►
From YouTube: IETF110-LSVR-20210309-1430
Description
LSVR meeting session at IETF110
2021/03/09 1430
https://datatracker.ietf.org/meeting/110/proceedings/
A
A
Okay
with
at
the
start
of
our
slot,
so
we
only
have
like
one
hour,
so
you
know
maybe
let's
get
started
there,
so
you
have
landed
into
the
lsvr
working
group
and
we
have
a
few
interesting
topics
here
now
before
we
start.
Let
me
see
if
my
slides
actually
work.
Okay,
so
we
don't
need
the
scribe
thing,
because
we
have
the
scribe
automatically
building
into
the
tool
nowadays,
which
is
very
handy.
A
If
there
are
like
any
minutes,
then
you
know,
please
feel
free
to
use
the
menu
tab.
You
know
on
top
of
the
meet
echo
screen,
you
don't
have
to
do
this,
you
know
just
by
yourself.
We
can
actually
all
do
this
together.
So
that's
very
hands
handful.
You
know
very
handy
and
very
convenient,
so
that
would
be
appreciated
next,
is
you
have
seen
this
note
well
already
a
few
times
because
we're
already
on
the
second
day,
if
any
questions
you
know,
please
be
aware
of
this,
then
the
agenda
bashing.
A
So
what
we
have
on
the
agenda
at
this
point
in
time
is
three
topics.
Three
main
topics
here:
first,
we're
gonna
get
like
an
l3dl
status
update.
Then
we're
gonna
have
paul
speaking
a
bit
about
ldpv2,
which
is
you
know
what
is
happening
in
the
ieee.
A
So
where
are
we
within
the
in
our
working
group
here?
So
we
have
sent
two
documents
towards
the
id
the
id
reviewed
it
and
actually,
you
know,
send
it
back,
but
you
know
together
when
sending
it
back,
he
actually
sent.
You
know
it
back
with
like
a
thousand
one
pages
of
feedback
which
the
authors
have
been
working
upon
very
intensively.
You
know
over
the
new
year's
period
until
now
and
ac
is
going
to
give
like
an
update
on
where
we
are
at
this
point
in
time
with
these
particular
drafts.
A
Now,
in
the
meantime
also
last
month
we
have
done
a
working
working
group
last
call
on
the
l3dl
documents,
while
no
major
concerns
were
erased.
You
know
there
was
also
not
really
a
lot
of
feedback
on
that.
So
you
know
we
need
to
figure
out
how
to
deal
with
that
going
forward
now.
That
being
said,
there
is
another
element
playing
also,
and
that
has
to
do
with
the
fact
that
l3dl
is
a
little
bit
out
of
scope
from
our
shorter.
A
So
what
we
are
thinking
about
doing
is
to
reshorted
the
working
group
by
explicitly
adding
the
l3dl
components
into
the
charter
itself,
which
would
give
us
you
know
a
better
foundation
to
keep
progressing
on
this
particular
technology
going
forward,
and
that
being
said,
you
know
for
in
an
effort
for
trying
to
enhance
the
feedback
required.
You
know,
feedback
requested
and
reviews
requested
for
these
documents.
A
I'll
also
be
you
know,
launching
a
call
on
the
working
group
email
alias
for
some
volunteers
to
actually
give
feedback
on
this
on
these
documents.
Just
you
know
more
discussions
on
that
going
forward.
A
D
A
B
Okay,
this
sue
raised
an
issue
by
a
private
email
which
was
interesting,
but
first
just
a
little
bit
more
on
status
is
the
documents
the
three
l3
tl
documents
passed
within
the
class
call
with,
as
gutter
said,
light
feedback,
but
you
should
be
aware
that
also
that
rob
is
contemplating
upgrading
updating
the
lsoe
open
source
implementation
to
l3dl,
so
that
kind
of
wraps
the
status
so
sue
raised
an
issue
about
the
hello
pdu,
possibly
causing
a
layer
2
storm
next
slide.
Please.
B
B
B
So
the
document
says
that
that
hello
should
be
repeated
at
a
configured
interval
by
the
default
of
60
seconds.
