►
From YouTube: IETF103-SPRING-20181107-0900
Description
SPRING meeting session at IETF103
2018/11/07 0900
https://datatracker.ietf.org/meeting/103/proceedings/
B
So
if
you,
if
you're
not
looking
for
spring,
this
is
probably
a
good
time
to
leave
the
room.
If
you
are
looking
for
spring,
we'll
start
it.
This
is
the
note.
Well,
it's
Wednesday
I
hope.
You've
seen
it
already,
if
not
note
it
well,
we
will
need
both
a
jabber
scribe
and
a
minute
taker
and
we're
blocking
on
continuing
until
we
have
both.
So
do
we
have
any
volunteers,
4
minutes,
I'll,
try,
add
some
things,
but
another
minute
taker
would
be
useful.
B
B
B
So
last
time
and
a
few
for
the
last
few
ITF,
so
we've
been
going
through
a
recharter.
We
are
now
done
with
our
recharter,
who
we
you
can
see.
The
Charter
has
been
discussed
in
pale
length
on
the
mailing
list,
so
hopefully
it's
not
a
surprise
to
anybody
and
in
general,
we're
going
to
give
preference
in
meetings
to
drafts
that
address
the
charter
items
rather
than
those
that
are
outside
of
the
Charter
I.
B
Think,
there's
not
a
huge
amount
that
there
isn't
if
there
isn't
included
in
the
Charter
that
we
discussed
working
on
one
notes,
I
think
about
the
Charter
is
that
in
general
it
requires
a
lot
of
a
lot
of
collaboration
across
working
groups.
So
it
would
be
very
useful
to
have
a
view
if
you
have
a
draft
elsewhere.
That's
that
needs
some
attention
from
spring,
even
if
the
work
is
progressed
in
that
the
working
group
please
make
sure
to
include
this
spring,
alias
just
to
cover
anyone.
That's
interested
in
the
area.
B
There
was
some
discussion
around
the
implementation
requirements
policy
that
will
use
per
working
group,
Martin
esters,
to
have
a
look
at
this.
It's
was
discussed
in
the
plenary
f-102
and
then
by
the
ASG.
The
implementation
policy
is
essentially.
What
do
we
require
in
terms
of
plantations
before
progressing
a
document
it's
defined
on
a
working
group
by
working
group
basis.
There
are
a
number
of
different
suggested
implementation
policies,
ranging
from
we
need
to
interoperable
implementations
and
detailed
reports
of
them
down
to
we
don't
care.
B
There
is
another
thing
that
we
were
asked
to
highlight,
which
is
there
has
been
since
May
2017.
There
has
been
an
updated
requirements,
keywords
document
and
replacing
2119
and
the
ihj.
You
still
sees
a
bunch
of
documents
that
have
2119
in
them
rather
than
h,
174,
you
should
have
data
and
use
the
boilerplate
from
8174,
rather
than
the
bullet
form
2119.
Please
do
this.
If
you
are
the
editor
or
author
of
a
draft
in
spring,
so
bit
of
information
the
agenda
again,
we
had
significantly
more
requests
for
time
than
we
have
time
in
the
meeting.
B
I
think
this
time
it
was
something
like
200
minutes
for
four
hour
two
hour
slot.
This
means
that
we
have
to
disappoint
a
bunch
of
people
and
say
that
they
can't
present
in
the
meeting.
Our
preference
is
generally
given
to
working
group
items
if
they
have
been
through
some
update.
We've
got
no
working
group
items
on
the
agenda
today,
because
really
there
was
no
update
on
those
during
the
interim
period
between
meetings.
B
That's
kind
of
disappointing,
I
think
I
think
we
need
to
finish
the
dockstar
working
group
items
before
we
necessarily
progress
a
lot
of
other
things.
Then
we
give
preference
to
things
that
have
been
discussed
on
the
list,
which
has
been
again
relatively
relatively
quiet.
Please
discuss
your
drafts
on
the
list.
B
It
gives
you
more
chance
to
get
feedback
from
folks
who,
on
here
it
gets
more
detailed
feedback,
especially
what
we've
given
the
time
that
we
have
here
and
it
will
guarantee
it
will
give
you
a
better
chance
of
getting
a
slot
on
the
agenda
when
you,
when
you
are
presenting
things,
we
prefer,
as
we've
said
before,
to
use
these
meetings
for
time
for
discussing
open
items
rather
than
giving
overviews
of
new
traps.
This
time
we've
also
tried
to
accommodate
presentations
that
over
first
time,
I
think
last
time
Rakesh
had
30
slides
in
about
30
seconds.
B
So
we
try
to
also
be
include
that
at
this
time.
So
this
is
our
agenda.
For
today
it's
pretty
packed.
We
have
about
six
minutes,
speaker
shuffling,
which
I'm
going
to
imagine
that
we'll
use
most
of
the
slots
are
pretty
short.
We've
tried,
I
think
we'll
end
up
overflowing.
Please,
please
focus
your
presentations
on
things
that
we
need
to
address
inside
of
the
working
group,
rather
than
the
need
and
updates
which
can
be
gleaned
from
reading
the
document.
B
B
The
EP
draft
is
on
the
idea.
Draft
the
LDP
Interop
is
is
blocked
on
SR
MPLS,
SR
MPLS
itself.
The
authors
are
addressing
comments
so
as
a
forest
and
great
releasing
the
editors
and
and
helping
drive
that
work.
We're
aiming
to
get
those
comments
addressed
pretty
soon
and
get
back
to
the
lady
on
this.
This
document
we
really
need
to
finish
I
think
it
there's
a
bunch
of
spring
work
that
really
depends
on
this,
this
getting
there
and
it's
implemented
and
has
been
stable
for
some
time.
B
B
It's
fairly
substantive,
but
in
terms
of
what
it
includes
the
yang
model
that
for
spring
we
had
a
review
from
Lada
who,
as
the
young
doctor,
so
thank
you
very
much
for
that
and
I
think
that
it's
those
comments
need
to
be
addressed
and
then
we're
probably
ready
to
try
and
progress
this
after
this
meeting
will
sync
up
with
the
authors
on
that,
and
then
we
have
the
MSD
C
draft.
This
has
been
undergoing
a
lot
of
discussion
after
ISU
review,
particularly
on
some
some
wording.
Around
hashing
and
traffic
placement
of
flow
looks.
B
If
you,
if
you
read
in
certain
ways,
I
think
we
should
focus
our
drafts
on
things
that
need
to
be
addressed
in
spring
rather
than
trying
to
overstep
and
cover
every
particular
use
case,
so
both
that
abs
and
the
ASG
and
burner
myself
have
had
some
input
on
this.
So
hopefully
this
will
be.
We
can
clear
up
this.
It's
been
kind
of
nine
months
addressing
this
particular
comment,
so
we
really
need
to
close
there
and
finish
off
with
this
draft
in
terms
of
working
group
adoption.
B
We
have
a
couple
of
drafts,
a
couple
of
areas
that
I
think
we've
had
some
discussion
in
a
couple
of
meetings
and
we
have
interesting
competing
drafts
so
between
the
on
BFD.
We
have
these
two
drafts,
the
one
press,
our
policy
and
one
for
more
generally
spring
VFD
we've
asked
some:
we've
asked
the
others
to
consider
merging
or
at
least
ensure
that
these
two
drafts
are
consistent
with
each
other.
B
The
working
group
would
like
to
progress,
and
so
I
I
think
Bruno
and
myself
will
kick
off
the
discussion
on
the
list
and
then
we'll
we'll
reserve
some
discussion
time
in
Prague.
For
this,
that
will
mean
that
there
will
be
less
time
for
presentations.
So
again,
please
only
ask
for
a
slot
if
you
need
one.
A
So
we
have
in
your
interaction
and
coordination
with
working
group
and
so
as
a
v6
is
an
extension
to
86
and
is
expanded
six
panel.
So
we
need
to
call
in
a
tweet
six
memo.
So
as
spring
charter,
we
can
do
protocol
extension
unless
we
have
an
agreement
with
visit,
walking
protocol
working
order.
So
currently,
if
our
service
is
the
status,
is
that
extension
in
and
working
abreast,
coal
or
humans
as
a
progression,
is
good
and
most
of
the
issue
has
been
addressed,
but
there
are
still
a
discussion
and
something
like
six
or
eight
issues.
