►
From YouTube: IETF106-SPRING-20191121-1550
Description
SPRING meeting session at IETF106
2019/11/21 1550
https://datatracker.ietf.org/meeting/106/proceedings/
A
A
A
A
A
B
C
My
name
is
a
four
I
am
going
to
present
this
replication
segment
draft
on
behalf
of
the
authors,
unfortunately,
Dan
and
Rishabh,
and
and
Jeffrey
cannot
make
it
so
I'm
just
helping
with
the
presentation
in
terms
of
brief
history
of
this
draft,
so
work
actually
started
back
in
2018
and
it
was
named
a
drug.
Why
a
SRT
20
policy
it
was
presented
by
then
voice
in
Prague
ITF,
and
it
did
receive
comments
from
chair
as
well
as
from
working
group
and
based
on
those
comments.
C
There
has
been
lately
quite
a
few
comments
on
the
mailing
list.
Many
many
thanks
for
those
comments.
Authors
have
addressed
those
comments
that
have
done
a
revision
during
this
week
and
have
also
calls
with
the
people
that
have
comments
specially
Shasha.
Many
things
are
sure.
If
you
are
attending
for
your
comments-
and
we
have
made
tremendous
progress
in
terms
of
achieving
a
common
ground
and
addressing
comments,
and
if
there
the
need
there
could
be
another
revision
really
quickly,
let's
talk
about
the
scope
of
the
draft.
C
This
draft
is
specially
targeting,
in
a
spring
charter
item
and
I,
read
verbatim,
it's
pretty
shorter
to
define
new
type
of
segments
netting
to
forwarding
behavior
and
then
one
of
the
definition
of
those
is
local,
English
replication.
The
draft
is
no
more
than
local,
English
replication
replication
segments
deliver
packets
to
multiple
nodes
in
the
SR
domain,
but
it's
a
direct
application
from
ingress
to
the
set
of
downstream
egress
knows
the
a
copy
of
the
packet
is
unique.
C
A
stood
to
each
of
the
downstream
node
in
where
you
can
think
of
it
is
a
little
is
a
simple
application
segment
with
low
propagation
over
N,
as
are
policies.
But
the
point
to
note
is
that
in
this
case
there
is
no
multicast
groups
in
the
context
of
document
or
scale
with
document.
There
are
no
multicast
flow
that
we
talked
about.
There
is
no
receiver
join.
There
is
no
three
building.
All
of
these
things
are
outside
the
scope
of
this
draft
and
topic
for
discussion
with
the
working
group
where
multicast
expertise
exists.
C
C
This
is
more
like
binding
cid
and
when
a
packet
comes
with
that
sit,
then
it
get
replicated
to
each
of
the
e
grosses
using
the
unique
Ostrom.
So
then
it
perfectly
max
284
or
2
in
terms
of
when
the
packets
are
replicated
via
local
policy
is,
is
a
push
operation
as
Marie
de
SR
re-texture
80
402,
when
active
segment
is
the
application.
Id
is
a
simple
continue
operation
and
in
the
downstream
node,
is
a
simple
next
operation.
C
C
A
E
F
F
F
So
to
be
able
to
do
that.
This
controller
is
going
to
learn
the
EPE
seeds
from
the
ESB
ours.
So
it
can
also
learn
which
Sid
is
a
backup
for
is
being
used
as
backup
for
which
other
Sid.
So
we
define
this
F
bit
in
the
existing
flags,
and
this
this
fr
Sid
will
be
included
as
part
of
the
peer
adjacency
sit
advertisement.
F
What
we
get
is
the
BGP
multi-hop
session
local
addresses,
which
is
r1
and
r2
loopback
addresses,
so
the
controller
doesn't
really
know
which
are
the
actual
links
that
are
being
used
for
this
node
set,
so
a
controller
that
is
doing
traffic
engineering
and
doing
bandwidth
management.
It
can't
really
know
what
are
the
links
and
what
has
associated
bandwidth
with
respect
to
those
links.
F
So
the
proposal
is
so
when
you
advertise
appear
nodes
it
and
when
you
want
to
do
traffic
engineering
and
bandwidth
management
over
these
peer
adjacency
said
it
is
a
must
for
each
of
those
links
so
peer
nodes
that
can
be
used
for
load.
Balancing
like
you
have
one
set,
and
you
can
load
balance
your
traffic
across
the
true
links.
But
you
should
separately
also
advertise
peer,
adjacency,
said
and
associate
all
the
traffic
engineering
attributes
to
that
adjacency
set.
So
this
draft
mandates,
the
peer
at
Jason,
sees
advertisement
of
peer
at
Jason's.
F
He
said,
even
though
you
advertise
peer
nodes,
it
and
one
more
property.
Is
you
cannot?
You
should
also
advertise
as
an
attribute
in
peer
node
said
the
link
addresses
so
there's
a
new
TL
we
defined,
which
is
the
link
address
TL
v.
It
carries
the
interface
addresses
of
the
links
which
are
corresponding
to
the
peer
node
state.
So
the
TLB
looks
like
this:
it
has
type
length
and
number
of
ipv4
interface
address
pairs
and
number
of
ipv6
interface
addresses
and
the
actual
pair
of
interface
addresses.
F
So
when
both
so.
This
is
a
case
where
some
interfaces
are
ipv4
and
some
underlying
interfaces
are
ipv4
and
some
underlying
interfaces
are
ipv6.
