
group
document,
so
we've
previously
pulled
it.
So
the
next
step
would
be
to
finalize
the
work
and
then
go
take
it
to
working
group
last
call
and
if
there's
not
support
to
take
it
through
at
that
point,
the
working
group
can
decide
to
kill
it
at
that
point
or
we
can
decide
that
their
support
to
move
it
forward.
No.
A
One
of
the
things
that
I
think
is
important
to
note
and
I'm
happy
for
this
to
go
back
to
the
iesg
is
that
we
often
talk
about
that.
We
don't
have
enough
participation
of
actual
users
and
one
of
the
way
the
users
contribute
to
our
process
is
by
writing
the
use
case,
and
it
I
don't
see
a
reason
why
we
should
devalue
that
work
relative
to
the
solutions
that
they
they
drive
us
towards.
B
B
Why
are
we
publishing
it?
Just
put
it
on
wiki
so
make
sure
the
document
it's
a
strong
document,
so
we
we
can.
We
can
get
it
through
and
make
sure
you
know.
Maybe
we
have
some
I
forget
whose
daughters
on
this
make
sure
have
some
operators
on
it
or
something
and
really
show.
This
is
not
something
that's
gonna
expire
like
in
five
years.
It's
a
useless
document.
D
Thanks
for
all
the
inputs,
I'll
take
it
up
with
daughters,
but
I
think
the
immediate
aim
was
just
to
make
sure
that
all
the
text
is
correct
and
updated
and
there
were
some
more
requests
for
a
few
more
use
cases.
So
my
immediate
goal
is
just
to
keep
the
document
up
to
date
and
then
we
can
again
start
a
thread
whether
what's
an
exception
it
or
should
we
not
I,
think
you're
still,
okay,
they
open
to
that,
but
I.
My
aim
is
to
make
the
document
up
to
date
and
correct.
Firstly,
yeah.
A
G
Know
Bruce
can
Chloe
I
agree
whether
in
actually
I
think
that
the
use
case
document
is
basically
helps
to
understand
requirements
and
requirements
have
to
set
up
frameworks
at
hand
by
the
time
we
have
solution.
Basically,
I
I
don't
see
much
help
in
use
case,
except
for
education
purposes.
Some
like
that,
and
in
many
cases
we'll
have
more
use
cases
and
something
is
has
changed
from
the
original
stuff,
so
so
basically
Mahad
storm.
In
my
opinion,
it
shouldn't
be
like
eventual
document,
something
that
is
published.
It
just
helps.
C
Bring
your
thoughts
and
we
go
to
the
world
no
class
code.
Thanks
next
is
a
tea
metric
recording
document.
There
was
one
revision
published
for
this
my
night
or
real
changes.
There
are,
however,
a
fair
number
of
outstanding
Commons
that
need
to
be
addressed.
The
authors
did
talk
to
us
offline.
They
have
promised
that
they
would
address
these
in
we
in
the
coming
weeks.
So
we'll
keep
an
eye
out
for
that.
C
A
If
you
think
these
documents
are
important
to
you,
now
would
be
a
good
time
to
read
it
yeah
it's
going
to
change
so
be
willing
to
put
in
the
time
to
do
a
review
now
and
when
it
comes
to
last
call,
but
there's
an
important
gate
of
going
to
the
the
yang
doctor
review,
which
will
make
sure
that,
from
a
yang
perspective,
it's
alright,
but
from
a
functionality
standpoint.
The
authors
think
this
is
really
close.
So
keep
planning
to
use
this
now's
a
good
time
to
look
at
it
and
comment
on
on
it.
It.
B
I
just
was
thinking
on
back
on
is
G
statement.
One
aspect
of
not
publishing
these
cases
requirement
documents
was
that
people
saw
it
as
delaying
protocol
solution
work
and
if,
if
the
protocol
solution
is
already
published,
this
is
still
coming
through
as
long
as
it's
aligned.
You
know,
that's
fine,
but
I.
Think
one
thing
from
the
isg
statement.
What
I
saw
is
the
most
important
line
is:
don't
let
it
delay
solution,
work,
we've
seen
a
lot
of
groups,
they
sit
on
these
documents
forever
and
don't
even
begin.
B
C
H
H
H
H
So
the
first
update
that
the
model
was
to
incorporate
the
association
between
LSPs.
This
is
done
in
ITF
by
using
an
association
object.
We've
added
support
for
identifying
those
different
Association
types
as
identities,
and
this
makes
it
extensible
to
model
new
Association
types
that
are
IETF
will
define
the
future.
H
H
The
base
has
a
type
ID
source
and
a
global
source,
as
defined
in
RFC,
and
the
extended
has
an
additional
extended
ID
we've
identified
a
Yusuke
couple
use
cases
for
using
those
an
example
I'm
putting
up
here
is
to
associate
Alice
peas
for
resources,
primary
and
standby
Association
and
computing
diverse,
Alice
peas
that
originate
from
different
ingresses
I'm.
Sorry,
if
the
font
is
not
clear
but
I
tried
to
make
it
legible
buddy.
H
The
next
update
was
the
the
explicit
route
list
that
we've
defined.
We
combined
and
numbered
the
numbered
unnumbered
entry,
and
we
made
it
extensible
to
cover
the
case
where
the,
where
the
entry
is
a
node
ID
or
a
link
ID
so
and
they're.
In
the
numbered
case,
it
can
be
a
numbered
node,
ID
or
a
numbered
link
ID.
In
the
unnumbered
case,
it
can
be
a
combination
of
a
node
ID
and
a
link
ID,
so
we
made
it
more.
Generic
to
cover
all
the
cases.
H
H
So
these
were
the
updates
or
additional
features
that
we
we've
covered
in
the
latest
revision.
We
wanted
to
raise
one
point
about
the
model
we
have,
a
module
which
is
encompassing
all
the
types,
traffic,
engineering,
types
and
other
other
mod.
Other
models
that
are
in
different
drafts
have
a
dependency
on
this
module.
They
imported
and
use
groupings
from
it
and
there
is
a
specifically
one
document
that
is
being
blocked,
because
this
draft,
which
includes
this
module,
is
still
not
advanced
to
that
state.
H
H
A
Actually
have
an
RFC
editor
process
question
for
Deborah.
If
we
just
extract
that
the
types
module
into
a
new
document
with
a
new
name,
can
we
change
the
name
of
the
document?
That's
in
ref!
Wait
because
it's
really
just
waiting
for
that
that
piece,
because
if
we
can,
if
we
can,
it
becomes
really
easy
just
to
extract
that
into
a
new
document,
wrap
it
up
and
push
it
out.
If
we
can't,
then
this
we
have
to
preserve
the
the
name,
the
file
name,
that's
a
reference,
and
it
comes
a
little
more.
A
H
Next
slide
will
show
in
more
details
what
we're
proposing
so
write
the
draft,
as
is
as
it
stands
now.
It
covers
the
the
green,
yellow
and
red
modules.
What
we're
proposing
is
to
take
the
red
module
and
put
it
on
its
own
in
a
document,
and
this
is
what
is
blocking
other
graphs
from
progressing
right
now.
The
green
modules
will
still
advance
in
the
define
and
document
but,
as
though
I
think
is
mentioning,
we
have
to
connect
these
documents
so
that
we
have
the
history
how
it
progressed.
That's.
A
Important
because
you
can
do
linkage,
but
we
can
do
linkages
that
we
need
to
replace
by,
but
we
can
do
so
make
interesting
link
with
linkages,
but
it's
more
of
the
document.
That's
in
the
RFC
editor
queue
since
it's
in
the
ref
weight
state,
it's
referencing
a
particular
basically
file
name
right
draft
name,
oh,
and
what
I
think
this
is
proposing
is
is
a
new
main.
Oh
and.
H
H
A
H
I
A
I
A
A
A
H
On
terms
of
next
steps,
we
think
once
the
split
is
done,
that
drafts
will
be
ready
for
young
doctors
of
you
and
a
working
group
last
called
the
last
draft
here.
We'll
need
will
require
one
more
round
of
review
and
to
be
ready
for
further
reviews
from
the
yank
doctors
and
asking
the
working
group
for
last
call
that
was
it
any
questions.
Welcome.
D
D
H
H
H
D
Got
it
thank
you,
and
also
I
could
not
really
understand
why.
What
was
the
need
to
define
two
different
objects?
I
understand
that
in
our
in
our
implementation
of
RSVP,
we
have
two
different
objects,
but
in
the
yang
model,
is
it
really
needed
to
differentiate
it?
So
much
can't
me
just
make
extended
ID
and
global
source
optional
and
keep
a
single
object
so.
F
A
I
also
figured
that
the
yang
doctors
will
catch
you
on
that
and
they'll
be
good.
They'll
figure
they'll
help
you
figure
out
the
best
way
to
encode
that
there's
a
couple
of
other
places
where
you
we
have
options
and
they're
now
encoded
completely
separately,
and
it
doesn't
seem
that
that's
the
most
efficient.
But
that's
why
we
have
yang
doctor
reviews
is
the
context.
A
G
G
So
I
will
not
introduce
it
again.
I
just
remind
that.