A
A
So
we
not
adopt
Ania
I
mean
neither
spring
Clement
will
adapt
a
document
with
a
v6
extension
before
a
search
is
a
Buddhist,
PC
final
and
then
we'll
see
how
we
work.
So
our
options
1
as
polish
after
sprint
as
a
functional
specification
and
then
6
tons
of
protocol
work
or
we
can
do
both
in
6
men
or
boys
in
spring.
After
agreement
between
shelf.
A
C
Ok,
so
the
struggle
that
I
have
is
that
there
are
documents
less
than
my
own
name
document
is
one
of
those
that
would
completely
block
with
go
despair
in
this
is
from
SRH
various
times.
We
have
understanding
where
things
are
and
spring
working
group
can
we
start
defining
it's,
not
a
blocking
factor.
That's
what
I'm
trying
to
say
there
is
a
lot
of
work
that
has
happened
in
SRA
draft
during
the
last
times
in
like
since
the
last
call
as
well,
and
there
was
a
very
good
discussion
yesterday.
We're
working
group
ii
was
agreeing
on.
C
D
A
E
C
Implementation
happening
and
then
to
the
premium
will
be
based
on
an
individual
document
or
an
individual
document
goes
through
a
lot
of
private
discussion
and
and
working
group.
Not
so
much
in
modern
bertie
group.
There's
a
lot
of
private
discussion,
a
person
in
jail
document
progression,
so
I
think
from
a
technology
point
of
view.
It
is
not
a
good
good
thing,
I
mean
I,
don't
see
anything
why
we
have
a
good
base
for
a
cirrage.
Why
we
cannot
take
more
I
mean
we
start
to
close
on
the
rest
of
the
world.
B
Discussion
of
an
individual
draft
is
entirely
down
to
four
authors,
where
authors
want
to
discuss
it
right.
It's
not
to
do
with
this
is
an
individual
draft.
Therefore,
it
cannot
be
discussed
on
the
spring
list.
Please
discuss
individual
drafts
on
the
spring
list,
I
think
with
this.
They
should
already
be
gating
whether
we
adopt
no.
C
D
F
Hi,
everyone
good
morning,
happy
Diwali
to
everyone
celebrating
Diwali
today,
Americus
Gandhi,
presenting
a
couple
of
graphs
on
performance
measurement
for
segment
routing
networks.
So
these
graphs
are
for
the
probes
based
performance
measurement.
We
also
have
couple
of
other
drops
for
the
in-situ
where
the
data
traffic
carries.
The
p.m.
information
I
think
so
far
is
presenting
one
later
today.
F
So
we
discussed
that
and
next
steps,
so
the
requirements
is
to
performance
measurement,
delay
and
loss
for
the
SR
links
and
end-to-end
P
to
P
and
P
to
MPs
are
policies
applicable
to
both
SR
MPLS
and
SR
v6
data
plane.
We
want
to
avoid
bootstrapping
PM
s--
essence:
that's
the
spirit
of
s
are
also
stateless.
On
egress
note,
again,
spirit
of
s
are
both
one-way
and
two-way
measurements.
I
want
to
handle
the
AC.
F
F
So,
as
mentioned,
I
PA
DB
part
is
defined
for
the
probe
messages
for
Sol
links
and
policies,
and
they
are
sent
with
the
label
stack
or
the
seedless.
4S
are
MPLS
and
sr6
policies
and
the
payload
is
basically
RFC.
Six.
You
see
sorry
for
so
not
inventing
a
new
performance
measurement
protocol
here,
so
we
are
requesting
to
Ayane
ports,
one
for
the
diem
in
one
for
LM.
F
So
for
the
prob'ly
sponsor
message,
there
is
a
lovesick
7876
that
defines
the
UDP
written
path.
So,
if
you're
doing
one
way
measurement,
you
can
use
information
from
there
for
two-way.
We
can
use
the
information
from
the
receive
probe,
query
message
for
the
IP
and
UDP
port,
so
this
table
does
define
return
path
TLV.
This
is
for
the
two-way
measurement,
and
this
way
we
can
avoid
the
state
on
the
egress
node,
the
TL.
We
may
contain
the
cid
list
or
binding
c4s,
our
MPLS
or
a
service
6.
F
As
we
know
s
our
policy,
that
is,
there-
are
a
CMP
between
ingress
of
transit
and
transit
and
egress
notes.
So
the
UDP
header
does
allow
us
to
take
advantage
of
the
hashing
function
to
go
over
the
various
ecmp
parts,
and
these
are
the
standard
hashing
that
doesn't
define
any
new
forwarding
behavior.
F
G
H
G
Yeah
and
next
here,
okay,
so
you
say
that
no
need
to
boost
rpm
session,
negotiate,
UDP
port,
and
then
you
mention
that
you
request
to
EDP
ports
well-known
so
using
well-known
UDP
ports
creates
a
dart
vector
attack
and
your
draft
doesn't
have
any
security
consideration.
How
you
do
how
you
address
the
possible
does
attack
because,
for
example,
if
we
take
t1
protocol,
it
uses
dynamic
ports
on
purpose
so
to
hide
and
basically
use
dynamic.
Port
numbers
negotiate
them
over
time.
We
learned
that,
for
example,
in
t1
white.
G
Yes,
there
is
advantage
of
using
well-known
destination
port
on
a
reflector
and
we
already
have
RFC
for
that
and
port
being
rededicated
862.
So
my
question
to
you
is
why,
to
reinvent
the
wheel,
when
we
have
already
t1
white,
and
now
we
have
a
working
group
document
in
IP
p.m.
which
you're
reasonably
soon
will
go
in
the
working
group
last
call
on
the
stamp,
which
has
all
the
properties
that
you
described
here.
F
Good
question
so
the
RFC
63
74
has
been
around
and
has
been
implemented
and
used
for
MPLS.
So
if
it's
good
to
bring
that
into
segment
logging
and
make
it
4s
a
v6
as
well
as
SR
MPLS,
so
the
main
objective
is
to
use
the
RFC
successfully
for
its
radical
RFC.
Why
Stewart,
who
is
there
again?
7876,
has
been
there
as
well.
So
the
motivation
is
to
continue
to
loudest
those
two
implementations.
That's
out
there,
but.
G
F
That
so
the
segment
routing
network
has
both
the
SRM
TLS
and
SR
v6
data
plane.
So
you're
not
gonna
run.
You
know
twelve
on
one
some
nodes
and
60s
early
for
on
some
nodes
or
or
some
parts
of
the
network.
Is
you
just
want
to
use
the
same
mechanism
6624
in
your
entire
segment
routing
network,
irrespective
of
what
data
plane
that
some
nodes
may
have?
If
somebody
wants
to
run
the
same
protocol.
F
G
I
I
The
port
allocation
people
and
the
general
sort
of
policy
is,
unless
you
absolutely
have
no
way
of
having
a
well
known
of
having
a
dynamic
port.
Well,
you
really
need
to
find
a
way
of
having
a
dynamic
port.
So
I
don't
understand
why
you're
asking
for
any
static
port
allocations,
hey
I,
think
you're
gonna,
find
it
very
difficult
and
be.
I
Yeah
III
was
just
a
very,
very
early
discussion
with
the
UDP
port
allocators.
Before
you
end
up
doing
what
we
did
when
we
wrote
seventy
seventy
eight
seventy
six
and
re-engineering
it
to
put
the
ports
in.
If
you
need
extra
ports,
you
could
extend
that
to
to
define
them.
So
if
you
want
them
for
your
other
oal
operations,
we.
J
J
F
F
J
G
There
will
be
that
we're
in
the
squad.
I
can
point
that
there
is
already
proposal
for
return
path,
TV
in
spring
PFD
draft,
so
I
encourage
you
to
look
at
and
let's
keep
it
number
of
new
entities
to
the
minimum,
because
it's
the
same
idea
of
controlling
the
return
path
in
traffic
engineering,
environment,
yeah.
F
E
G
A
F
F
So
look
at
the
requirements
and
scope,
probe
messages
for
the
SR
links
and
the
P
2
P
and
P
2
MP
policies.
We
are
talking
about
delay
and
last
measurements
again
probably
spawns
messages
and
how
we
compute
the
various
PM
matrix
and
flood
them
in
the
network
in
next
steps,
so
delay
in
lost
measurement
for
segment,
routing
for
links
and
policies
and
flooding
the
T
matrix
in
the
network.
