►
From YouTube: IETF97-SPRING-20161117-0930
Description
SPRING meeting session at IETF97
2016/11/17 0930
A
A
A
That
would
be
really
welcome
and
just
to
let
you
know
also
that
we
might
have
a
virtual
queue.
I,
don't
know
somebody's
there,
but
it's
not
a
crowded
meeting,
so
we
should
be
able
to
manage
so
working
hope
progress
if
I
may
call
that
poor
guys,
we
don't
have
any
new
RFC.
Since
the
last
meeting,
we've
managed
to
close
only
two
out
of
seven
working
hopeless
calls.
A
A
Strictly
speaking,
it's
to
the
resiliency
a
use
case.
We
managed
to
get
it's
in
orange,
because
the
last
answer
came
in
like
less
than
a
week
ago
and
but
to
address
that
point
we
we
had
to
run
after
that
person
until
I.
In
fact,
we
ended
up
removing
him
from
the
list
of
authors,
so
be
aware
that
something
like
that
could
happen
and
for
the
architecture.
I
think
we've
received
the
last
missing
IPR
answer
two
days
ago.
A
So
it's
orange,
it's
green,
but
I
wanted
to
emphasize
that
the
the
latest
answers
came
in
very
recently
for
the
ipv6
were
still
missing
some
answers
and
we
have
to
work
in
go
class,
call
to
go
the
target
that
I'd
like
to
set
is
that
we,
we
press
the
button
to
send
these
documents
to
iesg
before,
while
in
one
month
before
the
Christmas
period.
So
I
will
really
expect
everyone
to
work
on
that
for
the
other
working
of
documents.
A
A
A
Adoption
calls
so
not
much
difference
here
compared
to
working
or
glass
goals.
The
level
of
support
and
interest
expressed
on
the
list
is
quite
questionable,
and
so
I
will
really
like
this
to
change
and
we
have
also
the
same
IP
I
issue:
we're
still
missing,
IP
answers
on
large-scale
interconnect
for
the
early
cursing
info
as
defined
we'll
talk
about
it
today.
A
So
we
have
some
laughs,
we
are
which
are
Conda
late
for
work
and
go
bad
option
in
the
sense
that
the
authors
have
informed
us
that
they
would
like
to
us
to
do.
A
poll
for
adoption
situation
we
are
in
today
leads
us
to
prefer
clearing
the
situation
before
adopting
new
document.
It's
not
an
absolute
rule,
but
it's
a
very
likely
one,
at
least
over
the
next
three
months,
I'd
say
so
a
conclusion
on
that.
We
have
a
challenging
old
map
ahead
for
documents
to
ya,
to
iesg,
before
Christmas
and
for
other
one
before
chicago.
A
So
I
would
really.
I
really
expect
the
authors
to
be
extremely
reactive.
Both
is
in
answering
the
IPR
polls,
but
also
in
addressing
the
comments
that
will
pop
up
during
the
during
the
rascals
and
the
working
group
itself
to
feel
concern
and
review
those
documents
to
make
sure
that
in
the
pace
that
will
be
working
at
is
not
detrimental
to
the
quality
of
the
document
that
we
want
to
submit.
A
A
15
minutes
less.
B
C
C
We
discussed
this
in
the
ospf
and
is
is
working
groups.
I
initially
said
that
yeah
we
probably
have
to
change
the
segment
rather
attached
to
drive,
but
in
fact
I
don't
think
we
need
the
HR
lb
is
not
something
essential
to
the
architecture.
It's
a
small
IGP
extension
just
to
advertise
a
label
range
to
a
controller,
so
it
doesn't
impact
in
any
way
the
architecture
so
no
need
for
changing
for
changes
in
the
architecture.
D
Okay,
this
draft
has
been
progressing
I
guess
for
about
a
year,
I'd
like
to
think
we're
getting
close
to
the
resolution,
but
there
are
some
important
issues.
I
do
want
to
talk
about
today,
just
a
quick
summary.
Why
are
we
doing
this?
D
D
There
have
been
multiple
options
discussed
in
the
draft
over
the
past
year.
The
latest
version
of
the
draft
specifies
one
of
the
options
as
the
one
that
the
working
group
has
selected
and
I'm
going
to
talk
about
that
in
more
detail.
Next,
since
Berlin,
we
got
some
input
from
principally
one
service
provider,
with
some
support
from
a
second,
we
got
an
indication
that
one
vendor
has
implemented
ignore
overlap
only
which
was
one
of
the
policy
choices
and
we've
received
no
other
input.
D
D
Sms
advertisements
are
not
to
be
used.
This
gives
an
operator
a
way
to
start
configuring,
something
and
advertise
and
verify
it
before
it
actually
gets
used.
1
through
2
55
is
an
actual
preference
value
higher
being
preferred
by
default.
An
SMS
advertisement
is
has
a
value
of
128.
That
means,
if
you
don't
advertise
this,
if
there's
no
igp
advertisement
since
this
is
a
new
advertisement,
we
obviously
have
implementations
that
don't
support
it.
You
will
assign
128
those
advertisements
for
the
SIDS
that
are
advertised
in
prefix
reach
ability
advertisements.
D
We
have
not
changed
the
format,
so
we
haven't
added
any
fields
to
actually
advertise
a
preference
value
for
them,
they're,
just
all
assigned
a
value
of
192.