So
when
the
device
wakes
up,
it
should
find
its
friends
within
a
mean
of
30
seconds
and
a
max
of
60
seconds,
but
it
further
warns
that,
if
you're
in
the
multi-link
scenario,
in
other
words
you're
allowing
for
intermediate
switches,
be
aware
of
a
trade-off
between
the
timer
tuning
and
network
noise
and
maybe
tune
the
timer
up
or
down
accordingly,
but
sue
raised
a
further
issue
next
slide.
Please.
B
B
E
By
not
that
means.
B
F
E
Okay,
I
would
go
through
the
brief,
because
if
people
understand
multicast
storms,
this
may
be
a
simple
issue.
If
the
brief
doesn't
make
it
we'll
go
through
the
more
lengthy
approach.
E
So
folks,
the
diagram,
I'll
explain
and
then
I'm
gonna,
let
randy
go
through
the
discussion
because
he
tends
to
be
a
bit
briefer
than
I.
The
concept
is
you
don't
have
the
topology
described
in
the
document?
Okay
and
randy
will
go
through
it,
but
you
have
this
more
interconnected,
switch
topology.
How
you
got
there,
not
my
problem.
E
Perhaps
it's
recon
errors,
and
so
this
is
the
aero
topology
note
I
realize
it
is
not
a
standard
class
topology.
It
would
be
created
by
air
or
by
something
else,
so
randy.
Why
don't?
Why?
Don't
you
go
ahead
or
I
can
explain
the
next
step.
B
It's
fine
so
essentially
the
routers,
the
things
labeled
routers
are
l3dl
speakers,
the
things
not
labeled
routers,
the
little
square,
cubes
oops,
that's
bad!
The
little
cubes
are
switches,
whereas
if
you
use
the
multicast
to
be
assigned
multicast
packets
will
go
not
only
through
them,
but
it
will
as
it
receives
one
next
slide,
please
it
will
re-announce
it.
B
Okay,
you
see
the
multicast
hello
coming
up
from
router
a
and
switch
one
re-announcing
app
on
all
its
other
interfaces.
Next
slide,
please
those
other
destinations
generously
do
similarly
and
we
can
see
more
pretty
colored
little
boxes.
Multiplying
the
network
next
slide,
please
those
in
turn
echo
back
now.
What
was
fun
here
is
that,
because
the
echoes
caused
regeneration,
they
go
back.
B
B
G
E
That's
a
good
point:
this
is
an
edge
case
bill.
The
spanning
tree
is
most
likely
not
running.
You
would
have
some
interconnections
that
are
are
normal.
So,
as
I
started
out
in
the
beginning,
this
broadcast
storm
only
works
with
the
compine
with
a
set
of
edge
cases.
Okay,
you
don't
have
something:
that's
living
running
a
spanning
tree
you're
using
the
multicast
transmission,
which
goes
through
switches
and
then
redistributes
widely
paul.
I
don't
know
if
this
is
even
possible.
B
Just
just
bill,
let
me
see
if
I
can
cut
to
the
chase
here-
excuse
the
spoiler
the
final
scene
of
the
movie.
Is
this?
Isn't
our
envision
topology
in
this
working
group?
This
isn't
a
club
and
there's
tons
of
intermediate
switches
which
we
don't
envision.
Envisions
ls
are
enabled
device
ls
vo
to
enable
devices
all
on
link
to
link
just
in
case,
but
we
can
move
forward.
Let
me
move
forward
next
slide.
Please.
H
Yeah
I
just
was
gonna
echo
kind
of
what
bill
said.
We
have
it
obviously
kind
of
an
invalid
layer,
two
topology
here
there's
besides
multi-caps,
there
would
be
other
flooded
packets
that
were
not
learned
yet
or
various
things.
So
I
also
wanted
to
point
out
spanning
trees,
not
your
only
topology
management
protocol
at
layer,
two
there's
other
approaches,
including
software
defined
networking
things
or
as
well
as
shortest
path,
bridging
and
so
to
make
this
work.
There
must
be
some
kind
of
loop
free,
topology
management
going
on
that
would
prevent
this.
I
mean.
B
B
B
But
isn't
this
isn't
our
model?