This
is
a
document
that
is
meant
to
be
sort
of
Wikipedia
for
everything
we
are
doing
in
terms
of
tano
and
apology,
modeling
and
and
all
extended
definitions,
figures,
modeling
code,
subs
use
cases.
All
those
things
basically
are
meant
to
go.
There.
G
K
G
Itf
this
document
was
adopted
as
a
working
group
document
and
it
was
basically
progressed,
but
not
as
much
as
I
would
want
to
see
progressing.
So
we
had
a
lot
of
pre
published
republished
movements,
basically
for
the
IT
topology
model,
and
many
of
them
were
something
like
that.
You
know
it
would
be
nice
to
have
like
more
text
describing
this
concept
or
des
concept,
and
we
would
like
to
see
it
in
the
document
and
our
standard
answer
was
go
and
look
into
this
model
in
tutorial
and
people
like
this
approach.
G
G
So
we
all
know
that
basically
things
like
link
tunnel
and
connections.
These
words
do
not
have
intrinsic
definition.
It
just
labels
to
concept,
and
it
takes
a
lot
of
standardization
effort
to
basically
to
get
this
concept
so
right,
aligned
and
to
assign
a
right
labels
to
this
concept.
And
if
you
do
it
right,
then
the
discussions
and
move
very,
very
fast.
G
As
we
know,
conversely,
if
you
defined
differently,
if
you
assign
different
names
to
the
same
things
or
cool
with
the
same
name,
different
or
even
worse,
overlapping
things,
then
it
takes
a
lot
of
time
to
move
anywhere
right.
So
I
think
we
did
a
reasonably
good
job
with
a
cover
interpolation
model,
but
there
is
a
lot
of
work
that
still
has
to
be
done
with
a
ton
of
modeling
okay.
So
since
last
right
here,
we
provided
updates,
for
example,
and
description
as
to
how
do
you
provision
multi-domain
tunnels?
What
does
it
mean
Tunnel
segment?
G
What
does
it
mean
connection
segment?
How
do
you
configure
how
to
manipulate
this
configuration
so
forth?
That's
the
it
alone
who
provided
the
text
so
basically
in
the
future,
we
would
like
to
ask
people
to
point
out
good
things
in
the
models
are
not
covered
with
the
description
and
even
better
to
provide
some
text
so
that
we
can
review
and
put
it
into
into
this
document.
Again,
it's
very
important.
We
put
a
lot
of
effort,
developing
these
models
and
the
difference
between
publishes,
cirrhosis
or
a
publicist
and
what
was
deployed
in
the
operator
networks.
G
But
this
could
be
the
quality
of
this
document.
So,
for
example,
recently
I
have
a
question
from
the
client
who
said:
can
I
do
things
like
that?
I
want
a
townhome.
That
would
be
the
best
compromise
between
the
course
and
the
delay
and
I
said
sure.
Here's
what
you
do
you
you
go
and
basically
configure
your
optimization
criteria.
Society
stuff
like
that.
Now
it's
very
clever,
but
what
about
where
it's
described
so
and
basically
I
cannot
answer
anything,
because
this
is
exactly
the
place
where
things
like
that
needs
to
be
described.
L
G
No,
it's
a
well.
We
are
talking
about
a
single
tunnel
which
goes
through
multiple
domains.
Okay,
so
basically
it
starts.
It
could
start
on
the
client
and
then
go
through
several
provider
domains
and
and
yeah
on
another
client
device.
So
basically,
if
you
have
a
super
controller
which
basically
oversee
the
entire
network,
then
it
takes
a
request
as
a
tunnel,
but
when
the
super
control
is
a
client
of
the
domain
controllers,
then
this
application
needs
to
split
their
configuration
and
to
instruct
each
domain
controller
as
to
what
to
do
okay.
G
Depends
so,
for
example,
the
tunnel
as
we
defined,
starts
on
the
tunnel
termination
points,
and
this
is
where
the
tunnel
is
used
right
way.
It
adopts
the
client
traffic
but,
for
example,
transit
domain.
It
doesn't
have
any
termination
points.
It
just
has
access
links
and
labels
assigned
resources
allocated
access
leaks,
so
it's
a
big
difference
right
and
that
could
be
very
a
lot
of
different
cases
where,
for
example,
you
have
a
segment
for
working
connection
but
end-to-end
connection
for
protection.
So
all
these
things
are
coward,
so.
L
G
G
G
M
G
L
G
You
say
that
okay,
but
our
clients,
a
different
things
right
there
and
I,
can
tell
you
how
many
time
we
spent
explaining
how
to
provision
the
main
segments.
For
example,
the
domain
could
just
carry
protection
connection
I'm
not
working
connection
right.
So
how
can
I
instruct
domain
controller
to
just
provision
protection
connection
or
protect
a
segment
of
a
production
connection?
I.
A
Think
you
actually
provided
the
answer
to
your
own
question.
Is
that
there's
different
ways
to
do
it,
and
this
is
just
an
informational
document
talking
about
one
way
to
do
it
so
for
those
that
want
to
look
at
it?
This
way
I
want
to
understand.
This
is
helpful.
There
now
the
case.
The
other
case
that
you're
talking
about
that
should
be
covered
in
here
too.
If.
L
A
A
A
G
So
basically,
his
question
not
about
this
particular
document
per
se.
It's
basically
how
we're
doing
tea
tunnel
model
and
the
the
confusion
is
the
level
at
which
you
control
tunnels
right
so
so
there
could
be
at
the
client
level
where
the
client
doesn't
care,
whether
it's
a
single
domain
or
not
domain,
and
there
is
a
here,
a
controller
which
basically
cares
a
lot
about
provision
in
a
particular
pieces
of
this
tunnel
right
and
the
point
is
that
we
use
the
very
same
model
to
do
the
both
things
I.
A
Think
the
answer
of
there's
different
ways
to
do
it,
and
then
vendors
choose
to
implement
differently
and
and
carriers
choose
to
deploy
differently
and
covering
both
is
a
worthwhile
thing.
I
mean
we
already
have
the
definitions
that
do
is
support
both.
So
why
not?
We
have
the
description
that
supports
it.
Oh
yeah.
G
The
point
is
that
we
do
not
describe
it
well
enough,
and
this
is
actually
question
for
this
document,
and
this
is
our
update
for
this
particular
idea.
Okay,
but
the
similar
things,
it
was
just
an
example
similar
things.
There
is
a
lot
of
other
things
that
needs
to
be
described
as
tediously
as
to
this
one.
A
Before
we
leave
this,
one
I
had
a
question
I'm,
not
quite
sure,
if
it's
better
for
the
model
documents
that
Turk
discussed
or
for
this
one
one
of
the
things
that
happened,
that
towards
the
end
of
the
tea
topology
is
sort
of
because
of
its
complexity.
We
added
a
section
that
described
how
it
should
be
augmented
how
it
should
be
added
to
by
for
technology,
specific
extensions
or
other
uses
in
the
future,
and
we've
already
heard
that
that's
that's,
being
helpful
and
being
used
in
other
working
groups.
A
G
I
would
say
that
I
actually
erase
many
times
while
uu
is
not
asking
for
the
same
thing
that
he
has
asked
for
a
teen
topology
model
and
he'll
immediately.
Ask
for
that.
But
the
other
thing
is
he's
the
only
one
who
raised
this
issue
because
in
theory
any
model
could
be
augmented,
and
this
means
that
anything
that
we
produce
has
to
have
this
session.
Basically,
how
do
you
mend
this
and
how
to
use
this?
How
do
you
actually
build
admin
tations
based
on
this
model?
G
H
F
H
H
A
Unlike
other
groups,
you
know
I,
don't
think
I've
come
across
a
module,
that's
as
complex
as
teeth
apology,
for
example.
So
the
fact
that
we
have
to
have
this
type
of
guidance
and
other
modules,
don't
you
know
I'm
completely?
Okay
with
and
that's
with
any
hat,
including
net
bought
chair
by
the
way
which
document
we
do
it
in
I.
Think
you
know
it's
it's
up
to
the
author's.
F
G
Just
so,
basically
to
point
out
over
half
sections
describe,
how
do
you
do
obstruction
of
topology
modal
right?
How
do
you
do
topology
negotiation
between
client
that
so
you
know
I?
So,
basically,
how
do
you
do
advantageous
falls
under
the
same
category
as
as
this
kind
of
so
things
right?
So
we
need
a
section
that
says:
okay,
this
has
our
proposed
guidance.
How
do
the
augmentation
technology,
specific
or
domain-specific
or
layers
or
whatever
right
yeah.
A
F
J
First
credit:
we
would
thank
a
lot
people
like
Carlos,
Naraku
and
Francesco
Azeri
from
erikson
for
their
great
contribution,
review
and
also
proposal
for
enhancing
in
the
young
model
and
all
the
people
that
weekly
with
the
T
eternal
try
to
solve
the
common
issue
between
path,
computation
and
tunnel
model,
Tarek
Igor
su
phangan,
a
van
and
the
other
that
is
attending
with
a
weekly
cool,
ok,
summary
of
the
changes.
We
just
need
just
a
few
modification
about
the
text
because
we
made
a
greater
restructuring
that
was
triggered
by
last
ITF
by
Michael
Schaff
comments.