So
by
default
the
prefix
reach
ability
advertisements
are
going
to
be
preferred
over
SMS
advertisements,
but
given
the
values
we've
chosen,
if
somebody
wants
to
prefer
SMS
advertisements,
they
still
have
room
above
192
to
to
make
that
choice.
D
Next,
okay,
the
the
representation
of
a
generalized
mapping
entry
was
modified
slightly.
The
first
field
it
used
to
be
is:
this
is
the
source
from
a
prefix
reach
ability,
advertisement
or
ms
or
srms
advertisement.
It's
been
now
replaced
by
the
preference
value
other
than
that.
This
is
the
same
that
has
been
in
earlier
versions
of
the
draft.
So
if
you
look
down
at
the
bottom,
there's
two
examples
by
default:
if
you
have
a
preference
of
192
that
would
be
from
a
prefix
reach
ability.
Advertisement,
128
is
from
an
SMS
advertisement.
D
Next,
a
quick
review
of
terminology.
There
are
prefixed
conflicts
and
syd
conflicts.
The
prefix
conflicts
refers
to,
in
both
cases
we're
referring
to
the
object
that
has
a
conflict.
So
in
a
prefix
conflict
you
have
two
different
SIDS
assigned
to
the
same
prefix
in
a
CID
conflict.
You
have
the
same
CID
assigned
to
two
different
prefixes.
Next.
D
How
do
we
achieve
consistency?
Clearly,
everybody
has
to
have
the
same
database
or
it's
just
impossible
to
guarantee
that
you
will
come
to
the
same
conclusion
just
because
you've
configured
something
locally
doesn't
mean
that
you
include
it
in
your
conflict
resolution.
It
has
to
be
advertised
so
that
everybody
in
the
network
sees
this
value
as
well,
and
the
preference
for
the
advertisement
is
based
on
the
advertisement
itself,
not
where
it
came
from.
D
So
we
have
two
sources,
one
from
the
prefix
reach
ability
advertisement,
one
from
SRS
advertisement
and
again,
in
both
cases
you
you
only
use
things
if
they're
actually
up
for
advertised,
not
just
because
they
were
locally
configured
in
the
case
of
remote
advertisements.
It's
what
is
reachable
that
you
can
use.
If
you
can't
reach
it,
it's
potentially
a
stale
advertisement.
D
D
This
is
the
preference
rule.
The
preference
rule
has
not
changed
since
previous
versions.
Other
than
previously
we
had
specified
prefer
prefix
reach
ability,
/
SMS
advertisements.
One
point:
we
were
experimenting
with
the
other
way
around.
Now
we
have
the
preference,
so
we
just
substituted.
The
preference
value
was
the
first
tiebreaker
next.
D
This
is
a
working
example.
I
presented
this
the
last
time,
I'm
not
going
to
spend
a
lot
of
time
on
it.
I
left
it
in
the
slides
so
that
folks,
who
had
not
seen
it
before
I,
could
go
through
it.
So
if
we
could
go
next
next
next
next
and
one
more
okay,
this
is
just
a
summary
of
the
example
test
results
and
based
on
that
particular
example,
you
can
see
that
ignore
overlap
only
resulted
in
the
greatest
number
of
prefixes
that
still
had
a
CID
associated
with
them.
D
So
this
is
this
is
what
I
really
want
to
spend
a
little
bit
of
time
on
here
again,
based
on
the
input
we've
received
so
far,
ignore
overlap.
Only
is
what
we've
chosen,
who
we
don't
get
further
inputs.
That's
the
way
the
draft
is
going
to
proceed.
I
just
want
to
emphasize
some
points
about
the
complexity
that
this
introduces
in
your
implementation
and
in
achieving
interoperability.
D
Interoperability
means
you've,
got
multiple
implementations
in
your
network
if
they
don't
come
up
with
the
same
answer
based
on
the
same
database,
because
either
because
of
a
bug
or
because
of
a
misinterpretation
of
the
specification,
then
we're
not
going
to
achieve
our
goal
of
getting
a
consistent
resolution
network-wide.
So
here's
a
couple
of
examples
when
I
say
order
matters,
so
there
remember
there's
two
types
of
conflicts:
there
are
prefix
conflicts
where
you
have
different
SIDS
assigned
to
the
same
prefix,
and
these
these
conflicts
are
constrained
to
a
single
topology
and
the
single
algorithm.
D
There
are
Sid
conflicts
where
you
have
the
the
same
Sid
as
assigned
to
different
prefixes,
and
what
this
is
illustrating
is.
It
makes
a
difference
whether
when
you
apply
your
the
preference
rule,
whether
you
do
prefix
conflict
resolution,
first
followed
by
Sid
conflict
resolution,
or
vice
versa,
and
basically
what
happens
here
in
the
first
example,
we
take
the
first
entry
and
we're
looking
at
prefix
conflicts
first,
so
we
have
two
entries
for
192
021.
We
compare
them.
D
D
If,
however,
we
were
to
take
the
same
database
and
process
the
SID
conflicts
first,
then
we'd
start
with
the
two
entries
that
have
Sid,
200
and
we'd
pick
the
first
entry
because
it
has
lower
prefix,
and
then
we
compare
it
to
the
third
entry,
and
now
we've
got
two
entries
with
the
same
prefix,
so
we
have
to
choose
between
them.
We
end
up
choosing
the
one
with
the
lower
Sid.
D
D
Here's
another
example-
and
this
is
peculiar
to
the
ignore
overlap.