Okay,
we
don't
have
a
model
with
all
these
switches
with
this
dumb
switch
mesh
next
slide.
Please
our
model
is
a
club,
a
standard,
clothed
fabric
and
they're
all
3dl
speakers
next
slide.
Please,
at
worst,
we
might
have
a
switch
that
wants
some
fan
out,
one
stage,
which
is
why
we
stuck
that
one
form
of
switch
piercing,
multicast
mac,
okay,
this
isn't
going
to
storm
next
slide.
Please
so
do
we
actually
have
a
problem
and
that's
the
end.
E
Comments
is
because
I
think
it's
an
edge
case.
I
didn't
perceive
anything
that
really
needed
to
be
fixed
after
our
discussion
with
randy
and
I
couldn't
think
of
of
a
shortest
path,
bridging
or
normal,
spanning
tree
topology,
where
this
would
actually
apply.
I
I
just
thought
outside
the
box,
so
I'll
leave
this
to
bill
and
the
rest.
I
didn't
think
there
was
a
problem
because
I
don't
think
the
scenario
I
described
is
something
you
would
normally
see
now
that
I've
shared
my
evil
nightmare
I'll.
Let
the
rest
of
you.
G
Just
there
are
other
packets,
like
broadcasts
that
also
need
to
be
not
create
a
storm
and
multicast
is
just
another
instance
of
this.
So
you
need
to
have
some
mechanism
spanning
tree
or
s
or
as
paul
said,
there
are
many
other
mechanisms
in
your
layer
two
to
prevent
storms.
It's
not
just
multicast
that
has
this
problem
so
to
have
a
layer.
Two
like
this
is
just
to
have
an
invalid
layer
too.
It's
not
just
this
particular
multicast
that
will
have
a
problem.
G
It's
broadcast
arps,
it's
unicast,
unlearned
flooding,
so
I
think
the
pro,
even
though,
even
though
we
don't
need
this,
you
know
we
don't
want
to
address
this
topology
because
it's
not
a
clo.
We
also
don't
want
to
address
this
topology,
because
it's
just
not
a
valid
layer
too,
to
run
over.
B
I
agree
first
of
all,
there's
a
hidden
issue
here
or
desire
here,
which
is
over
an
idr
they're
discussing
bgp
discovery.
Finding
your
friends
and
their
immediate
target
is
the
data
center.
B
B
A
Yeah,
I
was
just
wondering
about
the
fact
you
know
so
if
you
look
into
the
layer,
2
topology
what
we
have
here,
you
know
it's
a
little
bit
illegal
because
it
represents
a
layer,
2
environment.
Without
you
know
a
spanning
tree,
or
you
know
some
sort
of
a
protocol
to
do.
You
know
to
manage
the
multicast
and
broadcast
floating
within
the
environment,
which
is
a
little
bit.
I
think
illegal.
It
will
give
problems
always
so
I
think
we
should.
You
know
to
me.
A
B
I
think
everybody's
discussed
the
fact
that
that
topology
is
not
something
you
want
to
write
home
to
mother
about,
but
that
last
statement
you
made
is
that
l3dl
should
support
multiplayer
switching
topology
properly
configured,
of
course,
is
that
really
a
target
for
lsvr,
and
I'm
asking
you
as
chair.
A
From
lsvr,
I
don't
think
that
is
the
case,
but
then
again
I
think
l3dl,
you
know
is
a
technology
which
could
be
used
potentially
beyond.
You
know
the
lsvr
use
cases,
so
I
don't.
You
know,
I
think
if
we
constrain
it
too
far
in
the
applicability
at
this
point
in
time,
then,
in
a
later
phase
we
will
have
to
do
you
know
more
r
d
research
to
expand.
You
know
the
use
case
from
it.
G
H
Yeah
I
was
going
to
respond
to
your
comment
about
maybe
not
using
the
address
that
passes
through
switches
and
using
one
that
terminates
at
switches
like
lldp
might
and
there's
probably
a
couple
of
ways
that
you
could
do
that.