J
We
we
added
the
possibility
to
have
a
chest
level
in
case
of
multi-domain
scenario,
in
order
to
be
able,
for
the
domain
controller,
to
to
understand
the
labels
for
the
previous
and
following
domain,
we
have
date.
The
the
model
in
alignment
with
the
Etana
will
be
directional
bulan
for
the
tunnel
to
indicate
whether
the
part
is
a
unidirectional
or
bi-directional.
J
Then
there
are
a
couple,
some
plum
bugs
that
we
correct
most
also
for
for
alignment
with
the
use
of
TA
topology
identifier
from
TA
types.
We
use
generic
path
disjoint
as
grouping
both
again
for
TA
types
model,
and
we
important
to
remove
the
container.
That
previously
was
wrong
in
in
pot
computation,
because
what
competition
is
stateless,
our
PC
and
we
gathered
a.
They
are
a
container
providing
and
storage
information.
That
is
not
the
case
for
a
stateless
path,
computation.
J
We
in
the
github
you
can
see
in
the
31
issue
that
we
asked
to
diminish
some
t,
eternal
attributes
that
are
not
needed
in
our
view,
for
path
computation
line
like
a
name
of
the
tunnel
identifier,
the
tunnel
description,
provisioning
state
other
other
attributes.
That,
in
our
mind
it
is
not
an
easy
for
path
computation.
So
a
proposal
is
was
not
to
use
in
part
computational
PC.
J
These
are
tables,
but
it
was
rise
during
our
discussion
weekly
comment
about
the
fact
that
you
should
have
a
policy
based
path,
computation
that,
let
me
say,
make
important
to
live
again.
These
attributes,
because
linked
to
specific
way
to
make
the
path
computation.
The
policy
is
not
specified
yet
and
well.
J
G
G
Approach
of
dropping
configuration
parameters
from
the
computation
request,
such
as
tunnel
name,
for
example,
okay.
So
the
logic
is
that
when
you
request
a
path,
so
you
assume
that
you
will,
when
you
specify
the
same
constraints
for
the
tunnel,
the
tunnel
would
take
the
path
that
written
by
the
path
computation
right.
But,
for
example,
we
see
a
lot
of
use
cases
where
local,
so
the
sewer
basically
takes
the
configuration
parameters
and
applies
our
local
policies
based
on
the
configuration
parameters
to
put
more
constraints,
relax,
others
and
suffocate.
For
example,
there
could
be
a
policy.
G
They
said
that
if
you
all
done
also
that
go,
that
starts
with
the
name.
Blah
blah
blah
basically
has
to
use,
should
go
always
through
these
links
or
for
these
affinities
of
stuff,
like
that.
So
now,
if
we
drop
this
from
from
the
path
computation
request,
it
could
be
the
case
that
path
computation
will
return
one
path,
but
the
tunnel
will
take
totally
different
paths.
G
Therefore,
therefore,
what
we
are
saying
is
that
basically,
we
do
not
lose
anything
from
keeping
us
a
tunnel
name
in
the
in
the
request,
but
we
allow
for
the
server
to
do
exactly
the
same
thing
for
the
computation
for
the
tunnel
versus
basically
when
when
it
actually
sets
up
the
tunnel.
Okay,
in
other
words,
we
do
not
see
a
path
computation
request
as
it
is.
G
N
Dear
Pinocchio
I
actually
don't
understand
why
we
need
specific
parameters
to
actually
support
policy
based
or
constraint,
based
path
computation
without
the
modeling
it
explicitly
so
I
would
prefer
to
use
an
approach
where
policy
is
actually
clearly
defined
or
policy
rules
are
defined
and
then,
of
course,
you
can,
when
you
send
a
path.
Computation
requests
include
an
identifier
that
triggers
the
servitor
instructs
the
server
to
you,
apply
specific
policy
rule
I
think
using
input
parameters
that
do
not
obviously
indicate
that
this
is
related
to
policies.
It's
not
a
good
approach.
Mapping.
M
You
say
I
think
you
got
the
problem.
I
see
here
is
what
this
type
of
policy
didn't,
because
the
user
may
not
know
that
there
is
this
policy
and
even
if
we
add
the
dis
attitudes
there,
the
user
doesn't
provide,
and
you
get
exactly
the
same
problem
that
you
say,
because
even
if
the
part
composition,
PC
has
the
attribute
and
the
user
doesn't
provide
the
asset,
but
assuming
that
that
is
not
needed,
then
it
gets
exactly
force
response.
So
that's
why
we
have
to
have
a
very
clean
design
on
the
part.
M
A
G
G
Go
Chrisman
clothing.
So
this
exactly
the
point.
The
client
is
not
supposed
to
know
the
local
policies
that
set
up
in
the
operator
so
and
basically,
when
the
client
says
a
putana,
he
doesn't
assume
that
their
policy
in
place.
Okay,
the
only
purpose
here
is
to
get
exact,
same
paths
for
the
tunnels
and
for
the
path
computation
request
and
that's
all
you
can
achieve
and
Sarah
can
achieve
that
if
he
gets
all
the
information
put
the
tunnel.
G
D
Just
to
support
it
goes
point
from
a
different
point
of
view
in
piece
F
as
well,
when
you
can
just
simply
ask
for
a
path
computation
request,
or
you
can
also
include
the
LSP
object
and
when
you
include
the
listview
object,
you
kind
of
identify
that
this
path
computation
is
a
for
a
particular
LSP
or
particular
tunnel.
So
I
kind
of
see
that.
J
A
A
Would
like
to
go
with
the
recommendation
of
the
authors
and
or
and
who
would
like
to
include
these
parameters
as
as
Igor
said,
as
optional
parameter,
so
two
choices
so
first,
who
would
like
to
go
with
the
recommendation
or
thinks
the
recommendation
of
the
authors
is
correct,
so
a
few
people?
How
many
think
that
we
we
would
be.
It
would
be
better
to
include
these
as
optional
parameters.
A
I
would
say
it's
it's
it's.
You
know
relatively
more
percentage,
wise,
its
percentage
wise.
It's
it's
actually
a
big
difference,
but
neither
of
them
had
like
overwhelming
numbers.
So
it's
you
know
it's
a
small
sample,
so
I
think
the
indication
from
this
room
is
to
keep
it
in
I
good
to
bring
it
raise
it
again
on
the
Working
Group
list
to
get
consensus,
see
if
we
can
get
consensus
there.
But
if
we
go
in
on
this
room
we
would
we
would
not
go
with
this.
That's
my
read
you
on
you
agree.
C
J
A
That's
a
good
thing
to
take
to
the
list.
I'll
remind
you
as
at
the
editors:
it's
not
actually
the
authors,
but
an
editor
of
a
working
group
document
is
responsible
for
reflecting
the
consensus
of
the
working
group,
even
if
it's
not
their
own,
so
right
now,
it's
we
have
an
indication
of
a
directions.
Let's
take
the
list,
try
to
confirm
through
discussion,
both
sides
and
as
a
contributor
you're,
welcome
to
argue
your
point
and
see
if
you
can
convince
people,
no.
M
I,
let's
take
you
is
what's
important:
those
who
wants
to
disable
to
clarify,
because
what
is
carrying
me
if
they
are
optional
and
the
user
may
make
mistakes
in
using
those
attributes.
So
we
need
to
provide
a
good
guidelines
to
avoid
the
mistakes
in
the
usage,
so
I'm,
okay
to
go
and
do
the
working
group
consensus,
but
we
need
to
have
a
very
clear
description
on
how
they
are
used.
Yeah
I'm.
J
Just
a
complete
okay:
this
is
them
the
status
of
the
open
issue,
but
if
I
have
to
shrink
like
so
the
status
is
reported
in
the
gig,
our
github
and
actually
we
have
a
closed
at
10:00
open
issue.
And
then
we
have
a
still
ain't
open
issue
for
a
that
is
specific
for
our
PC
three
are
in
common
with
the
eternal
we
discuss
with
them,
and
one
is
related
to
the
text
for
the
new
appendix
on
the
draft
we
need
to
I
wrote.
J
It
has
been
requested
that
were
up
but
computation
example,
and
so
we
need
to
update
and
that
for
next
step
we
surely
we
need
to
resolve
the
current
of
an
issue
and
and
then
also
for
path.
Computation
is
open
the
door,
the
guideline
for
technology,
specific
element
ation
in
particularly
for
sync
up
with
the
actually
for
dota
turn
and
model.
J
P
P
So
the
principle
here
that
we
want
to
reuse
and
that
we
want
to
extend
the
existing
te
topologies,
so
layer
straight
to
poetry.
We
have
nursery
topology
already
defined
and
we
also
have
a
key
value
defined
here.
We
are
trying
to
link
them
together,
so
we
have
a
little
teeny
topology,
which
is
a
extension
to
the
layer
tree
topology
model.
We
have
all
commentation
there
and
our
contention
contains
that
all
the
leaf
drives
referenced
with
the
teapot
remodel.
P
So
this
is
our
strategy,
and
this
recap
the
changes
since
revision,
the
director
was
adopted
and
the
working
document
in
March.
Since
then
we
have
done
some
updates
along
with
the
latest.