Only
policy
remember
an
ignore
overlap.
Only
if
I
have
an
SMS
entry
that
say
covers
that
has
a
range
greater
than
1,
I'm
going
to
operationally,
and
I'm
not
talking
here
about
how
the
code
might
be
implemented,
but
I'm
going
to
only
exclude
the
individual
entries
within
that
range
of
1,000
that
actually
have
a
conflict
and
end
up
on
the
losing
end.
D
I'm
not
going
to
go
through
all
the
steps
here,
I'm
happy
to
answer
any
questions
on
it
on
the
list
part
or
offline.
But
the
what
you
can
see
here
is
that
you
end
up
with
different
results
depending
upon
how
you
do
it
and
the
point
I'm
trying
to
emphasize
here
is
ignore.
Overlap,
only
introduces
implementation,
complexity
and
two
vendors
can
say.
I
have
implemented
the
draft
and
have
good
reason
to
believe
that
they
they
followed
things,
but
they
did
things
slightly
differently
and
they
end
up
with
different
answers.
D
And
so
we
have
not
achieved
the
goal,
which
is
to
get
consistent
conflict
resolution
policy
throughout
the
network
and
I'm
trying
to
emphasize
this
point-
and
it's
it's
really
been
appointed
discussion.
I
think
for
the
past
year
in
the
context
of
this
draft
is
that
there
are
there
sort
of
competing
goals
here.
One
of
them
has
been
the
Express
goal:
let's,
let's
try
and
maximize
traffic
delivery.
D
Do
we
want
to
invest
a
lot
of
time
in
implementation?
Complexity
and
comparing
implementations
to
see
if
they
actually
come
up
with
the
same
answer
under
the
same
conditions.
This
to
me
is
a
significant
risk
with
ignore
overlap.
Only
and
I'm
not
standing
up
here
and
trying
to
say
absolutely.
We
don't
do
it.
The
feedback
we
have
gotten
thus
far
has
been.
D
So
this
is
just
a
summary
as
a
result
of
the
the
examples
that
your
implementation
has
to
do
things
in
a
certain
order.
You
have
to
do
prefix
conflicts
before
Sid
conflicts,
and
you
have
to
look
at
the
entries
based
on
the
the
order
in
which
the
attributes
are
specified
in
the
preference
rule
next
slide.
D
And-
and
we
need
to
clearly
state
that
there
is
some
text
in
the
draft
now,
but
I
think
it
needs
to
be
embellished
to
indicate
that
just
following
the
preference
rule
doesn't
guarantee
interoperability.
You
also
have
to
follow
the
order
and
I
put
a
little
side
comment
here,
which
is
the
complexity
is
associated
with
the
ignore
overlap.
Policy
ignore
overlap.
Only
if
we
were
to
choose
it'd
be
the
simplest
policy
which
is
you've,
got
things
that
conflict.
You
don't
know
which
one
is
correct,
ignore
them
both.
D
We
would
not
have
this
complexity
next,
or
is
it
that
the
last
one
now
so
look
I
think
this
is
the
last
I
just
want
I
would
really
want
to
get
more
feedback
if
we
can
that
this
is
the
direction
that
the
working
group
wants
to
go
if
there
are
people
out
there
who
have
been
silent
up
to
now.
This
is
really
the
opportunity
to
speak
up,
because
if
you
don't
I,
think
that
you
know
we're
done,
the
draft
will
will
get
polished
up
and
and
will
progress,
some
become
an
RFC.
D
E
Hand
this
is
a
much
enduring
so
let's
say
a
last
revision.
I
didn't
read
it
actually
so,
but
I
have
been
following
and
I
have
a
lot
of
comments
in
the
mailing
list
earlier
versions.
So
after
looking
into
more
this,
like
you
know,
keep
it
simple
is
extremely
important,
but
you
know
you
already
progress
so
far
that
you
know
you
want
to
make
it
complex
but
sokale
at
the
point.
What
I
might
say
is
one
question
first
for
question:
is
you
said
that
even
you
follow
the
preferences
and
order
you
see
still
see
interoperable
tissues?
D
E
G
Da
juniper
networks
I
have
a
comment
on
the
preference
rules
I
in
network
consolidation
cases
like
if
multiple
protocols
are
running,
then
we
should
have
a
protocol
preference
also
in
the
preference
set
of
preference
rules
and
I
think
it
should
be
on
the
top.
Because,
generally
we
apply
the
protocol
preference
I
mean
if
there
is
I
mean
when
network
operators
look
at
the
network,
they
think
of
protocol
preference
as
paying
as
getting
the
highest
preference.
So.
D
So
this
is
actually
explicitly
stated
in
the
draft.
The
affording
plane
is
protocol.
Stick.
It
does
not
care
whether
the
route
was
put
in
by
ospf
or
ice
is
so
the
source,
whether
it's
a
protocol
difference
whether
it's
a
router
ID
difference
isn't
deliberately
not
taken
into
account
in
the
preference
rule.
G
D
G
H
So
my
without
much
analysis,
my
my
feeling
is
that
you
know
if
you
have
a
cogs
like
something's
broken
and
whatever
allows
operations
to
find
it
and
fix
it.
The
fastest
is
goodness-
and
it
seems
like
it's
more
simple-
ignore-
achieves
that
much
more
than
this.
Somebody
might
look
at
the
prefixes
that
they
end
up
with
and
be
scratching
their
head
for
some
time
before
they
figure
out
what
combination
of
things
caused.