H
First
of
all,
in
802.1
we
do
have
regis
attribute
registration
protocols
that
allow
you
to
propagate
information
like
who
my
neighbor
might
be,
or
something
about
somebody
across
the
network
like
that,
you
showed
without
flooding
and
without
you
know,
without
causing
the
storm
so
there's
there
are
separate
registration
protocols
that
could
be
leveraged,
but
perhaps
even
easier
would
be
to
to
find
your
own
like
higher
layer
entity
that
propagates
information
using
you
know
hop
by
hop
kind
of
methodology.
So
in
other
words,
we
do
this
in
lldp
as
well.
H
You
you,
you
may
receive
some
information
on
one
link
and
then
you
generate
an
entirely
another
packet,
rather
than
just
using
the
forwarding
plane.
You
use
the
software
plane,
you
generate
another
one
that
goes
out
the
other
links
that
sort
of
propagates
that
information,
so
you
could
conceivably
build
some
higher
layer
entity.
H
When
I
say
higher
layer,
I
mean
above
layer
two,
so
you
know
some
software
forwarder
so
to
speak,
that
that
just
regenerated
the
l3dl
information
and
sent
it
in
a
whole
new
packet
rather
than
just
flooding
it
so
anyway,
just
a
couple
of
thoughts.
There's
there's
possible
ways.
I
think
you
could
do
what
you
were
proposing.
B
I
Hi
randy
jeff
has
juniper.
I
B
I
No,
I
I
agree
with
you
it's
one
of
the
things
I
do
like
about
l3dl
is
that
it's
good
plumbing
to
extend
for
a
lot
of
things,
including
this
use
case.
B
A
H
Okay,
I
I'm
assuming
you
can
hear
me:
okay,
yep
yeah,
so
it's
been
a
while
since
we've,
given
you
an
update
on
what
we
call
project
802.1
abdh,
which
we
we
call
in
802.1x
lldp,
but
we've
talked
in
the
past
about
it
being
called
sort
of
lldp
version,
and
the
main
the
main
objective
here
was
to
allow
us
to
send
more
information
in
lldp
than
can
typically
fit
in
a
single
frame,
but
do
that
in
a
way
that
is
backward
compatible,
so
it
I
I
have
some
well.
H
First
of
all,
we
have
an
editor
who's
been
developing.
The
draft
steve
haddock,
you
may
or
may
not
know
him
he's
been
around
for
quite
a
while
excellent
editor
and
the
pro
the
draft
has
made
quite
a
bit
of
progress
and
sort
of
getting
to
the
sign
of
final
stages.
If
you
will
so
next
slide,
because
first
of
all,
this
is
a
an
individual
contribution.
This
is
not
an
official
kind
of
quote
liaison
or
anything
from
ieee
802..
H
This
is
just
me,
however,
if
if
lsvr
was
interested,
we
could
do
something
a
little
more
formal
in
terms
of
document
sharing
and
those
kind
of
things
so
making
it
possible.
For
you
know,
your
lsvr
group
to
you
know,
comment
vote
whatever
on
the
documents,
we're
typically
very
liberal,
about
accepting
comments
from
from
anybody
anyway
in
the
earliest
phases.
Once
you
get
to
the
final
final
phase,
then
of
course
it's
a
things
are
processed
a
little
bit
more
tight,
but
in
general
we're
open
for
comment
so
yeah
next
one.
H
So
briefly,
this
is
the
I
didn't
want
to
get
into.
I
don't
have
you
know
in
five
or
ten
minutes
still
not
enough
to
do
a
lot
of
technical
discussion
here,
so
there
have
been
presentations
in
the
past.
H
I
point
out
that
the
the
motivation
for
originally
doing
this
work
came
from
looking
at
the
requirements
that
ls
l3dl
was
satisfying
and
trying
to
see.
If
it
was
time
to
update
lldp
it's
a
widely
used
protocol.
H
So
the
l3dl
motivation
helped
kick
us
off
into
this
new
project
in
general.
So
there's
some
background
presentations
here
and
again.
Last
time
we
did
an
update
was
itf
106.,
but
nothing
really
has
technically
changed.
Since
then,
I've
just
got
some
references
to
pointers
for
the
project
page
and
then
the
project,
scope
and
those
kinds
of
things
you're
welcome
to
read
and
then
more
recently,
the
the
sort
of
technical
contributions
as
steve
haddock
came
in
and
became
editor.