The
dependencies
draft,
revised
document,
wise
and
I
did
there's
some
description
to
the
model
structure
and
also
updated
the
Security
section
so
here
and
all
the
dependencies
that
we
have.
P
We
updated
with
the
Yangtze
and
MPA
and
we
have
to
RFC
our
publisher
for
the
taste
for
the
model
and
the
nearest
authority
model
and
the
team
tunnels
and
the
tea
party
model
has
been
updated
significantly
since
last
time,
so
we
also
sync
with
them.
So
here
is
the
way
a
total
referencing,
so
the
King
to
pardon
is
ready
to
party
model
may
or
may
not
be
congruent
with
the
three
topology.
So
so
we
have
this
reference
in
case
the
left
side
that
Kyoto
project
one
and
on
the
right
side,
the
1982
Padre.
P
If
they
have
too
much
conquer
the
mantle
and
maintain
or
non-current.
So
we
can
have
the
two-party
reference
from
one
to
two.
Then
we
don't
need
to
respect
afire
anything
repeatedly.
So
he
have
differences
between
the
we
need
to
have
a
new
reference.
Ii
then
we
can
have
Lissa
are
3-3,
Prime
and
similar
things
to
the
termination
points
and
the
link.
So
we
have
all
the
reference
in
here
another
piece
of
the
document.
P
We
describe
the
technology,
specific
extensions
to
the
te
link,
so
in
in
this
case,
or
the
PAC
packet
specific
extensions
for
the
technology.
Specific
intention
we
have
extension
to
the
performance
metric.
This
topic
still
be
discussed
in
the
eternal,
so
once
that
final
finalized,
we
will
finalize
this.
Also,
we
update
the
document
for
the
Security
section
and
there
are
few
other
edits.
We
need
to
complete
socially
as
a
list.
We
also
need
to
have
dr.
reviewers.
Now
we
hope
has
more
comments
from
the
group
and
that's
for
this
document.
P
P
So
if
we
wanted
to
the
second
watching
te
land,
we
can
have
multiple
inheritance
and
the
changes
since
last
time,
with
the
similar
things
to
the
equity
that,
with
the
T
little
project,
layers
ready
to
party
I
did
whatever
synchronization
with
the
latest
dependencies
and
we
covered
two
more
missing
features:
bands
the
link,
protection
under
the
boundary.
So
here
are
some
similar
updates
for
the
dependency
documents
we
described,
how
the
way
argument
that
the
eyes
are
parameters.
It
was
a
topology
wise.
P
P
Some
extensions
to
the
eyes
are
specific
attributes
and
missing.
We
used
to
miss
a
flag
turn
indicator
for
the
link
protection
for
eyes.
Our
progress
so
put
a
tag
here
for
the
boundary
is
sensitive
to
party
model
already
specified
the
component,
so
they
can
do
the
component
link
and
other
patterning
techniques
here
so
with
trying
to
really
use
that
and
maybe
didn't
hide
anything
specific
SR.
So
that's
the
current
approach
we
are
taking.
P
F
A
A
Q
Q
Q
Q
F
P
A
A
Q
P
R
A
R
R
We
have
base
theater
policy
and
we
basically
in
though
in
the
solutions
we
have
SF
resources,
how
they
are
connected
to
the
node
through
LTP
and
TTP's,
some
of
a
safe
functions,
internal
to
the
node
and
others
are
external,
so
those
things
are
captured
here
and
so,
based
on
the
topology
available
to
the
controller
controller,
can
set
up
T
paths
with
self
constraints
together.
So
it's
kind
of
joint
optimization
to
give
the
best
combinations
of
sf'
resource
with
RT
connectivity,
so
I
think
we
can
solicit
more
use
cases
if
you
are
motivated.
R
You
can
do
so,
there's
a
room
for
more
use
cases,
but
our
aim
is
to
stabilize
this
graft
to
be
ready
for
the
last
working
group
call
and
in
case
this
doesn't
goes
through,
but
this
might
be
a
good
information
to
store
somewhere,
maybe
in
the
solutions
as
an
appendix.
If
that's
the
way
to
go,
but
we're
gonna
press
on
to
mature
this
to
be
ready
for
the
last
call
in
the
next
to
revision.
Okay,
any
questions:
okay,
okay!
Now
this
is
the
solutions
draft
based
on
use
cases.
A
R
R
Okay,
good,
so
the
main
gist
of
the
young
model
for
a
self
enable
topology
is.
We
basically
describe
our
connectivity
matrices
with
additional
as
self
information.
There
are
3
different
types
of
connectivity,
matrices
that
we
need.
We
call
SF,
SF
connectivity
matrix.
Basically,
SF
resources
are
locally
available
or
interconnected
within
the
node,
and
then
we
cannot
describe
through
a
connection
point
how
those
resources
are
connected
using
connectivity,
matrix
construct
and
second
time
is
sf2
LTP
Sen.
R
In
this
case,
resources
are
under
neighbouring
T
node,
so
we
use
LTP
construct
to
associate
T
topology
with
this
external
node
constructs
that
is
originating
outside
of
T
node
and
another
one
is
as
self
to
TTP
connectivity.
Matrix
TPP
is
an
adaptation
point
from
client
to
server.
In
some
cases
the
resources
are
connected
through
or
TTP's.
So
we
basically
outline
these
three
types
of
extended
connectivity
matrices.
So
this
is
the
example
of
how
SF's
are
related
with
T
topology
elements,
as
you
see
here
yeah.
So
this
is
basic
T
node
construct.
R
We
have
LTPS
and
TTP's,
and
in
this
case
ssl
resources
are
reside
in
the
node
or
through
connection
point
to
construct.
We
have
a
bunch
of
connection
points
and
then
in
this
example,
we
have
three
as
wave
functions.
They
are
interconnected
through
cities
and
also
virtual
link.
You
know
construct
which
is
defined
by
NFB.
This
is
actually
modeling
recommended
by
NFB,
so
we
basically
put
it
together
with
T
construct
in
ITF
model.
R
So
this
example
that
we
added
example
of
SF
to
LTP
or
connectivity
matrix
in
this
case
we
try
to
you
know
give
is
an
example.
How
compute
node
can
be
constructed
using
T
topology
using
SF
to
LTPS
are
connected
in
matrices.
So
this?
Basically,
if
you
have
this
topology,
then
you
have
integrated
cross
rerum
resource
model
network,
closed
Ellison
or
computer
storage
together,
so
that
in
case
you
have
requests
that
make
advantage
of
that
needs
both
the
connectivity
and,
for
instance,
virtual
network
mobility.
R
G
G
Asked
you
and
Deborah
okay,
so
I
I
want
to
briefly
describe
the
challenges.
So
basically
the
idea
was
that
we
don't
want
to
overstep
the
tools
of
it.
C
and
other
S
duos
that
basically
try
to
describe
the
service
functions
defined
to
the
service
function,
types
or
just
a
self-aware
okay.
So
that's
basically
we're
hoping
that
we
identify
that
as
service
functions
by
global
IDs
and
the
curious
client
can
look
up
their
other
data
stores
like,
for
example,
describe
interest
car
by
Etsy
right
and
defined.
G
What
is
it
that
this
service
function
is
doing
and
how
close
in
functionality
it
is
to
the
other
service
function
that
we
want
to?
Actually,
you
know,
replace
or
who
puts
basically
to
direct
our
tunnels
in
the
different
direction,
but
achieved
the
same
actually
processing
on
their
packets
and
on
the
traffic?
So
now
what
we
try
to
we
start
doing,
is
we
start
to
actually
doing
exactly
that?
G
Overstepping
toes
we
defined
like
general
resources
like
in
terms
of
CPUs
and
memories,
but
we
are
still
not
going
into
the
details
of
what
exactly
this
function
is
doing
and
so
forth.
Okay,
now
we
have
two
choices.
Other
way
wait
until
the
work
is
done
in
it's
here,
which
seems
to
go
forever,
okay
or
we
can
start
basically
repeating
what
they're
doing
and
then
thrown
away
as
soon
as
we
notice
that
this
basically
is
not
needed
and
the
work
is
done
sufficiently.
They
exit.
G
G
F
E
F
F
G
R
F
E
A
A
Technical
approaches
we
have
to
sort
of
figure
out
where
the
right
group
is
that
right,
right,
right,
expertise
is,
but
it
I
don't
think
it's
this
group
and
going
back
to
Igor's
previous
question
on
the
last
presentation
you
know.
Maybe
this
is
where
you
you
put
in
oven.
Tation
guidelines
say
that
this
is
beyond
what
we're
gonna
do
here,
but
this
models
can
serve
the
foundation.
We
expect
good
to
go
here.
Yeah.
R
R
A
A
S
A
A
A
S
So
this
draft
has
been
presented
in
an
pillars
working
group
before
us,
but
recently
it
was
moved
to
es
so
so
I'm
are
will
be
presenting
first
time
in
two
years.
I'll
just
give
you
a
brief
overview
of
our,
mr,
so
with
our
mr.
We
were
planning
to
make
it
easier
to
use
MPLS
in
ring.
Topologies
so
rings
are
like
very
commonly
found
in
access
and
application
networks.