In
that
case,
the
the
ipv4
addresses
are
encoded
first,
followed
by
the
ipv6
addresses.
So
the
controller
is
responsible
for
looking
at
clear
notes
it,
and
these
attributes
and
correlate
which
peer
adjacency
sits
these
pianos.
It
corresponds
to
and
get
the
traffic
engineering
attributes,
such
as
bandwidth
and
other
color
link,
color
or
other
traffic
engineering
attributes
from
this
this
advertisement.
So
this
this
advertisement
is
mainly
used
for
correlation
by
the
controller.
F
H
From
agama,
you
know
for
the
increase
authority
in
order,
and
they
also
follow
either
the
arc
here
here,
I
charge
the
one
to
ask
the
yoga
if
you
based
on
the
EPA,
to
build
the
interest
about
you.
So
if
we
have
several
partners
or
SBR
between
given
s
so
doesn't
need
to
be
needed.
Every
SPR
to
run
the
PCBs
or
protocol
yeah.
H
Maybe
maybe
some
deployment,
the
Shiksha,
you
know,
because
you
know
in
our
network
there
are
almost
every
edge,
rotor
or
connector
is
our
main
network.
So
so
in
a
way
based
on
the
EPS
Rosen,
the
in
the
area
as
you
not
her
must
run
the
PCBs,
proto
I
think
the
he's
a
maybe
some
accessibility
issue,
yeah.
E
Sorry,
I
do
have
a
cold.
So
is
it
better?
Thank
you.
So
there
were
two
other
drafts
as
well.
The
163
74
MPLS
sr
and
pls
io
am
so
both
drafts
are
now
moved
to
the
MPLS
working
group,
so
they
were
on
agenda,
but
we
won't
be
presenting
them.
They
were
presented
Monday
in
the
MPLS
working
group,
so
the
agenda
is
requirements
and
scope.
The
history
of
this
draft,
the
updates
we
have
made
since
ITF
in
Prague,
based
on
many
review,
comments
and
suggestions,
the
summary
and
the
next
steps.
E
So
the
requirements
is
a
delay
and
lost
measurement
for
links
as
well
as
end-to-end
SR
policies,
both
SR
MPLS,
as
well
as
sr
v.
Six
data
planes
no
need
to
bootstrap
the
PMC
essence.
I
think
this
means
no
negotiation,
no
signaling,
that's
the
spirit
of
SR.
It
is
stateless.
On
egress.
Node
again
goes
with
the
SR
simplicity.
E
We
need
to
handle
the
ICI
mp4s
our
policies,
a
support,
direct
mode
loss
measurement
as
well.
So
the
scope
is
using
the
RFC
53
57
1212,
light
probe
messages
we
have
now
also
included.
The
stamp
step
is
was
approved
by
IES
D,
so
it
should
become
an
RFC
soon.
It's
a
fixed
length
version
of
prompt
light
and
basically
using
the
user,
configure
UDP
ports.
E
So
the
history
of
this
draft.
He
was
first
published
in
February
this
year,
but
it
is
based
on
the
UDP
PM
draft
that
we
had
published
the
year
before
so
uses.
That
is
just
an
evaluation
after
the
UDP
PM
draft.
So
it's
been
around
for
some
time
now.
This
was
presented
in
Prague
in
spring
working
group
and
then
it
was
also
presented
in
AI
ppm
working
group
in
Montreal.
E
So
based
on
the
review
comments,
suggestions,
many
many
thanks
to
everyone
who
contributed.
We
have
updated
the
drafts.
We
welcome
back
as
well
as
Bart
as
co-authors.
We
have
added
the
provisioning
model
that
useful
for
this
solution
now
also
include
the
Stamp
message.
Formats
also
define
the
look-back
measurement
modes.
E
We
have
added
a
TLB
for
to
a
measurement
mode.
We
have
clarified
and
added
the
message:
processing
rules
for
TTL,
a
router
alert
option
and
UDP
checksum
also
elaborated
procedure
for
p2
MPs.
Our
policy.
There
is
some
format,
alignments
between
loss
and
delay
measurement
as
well,
and
some
editorial
changes
so
just
to
give
a
summary
of
the
changes
that
we
have
made
since
this
was
presented
last
time.
So
this
is
a
provisioning
model.
E
So
the
it
now
defines
the
measurement
modes,
so
one
way
is
that
replies
sent
out
of
been
an
IP
UDP
pass
to
a
measurement
mode
reply
sent
using
the
return
path.
Tlv
that's
received
in
the
query
message
and
in
the
loopback
measurement
mode,
the
probe
query
message
contained
the
return
path
in
the
header
of
the
packet.
So
basically
packets
are
not
pointed
on
the
responder
side.
E
E
So
this
is
the
the
message
format
defined
for
direct
mode
loss
measurement,
so
they
have
fixed.
The
main
characteristic
is
that
it's
a
fixed
length
and
they
have
fixed
a
location
for
their
counters,
so
it's
it
makes
it
Hardware
friendly
and
it's
aligned
with
the
DM
message
format
as
well.
So
if
Hardware
knows
where
to
come,
timestamp
hardware
will
also
know
where
to
put
the
counter.
E
So
our
next
steps.
We
welcome
your
comments
and
suggestions.
This
has
been
implemented
and
deployed.
We
presented
this
talk
in
a
ppm
and
they
asked
us
to
keep
them
in
the
loop
about
the
milestones.