So
this
is
again
for
the
SL,
with
MPLS
data
plane,
probe
messages.
F
F
There
is
a
gas
value
defined
at
the
same,
is
used
for
the
SR
links
and
SR
policies
and
similarly
a
gas
defined
for
LM
and
same
value
used
for
Excellence
and
SR
policies
so
that
there
can
be
one-way
or
two-way
measurement,
so
for
one-way
measurement,
the
RFC
7876
uro,
written
parts
TLV
is
used
and
for
the
two-way
measurement.
The
responses
end
the
same
way
as
the
ready
message.
F
So
this
way
we
can
do
the
performance
measurement
for
sr
links,
a
delay,
an
loss
and
then
compute
the
extended
matrix,
T
matrix
to
the
minimum
delay
maximum
delay
and
so
on
and
as
well
as
the
packet
loss
matrix
and
there
are
RFC's
there
today
for
OSPF
eius,
eius
and
bt
pls.
It's
a
drop
to
flood
those
values
in
the
network.
F
I
Clarify
this
Stewart
I'm
Cree
clarify:
are
you
going
to
do
colored
packet
techniques
in
this,
because
the
374
worked
okay
in
MPLS
TP,
because
there
was
never
any
ecmp,
but
as
soon
as
you,
every
CMP,
you
end
up
needing
the
the
piece
of
work?
That's
almost
complete
in
the
MPLS
working
group
and
is
being
discussed
in
I
ppm
of
covered
packet
marking.
So
what's
your
approach
to
to
the
packet
marking
going
to
be
yeah.
F
F
G
G
G
But
this
is
a
passive
performance
measurement
when
we
are
talking
about
active
Oh
am
I,
don't
know
about
any
other
method,
just
to
try
to
ensure
that
you
are
in
band,
because
otherwise
you
are
measuring
something
else.
Yes,
so
you're
saying
that
this
method
is
in
band,
but
you're
worth
saying
that
other
methods
are
not
in
band
our.
F
L
L
M
The
other
thing
is,
we
want
to
make
this
completely
self-defining
as
much
as
possible,
so
the
goal
was
to
have
a
lot
of
help
from
the
IGP
to
build
these
LSPs.
So
basically,
LSP
is
a
built
as
counter-rotating
pairs.
So
for
those
of
you
who
know
blsr,
it
should
sound
familiar
and
the
goal
is
again
also
not
to
configure
our
speeds
and
not
to
have
to
do
a
lot
of
work,
especially
if
you're
going
to
use
this
on
lots
and
lots
of
access
things.
M
M
So
here
you
see
a
and
LSP
from
r1
to
r4
and
the
clockwise
LSP
is
just
the
node
set
for
our
for
the
anti-clockwise
LSP
you
put
in
an
said
sort
of
halfway
there,
and
then
you
go
the
rest
of
the
way.
So
the
IGP
doesn't
defeat
you
so
from
our
to
our.
For
again
you
just
put
in
one
direction,
you
just
say
r4
in
the
other
direction.
M
You
may
pick
the
same
point
or
a
different
point,
but
you
want
to
go
about
half
way
and
then
go
the
rest
of
the
way
and
the
really
exciting
part
comes
for
protection
where
suppose
I'm
on
the
r1
to
r4,
LSB
and
that
are
I
realized
that
I
can't
go
to
it.
R4
in
the
clockwise
direction,
anymore.
I
will
jump
on
to
the
counter
clockwise
LSB
or
the
anti-clockwise
LSB
to
r4,
which
basically
means
swap
the
r4
note
said
with
the
note
said
that
goes
to
our
4
in
the
other
direction.
M
So
that's
the
blue
LSP
that
you
see
there,
and
so
you,
essentially
you
go
from
a
one
label
stack
or
two
label
stack.
There
are
cases
where
you
need
three
node
sets.
So
this
is,
you
know
painfully
construct
an
example
for
this,
but
Bruno
assures
me,
that's
the
most.
You
would
need
so
using
this
approach
in
one
direction.
You
need
one
node
set
in
the
other
direction.
You
might
end
up
through
three
notes:
it's
but
I,
don't
know.
M
M
Things
you
start
with,
it
could
be
four,
but
I
think
he
was
probably
the
maximum,
but
I'll
leave
it
to
address
theorist
or
Eric.
Whoever
comes
first
I'm,
so
an
alternative
approach
is
to
use
a
flex,
algo
type
approach,
not
necessarily
that
mechanism
for
flex
I'll
go
where
you
have
a
single
set.
That
says
I'm
a
clockwise
said,
and
you
have
another
set
that
says:
I'm
an
anti-clockwise
head
and
no
matter
what
the
metrics
say.
Because
part
of
the
ring
discovery
says
this
is
clockwise.
M
Everyone
in
the
know
on
the
ring
agrees
that
this
clockwise
and
this
is
anti-clockwise.
So
if
puts
down
the
forwarding
in
the
right
direction,
then
I
just
throw
that
sit
on
the
stack
and
the
packet
is
guaranteed
to
go
to
the
right
place
in
the
right
direction.
Again,
if
I
encounter
a
failure,
I
flip
it
to
the
shape
that
goes
in
the
other
direction
and
then
I'll
end
up
at
the
egress.
M
Without
having
to
worry
about
the
metrics
working
against
me
until
the
ring
has
defined
what
is
clockwise
and
anti-clockwise,
you
should
be
dropping
any
packets.
You
see
with
these
sits
on
top
once
the
definition
is
complete,
then
you
can
put
in
the
forwarding
path
that
says:
go
this
way
of
us
go
there
the
other
way.
So
there
is
a
question
of
how
to
allocate
these
heads.
I
was
a
private
conversation
with
both
Bruno
and
Shraddha.
M
Convinced
me
that
there's
configuration
involved
in
doing
all
this,
which
is
not
a
great
thing,
because
we're
trying
to
make
the
Rings
come
up
on
their
own.
There
might
be
asaji
bees
as
well,
because
unless
you
want
to
use
the
common
SI
TV
for
everything,
you
might
say,
here's
srgb
for
brains
and
here's
srgb
for
regular
note
sets.
So
this
is
definitely
a
problem
that
I
would
like
to
address
and
make
it
easier.
So
we
need
more
details
in
the
draft
for
sure
we
need
to
bring
this
approach
into
the
draft.
M
We
need
to
define
how
to
manage
in
a
better
way,
asaji
bees
as
well
as
notes
it's
or
these
clockwise
anti-clockwise.
It's
a
few
more
words
about
how
to
install
primary
routes
and
production
rods
when
to
install
them
when
to
remove
them.
For
example,
the
ring
is
being
redefined,
so
if
a
ring
change
is
occurring,
maybe
you
should
again
drop
all
packets
until
everyone
is
clear
on
what
directions
and
I'm
done.
For
now,
we
will
post
another
version
of
the
draft
with
a
lot
more
details.
H
Okay
morning,
everyone,
this
is
an
update
of
the
Sigma
routing
for
enhance
we
PM
draft
okay,
a
little
brief
review
of
the
motivations.
You
know
that
VPN
has
been
widely
deployed
in
to
support
Maori
Tennessee
in
the
operating
networks.
Basically,
it
provided
the
separate
attendant
address
spaces
and
also
the
separated
routing
and
forwarding
tables.
H
Well,
there
are
emerging
services
which
are
more
demanding
requirement
on
the
SLA,
for
example,
the
more
strict
requirement
on
the
banner
is
latency
cheater.
So,
basically,
we
need
an
enhancement
to
the
existing
mechanism
to
meet
this
demanding
requirements,
so
enhance
the
appeal.
Actually,
it
aims
to
enable
multiple
tenants
which
have
this
demanding
service
requirement
in
a
shared
network,
and
we
need
to
ensure
the
high
performance
was
a
reasonable
cost
and
the
scalability,
and
it
is
also
considered
as
a
solution
for
the
network
slicing.
H
H
For
the
enhancement
it
is
actually
the
foundation
is.
We
need
the
enhancement
in
the
forwarding
plane,
because
the
current
it
sra
comes
from
the
guaranteed
resource.
Our
deterministic
scattering
in
the
forwarding
plane,
which
currently
we
have
several
candidate
technologies
like
listed
here,
will
not
go
to
the
details
and
for
the
control
plane.