S
The
base,
our
mr
draft
by
katie
it
defines
the
the
mechanisms
to
bring
up
ring
Gillespie's
and
the
also
the
auto
discovery
of
those
rings,
and
once
the
auto
discovery
is
complete,
auto
disk
or
process
defined
in
the
base
draft
is
complete.
Then
this,
which
each
node
initiates
are
to
LSPs
one
in
clockwise
direction
and
one
in
the
anti-clockwise
direction.
S
S
Similarly,
each
node
will
also
create
two
LS
piece
to
itself,
and
so
by
doing
this,
so
a
ring
of
n
nodes
will
have
to
n
LS
piece
and
another
differentiating
point
of
these
LS
B's
are
these:
are
multi
point-to-point,
LS,
B's,
so
the
by
multi
point-to-point
we
mean
that
the
transit
node
can
also
be
the
ingress
for
that
ring
Gillespie's.
So
in
this
diagram,
so
although
the
LSP
signaling
wise,
it
is
starting
and
ending
at
r1,
there
will
be
also
each
transit.
Router
will
also
can
also
send
traffic
in
that
LSP
ending
at
r1.
S
This
is
like
what
what
happens
during
failure
scenarios,
so
the
link,
n3
and
n4
in
this
diagram.
Let's
say
the
rink.
That
link
is
failing.
So
then
the
traffic
moves,
so
we
don't
have
any
bypass
LSPs
pre-program
like
our
receipe,
vanilla
recipe
tea.
So
we
just
use
the
LSP
going
in
the
opposite
direction
to
divert
the
traffic
so,
for
example,
the
clockwise
RSP
starting
at
n1,
when
the
link
at
n3
n4
fails
I
will
switch
traffic
to
the
anti-clockwise
LSP
in
the
reverse
direction.
S
S
So
to
signal
these
LSPs,
we
have
the
extensions,
so
we
have
a
new
session
object
with
class
this
rmr
tunnel
and
it
will
have
the
instead
of
the
destination
address.
It
will
have
the
rink
anchor
node
address,
which
is
the
address
of
the
the
source
LSP,
which
is
where
it
is
originated,
and
it
will
have
the
ring
Flags
the
whether
that
recipe
is
going
clockwise
and
anti-clockwise,
and
also
it
will
have
the
ring
ID,
which
is
which
is
constant
throughout
the
existence
of
ring.
So
the
ringer
ID
is
basically
whatever
the
ring
is
configured
with.
S
So
in
this
diagram,
the
ring
guard
is
17,
so
you
just
configure
that
ring
Rd
and
for
all
of
the
recipes
going
through.
That
ring
will
use
the
same
ring
Rd
and
so
the
sender
template
object.
We
are
not
really
changing
anything
in
Center
template,
but
the
semantics
are
slightly.
We
are
slightly
changing
the
semantics,
explain
that
in
the
later
slide,
so
for
the
bandwidth
management
we
need
some.
S
We
need
the
different
semantics
for
the
sender
template.
So,
as
I
was
saying
earlier
for
this
ring
Gillespie's,
the
bandwidth
can
change
hop
by
hop.
So
the
LSP
is
starting
and
ending
at
r1,
so
it
can
initially.
Originally
it
will
signal
it
with,
let's
say,
0
bandwidth,
and
then
some
service
comes
at
node
r5
and
it
wants
to
increase
the
bandwidth
for
the
segment
from
our
five
to
our
one.
For
the
for
the
LSP,
and
also
some
service
comes
up
at
r6,
which
wants
some,
which
wants
increase
bandwidth
from
r6
to
r1.
S
Then
the
the
effective
bandwidth
is
like
from
R
1
to
R
5.
There
is
no
bandwidth
reserved
for
that.
A
recipe
from
our
five
to
our
one.
There
is
a
one
gig
bandwidth
reserved
and
from
our
six
to
our
one
it
that
that
segment.
Is
it
question
to
gate,
so
it
will
be
edition
of
one
day
plus
two
B,
so
there
will
be
three
we
bandwidth
reserved
for
that.
So
the
way
we
do
it
is
each
node
add
its
on
the
sender,
template
object
and
then
our
five
wants
to
increase
the
bandwidth.
S
Then
our
five
will
send
an
additional
sender.
Template
object
with
the
sender
address
of
our
five
and
similarly
in
the
path
message
that
is,
and
similarly
our
six
will
also
another
sender
template
object
in
the
path
message.
So
for
the
list
for
the
selected
segment
from
our
six
to
our
one.
There
are
three
sender
templates
and
similarly,
the
corresponding
residual
message
from
our
one
will
have
three
filter:
specs
from
R
1
to
R,
6
and
then
from
our
sixth
or
Phi.
There
will
be
two
frite
respects.
A
S
So
yeah,
and
if
there
is
let
us
say
from
link
r8
to
our
nine,
there
is
only
one
gig
bandwidth
available
and
we
are
requesting
three
weeks
for
the
LSP
from
our
six
to
our
one
for
segment
or
six
to
our
one.
Then
in
that
case
nor
R,
it
will
send
a
path,
error
message
with
just
the
sent
sender
address
for
our
six,
so
that
only
that
particular
reservation
request.
We
will
get
denied
and
the
existing
reservation.
S
So
you
just
continue
to
work
as
it
is
and
when,
let's
say
this,
alpha
no
longer
needs
that
bandwidth
reservation.
Then
it
will
just
send
a
modified
sender
template
with
zero
bandwidth,
and
another
point
is
that
I
will
not
actually
change
any
bandwidth,
but
it
will
sorry
will
not
actually
change
any
labels.
Only
the
bandwidth
accounting
will
change
when
we
are
doing
this,
and
so
these
are
the
extensions.
S
So
the
next
steps
are,
we
still
have
to
identify
how
to
use
the
Express
links
expects
Express
links
is,
let
us
say
there
is
a
link
between
our
phone,
our
ring
link
connecting
our
phone
and
our
seven.
Then
how
to
signal
that
link
that
thing
we
still
have
to
identify
the
best
way
to
use
it,
and
we
need
some
more
feedback
from
the
working
group.
F
A
S
A
S
S
S
A
Q
A
F
T
So
here,
I
just
a
kill.
Some
review
and
update
for
the
PC
in
neutral
IP
network
could
read
the
draft.
So
here
we
have
finished
the
three
drafts
Shay.
Ladies
often
toy
in
the
unit
is
working
or
and
whiny
in
the
PCB
working
group.
Here
I
just
gives
the
viewer
summary
and
the
plan
for
the
next
step
included,
save
had
the
Sinatra
in
the
solution
review
and
update
a
content
from
the
last
meeting
and
the
foreplay
awfully
soft
first
thing
is
this
snuggle
solution
review
and
then
you
know
we
held
in
two
large.
F
T
By
in
the
nature,
IP
network
and
another
in
here
as
well,
there
are
several
scenarios
we
want
to
do.
You
know
traffic
engineering
work
in
the
network
network
and
we
also
you
will
take
the
current
solution
for
the
trafficking
and
the
fund
as
they
are
not
very
suitable
for
the
native
Abnett
were
so
we
think
after
the
introduction
of
the
PC
or
Sdn
controller
area
network
of
Iike
finish
such
work
more
easily.
T
So
in
this
in
our
job
we
have
introduced
about
for
scenario
for
the
newly
your
case
of
the
PC,
our
network,
and
we
also
to
the
simulation
of
line
and
to
to
verify
it
of
the
PC
our
network
and
we
get
the
better
effect
than
the
traditional
teacher
to
a
calculator.
So
we
think
the
if
we
put
the
controller
our
PC
our
network,
because
the
wonderful
is
out
for
our
traveling
in
our
network,
we
see
the
neural
network
in
our
scenario.
Java
and
honor
is
home
and
already
the
solution
trapped
in
this
solution.
T
There
are
three
key
point
for
the
solution.
The
first
thing
is
that
we
want
to
deploy
the
initializing
controller
in
our
network
and
PC.
A
steam
controller
is
responsible
for
the
complex
paths.
Combien
has
communication
to
optimize
the
natural
traffic
open
area
network
situation,
and
the
second
point
is:
we
want
to
build
multiple
image
sensor
by
the
his
episode,
pc,
proto
and
distributed
different
prefix
by
the
different
PGP
says
between
the
edge
rotor
and
the
IR
rooted
reflector.
T
The
third
point
is:
we
want
to
manipulate
the
path
to
PCB
next
up
with
all
these
prefix
by
the
also
by
the
PCF
protocol,
put
different
the
traffic
14
purpose.
So
if
we
use
such
philosophy,
we
need
only
change
or
change
or
update
the
control
plane
of
the
of
the
rotor,
and
we
need
not
update
the
fourth
player,
such
as
the
solution
for
the
SR.
T
So
we
can
keep
our
has
root,
not
actually
and
still
still,
we
can
exploit
our
and
our
daughter,
you
already
our
network.
So
this
is
the
solution,
key
point.
T
So
for
the
two
jobs,
the
realities
we
reject
to
some
editor
update
from
the
last
meeting
and
for
the
first
staffed
way
and
the
we
end
some
description
for
the
security
consider
consider
in
Teresa
and
the
first
draft
we
and
the
culture
from
mr.