We
do
believe
that
the
draftee
is
ready
for
working
or
production
in
spring
Thanks.
Any
comments,
questions.
A
Okay,
nobody
so
discuss
with
a
with
rod.
We
have,
as
you
have
noticed,
we
have
a
few
back
roads
who
drove
the
back
of
a
truck
to
our
doctrine.
We
are
working
to
advertise
that
on
a
public
webpage
that
it's
more
or
it's
clarified
with
everyone,
but
we'll
consider
that.
Definitely
there
is
a
support
in
in
the
room
and
we
need
to
confirm
Alonso
on
the
list.
Okay,
thank
you.
Thank
you.
C
Thank
you
know,
so
this
is
a
document
that
we
presented
last
time
in
spring
working
group
and
at
this
ITF
as
following
spring
working.
Sorry
last
time
in
MPLS,
working
group
and
this
time,
as
well
as
in
MPLS
working
group,
is
certainly
something
that
we're
looking
at
the
chairs
in
terms
of
the
home
for
this
work,
the
document
talks
about
generic
FAC
TVs
for
less
beeping
and
I
am
presenting
on
behalf
of
my
quarters.
C
So
problem
statement
goes
as
follows:
in
MPLS
for
O&M
purposes,
for
ping
and
traceroute,
we
define
effect
stack
sub
Calvi
and
to
date,
every
time
we
define
a
new
set
type,
we
have
to
go
and
define
a
new
stack,
tell
me
and
define
procedure
associated
with
it.
It
the
way
that
do
I
name,
because
I
define
that
it.
C
It
is
a
complexity
from
the
point
of
view
of
ingress,
note,
being
able
to
gather
this
information
and
then
carry
in
the
packet
four
probe
for
onn
probe,
and
it's
also
in
well
has
complexity
for
the
egress,
because
the
has
to
validate
a
lot
more
information
read
the
information
gather.
The
information
do
lot
more
processing
again
that
information.
C
C
C
As
the
motivation,
we
started
to
see
that,
okay,
maybe
you
look
at
the
SR
CID
data
model
and
based
on
the
data
model,
define
one
fact
that
consistent
with
the
data
model
and
we
focus
on
data,
plane,
validation
and
and
leave
the
data
plane
to
controller
and
consistency
check
King
to
a
local
procedure
at
the
node,
every
node
has
procedures
to
locally
check
its
own
data
plane
with
control
plane.
No
remote
node
has
to
tell
them
oh
by
the
way.
Can
you
check
this?
C
C
So
we
included
that
as
well.
So
your
Eric
fact
is
label
fact
is
nothing,
but
what
is
yes
all
said
and
who
has
assigned
it
and
optionally,
adding
the
endpoint
address
and
and
Syd?
Obviously,
as
did
the
label
the
20
bit
label
and
the
assignor
is
the
node
ID
of
the
designer
that
has
signed
it
and
unless
the
endpoint
may
be
set
to
zero
when
the
initiator
doesn't
know
who
would
be
the
endpoint
but
maybe
set
to
the
end
point,
let's
say
binding
Syd
if
the
ping
is
for
the
binding,
said
and
endpoint
is
known.
C
C
So
here
we
say
that
16000
0,
0
X
is
the
H,
is
the
prefix
it
associated
with
algo
0
and
it
and
the
example
here
is
using
the
same
fact
to
address
ping
or
tray
straight
prefix
it
for
elbow
128
or
any
other
elbow.
Today
we
don't
have
this
procedure
today.
We
don't
have
this
coverage,
but
the
way
it
will
work
here
is
that
we
will
the
initiator
1
will
carry
the
facts.
So
let's
say
take
one
example
at
time,
so
we
want
to
send
to
paint
prefix
it
with
algo
0.
C
C
C
Now
the
same
procedure
would
apply
for
flexible,
based
prefix
set
labels,
so
that
says:
16
128
it
with
a
less
the
end
point
eight
and
the
assignor
are.
It
will
well
exactly
same
way
as
for
excel
go,
so
we
can
have
coverage
across
the
Fraxel
group
without
having
to
define
a
new
set
type
or
extending
the
flex
elbow
ID
in
the
existing
track.
C
So
if
you
take
thing
with
the
parallel
agency
said,
the
payroll
agency
said
case
today
is
not
covered
by
RFC,
because
the
initiator
one
does
not
know
which
interface
the
parisian
see
will
hash
to.
So
it
is
a
local
madras,
seven,
so
initiator
from
one
would
not
know
to.
Hence
it
put
zero
zero
zero,
as
as
the
as
the
target
value
for
the
for
the
interface
where
it
would
be
receiving
the
packet
here
is
with
this.
C
Factoid
would
be
easy,
because
what
will
be
carried
in
the
in
the
fact
is
the
city
which
is
the
agency
said,
and
then,
in
this
case,
parallel
agencies
in
1993
78
and
the
signer,
which
is
are
seven,
and
the
LSP
point
in
this
case
could
be
are
eight,
but
it
could
be
zero.
Zero,
zero
now,
regardless
of
where
the
packet
will
go
to
when
the
packet
comes
to
eight
eight
has
a
local
table
that
basically
say
that.
Oh,
this
is
not
my
local
label,
but
I
receive
it
from
my
neighbor.
C
Is
the
assignor
my
upstream
neighbor?
Yes?
Does?
This
label
belongs
to
the
of
him?