It
I.
I
Lobbying
for
hobby
I
have
a
I.
Have
some
comments
on
this
dropped.
Okay,
the
first
one
I
think
I
agree.
They
see
the
problem,
but
I
think
in
fact
that
the
problem
we
should
not
be
you
exist
if
we
use
the
caliber
label,
yes
and
also
I,
know
the
major
vendors.
The
not
know
I
am
here
today
asking
a
lot
of
vendors.
The
only
directly
use
the
label
instead
of
SRD,
be
okay,
there's
a
person.
This
is
a
fact.
Okay,
any
sense.
This
problem
can
be
very
simple,
the
second
a
while.
I
Even
you
can
you
the
srgb
or
converse
this
global
second
idea
to
the
local
label.
Okay,
no
problem!
You
want
to
make
it
a
complex,
okay,
but
I
sing
from
the
point
of
engine
engineering
real.
We
can
also
make
it
as
simple
either
justice
that
configuration
usual
in
an
order
taker
so
much
so
much
time
to
talk
about
this
problem.
That's
my
comment.
Okay,.
D
So
let
me
see
I
respond
to
both
the
the
problem
that
arises
here
has
nothing
to
do.
We
are,
in
fact,
using
a
you
know
the
srgb,
but
if
you
have
a
configuration
on
one
node
that
assigns
index
1004
one
dot,
100
101
and
on
another
note
it
assigns
index
2004,
wonder
wonder,
wonder
one:
you
have
a
conflict,
that's
the
problem!
We're
trying
to
solve
I.
J
Jasa
from
Huawei
I
have
some
questions
about
this
draft
because
o
in
this
draft,
we
have
listed
some
scenarios
for
the
comfort
of
conflict,
but
I
single
person.
There
is
a
lot
of.
There
is
some
other
scenarios
of
confidence
like
if
we
have,
if
the
same
prefix
have
a
different
ID
how
to
solve
this.
So.
J
D
J
And
and
another
scenarios
about
is
because
we
do
because
we
define
the
the
in
the
in
the
in
the
domain.
We
have
those
SIDS
global,
but
it
is
different
domain
like
a
different
area
or
different
process.
J
How
how
how
to
consider
this
scenario,
because
sometimes
we
have
to
import
the
prefix
from
a
lot
one
area
to
another
area
right.
This
is
another
scenario
and
and
non
scenario
is
the
different
IP
protocol.
If
you
have,
if
you
have
the
is
easy,
I
I
size,
the
protocol
and
ospf
protocol,
if
they
define
the
same
o,
SI,
d44,
prefix
or
four
different
prefix
how
to
resolve
this
problem
or
resolve
this
conflict.
So
maybe
we
can
give
some
clarification
for
such
scenarios,
so.
D
That
they
transfer
your
last
point.
The
scenario
is
covered,
it's
already
covered
in
the
draft.
As
far
as
your
previous
point,
what
we're
dealing
with
here
is
the
fawning
plane
for
a
particular
domain,
because
if
we
get
to
the
edge
of
the
domain,
then
we're
going
to
to
potentially
use
a
completely
different
Sid
and
different
label
to
go
into
the
other
domain.
That's
a
different
database,
we're
not
trying
to
resolve
conflicts
between
domains.
I
We
we're
quick
I,
just
a
comment:
okay
I
think.
Maybe
I
tell
you
that
other
working
group
user
doing
the
PDP
lingus
state
and
also
that
the
and
also
the
pc
union
state
shutdown
had
proposed
the
policy
usual
about
this
overlap
between
the
SS
and
ospf,
and
also
we
maybe
have
this
the
overlap
issue
between
the
ospf
OSS
between
in
the
state
p
ceiling
state.
You
will
get
more
and
more
time
to
solve
this
usual.
Okay,
so
introduce
more
and
more
complex
policy.
I
K
K
Smoky
eye,
I
think
what
is
important
here
I
think
you
have
to.
We
have
to
decide
between
as
less
stated
before,
between
minimizing
the
impact
when
you
have
conflict
first,
the
complexity
of
implementation
and
I
think
well,
I
would
advocate,
is
two
operators
who
want
to
deploy
this
technology
what
they
really
would
like
to
get
and
achieve
in
this
situation.
K
Of
course,
you
would
like
to
avoid
this
conflict
by
all
means,
but
there
is
manual
things
which
will
result
into
this
situation
and
I
think
what
is
important
element
would
like
to
stress
this:
to
get
operators
input
to
this
question
because
I
think
fenders.
We
all
prefer
to
do.
Keep
it
simple,
although
we
implement
it
to
more
complex
schema
at
the
moment.
I
think
it's
important
to
understand.
What
is
the
operators
input
on
this,
because
that
will
probably
determine
where
we
have
to
go
right?
K
C
Exactly
I
just
want
to
emphasize
what
wind
said,
so
I
am
happy
to
see
that.
Finally,
people
realize
that
simplicity
is
a
good
thing,
unfortunately
so
far,
but
has
been
very
silent.
Doom
I
cannot
in
the
mailing
list,
so
it's
good
to
converge
towards
a
request
for
simplicity.
It
would
even
better
if
more
operators
can
speak
up
and
tell
us
what
they
want,
and
so
we
can
close
this
in
the
best
way,
but
so
far
we
didn't
have
much
voices
speaking
up
on
that
on
that
sign.