Chen
Chen
from
China
Mobile,
because
they
have
the
same
your
network
constructor
with
us.
T
T
T
Because
you
know
we
are
ready
Nikki
besides
such
solution
for
this,
for
this,
for
instance-
and
we
want
to
finish
the
rest
and
standard
of
this
draft
and
also
waiting
for
the
implication
for
the
rehabilitation,
so
we
can
deport
them
or
try
them
in
our
in
our
network.
And
the
final
is
thanks
for
alcohol,
author
and
expert
for
the
pasta,
videos,
okay,.
T
So
you
know
we
just
want
to
return
the
year,
our
agree.
She
enjoys
our
network
and
to
not
want
to
change,
because
you
know
there
are
lots
of
water
in
our
network
and
located
in
different
place.
So
we
every
think
the
dough
needs
to
be
adjusted
to
change
the
console
player,
so
maybe
the
to
some
to
the
potato.
She
said
the
protocol
update
may
be
more
easily
for
for
Department.
A
Okay,
well,
something
to
think
about
is
what's
coming
down
the
pike
and
from
the
working
group
standpoint.
I
know,
that's
something!
That's
gonna
probably
come
into
this
working
group
or
be
discussed,
probably
not
next
ITF,
but
two
or
three
IETF
s--
from
now.
I
would
expect
that
that
to
come
in
and
it
that
that'll
be
traffic
engineering
for
IP,
and
this
is
traffic
engineering
for
IP.
D
Hi,
this
is
trove
and
I'll
be
presenting
a
CT
and
VN
yank,
so
quick
update
we
got
adopted
during
the
last
ITF.
There
was
a
comment
that
there
is
some
work
in
ops
working
group
as
well,
which
deals
with
network
over
later
source
model.
So
we
had
a
quick
discussion
and
we
kind
of
convinced
that
the
work
that
we
are
doing
in
a
CT
and
VN
can
meet
the
requirements,
so
we
integrated
and
consolidated,
and
that's
why
we
have
some
new
quarters
we
kind
of
did
not
update
the
yang
model
per
se.
D
This
time
we
have
some
things
in
the
pipeline
that
we
will
be
doing
just
after
this
idea.
We
wanted
to
have
a
face-to-face
discussion
and
resolve
those
things
during
this
meeting.
So
the
update
basically
clarified
a
few
things
like
what
is
the
scope
of
the
tea
service
mapping,
which
is
anyway
being
presented
in
the
next
in
the
gender?
The
next
and
we
added
the
reference
and
clarity
with
respect
to
that.
D
These
bian
models
could
be
created
because
of
the
service
things
coming
from
l3
SML
to
sm,
etc,
and
we
clarified,
because
there
were
questions
with
respect
to
where
is
this
VN
topology?
So
we
clarified
that
the
VN
topology
is
implemented
with
respect
to
the
connectivity
matrix,
there's
abstract
node
connectivity,
matrix,
which
anyway
I
presented
during
the
last
ITF
as
well.
The
big
change
that
we
did
before
we
got
adopted
and
the
tunnels
would
be
not
in
this
model.
You
have
to
go
through
connectivity
matrix
and
from
there.
D
You
can
refer,
though,
and
find
the
the
eternal
model
so
status
wise.
There
is
a
new
update
coming
in,
so
we
look
out
for
and
we
will
try
to
refine
the
document
and
prepare
the
document
along
with
as
the
tea
tunnel
and
other
documents
with
the
young
are
moving
along.
We
would
also
like
to
move
lacy
T
and
B
and
young
as
well
others
I
have
back
up
if
there
is
any
questions
regarding
the
model,
but
that's
it
from
I
said
so.
I
think
this
was
quick.
L
So
I
wanted
that
too,
to
give
some
clarification
based
on
on
those
feedbacks,
they
mean
a
which
being
the
fact
that,
when
we
speak
about
mapping
that
can
be
source
for
confusion,
actually
we
use
the
term
mapping
there,
but
what
we
really
mean
is
binding
in
the
sense
that
this
draft
is
provides
the
tool
to
create
a
binding
relationship
between
an
overlay
service.
In
this
case
here
we
have,
we
provide
the
support
for
layer,
3,
sm
layer
to
assembly,
Awan
CSM
and
the
underlay
infrastructure
traffic.
L
Engineered
infrastructure
use
the
to
carry
the
services
being
in
being
it
in
a
city
and
virtual
network
to
start
with,
but
we
now
also
provide
the
support
for
direct
map
died,
it
binding
against
the
T
topology
and
the
T
tunnel.
What's
the
main
rationale
for
for
this
work
and
for
it
is
the
kind
of
binding
between
over
and
underlay
the
first,
the
berryfest.
The
use
case
is
natto
slicing.
Basically,
we
wanted
to
provide
the
overlay
services
with
a
guaranteed.
L
Kpi's
SLA
is
guaranteed
the
requirements
that
can
only
be
provided
by
a
traffic
engineer,
the
infrastructure.
So,
for
example,
if
you
wanted
to
have
a
classical
year
free
VPN
without
any
KP
guaranteed,
kpi's
and
so
on,
you
just
go
for
the
layer,
3
sm
you
ask
for
a
layer,
3
VPN,
you
don't
ask
to
bind
it
to
any
tunnel
or
infrastructure
if
happen.
L
On
the
other
side,
you
wanted
to
offer
a
different
type
of
service,
which
is,
for
example,
a
layer,
3
VPN,
with
an
amount
of
reserve
bandwidth
minimum
delay
for
the
traffic
of
that
VPN.
What
you're
going
to
do
is
to
bind
it
that
to
a
tea
tunnel
or
a
virtual
network
that
was
computed
and
provisioned
with
the,
for
example,
a
minimum
delay
a
path
along
using
the
resources
along
the
path
with
a
minimum
delay.
This
is
the
rationale,
the
main
reason
behind
the
Andes
worker.
Beyond
the
draft
we.
G
G
G
There
are
also
such
cases
like
for
protection
purpose
stuff,
like
that
there
should
be,
in
my
opinion,
a
framework
which
describes
you
know:
here's
the
tunnels
and
here's
the
services,
and
here
how
we
can
actually
map
a
bank
and
a
service
to
the
tunnels,
either
specifically
or
based
on
certain
policies
or
basically
to
express
preferences
and
stuff
again.
I
think
it.
G
Addition
to
that,
what
I'm
saying
is
that
I
won't
have
something
more
generic,
so
you're,
describing
only
based
on
the
boundaries
between
clients
and
network,
but
but
there
could
be
such
bindings
happening
at
any
node
in
the
in
the
operator.
Networks.
I
owe
some
some.
You
know
routers
or
not
sitting
inside
there,
the
network,
okay,
so
so.
A
We're
actually
rehashing
a
little
bit
of
a
discussion
we
had
at
the
last
meeting
and
at
the
last
meeting
we
said
you
know,
can
you
guys
work?
You
know
talk
offline
and
come
back
and
report
from
that
discussion
and
based
on
what's
happening
here.
It
seems
like
that
discussion
didn't
happen,
so
we
ended
up.
R
R
R
I
A
To
in
the
context
of
this
specific
document
get
with,
it
was
Igor
and
Adrian
and
I,
don't
remember
who
else
I
can
look
at
the
minutes,
get
with
them,
spend
some
time
and
see
if
you
can
come
back
with
an
agreement
on
the
comments
that
were
raised
last
time
in
the
comments
that
are
coming
back
up
this
time
and
if
the
agreement
is
hey,
we're
talking
about
different
things?
That's
okay!
Yes,.
A
E
About
me,
it
said
Lou
has
asked
me
to
put
on
technical
adviser
hat
and
look
at
this
and
the
previous
draft
and
and
the
relationship
with
other
models
which
I
will
do
I
also
just
went
and
looked
at
the
minutes
from
the
last
meeting,
and
we
are
all
fumbled
this,
because
the
people
who
commented
should
have
gone
on
further
and
the
authors
should
have
come
back
to
them.
I
think
that
something
is
missing
from
this
document,
which
would
really
really
help,
which
is
to
understand
the
flow
of
information
between
logical
components.
E
E
L
L
A
L
K
A
R
Sum
of
key
performance
index
for
their
virtual
network,
they
are
creating,
so
this
young
model
basically
is
on
a
CMI
model
that
when
customer
request
VN,
they
also
ask
for
performance
monitoring
on
a
VM
virtual
network
level
and
once
the
virtual
network
may
have
multiple
Fulcher
members,
which
is
kind
of
tea,
a
tonal
model.
So
we
have
those
two
layers.
R
This
is
based
on
the
use
case,
so
we
have
a
two
young
modules:
key
KPI
telemetry
and
also
a
CT
NT,
a
telemetry
one
is
just
one
single,
a
point-to-point
Ono
level
and
the
other
one
is
more
virtual
network
that
consists
of
multiple
virtual
members
and
each
of
which
is
corresponding
to
TA
tono.
Also,
we
have
a
network
autonomic
scaling,
scale-out
mechanism
that
based
on
the
performance
metric
they
want
to
subscribe
to.
R
They
can
ask
for
multiple
combinations
of
a
performance
metric
to
be
combined
through
end
operation
or
or
or
max
min,
depending
on
the
the
metric.