Neighbor
agency
said
on
the
interface
I
receive
the
packet.
Yes,
so
it's
good.
Otherwise
it
feels
it
now.
This
is
improvement
from
the
current
condition,
because
in
current
condition,
because
they
put
the
link
ID
to
be
zero,
zero,
zero,
it
can
receive
the
packet
from
any
into
any
other
neighbor
and
will
still
think
that
it
has.
It
has
reached
from
the
right
neighbor
and
the
validation
will
still
pass.
C
So
this
is
strict
compared
to
that
and
then
the
same
thing
will
work
in
the
case
when
you
have
parallel
agency
to
different
neighbor
in
this
case,
teledensity
to
neighbor,
8
or
neighbor
88,
and
the
agency
said
associated
with
it.
97
78
7
will
do
hashing,
based
on
whatever
load-balancing
parameter
that
it
carries.
Packet
may
end
at
8
or
88.
C
Then
thing
just
to
recap
is
that
motivation
here
is
to
really
go
and
simplify
Zack
type
for
onm
operation,
for
segment
outing
with
this
and
then
in
such
a
way
that
we
can
cover
all
different
seat
types
that
are
today
or
in
future,
and
we
can
extend
this
for,
for.
We
can
easily
support
it
once
and
then
we
don't
have
to
do
software
upgrade
to
define
in
fact,
I
have
etcetera.
C
B
B
I
I
I
B
I
B
I
B
C
Is
for
then,
the
types
that
are
not
defined
yet
less
safe
as
a
flex
I'll
go
if
not
covered
by
name,
serious
or
covered.
There
are
other
use
cases,
but
that
are
not
covered,
but
what
is
already
defined
and
in
RFC,
if
is
a
procedure
that
will
continue
to
be
like
this
is
in
your
depreciation
here,
but.
B
B
J
B
C
But
I
just
do
different
things
here,
like
there
are
additional
features
that
come
with
the
existing
fact
type
and
those
features
are
welcome
to
use
for
people
so
I
think
having
an
option
is
better
than
depreciation.
I,
pick
more
for
that
and
and
share
theirs
on
the
mic.
Like
some
discussion,
we
share
there
as
well
as
going
on
for
the
new
definitions
that
are
coming
and
we're
cooperating.
So
that's
why
we're
not
asking
for
adoption
here,
but
more
feedback
and
working
with
with
with
other
folks,
like
shadow
and
and
Tariq
and
others
yeah.
F
C
K
The
signer
is
correct
and
we
reach
the
end
point,
but
you
don't
validate
what
this
sit
is
assigned
to
it's
missing
and
and
the
last
statement
you
said
you
know
I'm
just
you
know
capturing
what
you
said
this
we're
doing
data
plane,
we're
really
validation
of
you
know.
The
label
is
correct
or
actually
the
label
is
allocated
correctly
on.
The
egress
is
allocated
correctly
or
not.
So
it
could
be
an
option
that
you
want
to
just
validate
that
the
label
is
as
valid
and
but
I.
K
L
M
M
M
But
in
control,
but
then
why
would
we
call
it
generic?
Because
if
we
have
other
facts
that
already
being
specified-
and
we
introduced
fact
that
I
agree
it's
more
flexible
and
a
future-proof?
Why
wouldn't
we
consider
at
least
consider
and
discuss
whether
just
make
it
through
the
generic
for
a
sir
yeah.
C
K
Hello,
my
name
is
tarick,
I'm
going
to
introduce
to
you
segmenting
/
forwarding
adjacency
links,
so
this
work
is
a
collaboration
between
myself,
my
colleagues,
Pavan
and
Kobi
I'll
go
over
the
motivation.
Give
you
a
little
bit
of
overview,
I'm
sure,
you've
heard
about
these
FA
links
in
the
past
and
how
we
can
enable
segments
routing
on
them
and
how
we
can
use
them
and
we'll
talk
about
the
next
steps.
K
So
we
wanted
to
improve
scalability
in
the
SR
Network
by
creating
subpaths,
as
our
sub
paths
and
the
concatenation
of
these
sub
paths
will
give
you
the
end-to-end
path.
We
want
to
reduce
the
size
of
the
topology
by
you
know,
creating
these
summarized
or
compressed
links,
if
you
can
think
of,
we
want
to
allow
seamless,
reroute
and
we
optimization
of
the
sub
path.
That
does
not
impact
the
end-to-end
SR
path
and
we
want
to
allow
stitching
of
domains
that
are
running
different
technologies.
K
K
K
Such
links
have
the
endpoints
as
the
ingress
and
egress
of
the
fa
LSP
the
forwarding
over
the
fa
LSP
is
bound
to
one
or
more
F.
Sorry,
the
forwarding
over
the
fa
link
is
to
one
or
more
fals,
be
the
fals
visa.
As
I
mentioned,
they
can
be
signaled
using
different
technologies.
It
can
be
SR
or
RSVP
or
static
LSP.
For
that
matter
they
can
have
traffic
engineering,
attributes
or
parameters,
and
we
can
obviously
assign
them
segment
routing
segments
and
and
enable
steering
over
them
in
terms
of
creation
and
management
of
such
logical
fa
links.
K
We
we
think
they
can
be
configured
with
operator
can
decide.
They
want
to
configure
such
an
F.