C
L
Same
question
too:
yes,
I
jungle,
joe
from
google
I'm,
not
really
interested
in
the
solution
to
this
problem.
To
be
honest
with
you,
I
would
rather
take
the
simple
approach.
We
manage
a
number
of
number
of
spaces
across
various
routers
in
our
domain
and
I.
Think
we'll
take
the
risk
that
there
is
a
an
issue.
If
you
do
miss
configure
a
router
that
we
can
then
correct
very
quickly.
M
N
O
N
M
Light
ok,
so
this
presentation
is
about
an
update
for
the
segment
watching
young
mother
draft
for
next
slide.
Please.
So
it's
been
a
very
long
time
since
we
presented
via
latest
update,
so
just
a
good,
but
it's
a
very
straightforward.
So
this
is
a
configuration
and
operation
model
for
segment
wedding,
so
just
the
base
of
segment
routing
or
the
igp
extensions,
so
activating
signal
routing
for
ISAs
an
ospf.
It's
in
a
separate
document
that
are
in
a
dedicated,
walking
home
next
slide,
so
were
about
the
change.
M
M
M
Normally,
it
cannot
happen
because
the
sender
is
not
allowed
to
propagate
multiple
overlapping
srgb,
but
at
the
same
time
we
conflict
resolution
drafted-
if
I
remember
correctly
say,
is
that
the
receiver
must
check
that
it
does
not
happen.
So
if
this
check
is
really
necessary,
I
think
it
would
be
interesting
to
have
a
notification,
but
we
can
discuss
on
this.
M
We
added
also
support
for
a
different
algorithm
for
the
prefix
seed,
so
especially,
we
added
via
Street
SPF.
That
is
still
under
discussion,
but
at
least
it
is
where
former
configuration
point
of
view,
we
also
added
the
support
of
MSD.
So
we
have
also
a
feature
for
this,
so
it's
optional
and
we
have
both
configuration
of
the
MSD
and
also
operational
state.
M
So,
although
I
add
some
discussion
this
week,
weaver
with
jeff
who
raised
visa
MSD
stuff
so
today,
things
that
were
some
may
be
other
issues
that
are
not
able
to
expose
directly
via
MSD
22.
We
control
planes
which
it
may
be
interesting
to
have
a
kind
of
configuration
if
we
cannot
directly
retrieve
the
information
of
the
other.
M
M
We
were
requested
by
some
people
from
aight
where
her,
so
these
that
are
working
and
topology
modeling
to
be
able
to
reuse
some
of
our
groupings
and
at
the
same
time,
they
did
not
want
to
import
the
old
segment
routing
model,
because
it's
not
rest
aside,
it's
not
necessary
for
when
they
only
want
some
really
piece
of
I
would
say
encoding
on
some
stuff.
So
we
separated
the
groupings
in
two
parts.
M
We
have
still
some
work
to
align
with
all
the
latest
draft,
especially
including
via
SLB,
via
SMS
preference
is
not
very
yet
maybe
adding
also
a
new
notification
from
the
prefix
conflict.
So
we
have
some
alignment
to
our
to
do,
but
it's
it's
in
good
condition
to
maybe
close
it
soon
that
I
I
don't
know.
If
we,
you
will
need
to
wait
to
maybe
close
via
so
config
draft
and
son
before
going
to
last
call
this
document.
So
we
are
welcoming
comments,
reviews,
critics
or
whatever
so
feel,
free
to
raise
your
point.
M
M
I
So
I
think
that
either
you're
from
this
operator
I
think
that
you'll
have
this
the
experience
enough.
You
view
the
maintenance
and
the
management
of
the
network.
Ok
I
wanted
to
do
this.
One
now,
I
think
the
most
important
challenger
for
the
operator,
either
to
a
a
most
importantly,
is
a
faster
service
provini
to
get
more
revenue.
Is
that
the
right?
Yes.
I
I
I
M
I
I
And
now
I
identities,
castle,
young
I
tell
you
what
will
happen
for
the
young
model,
because
the
faster
for
the
past
three
years,
I
probable,
other
ideas
about
the
sec-
routine
van
wise,
the
tyrant
here,
the
segmented
tunnel
segment,
Marty
topologies
segment,
because
this
may
be
the
new
features.
Yes,.
M
M
I
M
M
If
you
have
five
vendors
and
each
vendor
is
taking
in
twins
its
implementation,
a
different
approach,
and
when
you
have
a
multi-vendor
network,
we
will
not
be
able
to
build
an
end-to-end
service,
because
all
these
options
may
be
optional
in
terms
of
implementation,
I
and
I.
Never
seen
a
vendor
in
commenting
already
possible
option,
so
we
need
to
have
a
kind
of
recommended
solution
that
everyone's
everyone
need
to
support
and
then
it
would
be
up
to
implementation
to
add
other
options
as
well.
But
at
least
this
one
may
be
monitoring.
M
So,
first
one
to
goal
of
the
draft,
so
we
need
to
present
the
various
options.
So,
for
the
moment
it's
really
opened.
We
don't
have
any
recommended
solution
and
please
don't
blame
me
about
the
content
of
a
draft.
I
wrote
it
to
be
honest.
I
rotted
couple
of
minutes
before
going
to
vacation,
so
journalist
is
not
really
complete.
It's
really
far
from
this,
but
I
wanted
absolutely
to
start
this
discussion
within
this
idea.