He
had
a
couple
of
questions
and
he
noted
a
few
things.
H
We
had
resolved
those
those
are
discussed
in
here
and
there's
a
paper
written
by
myself
and
paul
bottom,
who
kind
of
goes
into
really
more
of
the
details.
That's
the
fundamentals
for
the
for
the
the
draft
so
feel
free
to
take
a
look
at
those.
I
think
those
are
all
publicly
available
documents
next
slide.
H
So,
since
the
last
the
status
and
since
the
last
ietf
we've
completed
what
we
call
the
working
group
ballot,
first
working
your
ballot
and
that
was
very
recently
completed,
it's
kind
of
roughly
equivalent
to
an
ietf
last
call
and
we're
resolving
those
comments.
Now
this
week,
as
unfortunately,
ietf
and
ieee
overlapped
at
the
same
time
so
made
difficult
to
attend
both,
but
really
the
changes
are
we
put
all
the
content
into
the
format
of
a
standard
we?
H
That
draft
is
complete
with
all
the
normative
texts
and
state
machines
and
conformance
we're
using
this
xlvp
terminology
to
to
express
the
extension
of
lldp.
It's
not
necessarily
a
version
two,
because
what's
important
is
that
a
version?
One
implementation,
that's
probably
already
running
in
the
network,
is
capable
of
communicating
with
the
version
2,
but
obviously
only
the
first
pdu's
worth
of
information.
H
But
in
that
pdu
we
we
defined
a
new,
a
new
tlv
that
basically
describes
a
set
of
extensions
and
and
all
that
information
allows
you
to
send
up
the
like
64
or
more
pdus
worth
of
of
data,
and
one
technical
thing
that
came
up
lldp
supports
the
concept
of
multiple
agents
per
interface
and
really
there's
different
scope
in
l2
networks.
H
You
might
have
provider
bridges
or
provider
backbone
bridges
which
are
kind
of
layers
of
layer
two,
if
you
will-
and
we
want
to
be
able
to
communicate,
lldp
messages
between
different
endpoints
across
that
layer
to
topology,
and
so
there
was
some
there's
some
trickiness
in
how
that
works,
but
we
made
some
clarification
of
that
in
the
standard.
So
it's
always
been
there.
It's
not
new
support,
it's
just
when
we
added
xldp
capability.
It
got
a
little
bit
more
interesting
so
next
time,
so
the
next
steps
is
again
we're
resolving
comments
on
thursday.
H
I
guess
it
is
this
week
and
then
we'll
have
what
we
call
recirculation
ballot.
If
that
passes,
then
it
goes
to
what
we
call
standards.
Association
ballot,
which
is
kind
of
you,
know
the
equivalent
of
the
end
of
the
road
and
what
would
be
great
is
to
see
an
open
source
implementation.
You
know
the
current
lldp
open
source
implementation
would
be
would
be
great
to
see
that
augmented
to
support
this
and
then,
like
I
mentioned
early
on,
if
there's
a
desire,
we
can
do
some
liaison
and
document
sharing
with
lsvr.
H
If,
if
that's
of
interest,
we
can
take
that
offline
to
to
roll
those
wheels,
so
if
anybody's
interested
in
doing
that
so
anyway,
that's
that's.
Really
the
update.
If
there's
any
questions
happy
to
take
those.
J
Yeah,
no,
this
isn't
just
a
question.
It's
ac
linden,
cisco.
I
want
to
say
that
I
think
this
is
a
good
work.
I'm
hoping
I
recommended
that
cisco
support
this,
even
though
I'm
not
active
in
the
ieee,
so
I'm
hoping
that
happened.
H
Yeah
we
did
get
your
support
to
make
it
all
happen.
Thank
you.
B
I
did
come
over
to
ieee
and
try
to
help
and
join
this,
and
nothing
much
was
happening
if
things
are
spinning
up
again,
I
again
volunteer
to
help.
H
Okay,
let
me
know
how
we
make
that
happen.
I
can
I'll
bring
that
back
to
the
group
and
mention
somehow
we
can
as
an
individual
work
you
in
or
or
again
we
can
do
a
more
official
cooperation.