A
link
or
a
controller
can
use
any
northbound
interface
or
of
our
signaling
protocol
like
P,
subnet,
confit,
RPC,
your
favorite
and
automatically
or
dynamically
create
such
links
in
terms
of
the
state.
If
the
underlying
LSP,
or
at
least
one
of
them
is
valid
and
signaled,
then
the
FHFA
link
is
operational
and
ready
to
accept
traffic.
K
If
a
link,
the
attributes,
they
can
be
dynamically
derived
from
the
underlying
LSPs
or
they
can
be
overridden
by
the
operator
statically
now
they
have
a
link,
it
can
be
numbered
or
unnumbered
for
numbered,
obviously
will
be
assigning
addresses
to
them
local
and
remote.
For
unnumbered,
there
will
be
an
identifier
unique
on
the
on
each
endpoint.
K
Typically,
depending
on
the
underlying
signaling
of
the
LSP,
you
might
be
able
to
gather
the
remote
ID
for
each
side.
For
example,
in
our
an
RSVP,
there
is
a
way
to
collect
it
in
SR.
There's
no
such
way
to
identify
the
remote
ID.
It's
enough
for
Anna
numbered
link
unidirectional
to
have
the
local
and
remote
no
descriptors,
as
well
as
the
local
identifier,
to
pin
down
the
link.
Now
there
are
these
symmetrical
FA
links.
Basically,
there
are
links
that
are
Co
routed.
K
K
K
K
We
can
assign
any
JCC
segments
to
them
or
parallel
adjacency
segment,
where
your
load-balancing
equally
or
unequally
for
BGP
FA
links
such
that
they're
connecting
to
be
GB
nodes
speakers
we
can
assign
an
JCC
segment
that
takes
us
to
specific
next
hop
BGP
next
op
or
a
a
peer
note
said
it
load
balances
on
multiple
possible
paths
to
the
same
node
and
the
peer
set
can
actually
take
you
to
multiple
different
next
hops
in
the
set.
So
all
these
are
possible.
They're
all
defined
already
with
just
a
reiterating
that
they
are
possible.
K
There
is
an
application
for
this
to
enter
the
main
I
slightly
hinted
to
it.
We
can.
We
can
provide
the
connectivity
ingress
to
egress
using
such
fa
links
once
we
have
that
the
end-to-end
is
just
a
concatenation
of
this
fals
piece
to
provide
you
the
intent
path.
You
can
do
a
full
mesh
of
FAA
links
between
the
egress
and
ingress
border
notice,
or
you
can
do
a
hub-and-spoke
type
of
FA
links
and
export
that,
and
in
fact
you
can
actually
decorate
these
links
with
certain
tags
and
we're
thinking
of
ug
or
using
BTW
pls
Maps.
K
K
So
in
terms
of
next
steps,
this
currently
is
an
informational
draft.
I
did
mention
the
remote
ID
is
useful
to
identify
the
core
outed
FA
link.
We
are
investigating
this
at
the
moment.
We
might
push
a
new
extension
and
if
that
happens,
we
will
transition
at
the
standards
track.
We,
this
is
zero
zero
version.
We
do
for
a
solicit
feedback
from
the
working
group
and,
if
you
have
any
question,
I'm
I'll
be
happy
to
answer.
K
So
yes,
and
no,
if
a
link
has
an
acid
associated
to
it,
it
can
be
a
binding
set
of
the
underlying
as
our
policy
it
can
be.
But
if
the
underlying
is
an
RSVP
LSP,
it
would
be
a
binding
label
or
it
can
be
multiple
LS
fees,
in
fact,
so
I
didn't
mention
that,
but
in
the
draft
they're
mentioned
but
binding
said,
is
one
option.
Yes,
what
I
mean
was.
E
K
So
the
idea
is
that
we
need
advertise
this
in
BGP
LS
as
a
link
and
import
it
into
the
Ted
and
allow
anything
that
we
can
do
and
on
any
BGP.
Ls
advertisement
like
tagging
on
the
link
and
Al
are
I
and
make
it
appear
in
the
Ted
right
now
the
binding
set
by
itself.
There
is
no
standard
way
how
you
push
it
into
the
Ted,
so.
N
K
N
K
N
K
N
I
E
I
Tarik
to
add
to
what
drew
was
bringing
I
wanted
to
bring
to
our
attention
that
we
have
a
draft
which
is
a
working
group
document
for
exporting
such
links
via
bt
pls.
The
only
difference
is
it's
not
done
as
a
link
NLRA,
it's
done
as
a
tea
policy
and
indirectly
but
functionally
it
can
give
you
the
same
thing
so.
K
I
O
K
J
A
J
A
So
UNIX
is
G,
so
we
have
three
slots
related
to
network
slicing,
just
to
be
aware
that
there
is
a
work
going
on
in
cheese,
working
group
related
to
network
slicing,
so
I'm,
assuming
that
we
are
going
to
work
on
the
generic
work
and
I've
yesterdays
have
several
description
related
to
design
Tina,
so
since
I
are
moving
in
T's
working
group
and
finally,
with
a
bit
out
of
short
of
time,
so
please
focus
on
spring
related
stuff
or
the
rather
than
arrest
generic
slides,
which
are
going
to
be
working
Jesus.
Thank
you.
P
Okay,
here
are
some
analysis
about
the
current
segment
routing
using
similarity
was
designed
for
the
South's
regime
and
the
see
the
least
is
used
to
specify
the
nodes
and
links
should
be
the
traffic
rule
along
the
path
and
the
current
status.