That's
why
I
wrote
with
draft.
M
We
need
to
remember
that
with
spring
we
have
some
constraints
today.
Maybe
these
constraints
will
disappear
in
couple
of
years,
but
we
still
have
issue
with
MSD,
so
capabilities
of
a
particular
not
to
push
a
number
of
flavors,
and
we
have
also
some
load
balancing
issue
is
router
capable
to
inspect,
for
example,
the
IP
payload
or
ashing
of
stack
with
10
20
25
labels.
M
Is
it
really
realistic
or
not
so
these
constraints
we
need
to
take
them
into
an
into
account
when
we
will
decide
we
recommended
solution,
so
we
are
trying
to
highlight
vid
pros
and
cons,
I'm,
not
sure
it's
really
good
today,
but
we
need
to
progress
on
this.
It's
really
just
first
overview
and
the
goal
is
really
to
provide
as
an
outcome
a
recommended
solution.
M
So
we
have
a
Saudi
flag
when
we
are
able
to
have
multiple
edges
on
CC,
don't
particularly
so
we
can
do
it,
but
I
think
everybody
can
see
that
there
are
some
challenges,
because
if
I
have
20
up
to
cross,
I
will
need
maybe
19
19
labels
just
for
the
transport
and
then
I
will
need
to
add
entropy
and
also
service.
Libel
suits
does
not
seems
to
be
realistic
honestly,
so
we
can
try
to
compress
the
stack
by
using
a
combination
of
nodal
segments
and
adjacent
segments,
but
here
we
have
also
multiple
sub-options.
M
First
one.
We
need
to
not
see
that
is
not
protected.
So
why
not
relying
on
this
new
street
SPF
not
seen.
So
there
was
some
discussion
on
this
and
with
definition
of
the
strict
SPF
not
seen
so,
if
I'm
using
the
strict
SPF
not
see,
does
it
prevent
it
to
be
fast
rewarded.
So
it's
an
open
question
and
I
think
we
need
a
clarification
on
this
because
normally
strict
SPF
it
it's
following
the
SPF
and
it
cannot
go
away
alviso.
Is
it
fast
throughout
included
in
this
or
not
such
a
question?
M
M
We've
not
said
is
that
we
can
create
as
an
operator
to
not
sit
and
we
managed
by
local
policies
the
activation
of
fast
free
world
for
a
particular
node
seed
or
not,
but
we
will
need
different
prefixes,
so
it
may
increase
our
IGP
table
and
so
on
so
I
some
impactor
as
well.
We
there
was
a
last
option:
Martin
I,
no
worry
so
last
option
is
to
use
a
combination
of
adjacent
CC
than
binding
seed,
and
we
may
be
able
to
use
binding
seed
to
represent
a
list
of
adjacent
cc'd.
M
So
let's
say
that
I
have
a
controller.
Some
very
smart
controller.
You
may
be
able
to
place
some
transit
tunnels
within
the
network,
associating
a
binding
seed
with
this
and
then
I
will
be
able
for
my
aunt
went
LSP
to
reuse.
This
painting
sheet
sets
an
option
as
well.
Can
I
take
the
question
after
the
presentation.
I
almost
finished,
so
the
next
step.
So
clearly
we
need
to
go
ahead
with
the
analysis.
Don't
really
take
care
about
what
is
written
to
the
in
the
draft,
so
we
need
to
progress
on
this.
We
need.
M
We
really
need
feedback
from
the
working
group
on
the
pros
and
cons
and
also
is
there
some
other
options
that
we
missed.
Maybe
the
really
important
point
is
that,
as
an
operator,
we
have
some
deployment
requirement,
so
we
need
to
have
running
code
as
soon
as
possible.
So
we
need
to
come
with
this
recommended
solution
in
a
really
fast
manner.
M
N
M
N
A
service
point
of
view
or
for
transporting
transfer
point
of
view,
each
other
you'll
have
VPN
or
any
other
service
on
cup,
which
will
use
the
same
label
and
same
IP
address
in
the
transport
network.
At
some
point,
he'll
have
to
customers
one
asking
for
protected
path
while
another
one
asking
for
not
protect
pass.
That's.
M
N
N
A
N
If
your
customer
wants
to
use
their
own
ingress
protection,
so
be
they
just
need
to
make
sure
that
it
kicks
in
before
you
try
to
reactivate
your
transport
or
whatever.
It
is
yes,
but
we
don't
really
know
what
are
the
times
that
are
used
by
all
customer.
So
sup
my
point,
the
problem,
spaces
slider
and
we
need
serious
to
consider
it's
a
valid
problem
and
what's
working
on
absolutely
okay,.
H
B
Crisp
errors,
so
can
you
go
back
to
the
slide
with
on
the
potential
solutions
you're
looking
at
with
ya,
even
one
back
I
think
no,
no!
Ok
forward
forward.
Sorry,
so
you
mentioned
not
knowing
the
timers
that
the
customers
require
I
mean
like
with
the
node
Sid
solution.
B
It
seems
to
me
like
eventually
it's
going
to
converge
with
the
igp
again,
that's
true
now
you
know
that
could
this
occur
in
a
second,
so
is
that
is
that
a
problem
I
mean
what
are
the
requirements
on
not
protection?
Is
it
not
protection
forever
or
not
protection
for
five
seconds
or
10
seconds?
What's
the
goal
here
in.