It
actually
need
to
take
advantage
of
the
forwarding
plane
enhancements
so
to
build
virtual
networks
where
they
enhance
the
characteristics
and
the
role
of
the
control
planes,
also
to
give
a
good
abstraction
of
the
different
tech
forwarding,
plane
technologies.
H
So
I
have
a
generic
mapping,
something
the
control
plane.
Also,
the
scalability
is
an
important
teaching.
We
need
to
consider
in
the
design,
so
we
can
see
whether
Sigma
routing
can
support
the
has
maybe
any
requirement.
We
know
that
countries
and
routing
is
mainly
designed
for
the
explicit
routing
we
use
different
seeds
as
topological
or
service
instructions.
H
Then
we
can
see,
as
as
shown
in
this
diagram
different
service
flows.
They
can
share
the
same
seed
or
seed
list
using
the
packet
forwarding
and
how
about
the
performance
in
a
guarantee
in
current
eyes
are.
It
is
based
on
the
tips
of
cues
to
provide
a
class-based
traffic
different
differentiation.
This
may
be
enough
for
many
of
the
existing
services,
but
we
can
see
it
does
not
provide
the
solution
for
the
perb
tenant
of
a
service
traffic,
a
differentiation
or
treatment,
so
it
may
not
meet
the
requirement
of
the
demanding
services
in
the
future.
H
So
here
we
give
define
some
enhancement
to
the
second
routing,
which
is
to
put
it
resource
information
associated
with
the
segment
in
seeds.
This
is
based
on
the
forwarding
plane
resource
partitioning
technologies.
For
example,
we
can
allocate
it
allocate
dedicated
link
resource
for
different
virtual
networks
on
each
link,
and
then
we
use
different
seeds
to
identify
the
different
set
of
resource
allocated
and
have
the
multiple
adjacency
cedar
for
one
link
each
represented
a
partition
of
the
resource.
H
We
can
also
have
multiple
prefix
say
for
each
node,
then,
based
on
this
mechanism,
we
could
build
separated
segments
in
virtual
networks
with
different
set
of
seeds.
As
shown
in
this
diagram,
we
can
build
three
and
separated
the
virtual
networks,
which
was
a
group
of
state
used.
In
this
way,
the
traffic
in
different
virtual
networks
will
be
steered
to
the
isolated
resource.
According
to
the
seed
used
in
the
VPN
service,
can
map
to
either
dedicated
a
virtual
network
or
sharing
a
virtual
network.
According
to
the
requirement.
H
This
overview
of
the
controlling
procedures
basically
did
rely
on
a
hybrid
mode
of
the
control
brain.
We
need
a
controller
to
compute
the
virtual
network
topology
and
the
resource
needed.
According
to
the
service
request
received,
then
we
need
a
controller
to
instruct
the
nose
involved
to
allocate
per
segment
resource
and
seed
for
the
virtual
network
link
in
the
notes.
The
distributed
control
plane
will
be
used
to
advertise
or
report
the
information
about
the
resources
allocated
and
seed
associated
with
the
resource.
H
H
Here
we
summarized
the
updates
since
last
the
presentation
in
the
recent
two
versions.
We
had
more
details
about
mechanism
in
the
draft
and
we
described
the
procedure
with
an
example
diagram.
Also
the
control
brain
considerations
are
ideal
in
the
recent
versions.
We
also
clarify
that
the
applicability
of
this
solution-
post
SRM,
Harris
and
I
service
six.
There
are
also
many
editorial
changes
to
improve
the
readability.
H
We
have
received
a
many
comments
on
this
draft
here,
just
at
least
at
some
of
them
and
and
the
answers
to
the
questions
the
first
days.
What
extension
to
Sigma
routing
is
needed,
I
think
the
extension
here
is
to
use
some
routing
IDs
to
represent
the
different
set
of
natural
resource
on
the
narrow
segment
and
how
to
reserve
resource
in
the
network.
H
There's
also
concerns
about
whether,
as
we
add
the
state
to
the
state,
morality,
networks,
I
think
the
answer
is
yes,
when
you
add
more
state,
but
the
states
will
be
necessary
if
you,
the
purpose,
is
to
provide
a
guarantee,
there's
always
the
same
routing
technology
and
how
about
scalability
compared
to
existing
technologies,
because
this
Maxim
is
still
compliance
to
the
same
routing
paradigm.
We
will
not
introduce
per
pass
state
in
the
network
just
at
the
present
state
for
scalability
will
be
better
than
the
end-to-end
TRS
behemoth.
Listen!
H
H
K
Maybe
there's
something
I'm
missing
because
I
have.
Has
it
also
sorry
for
that
I'll?
Do
you
partition
resources
by
just
allocating
new
seeds,
I
mean
because
you're
telling
that,
on
the
same
link
you
are
allocating
adjacency
seed,
but
this
does
not
provide
any
resource,
partitioning
yeah.
So
what's
the
mechanism
behind
you.
H
E
H
N
Didn't
allow
liquor
Cisco,
how
are
so
so
there
is
a
virtual
network
which
is
defined,
and
there
are
this
node
and
prefix
seeds,
which
are
you
know,
additional
state
which
is
created
to
map?
What
is
this
construct
of
this
virtual
network
and
I
mean
what's
the
way
in
which
let's
say
here,
you
have
colors
blue,
for
example,
or
yellow.
How
are
they
associated
or
grouped
together,
yeah.
N
H
H
H
N
H
N
So
I
think
it
would
help
if
you
clarify
that
this
mechanism
is
about
partitioning
networks
with
multi
topology
I
think
that
clarity
would
help
that
that's
one.
The
other
thing
is
some
scalability
analysis
of
how
many,
if
you
are
doing
multi,
topology
routing
how
many
such
partitioning
can
be
done,
and
you
know
what's
the
proposal
that
that
would
okay.
O
Lu
program
actually
can
respond
to
both
comma
previous
speakers
comments.
This
the
fun
he
talking
about
whether
you're
modeling
the
resources
as
links
or
not.
It's
really
a
question
of
model,
a
and
there's
options
here.
You
can
represent
it
as
a
queue,
because,
fundamentally
that's
actually
what
it's
going
to
do
with
the
hardware
is
you're
gonna
have
some
allocations
of
a
logical,
virtual
or,
however,
you
implement
your
queuing.
You're
gonna
get
different
resources
and
you
can
model
it
as
lengths.
O
O
So
there
there
are
different
options,
and
you
know
you
chair
mentioned
that
there's
a
related
draft
and
T's,
and
we
would
hope
that
that
tide
of
trade-off
actually
is
that
defined
in
here,
not
in
T's,
because
it
should
be
driven
by
the
actual
use
cases
that
this
group
has
versus
that
what's
possible
in
an
architecture
is
from
an
architectural
standpoint.
You
can
go
many
different
directions
and
that
got
a
shoes,
so
multi
quality
routing,
that's
great,
that
works,
but
it
has
certain
scaling
properties,
controller
based
model.
If
that
works.
K
O
A
H
H
P
Heaven
builders
think,
five
years
ago,
another
company-
you
really
want
to
start
looking
into
decoupling,
your
non-connected
labelled
next
help
at
egress,
not
to
the
service
itself,
because
in
current
proposal
soon
you
are
going
to
map
next
help
in
your
LG
VPN
to
the
remote
look
back,
you
really
need
to
figure
out
how
to
provide
some
degree
of
interaction
between
number
of
services
and
number
of
remote
log
back
in
a
way.
Otherwise,
it's
going
to
kill
you
mapping,
you
mean
from
the
yeah.
E
C
F
Q
So
here's
a
motivation
for
this
tract
so,
as
we
know,
signal
thing
is
didn't
as
a
signal
as
a
source
tracking
technology
is
that
directly
encoding
for
what
in
the
instruction
in
packet
header?
That
means
the
pack
apostate
is
only
maintained
at
the
ingress
node
there's
no
path,
state
maintained
at
the
transmit
and
egress
out.
So
that
means
when
a
packet
reaches
the
egress
loud
summer.
Most
of
the
pack.
Labels
has
been
popped
off
so
for
the
egress
there's
no
way
to
determine
from
which
is
her
path
of
packets
come,
but
for
some
use
cases.
Q
For
example,
the
past
performance
measurement
down
real
traffic
bi-directional
pass
correlation.
The
passer
implication
at
the
egress
is
necessary,
so
this
trap
introduced
the
path
segments
that
can
be
used
to
identify
her
as
her
pass
in
the
context
of
the
egress
load.