The
services
will
share
the
same
group
of
sees
and
currently
they
cannot
differentiate
Rafic
using
the
seeds.
But
the
question
is
a
second
routing:
can
we
create,
as
it's
customized
the
virtual
networks,
but
the
second
question
is
setting
the
routing.
P
Currently,
it
does
now
to
take
the
resource
reservation
into
the
consideration,
so
that
this
is
just
a
steal
of
traffic
along
the
past.
It's
a
link
level,
granularity
traffic
steering
and
the
basically,
it
can
rely
on
the
tips
of
us
for
the
traffic
differentiation.
It
may
be
good
enough
for
some
existing
services,
but
always
think
there
are
gaps
and
the
challenges
for
them.
Services
like
the
5g
critical
applications,
so
that
the
other
question
is
how
can
we
achieve
the
guaranteed
are
more
predictable,
SLA
with
similarity.
P
Okay,
here
is
the
proposal
in
this
draft.
Basically,
we
introduced
the
topology
and
resource
information
into
the
same
semantics.
We
can
use
the
seeds
to
identify
the
topology
and
the
data
playing,
so
we
can
use
differences
to
identify
the
same
node,
a
link
in
the
different
network,
topologies
au
notation.
We
can
also
use
the
C's
to
represent
a
subset
of
natural
resource
allocated
to
a
particular
virtual
network
or
network
slice.
P
We
see
this
is
an
abstraction
of
the
different
mechanisms
in
the
end
early
to
provide
you
to
guarantee
the
resource,
isolation
and
it'll
guarantee
the
performance.
So
this
is
a
good
abstraction
of
different
mechanisms
in
underlay
and
with
this
mechanism,
we
can
use
these
different
guru
prophecies
to
create
a
different
network
slices
or
virtual
networks
and
which
lists
in
different
virtual
networks.
The
traffic
will
go
through
the
isolated
topology
with
isolated
resource
so
that
we
can
keep
other
so
traffic
in
different
slices,
separated
and
not
interfere.
P
Each
other
okay
here
is
a
similar
mechanism
for
SR,
v6
I.
Think
every
six.
We
need
to
use
different
two
locators
to
identify
the
different
network
topologies
and
on
the
same
node
or
links,
and
the
C's
I
thought
we
succeed
basically
to
be
inherits
the
topology
identification
from
the
locator
and,
in
addition,
we
can
also
add
this
resource
semantics
to
the
locator
and
C's
to
represent
the
subset
of
resource
in
the
an
array.
Here
are
the
updates,
since
last
presentation,
ITF,
we
had
new,
closers
and
updated
affiliations
of
some
authors.
P
We
also
add
the
references
to
the
narrow,
slicing
definition
3gpp
and
also
doing
the
beeping
plus
framework,
and
this
one.
We
also
clarify
that
the
C's
can
identify
the
topology
and
may
represent
the
Associated
resource
so
that
we
provide
a
unified,
a
diplomat
nism
for
both
soft
and
as
hard
isolation.
They
also
resolve
the
receivers
review
comments
and
there
are
some
editorial
changes
to
make
it
more
readable
and
updates
to
the
reference.
P
Okay
for
the
next
steps,
I
think
we
received
some
further
comments,
which
are
we
can
solve
them
in
the
next
version
and
we
think
the
framework
has
been
adopted
in
TS
and
it's
gets
more
true,
a
towards
the
working
group
last
call
and
so
for
this
document.
We
already
work
on
this
since
early
2018
and
the
contents
of
this
draft
is
also
getting
stable.
We
already
see
many
interest
from
operators
to
have
a
signal
routing
based
natural
slicing
solution.
P
A
P
Can
be
relatively
hot
from
the
work
in
the
design
team
because
we
already
have
the
framework
of
the
vaping
past
in
the
teeth,
and
we
think
this
is
a
solution
for
the
iam
class
and
one
of
the
use
case
is
natural
slicing.
So
this
is
based
on
this
current
current
status.
Maybe
we
can
no
need
to
rely
on
this
progressing.
The
design
team
I
also.
A
Am
at
first
and
yes
also
some
comments
on
the
draft
starting
with
an
Emma,
so
crew
sees
a
bit
of
marketing
its
could
be
useful
to
know
what
is
added.
So
my
guess
is
a
network
partitioning
VPN
against
is
not
really
the
scope
of
the
working
group
and
I.
Not
sure
it
worked
is
about
VPNs,
so
maybe
the
introduction
on
the
title
and
also
best
fit
so
there
are
different
terms
used
and
the
shoulder
means
the
same
single
notes
such
as
network
slicing
VPN,
yes,
will
network.
As
a
consequence,
it's
not
very
clear.
P
P
J
So,
as
a
member
of
design
team,
it's
not
the
intention
to
define
technology
as
such.
The
focus
is
northbound
API
and
programmatic
interface
between
higher-level
business
logic
and
underlying
controller
catherine
de
tomate
associate
with
it.
So
we
are
not
going
to
work
on
this
specifically
and
since
certainly
there
is
do
we
are
trying
to
align
on
meaning
of
transporter
networks.
Lives
would
be
actually
good
to
kind
of
Express
same
meaning
when
we
use
same
words
and
please
think
about
it,
you're
part
of
design
team
as
well.
So
we
we
work
together
right,
nothing.