M
Terms
of
requirements,
it's
a
question
fairy
world
is
not
acceptable,
but
I
think
regular
conversions,
maybe
because
there
will
be
a
kind
of
fast
free
world
like
mechanism
at
just
an
outside.
So
if
the
HP
concerns,
I
think
it's
fine,
because
we
need
absolutely
to
reestablish
the
path,
so
the
automatic
reconversion.
So
the
path
is
needed.
O
Acm
I
think
I
I
haven't
read
the
drafters
thought
much
about
it,
so
I
won't.
I
may
have
already
disclosed
that
I
have
an
idea
which
of
the
solutions
I
prefer,
but
I
think
it's
good
you
can
tell
it.
I
think
number
two
is
is
the
best
right
now,
which
flavor
of
nimble
too,
adding
the
protection
flag
to
a
CID
and
having
multiple
states,
but
I,
but
I
got
to
think
about
it
more.
I
can
see
less
shaking
his
head
so
because
I.
N
You
so
if
you
see
my
tickle
stick,
SPF,
meaning
there's
going
to
be
a
GP
convergence
and
potentially
you'll
get
unto
the
shortest
paths
again
right
right
now
in
jail.
If
a
you
could
pre
calculate
button
worst
apology
and
have
your
back
up
following
your
post
conversion,
so
that's
no
really
different!
You
just
do
it
faster!
That's.
M
Varies
because
the
you
have
to
take
the
path
from
an
end
to
end
point
of
view,
so
even
if
transiently
is
strict,
SPF,
not
CDs,
rerouted
on
the
same
path
that
tilf
a
may
take.
The
controller
also
ended
the
head
end.
If
so,
then
is
responsible
for
end-to-end
path,
computation
may
reroute
and
may
not
use
anymore.
This
not
seen
because
the
constraint
path
has
changed.
G
Stiffened,
the
real
requirement
is
not
very
clear
to
me
an
end
customer.
He
need
not
be
bothered
about
it's
a
protected
path
or
non-protected
part
I'd
really
bothers
whether
it's
his
bandwidth
requirements
are
met,
or
you
know
the
requirement
has
to
be
something
of
that
sort
and
not
really.
I
don't
want
and.
M
We
it's
it's
like
transmission,
like
services
today,
for
example,
as
an
operator
when
we
are
ordering
some
WDM
links,
we
can
ask
for
protected
links
or
non-protected
links,
because
we
are
doing
the
backyard
on
our
side,
because
we
are
ordering
another
link
which
is
using
a
completely
another
path.
So
this
is
really
similar.
So
sometimes
we
are
setting,
for
example,
Ethernet
links
about
our
mpls
networks
that
we
resell
to
evaporators.
So.
I
Will
be
our
house
of
having?
I
have
several
comments.
The
first
is
that
it
can
you
clarify
your
the
application
scenario
of
these
draft,
the
first
active
user,
you,
the
first
you
the
code.
You
got
these
two
of
the
srt
and
SRB
this
one
scenario:
a
secondary
school
is
a
tea
and
what,
secondly
routine
traffic
engineering.
Yes,.
I
I
mean
such
a
scenario:
y
si
Nardo
I
accuse
the
three
option.
You
tell
me
which
way
either
you
get
this
usual
and
you
one
solve
this
usual
okay
in
these
applications
in
our
the
first
one
is
a
coexistence
that
IP
network
a
coexistence
of
the
SRT
and
SRB
okay.
This
is
one
scenario.
Second,
is
out,
be.
M
M
I
Word
may
be
different
from
your
world.
Ok,
the
second
way
is
that
the
coexist
of
our
DB
and
sr
PE
as
our
past,
the
effort-
ok,
our
DP,
also
your
eyes,
rdp.
Also
you
can
cease
to
you
in
this
network
life
in
a
second,
the
third
way,
totally
new
network
for
sr.
So
three
scenario:
what
a
scenario
you
get
this
usual
and
probe
hold
the
solution.
It's.
I
I
I
L
M
I
I
M
I
C
Hi
Stefan
Oprah
Cisco,
so
these
draft
came
up
after
I
would
say.
A
description
of
a
use
case
of
a
requirement
has
been
expressed
in
the
Middle
East
and
the
requirement
was
next
slide.
The
ability
to
deploy
on
a
single
router,
multiple
addresses
of
the
router
associated
with
different
services,
but
all
those
addresses
would
use
the
same
set
now.
The
segment
erotic
architecture
is
pretty
clear.
C
One
graphics
one
said,
and
we
don't
want
to
change
the
architecture
also
because
we
think
there
is
obviously
an
architecture
compliant
solution
to
this,
and
this
is
where
the
solution
is
documented.
In
this
draft.
Now
we
send
this
draft
I
mean
we
submit
it.
We
send
it
to
mailing
list,
and
surprisingly,
the
individual
that
expressed
the
requirement
in
the
use
case,
never
came
back
so
I
don't
know
so
next
night.
C
The
solution
is
very
simple:
instead
of
advertising
the
prefix
with
the
set
you
advertise,
the
prefix
with
a
new
piece
of
information
into
which
you
encode
would
ease
the
sage
and
the
prefix
owning
that
said,
you
want
to
recurse,
so
it's
pretty
simple:
we
defined
a
new
sub
TV
in
both
protocols,
ospf
an
ASIS.
We
call
this
the
segment
rodding
recursing,
the
information
next
slide
and
we
encode
the
the
global
said
that
this
preface
should
use,
and
this
global
stage
in
fact
belongs
to
another
projects.