So
here's
the
reviewers
of
this
path
segments
so
passing,
will
actually
is
for
SRM
PS
network.
A
passing
is
a
label
that
is
a
signed
by
the
egress
load
of
surpass.
They
can
uniquely
identify
the
SR
pass
and
the
u.s.
code,
and
when
passing
we
used
it
must.
Q
You
must
be
inserted
at
ingress,
node,
and
it
must
be
immediately
follows
the
last
label
appears
of
Hearts
Serpas
here
can
be
an
either
or
sub
pass
or
+
2
n
pass.
So
when
they
and
the
past
segment
must
not
be
popped
off
until
it
reaches
to
the
it
was
loud.
That
means
the
PHP
function
must
not
be
enabled.
Q
At
the
very
beginning,
we
proposed
two
options
for
positivity
identification
and
considering
that
the
use
case
and
better
use
cases
for
this
path,
segment
needs
to,
for
example,
to
pass
back
to
the
performance
measurement
and
bi-directional
correlation.
Most
of
this
same
should
be
done
in
the
egress
node,
so
we
think
is
suitable
to
to
use
the
ones
as
one
label
solution
that
the
solution
is
are
assigned
by
the
egress
and
the
second
one
is
to
add
a
new
section
to
to
describe
how
to
support
an
listing
of
the
path
segment.
Q
This
is
Corinne
such
a
suggestion.
The
idea
is
to
use
the
finding
a
pile
inside
and
together
with
spy
segments,
and
with
this
it
can
reduce
the
depths
of
the
label
stack
and
also
you
can
support
those
sudden,
a
sub
pass,
a
monitor
and
an
to
end
part
monitoring.
You
can
see
that
in
the
next
slide,
and
also
we
do
coddle
of
actor
in
changes
to
the
draft
and
to
make
it
more
readable.
Q
So
here's
the
histogram
shows
the
details:
how
to
combined
used
a
path,
segments
and
the
binding
site
to
support
sub
path,
monitoring
and
Antwon
monitoring.
You
can
see
that
in
the
figure
there's
a
path
from
A
to
D
that
spans
multiple
scenarios,
multiple
domains,
the
access,
the
aggregation
and
coda
mean.
This
is
a
very
typical
scenario:
mobile
backhaul
network.
Q
The
hair
just
stares
at
always
way
what
would
like
to
select
a
lacy,
more
review
and
comments
on
the
draft,
and
then
we
find
grab
the
conley
and
attract
is
the
foundation
server
ongoing
work
with
them
for
the
past
segments
based
performance
measurement,
Patrick,
no
answer
policy,
PCE
extension
for
path
statement
and
also
there's
apparently
work
that
is
different
services
per
segment
and
also
currently.
How
is
implanting
the
path
segments
based
performance
measurement
and
the
bedrock?
Q
E
G
Sr
MPLS
and
a
service
6,
because
I
think
that
they're
issues
that
you
need
to
address
will
be
the
same.
So
basically,
you
want
to
characterize
their
attack
between
two
endpoints.
Second.
Is
that
I'm
not
sure
that
removing
their
two
ID
label
to
characterize
specific
path
between
two
endpoints
brings
advantages,
so
sub
segments
might
create,
might
require
more
SIDS
to
define
explicitly
which
path
between
two
endpoints
being
taken,
so
we
can
discuss
it
offline.
J
J
The
reason
is
you're,
adding
now
more
levels.
Actually
you
lost
me
at
when
you
said
this
is
gonna,
be
another
label
in
the
mix.
If
you
really
have
to
solve
those
kind
of
problems,
then
adding
more
levels
is
not
the
solution.
Yeah
I,
think
by
the
time
we
actually
add
mod,
6
or
levels,
there's
actually
no
mores.
You
know
label
space
to
actually
in
a
sense,
we
can
push
more
levels
to
the
start.
I.
Q
Just
saw
in
the
beginning
of
this
presentation:
the
use
cases
for
this
hot
segment
is
about,
for
example,
the
past
vironment
past
a
performance
measurement
real
traffic,
so
because
you
want
to
pass
measurement
and
a
need
to
know
which
packet
belong
to
which
passed
and
king
counting
based
on
that
counting.
So
with
the
current
macadam
you
can
do,
you
cannot
do
that
sure.
J
Q
J
Q
J
Q
N
Get
ontological
Cisco,
so
I
think
we
should
review
more
closely
the
use
cases
and
where
this
is
really
applicable
or
needed.
One
comment
is,
for
example,
for
the
delay
measurement.
I,
don't
really
see
that
this
is
necessary,
we
can
discuss
more
offline.
The
other
part
was
that
the
path
segment.
The
draft
says
that
it
has
to
be
allocated
from
the
sRGB
or
the
srl,
be
I.
Think
you
know
if
where
it
is
really
needed,
it
could
also
be
a
local
level,
so
I
think
it
will
help
if
you
plot
it
yeah.
N
P
Japanese
Robster
two
points
one
you're
explicitly
stating
that
it
has
to
be
bottom
of
stack
label,
so
anytime
I
need
to
look
at,
but
I
need
to
traverse
how
label
stuck
look
up
for
the
both
feet
set
and
then
it's
very
expensive
operation.
You
won't
be
able
to
do
this
on
any
loan
platform,
so
doing
entropy
label
work.
We
thought
about
this
as
well.
The
easiest
way
to
put
it
the
watermark
is
a
stator,
but
then
half
of
the
devices
in
the
past
might
not
support
it.
You
really
need
to
think
where
you
keep
the.
Q
Know
did
this
Passavant
only
integrated
a
performance
measurement
and
the
egress
or
pass
by
directional
correlation?
So
when
the
packet
reached
the
egress,
the
labor
labor
has
already
popped
off,
so
that
has
been
in
the
past
segment
directly
present
to
that
egress.
So
it
only
necessary
to
scan
they
never
start
to
to
found
a
path
segment.
I
thought
you
were
going
to
use
it
for
counters
and
country
and
the
egress
to
accounting
to
account
at
the
u.s.
okay.
Second,.
A
P
R
If
I've
got
you,
if
I
understand
you
correctly,
the
purpose
of
the
path
segment
is
to
reconstruct
what
the
path
looked
like
when
it
left
the
source
before
labels
were
popped,
given
that
in
SR
v6
you
don't
pop
labels
and
you
don't
throw
away
anything
from
the
segment
routing
header.
What's
the
motivation
for
doing
this
in
SRV
sex,
okay,.
Q
I
know
this
trap
is
not
about
SRV
Six,
Sigma
Pi
segment,
bad
for
that
direct
and
yes,
I
agree
with
you
at
the
US.
They
can
use
dice
our
edge
to
recover
the
recover
the
past
right
identify
the
past,
but
from
the
implement
point
of
view,
it's
very
maybe
it's
forgiving.
It's
not
friend
to
the
hardware
implementation.
You
know
to
compare
of
very
naughty
of
arable
lands
field
to
compare
with
to
determine
the
package
from
which
pass.
That's.
You
know
that's
a
optimization
to
that.
Okay,.
R
Second,
second
part
of
the
question:
the
destination
node
can
use
the
path
segments
to
reconstruct
the
entire
path
as
it
was
when
it
left
the
source.
I
assume
any
other
net
node
in
the
network
can
make
the
same
transformation.
It
has
access
to
the
same
I
GP.
Given
that
what
do
you
need?
The
rest
of
the
note
labels
for
every
node
could
generate
the
whole
label
stack.
R
E
S
S
Good
morning,
everyone
and
Shanta
a
and
presenting
TTL
procedures
for
a
society
paths
on
behalf
of
quarters
stack
to
start
with
the
the
let's
J
and
see
what
what
is
the
use
case
you
are
talking
about.
So
we
have
this
network
R,
1,
R,
2,
R,
3,
R,
4,
R
5,
and
we
have
a
srte
path,
which
is
represented.
Using
two
labels
in
the
label
stack
one
zero,
zero,
four,
which
is
going
to
R
4
and
then
from
our
photo
are
fine.
S
So
the
label
stack
has
one
zero
zero,
four
on
top
and
one
zero
zero
five
on
top
at
the
bottom.
So
what
we
want
to
do
here
is
to
be
able
to
trace
this
path,
to
see
where
exactly
the
packets
are
going.