J
P
Q
Phone
teen
farmer,
trainer
mobile
now
fabric
you,
my
family,
is
about
to
into
the
commercial
yesterday
tree
in
China
Mobile.
We
have
ratified
the
many
other
case,
also
Network
us
lysine
satirize.
Last
month
great,
the
idea
with
your
video
are
into
the
mobile
no
surgery
and
so
on.
Kudus
record
cooler
with
I
was
helping
you
achieve
that.
They
are
many
always
asking
you
for
network
us.
Try
seeing
these
nice
draft
provides
the
eye
surface.
The
teat,
applying
I
think
a
third
day.
The
good
idea,
Thank
You.
R
Hi
I'm
Afra,
Maliki
I
agree
with
Jeff
I'm,
a
member
of
the
SLS
Dillon
team.
This
is
out
of
the
scope
of
the
design
team
and
we
have
on
neurotechnology
about
I
saw
later
we
will
do
the
prairies
come
here
with
yes,
I
have
a
three
announced
the
requestors.
The
first
is
compared
with
your
activity,
our
jobs
as
soon
as
the
air
is
no
more
new
technology.
I,
don't
know
what
she
young
meaning
of
this
disrupt
as
a
standard
structure,
and
it's
a
second
question
is
for
ITP
extensions.
R
R
P
The
list
just
I
want
to
maybe
emphasize
that
this
is
a
draft
about,
as
Sigma
out
in
theta
playing
the
controlling
belongs
to
other
separate
documents.
This
is
we
want
to
show
the
extension
to
the
semantics
of
the
see
we
got
previously
is
just
to
identify
no
darlink.
We
give
the
finer
granularity
of
this
resources
on
the
link
and
or
the
virtual
topology
on
the
node.
So
this
is
the
extension
to
the
data
plane.
A
S
E
In
T's
and
one
watching
scope
is
the
requirements,
as
well
as
the
framework
for
network
slicing,
so
we
need
to
make
some
progress
and
I.
Think
G
is
also
part
of
the
networks
like
the
same
design
team
as
well.
So
G
knows
very
well
that
we
need
to
make
progress
on
the
requirements
as
well
as
the
framework,
so
I
think
we,
we
agree,
agree
with
Bruno
that
let's
wait
a
little
bit
more
to
solidify
the
those
aspects.
M
P
A
Right,
nice
didn't
Curtis,
I,
know
we're
late
Zafar,
so
we
need
to
discuss
on
the
list
between
definitely
need
to
sync,
with
this
working
group
on,
you
need
to
refine
the
scope
of
your
draft
on
make
the
title
abstractor
introduction
in
line
with
the
scope.
So
if
it's
about
the
piano,
maybe
it's
not
for
spring,
if
it's
not
about
the
piano,
maybe
refrigerator
partitioning
network
resources,
I
have
some
other
commands,
but
I'll
Alice
a
little.
Thank
you.
C
Okay,
try
to
do
it
in
five
minutes,
so
the
objective
of
this
draft
is
to
outline
a
start
by
outlining
the
building
block
for
slicing
that
already
exists
in
SR
Network
and
then
see
the
gaps
and
how
we
can
address
those
gaps.
So
it's
not
this
elaborate.
What
are
the
building
blocks
and
how
we
can
use
it
for
network
slicing?
So
here's
the
list
initial
list,
so
we
have
support
for
SR
policies,
vigor
without
flex
I'll
go.
Then
we
have
notion
of
flux.
I'll
go
now.
C
First
I
see
we
also
have
notion
of
doing
TLS
a
within
a
slice
under
less
slice.
We
have
notion
of
carrying
various
sleeping
services.
That
is
also
provides
separation
among
the
customers
and
then
support
for
service
programming,
onn
p.m.
and
the
QoS
aspects
and
orchestration.
So
these
are
some
of
the
building
block
up.
I
have
shorten
time
so
go
fast,
so,
let's
think
of
as
a
policy
first.
C
We
have
ability
to
do
path,
calculation
for
different
optimization
criteria
and
ability
to
isolate
certain
type
of
path
from
the
other.
We
have
support
for
the
ODN
and
support
for
automatic
steering
into
the
SL
policy,
and
SR
policies
can
spend
multiple
domains,
and
this
is
something
which
has
been
widely
deployed
in
segment
outing
Network.
C
Then
we
have
notion
of
flexible
algorithms,
which
associate
intent
to
a
prefix
it
and
it
allows
IDP
to
compute
path
constraint,
part
based
on
the
intent,
so
let's
say
low,
latency
flex,
Algol
hooligans
is
right
and
then
IDP
is
able
to
run
csdf
to
satisfy
the
intent
in
in
in,
and
it
is
a
way
to
create
provision
networks
price
with
different
intent.
Then
we
have
notion
of
LFA
and
micro
avoidance,
but
the
beauty
here
is
that
when
we
do
TL
FA
for
flexible
I'll
go
the
backup
part
only
uses
the
resources
that
belongs
with
there.
C
Then
we
have
believed
to
carry
very
different
type
of
SR
VPN
services
like
l3
VPN,
we
PWS,
VPLS
and
even
TN
services
that
provide
various
level
of
separation.
Of
services
from
a
customer
point
of
view
or
different
customers
can
can
have
different
different
type
of
service
differentiation
or
service
isolation
within
the
within
the
from
a
common
underlay
notion
of
service
programming
allows
customers
to
do
traffic
engineering
or
in
ten
days,
routing
not
only
through
topological
segments,
but
also
through
programmable
service
programming,
like
network
appliances
and
support
for
the
NF
ease
and
service
function.