B
I'm
just
an
old
hippie
engineer,
so
you
can
wrap
whatever
you
want
around
it,
but
email
works
well.
H
A
J
Let
me
adjust
here
good
morning:
everyone
I'm
going
to
be
presenting
one
thing:
I'm
only
good.
We
we
well
both
documents
went
back
to
the
working
group.
We
focused
on
the
protocol
document
with
updates.
We
were
doing
during
starting
about
thanksgiving
care
victor
and
I
met
weekly
and
and
went
through
the
comments.
J
J
So
there
was
a
we
spent
quite
a
bit
of
time
on
this
on
the
first
document.
The
reason
we
didn't
look
at
the
apple
applicability
yet
is
because
why
don't
you
go
to
the
next
slide?.
J
Because
the
applicability
will
be
impacted
by
the
change,
the
the
protocol
in
a
lot
of
places
when
we
first
did
this,
we
thought
something
with
bgp.
We
said
yeah.
We
could
do
this
with
bgp
spf,
but
we
took
out
all
those
things
that
we
could
do
but
didn't
fully
specify,
because
we
wanted
to
have
a
nice
tight
core
protocol
that
could
that
can
be
augmented
with
future
giraffes
to
do
more
things
with
it
a
lot
of
the
comments.
J
These
are
just
things
we
did.
We
beefed
up
the
error
handling
we
added
the
security
section
with
the
parts
of
it
that
are
relevant
to
bgp
spf.
We
clarified
the
relationship
to
bd
bgpls
and
you
know.
In
addition,
we've
always
had
the
separate
address
family,
but
we
clarified
that
clarified,
which
nrli
and
which
attributes
we
were
specifically
using
in
the
base
protocol,
and
we
also
also
like
I
said
we
eliminated
things
in
bgp
that
we
said
you
know.
You
know
like,
for
instance,
route
optimized
route
reflection.
J
We
took
that
out
completely.
We
took
out
some
of
the
other
discussions
of
55
49
next
tops
and
things
like
that
and
I
rewrote
the
or
we
rewrote
the
spf
section
and
used
consistent
terminology
throughout.
I
think
I
think
it's
a
lot
easier.
I
mean
it
would
be.
J
There
were
actually
some
mistakes
in
it
too,
and
once
we
did
that
so
I
know
the
implementations
must,
I
hope
the
implementations
just
said.
Oh
I
we
know
it,
we
know
what
it
what
he
means
not
or
what
what
the
draft
means,
not
what
it
says
and
some
other
minor
nets
next
slide.
J
J
There
are
weren't
a
lot
of
changes
to
the
protocol
other
than
the
error
handling.
I
mean
that
that
really
really
crisp
that
up,
because
he's
familiar
with
all
the
different,
the
evolution
of
error
handling
in
bgp,
complete
with
the
the
draft.
I
forget
the
number
of
it
that
tells
what
you
do
when
you,
whether
you
ignore
an
attribute
and
continue
or
whether
you,
whether
you
take
it
as
an
implicit
withdrawal
or
bring
down
the
session.
J
We
tried
to
stay
away
from
that
last
one
of
course,
and
we
put
in
the
the
code
points
used
by
the
arcus
implementation
into
the
ant
ayana
section
and
have
requested
early
allocation
of
those
next
slide.
J
J
We
inherited
because
we
used
bgp
ls
nrli,
we
kind
we
pretty
much
inherited
that
you're
going
to
be
limited
to
one
nrli
per
update
because,
conceivably,
you
could
have
a
bunch
of
prefixes
that
all
have
the
same
metric
all
the
same
attributes,
but
if
once
you
start
like
that,
if,
if
anything
ever
changes,
you'd
have
to
keep
them
together
because
of
the
sequence
number
everything.
J
So
I
think
we're
going
to
go
ahead
and
say
really
you
really
really
only
can
have
one
nroi
per
update,
I'm
going
to
go
to
I'm
going
to
do
initial
synchronization
last
the
next
top
requirement,
okay
or
crisp
that
up.
So
it
follows
the
4760
rules,
and
I
think
this
issue
is
closed
completely
and
the
single
session
requirement.