So
when
those
things
are
met,
it
can
be
automatically
scale
in
or
scale
out
so
statuses
we
presented
it
while
ago,
and
this
time
major
comment
was
from
pub
on
weren't.
You
argument
existing
performance
data
from
existing
model,
so
we
took
that
action
and
then
we
had
many
conversation
with
shipping
and
ego
and
Tareq.
We
found
out
teeny
types.
R
A
module
has
actually
performance
metric
attribute
as
a
grouping,
so
we
make
use
that
grouping
to
describe
unidirectional
performance
monitoring
data
without
changing
that
and
then
because
sometimes
we
need
bi-directional.
So
we
augment
it.
We
added
actually
a
bi-directional
part
in
this
module,
so
hopefully
I
think
we
had
a
weekly
meeting
on
the
theatre
policies
designed
Tim
now
taped
on
Alton.
So
hopefully
this
is
answering
a
bubble's
question.
So
basically,
as
you
see
here,
Tek
telemetry
weari
uses
its
grouping
defined
in
T
types
through,
you
know
referencing
for
unidirectional
and
then
for
bidirectional.
A
R
A
F
F
A
R
You're
not
reinventing
in
a
new
wheel
here.
Are
you
just
referencing
performance
actually
from
T
types
for
link
and
tunnel
and
LSP
that
can
be
applied
so
I?
Don't
think
we
are
bulleting
what
this
is
customer
perspective?
They
were
monitor
this
data
for
their
own
sake,
so
they
kept.
They
can
subscribe
to
it
and
then
network
we
provided
and
MDS.
It
has
to
concatenate
all
this
data
from
the
lower
level
to
give
this
meaningful
information
to
the
customer.
So
that's
the
scope.
A
R
At
that
reference,
if
Greg
can
point
out
exact
reference,
I
can
take
a
look
and
then
give
him
reply.
Okay,
okay,
so
we
believe
that
this
is
very
important
work
along
with
a
CT
and
V
engine
which
is
now
adopted
along
with
that
this
one
gives
customer
how
their
V
ends
are
performing,
so
that
they
can
program
telemetry
and
then
autonomics
as
a
part
of
zero-touch
automation
initiative.
R
R
A
Okay,
doesn't
look
like
Gregor's
gonna,
join
the
queue
or
follow
up,
so
with
the
understanding
that
there
may
be
an
additional
model
that
needs
to
be
a
module
that
needs
to
be
looked
at
to
deconflict.
How
many
people
think
that
this
is
the
type
of
work
that
we
should
be
working
on?
How
many
are
interested
in
this
work?
I
would
say:
that's
a
a
reasonable
number.
A
A
G
R
R
A
A
R
R
A
F
R
I
think
we
presented
this
one
I
think
Dan
King
did
in
the
one
of
the
puffs
about
Network
slicing
a
while
ago
and
now
I
think.
A
lot
of
things
are
now
little
bit
a
little
bit
more,
getting
clearer,
so
I
own
a
present
here
with
some
update
how
a
CTN
can
be
applicable
to
network
slicing
here.
What
do
we
mean
by
network
slicing?
We're
not
you
know
redefining
network
slicing
defined
by
somewhere
else,
but
we
are
talking
about
Tina
took
slicing
from
a
CTN
perspective,
so
we
are
talking
about
Tina
talk.
R
Slice
is
a
collection
of
resources
that
is
used
to
establish
or
logically
dedicated
virtual
networks
over
one
or
multiple
t
networks.
That's
actually
our
narrow
definition
of
Tina
talk,
slicing
and
essentially
network
slicing
in
the
context
of
a
CT
and
provide
traffic
engineering
connectivity
that
ensures
performance
guarantee
to
serve
the
customer
with
the
variety
of
service
constraints.
That
could
be
your
latency
reliability.
Capacity
and
service
function,
specific
capabilities,
okay,
and
we
put
some
requirements
here.
What
are
the
requirements?
Resource
icing
definitely
is
very
important
requirement
or
partitioning
so
that
it
can
be.
R
You
know
given
to
the
customer
and
then
also
network
function,
virtualization
whether
is
physically
or
virtualized,
but
that
the
resource
can
be
given
to
the
customer
and
also
isolation.
I
think
is
a
very
important
requirement.
Whether
I
want
mice,
tunnel
or
virtual
network
isolated
in
a
sense
that
I,
don't
know
be
impacted
by
other
virtual
networks,
for
instance,
and
that
can
be
also.
Customer
can
also
ask
for
very
strict
latency
with
no,
you
know
variations.
R
So,
as
I
said,
virtual
network
is
very
important
concept
and
then
we
almost
use
fortunate
as
a
slides,
fortunate
of
slice
allocated
to
the
customer
to
meet
isolation,
requirement,
resource
reservation,
guarantee
and
reliability
and
performance
monitoring,
program
ability
and
monitoring
and
underlay
visibility.
If
they
want
to
see
that
we
give
that
reference
point
so
that
they
can
look
at
operational
data
if
they
will-
and
another
thing
was
SS
f
en
f
NM
and
F
over
a
connectivity
which
is
I
just
presented.
Those
are
might
be
very
important.
You
know
down
the
road.
R
So
what
are
the
building
blocks
that
we
have
done
so
far
to
fulfill
Tina
talk
slicing
within
a
city
in
context?
We
have
architecture
framework
and
we
have
a
system
via
a
young
model.
We
have
a
good
basis
to
create
vien
and
then
to
crud
operation
of
the
end,
and
then
our
service
metal
model
that
we
believe
that's
very
important
glue
to
glue
or
customer
service
model
to
or
T
on
delay,
so
that
customer
can
so
that
the
service
can
be.
R
You
know
using
the
underlay
resources
and
we
the
visibility
and
requirement
like
isolation
and
reliable,
that
it
will
be
conveys
through
those
metamodel
and
then
and
the
customer
would
be
interested
how
there
are
Network
foreigner
talks.
Life
is
performing,
which
is
very
important
criteria,
especially
5g
ultra-low
latency,
with
it
emits
a
characteristic.
R
They
would
be
very
interested
how
their
services
are
performing,
so
performance
monitoring
is
very
important
aspects
and
SFO
topology
model,
because
it's
not
just
connectivity
anymore,
because
customer
may
want
to
do
have
some
dynamic
of
virtual
machine
mobility
tied
to
it
or
some
other
or
fortunate
to
function,
enabled
T
connectivity
as
a
slice.
So
those
are
the
four
aspects
that
we
believe
a
CT
and
have
has
discussed
in
this
working
group.
R
So
what
additional
ITF
work
might
be
needed?
I
said
IETF
for
all,
not
just
tears
may
be.
A
security
issue
might
be
very
important
how
to
assure
that,
and
the
police
control
of
slices
is
more
policy
driven
stuff
and
also
requesting
compute
storage
function
or
resources
during
Vienna
set
up.
Currently,
we
don't
have
this,
but
this
can
be
definitely
a
addressed
down
the
road
that
includes
the
N,
F
type
and
size
location
and
ensuring
correct
service
chains.
R
So
next
step
this
was
informational.
Draft
basically
put
all
the
ICT
on
and
then
as
self
in
every
topology
work,
which
is
scoped
to
address
that
hey.
A
lot
of
our
network,
slicing
from
transport
network
perspective
or
Tina
to
perspective
have
been
in
place,
but
we
are
kind
of
terrible
to
tell
other
people
outside
of
teams
working
group
and
even
outside
of
IETF.
This
might
be
a
good.
You
know
tutorial
type
of
information
or
drafts
that
can
be
very
useful
so
that
people
can
understand
better.
So
that's
the
purpose
of
destruction.
R
F
F
G
U
Thank
you
right,
so
this
craft
has
existed
in
a
number
of
incarnations
for
a
while.
So
the
purpose
and
scope
of
the
draft
is
to
describe
the
enhancements
that
we
need
to
make
to
VPNs
to
support
the
requirements
of
new
services
and,
in
particular,
we're
concerned
about
fight
the
5g
scenario
such
as
network
slicing.
So
what
we
want,
what
we
are
interested
in
is
the
design
of
an
enhanced
data
plane
that
is
needed
to
to
support
this.
The
hard
and
soft
isolation
at
the
data
plane
and
the
resources
immediately
above
the
data
plane.
U
Determinism
of
in
turn
determines
the
operation
so
there's
a
link
with
deterministic
networking
we're
interested
in
the
protocols
in
the
underlay
and
the
overlay
of
the
enhanced
p2p
P
n.
So
there
and
there's
a
an
integration
between
physical,
virtual
and
service
functions.
That
needs
to
happen
and
of
course
it
needs
to
be
simple
to
create,
delete
and
modify
I
with
some
very
interesting
on-the-fly
modification
requirements,
and
we
need
monitoring
and
instrumentation.
U
So
here's
what
we
think
the
service
model,
what
the
model
basically
needs
to
look
like
level
1
operation
is
kind
of
like
existing
VPNs
that
we
have
today.
So
we
have
logical
separation
or
own
and
differentiation.
We
have
the
possibility
of
interference
between
services,
and
this
is
what
we've
got,
and
it
is
one
of
the
services
that
we
will
need
to
still
support
on
this
network
going
up
to
the
next
degree
of
of
operation.