When
this,
when
we
have
the
society
path
constructed
on
the
head
end,
so
we
are
doing
an
MPLS
trace,
knock
procedure
from
the
head
end,
so
so,
according
to
RFC
it
to
870.
S
We
suggest
that
you
know
you
have
to
have
the
the
tunnel
that
you're
tracing
the
TTL
moves
from
one
two,
one
two
and
then
three
four
and
so
on,
and
then
the
bottom
labels
are
all
set
to
zero.
So,
let's
start
with
the
top
label:
1
0
0
4,
which
has
1
and
then
the
first
step
is
like
the
top
label
has
TTL
1
and
the
bottom
label
has
a
TTL
of
0
and
then
on
step.
S
S
So
when
R
one
receives
an
eco,
applying
it
increments,
the
TTL
to
2
for
the
top
level
and
then
in
the
step
2,
it
takes
the
packet
it
at
the
top.
The
TTL
of
the
top
level
becomes
2,
and
then
it
sends
the
packet
out.
So
in
step
2
the
packet
reaches
it.
Transits
are
2,
it
reaches
three
and
and
the
TTL
expires
on
our
tree
and
then
packet
gets
punted
to
our
tree
and
it
generates
an
eco
reply
that
it
is
a
transit
for
our
trees,
doing
the
PHP.
S
Here
it's
and
it's
short,
five
more
with
all
PHP,
so
at
r3
the
label
gets
popped
and
the
the
bottom
label
gets
exposed.
So
then
our
three
cents
transit
written
code
to
r1,
r1,
further
increments,
the
TTL
of
the
top
label
to
3
and
then
the
MPLS
based
or
precision
and
TLS
eco
request
message
is
sent
again
so
this
time,
when
it
when
the
packet
reaches
r3,
the
TTL
does
not
expire
because
it
is
3.
So
you
still
have
TTL
left
so
the
but
then
the
packet,
the
top
level
gets,
gets
popped.
S
So
you,
portable
label,
gets
exposed
and
packet
reaches
our
four
with
bottom
level
thousand
five
with
TTL
of
zero.
So
this
could
be
problem
because
incoming
TTL
is
zero,
so
packet
may
be
dropped
or
rate
limited.
So,
even
if
we
assume
that
it
packet
is
pointed
to
a
four
and
then
r4
generates
an
egress
egress
return
code,
the
next
step
will
again
have
a
problem,
because
now
our
fortune
rates
that
I
am
egress
and
then
so
the
the.
S
So
we
are
done
with
tracing
the
top,
the
first
tunnel,
and
so
now
we
start
tracing
the
next
tunnel,
which
is
one
zero,
zero
five,
and
so
we
set
the
TTL
to
one
four
one:
zero,
zero,
five
and
then
when
it,
because
our
trees
popping
the
label.
So
when
it,
when
the,
when
the
label
reaches
our
for
the
TTL
again
expires.
So
our
expectation
in
this
step
was
to
reach
to
our
five.
But
the
TTL
expires
on
our
four.
S
S
G
S
S
O
In
it
I
understand
what
a
stack
of
labels
are,
but
there's
it
if
it
had
said
that
I
I
would
have
understood
it,
but
it
didn't
it
kept
saying
srte
and
there
is
no
definition
SRT
and
it
gives
a
reference
and
I
get
lost.
So
they'd
be
good
to
clarify
that
that's
the
problem,
and
maybe
that
would
help
brag
to.
S
A
S
Let's
move
on
to
the
the
proper
to
describe
what's
the
problem
with
a
uniform
model,
so
uniform
model
is
where
you
copy
the
and
the
label
is
getting
popped.
You
copy
the
TDL
of
the
top
label
into
the
bottom
label,
so
so
the
first
step
and
the
second
step
pretty
much
same
no
difference.
So
you
set
top
label
to
top
TTL
top
label
TTL
to
one
and
it
expires
on
r2
sends
transit,
and
then
you
will
increment
the
TTL
on
the
ingress
to
two
and
then
send
the
packet
again
and
then
it
expires
on
r3.
S
And
then
it
says,
r3
also
says
it's
a
it's!
It's
transit
for
label,
one
zero,
zero!
Four!
So
at
this
point
the
the
label
is
getting
popped
from
r3
to
from
0:04
is
getting
parked
at
our
3,
so
it
is
supposed
to
copy
the
TTL
from
top
label
to
bottom
label
that,
but
that
doesn't
really
happen.
So
if
you
really
see
this
is
step
3,
where
we
have
incremented
the
TTL
of
1
0
0
4,
2
3.
So
it's
even
if
it's
getting
popped.
S
If
we
expect
that
it
copies
from
top
label
to
bottom
label,
that
may
not
happen
because
so
MPLS
TTL
copy
mechanisms
will
not
copy
the
TTL
from
top
label
to
bottom
label
if
the
bottom
level
has
a
lower
value
compared
to
the
top
label.
So
that's
that's.
That's
done
to
avoid
kind
of
you
know
in
MPLS
data
packets.
If
you
have
TTL
for
the
bottom
tunnels
test
set
to
1,
then
you
need
to
honor
that
you
cannot
copy
the
live
TTL
from
the
tunnels
that
you're
traversing,
which
is
you
know
your
outer
telex.
S
So
we
need
a
solution
for
this.
So
what's
the
solution,
so
we
need
when
we
reach
in
PHP
case
when
we
reach
the
PHP
node,
we
have
to
increment
the
bottom
label
TTL
to
1.
Instead
of
waiting
until
we
reach
the
egress.
So
that's
the
solution
we
need.
So,
let's
see
if
we
do
that
everything
works,
fine,
so
step.
One
is
same,
so
you
would
reach
our
to
Xandar.
Its
and
TTL
is
1,
so
it
would
expire.
You
send
back
transit
step
2,
you
would
increment
the
TTL
of
top
level.
S
You
reach
our
3
and
our
3
would
say:
I
am
PHP
and
then
once
it
says,
I'm
PHP
r1
would
increment
next
step.
It
would
increment
the
TTL
of
the
next
level
to
one
when
PHP
is
reached,
and
so
when,
when
a
task
3,
when
the
label
is
parked,
so
we
would
get
you
the
label
one
zero,
zero
five
would
be
exposed
on
our
4,
which,
which
has
a
TTL
one
and
now
the
next
step.
S
S
S
So
these
are
some
of
the
alternate
approaches,
evaluated
tht,
router
itself,
sending
saying
I
am
egress
and
second
approach,
TT
setting
inner
label.
One
third
approach
is
secure,
setting
inner
label
to
255
and
then
fourth
option,
setting
inner
label
on
reaching
egress,
so
PHP
router
itself.
Sending
egress
return
code
has
problems
because
you,
you
are
not
pressing
the
tunnel
and
to
the
octave
up
until
the
egress.
So
you
might
miss
some
of
the
things
that
have
you
know
that
that
are.
S
You
are
supposed
to
validate
on
the
egress,
so
that's
not
the
right
way
to
go,
or
also
it
would
not
detect
these
kind
of
errors
where
r2
as
misprogrammed
or
r4
has
programmed
a
swap
operation
instead
of
an
egress
operation,
and
also
it
does
not
guarantee
that
the
traffic
is
heating
r4
before
it's
going
to
r5.
So
those
are
the
those
are
the
reasons
for
which
this
option
was
not
considered
solution
and
then
setting
inner
labels
to
one.
S
So
if
you
to
begin
with,
if
you
stay
exactly
all
the
labels
to
one
all,
the
labels
in
the
label
stack
to
one.
So
this
works
fine
when,
when
the
network
is
programmed
correctly,
so
let's
say
everything
is
fine.
R1
r2,
r3,
r4
r5.
Everything
is
programmed
correctly
than
if
you
had
set
the
inner
label
to
to
begin
with.
S
So
in
that
case
it
may
not
correctly
detect
this
Empire
State
or
procedure.
Omaha
correctly
detect
this.
If
we,
if
we
use
like
the
inner
layer
TTL
to
1,
because
in
SR
the
labels
are,
you
know
pretty
much,
everybody
understands
the
labels
that
up
there
in
the
stack,
so
1
0,
0,
v
r3,
would
understand
1,
0,
0
v
as
well,
so
when
which
would
not
have
been
the
case,
most
probably
not
have
been
the
case
in
in
NDP
or
RSVP.
But
here
s,
RR
3
definitely
understands
1,
0
0
5.