This
was
from
melbaro,
but
we
he
said
he
said.
Are
you
limiting
it
to
a
single,
a
a
a
separate
session
for
bgpls
and
the
other
address
families?
J
J
What
alvaro
brought
up,
which
is
a
really
good
point,
is
the
igps
at
least
ospf
has
a
you
know
you
don't
advertise
a
neighbor
until
you
have
synchronized
with
that
neighbor
or
the
link
that
it's
you
know.
You
know
that
you're
connected
over
that
link
to
that
neighbor.
J
Now
we'll
have
we're
going
to
add
some
discussions
of
this.
I
guess
isis
has
a
a
similar
thing
that
you
don't
advertise,
but
it's
not
mandatory
in
isis,
I
forget
the
name
of
the
the
bit
in
the
in
isis
hellos,
but
anyway
we're
going
to
discuss
that
further
on
the
list.
The
one
thing
that
I
was
thinking
this
is
just
me
I
haven't
even
talked
to
my
co-authors
about
this
is
maybe
we
could
push
the
route
reflector
case
out
or
the
controller
case.
J
J
Like
I
said,
we're
gonna,
we
we
did.
One
working
group
last
call
we're
gonna
do
another
once
we
get
this
done,
we'll
start
working.
You
know
once
now
that
we
know
we've
sort
of
scoped
the
base
protocol
down
to
what
we've
fully
specified
we're
going
to
work
on
the
applicability
as
well,
and
there
are
now
multiple
implementations.
There's
a
there's,
arcus
and
one
underway
for
with
for
free
range
routing.
K
K
I
guess,
if
we
go
back
to
the
last
slide,
the
slide
before
this
one
insist
a
little
bit
on
the
separate
session
requirement
or
maybe
not
requirement
considerations,
because
not
only
are
we
carrying
different
information
inside
of
bgp
session,
but
we're
making
decisions
based
on
different
things,
meaning
running
spf
here
versus
the
decision
process
in
4271.,
so
the
potential
dynamics
of
how
the
updates
get
distributed
around
the
network
are
different
from
a
normal
by
php
session
to
a
bgp
spf
session.
K
So
if
the
working
group-
so
I
would
you
know
I'll-
obviously
really
want
to
see
a
requirement
for
software
session,
but
what
I
would
settle
for
is
if
we
could
include
a
draft.
You
know
some
considerations
around.
You
know
what
things
could
happen.
You
know
what
what
things
to
look
at
in
the
network-
I
don't
know
congruent
topologies
things
along
those
lines,
just
the
individual
contribution.
L
To
that
one
super,
first
of
all
cave
patel's
arkis,
I
was
gonna,
say
cisco,
following
ac
patella.
L
One
thing
I
wanted
to
first
acknowledge
was
that
it
was
a
fantastic
feedback
from
alvaro
on
the
draft
we
couldn't
have
asked
for
more,
and
so
thank
you
on
that,
particularly
on
a
single,
a
single
and
a
separate
session
requirement
to
me,
we
could
certainly
add
some
text
in
the
draft,
but
this
is
also
somewhat
an
implementation
detail
in
sense
that
today
certain
bgp
implementations
do
allow
you
to
run
multiple
processes,
wherein
you
can
listen
on
a
different
ports
over
and
above
179..
L
So
whether
you
want
to
do
that
or
not
is
a
very
implementation,
specific
details,
a
lot
of
people
do
it
for
a
lot
of
different
saffies
and
that's
why
I
think
we
can
add
something
in
in
draft
as
a
loose
text,
but
we
won't
want
to
make
it
as
a
requirement.
J
L
D
D
D
Perhaps
the
s
loop
or
maybe
check,
may
be
used
or
other
otherwise
of
the
ordinate
id
may
be
used
to
identify
the
rule.
So
I
think
this
scenario
perhaps
may
not
be
happened
in
the
bpspf.
J
Yeah,
I
I
remember
this
one
from
the
from
the
email
he
sent.
What
he's
talking
about
is
the
fact
that
we
we
have
handling
of
self
self-originated
nrli
and
by
self-originated
I
don't
mean
I
don't
mean
enter
a
lie
from
your
neighbor.