We
need
forwarding,
plane,
isolation.
U
U
Even
better
isolation,
that
is
for
further
study,
that's
not
the
area
that
we
want
to
address
here.
So
let's
look
at
some
a
spectrum
of
resource
isolation,
so
traditional
VPNs
produce
or
operate
with
statistical
multiplexing
with
all
the
resources
shared
absolute
isolation
is
only
really
achievable
through
having
dedicated
networks.
F
U
Which
would
be
too
expensive
and
when
I
talk
about
resources,
I'm
talking
about
links,
bandwidth
buffers
in
the
dedicated
buffers
in
the
devices,
possibly
dedicated
resources
in
the
forward
or
NPU
dedicated
queuing,
etc.
So
modern
routers
are
requiring
the
ability
to
carve
out
resources
for
the
most
important
services
and
when
we
need
a
structure
to
deal
with
that.
U
A
Did
read
your
before
this
week,
so
did
you
cover
the
case?
Are
you
aware
of
the
case
that
the
current
tool
set
does
allow
mapping
VPN
VPNs,
even
LDP,
controlled
VPNs
to
te
use
it
and
using
RSVP
to
give
the
current
level
of
service
described
by
the
current
documents,
and
some
people
actually
deploy?
That's
not
a
lot,
but
some
people
actually
deploy
that
or
you
wear
that
and
it's
that
covered.
A
U
U
So
expect
from
performance
requirements,
largely
VPN
is,
is
best
effort
and
as
it
was
saying,
when
you
try
and
give
provide
a
short
bandwidth,
the
VPN,
then
you
end
up
with
some
scaling
issues.
U
U
The
next
slide
bit
is
easier
with
your
room
asset,
so
so
here's
the
architecture
we
know
going
to
enhance
data
plane,
which
is
where
our
resource
partitioning
your
reservation
happens.
On
top
of
that,
we
construct
our
virtual
networks
with
guaranteed
resources,
and
we
will
have
to
have
a
tighter
integration
between
the
overlay
and
the
underlay
than
we
traditionally
have
with
VPNs,
because
we
need
specific
resources
in
order
to
assure
performance
and
then,
above
that
we
have
a
control
plane.
F
A
U
I,
don't
think
we're
when
you
say
internet
level,
internet
with
the
big
eye,
I,
don't
think
you
can
do
anything.
The
big
eye,
a
level
like
this,
so
I
think
we're
really
and
then
the
demand
is
really
for
large-scale
customer
networks.
So
the
networks
owned
by,
for
example,
a
mobile
service
operator
I,
don't
think
we
have
to
do
the
big
ice.
V
U
U
So
yes,
we're
talking
about
mobile
phones
or
airily
might
work
anyway
mobile
phone
areas,
not
factories
and
enterprises.
We're
not
we're
not
talking
about
anyone
can
connect
anywhere
and
yet
this
service,
okay,
so
candidate,
some
candidate
technologies.
So
to
the
underlay,
there's
a
bunch
of
technologies
that
we
can
use
flexibly,
thinit,
TSA,
I,
Triple,
E,
TSN,
sometimes
networking
we
can
build
an
enhanced
queueing
in
quas
model
and
we
can
have
provide
dedicated
nodal
resources
as
cpu
slices
and
a
network
processor
slices
in
the
network
layer.
U
There's
a
number
of
candidate
technologies
that
we
need
to
glue
together,
regular
mpls
and
our
SCP
te
mpls
sr
looks
quite
interesting
because
sr
provides
a
way
of
identifying
resources
without
finding
everything
to
the
to
the
path
assume.
There's
our
v6
is
getting
some
attraction
and
then
all
the
work
we're
doing
in
deterministic
networking.
But
our
view
is
that
that
will
the
second
reaching
world
needs
enhancement
to
be
usable,
because
segment,
routing
isn't
really
yet
talking
about
resource
reservation.
F
A
Rsvp-Te
is
in
a
network
layer,
its
righto
plane
right,
LSP,
sortie,
POS
fees
are
certainly.
K
U
U
U
We
certainly
need
to
identify
what
resources
are
available
and
send
them
up
to
a
centralized
control
controller,
there's
a
bunch
of
technologies,
many
of
which
you
would
discuss
here
earlier
today
or
to
SM
age
H,
CTN
pset,
BGP
LS,
which
are
good
for
having
the
conversation
between
the
lower
layers
and
the
centralized
controller.
What
we're
worried
about
is
whether
a
pure
centralized
model
at
these
scaling
levels
is
right
and
what
and
how
much
we
need
to
move
down
in
back
into
the
into
a
distributed
model
just
to
keep
the
thing
operating
in
a
timely
basis.
U
So
with
with
thinking
it's
not
thinking
about
scalability
right.
So,
if
you
think
about
what
we've
got
here,
we've
distributed
and
centralized
control
on
the
left
and
the
state
in
the
network
versus
the
states
in
the
packet
on
the
right
rsvp-te.
The
state
is
kind
of
in
the
network,
with
just
a
label
on
the
packet
and
and
it
uses
a
distributed
control,
whereas
MPLS
TP.
Just
as
an
example
of
just
simple
MPLS,
with
a
completely
centralized
control
plane,
the
the
state
gain
is
fully
in
the
networking
you
have
on
touch.
U
In
that
case,
all
the
nodes
in
the
network
to
set
your
path
are
honestly,
so
SR
took
an
opposite
approach,
which
is
how
to
move
all
of
the
state
into
the
packet.
Well,
don't
quite
do
that
actually,
because
they
have
binding
SIDS,
which
kind
of
break
that
model
and
move
it
closer
to
the
to
the
hybrid
approach
here.
U
So
what
other
work
is
not
really
good.
Is
it
what
other
work
is
going
on?
So
we've
got
a
draft
in
spring
we're
trying
to
get
adopted
spring
which
describes
the
segments
routing
aspects
of
this
of
this
work.
We've
got
another
draft
in
LSR,
describing
some
of
the
enhancements
that
we
think
need
to
happen
to
the
leaf
state
protocols
to
collect
and
manage
the
information
and
do
some
of
the
calculations.
U
There's
a
lot
going
on
in
debt
net,
which
we
see
as
one
of
the
main
underpins
for
for
this
work,
both
the
architecture
and,
in
particular,
the
the
MPLS
solution
document
we
see
as
very
applicable
and
the
actn
work
happening
here
and
there's
also
an
ASEAN
piece
of
work
that
was
taken
to
r2
gwg.
So
that's
all
an
IETF
work.
That's
related
this
and
we
see
this
is
supporting
the
framework
that
we're
putting
together.
U
Uri
wrote
a
draft
on
virtualization
in
the
work
the
ITF
was
doing.
This
is
a
study
of
virtualization
techniques.
I'm
we
are.
We
should
treat
that
and
make
sure
that
we're
not
missing
any
tricks,
but
I,
don't
see
that
as
being
in
the
same
space
and
draft
a
leaf
spring
Network.
Slicing
building
blocks
is
just
that.
It's
a
client,
it's
a
collection
of
building
blocks
that
may
be
used
to
underpin
this
sort
of
network
slicing
technology.
It
provides
a
catalog.
U
L
J
F
U
So
there's
two
dimensions
of
scalable
innovation,
especially
scalability,
there's
the
number
of
services
and
sure
it's
not
going
to
be
I,
don't
think
it'll
be
one
application.
One
service,
as
some
people
in
3G
in
3gpp,
are
suggesting,
but
considering
the
sort
of
a
source
binding
you're
going
to
have
to
do
it's
going
to
be.
You
know,
look
like
a
lot
more
from
the
resource
point
of
view
than
classic
of
VPNs,
where
basically,
a
classi
VPN
you've
got
like
eight
classes
of
service
and
that's.
L
F
U
A
E
Barrow
engine
peril
and
the
documents
found
the
right
home
in
my
opinion,
and
it
is
going
the
right
way
and
thank
you
for
an
architecture
that
gives
us
context
for
everything.
I
think
we
need
to
move
some
of
this
more
towards
a
framework
of
pointing
to
other
things
and
identifying
gaps
in
work
and
you've
sort
of
got
that
on
the
slide.
So
I'm
happy
to
talk.
Yes,
please
do
and
always
we're
willing
to
work.
X
With
anyone
who's
going
to
help
us,
then
MacDonald
in
two
plus
one
on
you
know
previous
comment
as
well
and
I
have
some
consideration
about
the
underlying
hardware
capabilities.
Then
I
have
taken
to
the
account,
and
that
will
be
some
of
my
concerns
in
other
houses.
So
some
of
the
issues
like
thank
you
take
that
it's
to
take
it
on
this.
Would
you
like
to
do
it
I'm.
A
Not
gonna
ask
question
to
this
is
only
one
question:
how
many
people
are
interested
in
hearing
this
topic
more
in
the
future.
You
know,
let's
continue
to
talk
about
it.
That's
a
really
good
endorsement.
That's
a
great
number!
So
thank
you
and
with
that
we're
over
time
and
done
made
it
through
our
agenda.
Thank
you
all
for
a
very
good
meeting
and
we'll
see
you
in
Bangkok.