S
So
when
it
reaches
our
two
eyes
pop
the
label,
top
level
is
1,
0,
0
5.
So
it
comes
to
our
3
and
then
it
looks
at
it's.
A
valid
lay
incoming
label
1
0,
0
5,
is
a
valid
incoming
label
and
also
the
fact
is
4
4
4
4,
which
is
transit
for
this
case.
R3
is
programmed
transit
for
4
4
4,
so
we
may
not
be
able
to
correctly
detect-
and
you
know
the
ste
trace
or
procedure
may
repeat,
like
just
keep
incrementing.
S
S
E
S
S
T
S
S
What
I'm
suggesting
here
is
the
practice
that
I'm
listing
down
in
red
line
here
is
the
best
practice,
probably
not
there
in
any
RFC,
but
I
think
that
that's
the
way
it
should
be
I'm,
not
I'm,
not
saying
you
know,
you
change
that,
because
if
you
change
that
things
won't
work
right,
because
if
you
change
that,
let's
say
we
change
that.
Okay,
we
may
we
say
you
know
you
have
to
copy,
then
this
doesn't
work
right.
This
doesn't
work
because
moment
you
copy
254
to
the
bottom
label,
the
bomb
the
traceroute
won't
work.
S
So
what
I'm
suggesting
is
keep
this
best
practice
if
it's
not
I'm,
not
fully
sure
if
it
features
in
any
RFC
or
draft
the
the
statement
in
the
bottom.
We
probably
need
such
a
statement
because
and
people
and
my
understanding
is
and
I'm
talking
to
people
and
pls
veterans.
My
understanding
is:
they
do
believe
that
this
most
implementations
do
honor
this
constraint,
probably
not
there
in
any
RFC
people.
If
you
have
opinion,
you
comment
on
the
sir.
J
S
We
can
discuss
actually
we
discuss,
we
thought
about
it
like
introducing
a
new
return
call,
but
we
were
concerned
about
interoperability
issues
see
one
of
these
things
writes
because
this
8
to
h7
RFC
has
been
there.
People
might
have
implemented.
You
know
having
this
one
bottom
label,
one
or
you
know
this
up
last
approach,
so
we
we
don't
want
you
know
if
somebody
people
are
getting
it
working
somehow
it
should
just
continue
to
work
for
them.
So
that's
definitely.
J
J
S
Can
do
this,
so
this
is
about
so
basically
we
have
this
e
PE
use
case
where
you
know
we
have
the
strap
for
century
te,
which
defines
these
three
different
kinds
of
sits
for
for
doing
egress,
PR
engineering
note
said:
I
just
insisted
and
Pierce
exit.
So
when
we
have
when,
when
there's
a
network
built
inter
a
yes
using
these
sits,
you
probably
want
to
validate
the
control
plane.
Data
plane
validations
for
this
network,
which
are
and
paths
which
are
using,
which
are
built
using
these
sits
so.
S
So
the
cross,
a
is
called
localization,
is
useful
and
in
certain
cases
you
know
the
cross.
A
s
om
packets
may
not
be
accepted
if
they
are
in
different
ownership.
So
so
we
are
trying
to
address
both
of
them,
and
then
we
need
the
target
extract
definitions
in
MPLS
OEM
protocol
to
identify
these
new
kind
of
sits
which
have
EPE
sits.
So
this
draft
basically
defines
what
are
these
six?
What
is
the
facts?
Technician
for
these
fits
in
MPLS
OEM.
S
So
so
we
have
this
for
piano,
so
that
fear
I
just
insisted,
so
it
basically
has
local
ace
number
remotest
number.
Basically,
the
definition
of
sits
is
included
in
the
target
effects,
TAC
definition.
So
when
the
when
the
label
comes
in
with
this,
the
defect
stack-
and
you
know
it
can
make
it-
can
verify
whether
you
know
it
matches
with
the
definition
of
this.
S
So
we
have
the
spear
set
6
Pierce
actor
is,
you
know,
represented
by
a
set
of
links,
so
you,
the
the
the
tech
stack
definition.
Has
you
know
a
set
of
links
which
is
represented
using
using
the
same
block
where
you
know
you
have
this
all
the
the
link
representations
using
the
local
ace
number,
remotest
number
and
interface
addresses,
as
well
as
it
out
radius.
S
C
E
C
And
indeed,
note
said
what
says
the
agency
said:
there
could
be
cases
where
the
interface
is
not
known,
but
notes
it
validation
work.
So
not
in
all
cases,
for
the
note
said
you
need
the
you
need
to
mandate.
The
interface
address
information
because
that
may
not
be
known,
may
may
not
be
known
to
the
sender,
but.
C
C
And
I
think
it's
going
back
to
the
my
side
of
things
lights,
but
but
I
think
the
other
thing
is
also
that
the
Engel
and
in
a
different
year,
so
we
need
some
consideration
with
that
because
in
the
reverse
path
may
not
be
there
at
all.
So
this
is
a
kind
of
thing
that
we
were
looking
into
it.
Incidentally,
we
were
working
on
the
same
thing.
One
of
the
things
that
we
were
looking
is
when
you
land
on
a
cross.
S
So
what
so,
if
you
did,
we
did
think
about
that
as
well,
but
what
I
was
thinking
was.
This
is
mainly
getting
generated
from
like
performance
management
system,
few
QC
RFC,
eight,
four
zero.
Three:
it
talks
about
a
performance
management
system
which
generates
you
know
this
ping
and
traceroute
packet.
So
you
could
have
a
situation
where
you
know
the
PM
s
has
a
connectivity
to
all
parts
of
the
network
and
then
it's
so
society,
so
some
IP
would
be
that
PMS
and
that
that's
reachable,
that's
where
we
were.
S
So
you
have
to
have
two
different
procedures
for
validation
because
it's
cross.
Yes,
so
in
certain
cases,
if
it
belongs
to
same
operator,
it
may
be,
he
may
be
willing
to.
You
know
allow
that
cross-validation
and
in
some
cases
that
cross-validation
like
if
a
packet
really
reaches
the
other
areas,
it
may
not
accepted.
So
we
have
two
different
modes
of
validation
and
that
that
can
be
controlled
by
a
config,
see
that's
what
I
have
so.
S
A
C
C
So
this
is
the
fairly
and
preventing
it
on
behalf
of
my
quarters,
so
we
were
looking
at
the
problems
that
82
say:
80
to
87
does
not
address
the
ship
types,
so
the
three
VB
prefix
it
is
not
mentioned.
There
be
PEP
that
Rochelle
just
talked
about
it's
not
mentioned.
They
are
in
finding
positive,
it's
not
mentioned
there,
so
we
try
to
address
and
it
happen
to
be
there
for
EPE.
We
were
looking
at
the
same
solution
this
so
I'm
going
to
skip
this
because
the
changes
are
minimal.
C
I
would
like
to
also
mention
that
we
are
talking
with
the
shelter
and
the
courses
on
the
other
draft
and
to
see
if
he
can
collaborate
or
or
manage
the
traffic
so,
but
so
that
is
a
depression
in
progress.
I
would
skip
these
because
in
the
question
I
already
mentioned
some
of
the
use
cases
that
we
try
to
address
now
for
the
PT
prefix
it
we
don't
need
a
new
fact,
the
fair,
but
we
needed.
C
We
need
a
description
that
we
can
use
so
I
skip
that
slide
already
then,
for
the
BOD
finding
said
we
did
put
in
effect,
which
would
be
the
Hedden
address,
the
color
and
endpoint,
and
but
the
really
question
is
that
the
path
binding
set
validation
procedure,
so
we
did
define
something
that
initially
should
set
ETL
to
one
for
the
path.
Binding
set,
validation
and
the
procedure
is,
are
there
in
the
draft
or
me
in
the
interest
of
time?
I
just
want
to
highlight
some
some
differences
in
the
same
thing
for
the
BT
PPE.
C
If
the
different
parts
of
the
Nexus
or
the
pairing
node
is
a
non
MPLS
router,
but
to
which
we
cannot
forward
the
packet,
then
we
also
are
recommending
that
the
TTL
for
the
label
could
be
set
to
one
also
a
text
file.
So
these
were
with
the
procedure
I
think
we
can
discuss
more
in
the
MPLS
working
group.
I.
C
U
U
Basic
intention
of
the
draft
gives
a
couple
of
application
scenarios
where
the
network
service
editor
can
work
with
segment
outing
together
deployed,
while
maintaining
the
separation
of
the
transport
and
the
service
plains,
which
was
the
original
intention
at
the
SFC
architecture.