I
mean
enter
a
lie
where
the
node
descriptor
in
the
bgpls
nrli
identifies
the
receiving
router
and
we
do
it
and-
and
I
had
added
something
to
do
it
like
the
igps
to
speed
the
convergence.
In
the
case,
I
don't
know
if
we
don't
think
we.
J
Let
me
talk
about
that
with
with
the
co-authors
on
the
list
we
can
see.
If
we
we
don't
need
this,
but
that
was
added
yeah
that
was
added
to
speed
convergence
in
the.
In
the
event
there
was
stale
nrli,
but
granted
it
should
all
go
away
as
it
permeates
from
the
you
know,
from
the
from
the
source
from
this
from
the
originator.
J
D
J
We
have
a
rule
that
you
always
if,
if,
if
the
this
is
to
cover
the
case
where
the
router
restarts
and
loses
all
its
stake,
we
have
a
rule
that
if
you
receive
nrli
from
the
originator,
you
will
you
will
in
fact
prefer
it,
even
if
it
has
a
lower
a
less
recent
sequence
number
to
take
care
of
this
case.
So
probably
maybe
we
don't
need
to
discuss,
discuss
the
the
self-originated
case.
I
thought
about
that
as
well.
F
J
Oh
yeah
yeah
right
right,
yeah,
the
yang
model
work
is
in
progress
in
care
care
and
associates
are
working
with
it.
We're
gonna
work
on
it.
J
We're
sort
of.
I
guess
the
bgp
yang
model
has
gone
through
one
last
call,
so
it
would
augment
it
would
augment
the
bgp
la
yang
model
in
idr,
which
is
pretty
big
already
with
a
new
address
family.
L
We
were
thinking
of
doing
it
just
specifically
for
bgp
spf,
but
it
would
be
generic
enough
to
extend
it
to
ls
eventually,
and
maybe,
if
we
have
an
agreement,
if
the
chairs
have
an
agreement
with
the
chairs
at
idr
that,
then
it
could
be
a
common
draft,
but
what
we
don't
want
is
to
have
a
perception
that
we
are
taking
over
idr's
work.
So
if
there
is
an
agreement,
we'd
love
to
do
it
at
a
common
place,
I
think
it
makes
sense,
also,
but
then
again
I'll
defer
that
to
chairs.
A
A
A
That's
it
victory
of
any
last
words
to
say
at
the
end
of
the
session.
C
Specifically
we'll
talk,
I
guess
we'll
take
that
as
a
output
sort
of
how
we're
going
to
deal
with
that
along
with
idr,
and
then
we
will
be
processing
some
of
these
updates
and
getting
and
we'll.
Let
the
working
group
know
by
mailing
list
when
ready
for
the
next
working
group
last
call
now
ac.
C
One
question
I
had
for
you
is:
I
guess
we
have
to
discuss
whether
we
want
to
last
call
the
b2b
spec
before
the
update
to
the
applicability
draft
or
if
we
want
to
get
that
rev
and
then
kind
of
send
them
both
in
as
a
pair.
C
If
we,
if
you
don't,
think
it's
going
to
take
too
long
to
get
the
applicability
updated,
then
perhaps
doing
it
at
the
same
time,
maybe
simplify
things.
Did
you
have
a
religion
on
that.
J
I
don't
no,
I
don't
I
I
wouldn't
mind
doing
separate
cycles,
but.
A
K
Hey,
I
love
that
about
you,
so
I
would
prefer
to
we're
going
to
publish
them
both
publish
them
both
together.
That
doesn't
mean
you
have
to
last
call
them
at
the
same
time,
you'll
progress
through
my
q
and
I'll
just
push
them
both
together
at
the
end.
C
Okay,
all
right
so
I'll,
just
because
of
gunpowder.
We
were
talking
there
about
how
to
deal
with
the
working
group
paul,
but
it's
understood
and
noted
that,
when
we're
ready
for
publishing
that
we
put
them
that
they
should
go
together
for
simplicity
reasons,
no.
K
A
Okay,
I
think
then
I
guess
we'll
finish
that
you
know
we
give
you
back
like
eight
minutes
of
time
before
stepping
into
the
next
session.
Thank
you.