C
That's
why
you
need
a
different
encoding
for
for
expressing
that,
of
course,
you
have
the
usual
flags
and
algorithm
and
optionally.
You
can
also
add
the
local
state.
Let's
have
a
look
at
the
picture
next
slide.
So
this
is
what
we
want
to
encode
in
in
writing
protocols.
But
let's
skip
this
one
for
now,
let's
focus
on
the
functionality.
C
So,
on
the
right
side,
you
have
a
rather
rather
a
which
has
a
prefix,
a
loopback
address,
2
/
32,
with
the
notes
it
eighteen
thousand
and
two,
and
then
you
may
have
multiple
other
prefixes
representing
different
services
and
the
requirements
that
has
been
expressed
at
the
beginning
was
I.
Want
those
services
to
be
reachable
through
the
same
set
so
that
I
have
taught
okay,
just
one
single
said
for
all
those
prefixes,
because
apparently
there
are
platform
that
still
have
problems
and
allocating
a
certain
range
of
labels.
C
The
is
is
LSP
illustrated
there.
You
can
see
that
prefix
200
10
has
the
the
new
sub
Calvi
srri
with
includes
the
IP
address
to
the
Torah
22,
which
is
the
loopback
address
of
the
router
with
its
own
notes
it.
So,
on
the
left
side,
you
have
the
the
fame.
Sorry,
it's
not
rudder
a
of
course
it's
another
router
and
for
the
prefixes
that
have
been
advertised
with
the
segment
rodding
recursing
info
sub
TLB,
you
see
there.
C
What
is
the
label
that
correspond
to
that
service
based
on
the
encoding
received
in
the
writing
protocol,
so
the
requirements
eased
rests.
You
have
multiple
services
that
in
fact
may
use
the
same
set.
Still
you
are
compliant
because
prefix
has
a
say
dundas.
It
belongs
only
to
one
prefix,
and
so
you
can
multiplex
by
doing
this
kind
of
recursion
in
fact
next
slide.
C
So
this
is
just
the
text
that
illustrates
that
mean
that
explains
the
the
picture.
So
next
slide
the
benefits.
As
I
said,
it
addresses
the
requirements
of
limiting
the
number
of
labels
you
want
to
allocate
in
a
box,
and
you
covers
all
the
scenarios
that
you
have
in
an
igp
multi
area
and
a
single
hole.
Multi-Home
practices,
and,
and
so
on,
so
we
submitted
this
draft
here
in
in
spring.
We
are
going
to
have
obviously
the
equivalent
drafts
in
writing.
Protocol
groups.
Ospf
I'm
is,
is
especially
that's
pretty
much
it
Russians
I.
I
Know
no
this
one
solution,
the
problem,
so
the
previous
one
yeah,
ok,
ok,
the
this
increasing
number
of
a
required
global
SIDS,
yes,
eleven!
Oh
yes,
okay,
I,
agree
that
one
since
I
read
one
meter
advantage
of
a
second
routine
use.
The
editors
days
are
only
meeting
at
the
head
and
under
a
no
stated,
is
necessary
to
be
meeting
at
the
midpoints
and
appearance,
but.
M
I
The
development
of
seven
regimes
there
will
be
more
segments
which
will
be
used
on
multiple
sorry,
not
here
for
multiple
purposes
such
as
any
casa.
Second
service
Cheney,
the
extra
other
than
limit
he
didn't
know
the
segment
an
agency
said
under
this
may
be
more
state.
There
may
be
most
days
in
the
secondary
routine
faced
network
and
the
flooding
feature
of
igp
will
make
all
nodes
in
the
network.
How
to
maintain
these
states,
which
may
be
unnecessary
to
some
no
Reuben.
Okay,
I
saw
I.
Just
tell
you.
This
is
my
draft
okay
I.
I
C
I
I
I
Q
C
C
You
do
PHP
when
the
packet
comes
to
you,
where
I
mean
you
are
the
rudder
who
are
located
that
local
label,
so
you
know
that
local
label,
what
is
the
meaning
in
the
semantics
of
that
local
level,
but
if
you've
also
received
the
same
label
value
as
a
global
level?
No,
but
if
you
do
PHP
and
you
are
you
receive
that
label,
you
know
that
that
label
is
on
the
local
range
label,
and
so
you
have
allocated
it.
So
you
know
what
is
the
function
associated
to
it?
Okay,
I'll.
Take
that
away
now
the.
C
I
Okay,
lobbying
from
hobby,
okay,
I
think
if
you're
like
I,
I
hover
because
of
the
memory
may
be
passed
away
with
the
year.
I
hope
you'll
read
the
mailing
list
of
a
spring
working
group
carefully.
Okay,
maybe
remind
your
memory.
I
also
I
can't
get
at
this
point
and
a
poster
to
you
again,
I
the
mainly.
I
C
I
R
Can
I
have
one
more
question:
meteor
Nassif,
Yandex,
actually
I
see
useful
cases
for
that,
but
I
have
an
impression
that
model
is
the
same
can
be
achieved
with
just,
for
example,
is
distributed.
Seeds
and
something
like
labeled
bgp
sessions,
for
example
between
local
node
and
I,
get
note
and
using
recurse
different
for,
and
so
probably
it
would
be
useful
to
have
some
additional
discussions
about.
Where
went
to
choose
what
case.
I
absolutely
agree.