A
Yes,
actually,
you
existed
you're
out,
but
I
think
your
first
slide
was
very
interesting
yeah,
it's
a
good
summary
of
all
the
tools
we
have
in
spring
to
fulfill
some
requirement
from
Network
slicing.
Maybe
one
thing
which
is
missing
is
what
is
missing
actually
in
spring,
what
is
missing,
what
tool
is
missing
in
spec
or
what
tools
are
missing
in
spring
to
fulfill
the
requirements?
So
again,
we
may
need
to
to
wait
for
all
the
feedback
from
his
design
team,
but
I
think
it's
a
good
start
to
see
what
we
are.
A
A
J
Similar
to
previous
command,
it's
a
greater
variant
routing
can
do,
but
it's
neither
correct
nor
complete
definition
of
what
network
slice
is
so
yes,
terminology
correctly,
it
could
be
used
a
subset
of
what
we
are
trying
to
do
is
network
slicing.
But
this
is
not
the
network
sizing
and
don't
call
it
this
way.
It's
really
wrong.
So.
L
A
A
Q
Read
about
the
an
authority
from
Hanna
mobile
now,
I
will
talk
about
Michael,
Annett,
Walker,
Selassie
you'll
see
a
second
mini
routine.
The
into
Internet
workers
rights
include
the
design.
The
pious
rated
PG
include
the
ranks,
light
red
cows,
land
and
cow
slaughter.
Now
I
will
focus
on
the
red
powder.
Network
catalyze,
the
F
I
will
in
a
major
the
cramming.
The
first
of
all
is
that
educator
what
children
works.
The
second
one
is
the
into
engines
Lyceum.
The
third
one
is
unifying
the
ESI
the
first
and
the
time
is
the
traffic
engineering.
Q
The
last
my
is
last
relate
to
as
well
as
the
last
training
results
of
Hattie
school.
They
are
also
some
Disney
ID,
the
first,
but
they
could
allowed
to
meet
all
of
the
a
parameter.
So
I
will
reduce
the
new
base
to
the
new,
the
new
slice
slice
paste
at
the
first,
how
to
administer
it
her
when
he
stains
the
added
affirm.
Our
AI,
for
short,
is
the
appreciate
the
work,
your
network
added
fiction.
It
could
be
used
as
a
trade
panel
and
workers
lives.
I
didn't
turn.
Q
It
turn
indicates
that
apology,
computing
and
the
storage
resources
also
the
educator.
What
he'll
network
yeah
is
the
identifier
of
the
dedicated
to
virtual
network
for
the
slice
it
kudus
about
into
end
slicing
a
decoder
also
identifying
the
Univ
an
in
at
the
arm.
It
is
a
wonderful
knocking
trying
to
criteria
of
color
template
and
the
tan
color
complains,
with
a
I
provide
more
exporting
to
it.
A
determinant
also
the
eyes
18
to
main
touching
in
Altoona
teaching,
kristopher
wooten,
not
aware
as
prefix
the
way
opa-locka
teacher
I.
Q
So
there
is
no
modification
for
the
corrodium
plan.
Where
is
the
Adama
Alfred
in
solution?
The
collectors
who
collect
the
collector
the
controller,
collects
the
results
information
with
the
AI,
while
PP
our
eyes,
and
establish
the
Congress
bounty
as
our
what
he'll
network
and
the
calculator
the
past.
According
to
the
color
with
the
AI
there,
the
truth,
the
politic
read
and
write,
and
also
a
is,
could
it
be?
A
defect
are
for
trade
apart
and
networks
alliance
results
at
the
first,
the
this
responsibles
soft
and
hard
and
hard
a
solution
for
the
lace
rate.
Q
Interface
lies
a
solution
in
a
GP
domain.
Each
neighbors
are
a
nevertheless
raining
could
be
kept
in
court
with
the
AI
information
and
recognize
them
among
the
addition
across
the
Aegean
Linga
state
database
narrow
to
contain
unless
readings
with
the
a
are
two
spouts.
The
te
pass
computation
taken
account
also,
the
aircrack
area
for
the
line
tool
in
the
faceless
as
a
nation
yet
relied
to
never
link
of
Orleans,
repellent
police
reported
link
could
be
crapping
word
with
the
AI
and
the
artisans
it.
Q
The
line
to
memory
in
cook
to
the
P
of
legacy
of
life
is
eternal.
All
his
Montano
and
I
know
other
such
as
other
other
in
the
face
type
in
the
curriculum.
The
routine
protocol
packet
following
the
last
readings,
select
the
line
tool,
never
link
with
the
highest
superiority.
At
the
same
time,
in
the
following
in
the
forwarding
applying
the
packet
repel
intruders
versus
vehicle,
what
he'll
network
we're
passed
along
the
line,
tool
and
burning
with
the
specifically
reduce.
Q
S
G
A
A
S
A
Sorry
for
the
next
slide
less
presentation
a
slicer,
much
improved.
There
was
an
audio
description
in
the
new
MPs
about
working
group,
so
it
has
been
presented,
but
the
drafts
that
means
now
to
be
aligned
with
the
slides
because
I
don't
think
they
are
aligner.
So
maybe
there
is
a
work
to
be
done
on
the
draft
following
a
discussion
during
this
meeting.