So
two
scenarios
there's
a
picture
I
showed
last
time,
so
I
won't
bore
you
with
that.
This
is
the
first
scenario
with
issues
in
NSA's
using
SRS
of
transport,
and
then
this
is
an
integration
of
the
two.
We're
basically
nsh
follows
the
segment
routing
stack.
U
The
encapsulation
details
are
here
everything
all
of
this
is
in
the
draft
anyway,
but
basically
the
NSA
follows
the
segment
routing
stack
and
we
would
use
something
like
draft
IETF,
MPLS
SFC
encapsulation,
to
indicate
that
NSA's
follows
the
label
stack
for
sr.
There's
the
details
there,
but
basically
we
added
some
deep
detail
in
the
new
version
of
the
draft
to
indicate
how
you
would
indicate
nsh
follows
the
second
route
in
stock,
so
changes
since
102.
We
changed
the
name
because
we
think
spring
is
the
right
working
group.
U
We
added
more
detail
on
the
SFF
to
SF
interaction
and
the
layer
in
between
segments
and
then
SH.
We
added
more
detail
on
the
SR
v6
encapsulation,
which
is
basically
indicating
a
UDP
in
the
next
header
fill.
So
we
need
new
UDP
port
or
indicating
an
IP
protocol
number
for
NSA
to
it.
Both
both
of
those
are
requested
in
in
the
document
from
Ayana
conclusions,
same
conclusions
as
last
time.
U
Basically,
next
steps,
we
look,
requests
working
group
adoption
into
spring,
but
I
heard
the
chairs
earlier
we're
going
to
have
a
conversation
on
this
in
Prague,
so
that
that's
fine
and
we
can
make
some
decisions
then,
and
then
we'll
continue
to
refine
the
technical
details
in
the
document.
Any
questions.
A
P
C
So
the
next
one,
SB
f
d4,
it's
our
policy,
but
chair
has
a
slide
and
that
is
partly
covered
to
that
direction.
That
share
one
want
us
to
take.
So
we're
going
to
talk
with
Craig
and
the
author
of
VFD
drops
well
for
the
S
equality
n,
okay.
So
here
what
I
would
like
to
discuss?
Is
the
SR
v6
Wynnum
draft
on
behalf
of
the
quarters?
C
E
C
That
when
we
presented
this,
so
there
was
last
call
for
the
SRS
draft,
and
one
of
the
comments
during
the
last
call
for
the
SRA
job
was
that
the
obut
definition
should
either
be
defined
fully
in
a
suresh
draft,
or
the
OB
definition
should
be
removed
from
the
authority
drop.
So
at
that
time,
I
think
we
made
a
mistake
in
terms
of
removing
from
the
asari
table,
but
then
later
on,
it
turned
out
that
there
are
many
use
cases
of
the
opet.
C
There
are
now
spreaded
around
different
drafts,
so
that
includes
the
PN
and
and
also
onm
for
the
service
programming
in
a
services
network
and
in
Menuhin.
So
based
on
this
and
this
one
and
so
based
on
this
it
it
it
make
more
sense
to
have
the
obit
definition
getting
back
to
the
side
across.
There
is
something
that
suggested
I
suggested
with
this
expansion
yesterday
and
would
be
follow
up
on
610,
meaning
mister
now
other
than
that
is
only
a
tree
changes
with
this
job.
C
Since
the
last
time
I
presented
this
so
in
interest
of
time,
I'll
remove
disco,
but
but
really
this
draft
does
not
define
any
changes
to
any
existing.
My
simple
procedure
it
they
were
seamlessly
for
a
savvy
six
Network.
The
job
does
not
work
changes
and
make
any
changes
with
the
SRH
processing
or
ipv6
diddly
and,
and
it
does
not
make
any
changes
to
so,
essentially,
the
only
difference.
C
The
only
thing
that
we
are
asking
for
is
some
new
code
point
for
the
ICMP
v6
type
numbers
so
that
we
can
have
some
more
specific
error
codes
and
during
doin
and
for
the
a
solid
I
serve
it
succeed,
that's
the
only
thing,
but
that's
not
sufficient
enough
or
or
big
dependents
on
six
men,
so
I
think
from
that
perspective,
what
you
would
like
to
see
is
that
this
draft
proceed
in
in
spring
working
group
and
get
adopted
into
in
working
group.
I,
don't
see
like
even
from
an
SRA
definition.
C
C
C
A
It's
not
specific
to
see
strapped
on
it's
coming
again.
It's
not
specific
to
this
draft
on
it's
coming
and
begin
again
again,
and
so
it's
better
to
be
before
the
trim.
Another
comment
in
regaining
I
on
our
section:
you
cannot
do
self
allocation
of
I
in
a
good
point,
so
they
need
to
be
to
be
definer,
and
then
we
can
do
adoption
with.
C
C
D
Okay,
okay,
so
thank
you.
Greg
Martin
speaking,
I
haven't
read
the
drop
sorry,
but
if
it
is
another
first
come
first
serve
cold
point:
please
remove
it
very
rapidly,
because
otherwise,
if
people
will
read
the
draft,
they
can
think
that
it's
been
allocated
and,
let's
start
implementing
and
that's
messy
afterwards.
I
fully
understand
and
they're
fully
agree.
E
G
City
I
will
not
ask
to
rewind
it
couple
quick
comments.
It's
it's
very
nice
draft
and
it's
nice
that
you
demonstrate
that
existing
mechanics
works
with
service
six.
The
question
I
have
is
that
what
is
real
need
for
all
bit?
If
you
have
ICMP
port
number,
so
you
identify
om
in
IP
through
their
well-known
port
numbers.
There
is
no
apparent
need
to
have
a
beat
in
SRH
to
identify
that
your
payload
is
belongs
to
the
certain
protocol
and
you
can
parse
it
accordingly.
C
So
when
a
segment
know
when
a
node
is
processing
a
segment
and
and
it
needs
to
perform
when
M
operation
you're
not
going
to
go
down
all
the
way
to
I
saying
beautify,
not
the
payload,
what
did
I
see,
MVP,
adore
and
and
and
and
what
it
contains,
and
it
needs
to
be
done
at
that
time.
Note
should
not
even
have
to
worry
about
whether
and
it
doesn't
it's
not
only
the
payload
for
the
use
cases
of
orbit
are
not
only
ICMP.
The
payload
could
be
UDP.
Payload
payload
could
be
some
other
constructed,
payload
n/a.
C
G
G
Mixing
in
many
things
in
the
one,
so
if
we're
talking
about
this
draft
and
during
ping
and
traceroute,
so
the
traceroute
works
through
TTL
expiration
and
if
node
has
TTL
expired
on
it,
then
you
apparently
look
at
the
payload.
If
no
doesn't
have
tto
expiration
on
it,
then
obit
is
irrelevant
because
you're
just
for
running
this
packet,
so
again,
I,
don't
really
see
apparent
need
for
obit
no
but
I'm
you're
saying
that
just.
C
G
C
G
C
C
G
C
All
no
they
don't
they
don't
end.
It
fully
explained
in
the
draft
crystal
clear
they
because
not
need
to
understand
when
a
node
receives
a
packet
where
TTL
expiry
happens,
that
TTL
price
body
responds
in
ICMP
error,
message,
time
exceeded
and
it
copies
the
invoking
packet
as
much
as
copy
it
can
make
it
is
this
is
this
has
been
like
20
years?
This
is
not
something
new
and
it
dad
no
does
not
have
to
read
this
chrisrip
yeah
I
mean
it's
mentioned,
the
drop
very
clearly.
C
C
C
G
You're,
defining
Orion
functionality
in
two
places:
you're
using
some
flag
and
you're
using
effectively
protocol
type
through
UDP
port
number
and
that
creates
ambiguity.
I
suggest
you
read
the
draft
that
I
put
before
Montreal
on
am
identify
which
analyzes
this
complexity
and
it
creates
more
confusion
rather
than
helps
things
so
in
IP,
om
being
identified
for
the
UDP
port
number
and
let's
not
create
the
second
mechanism,
which
only
confuse
that,
because
what
will
happen
if
you
have
or
a.m.
port
number
like
UDP
or
performance
measurement
and
old,
it
is
not
set.
It's
confusion.
No
I.