►
From YouTube: IETF104-LSR-20190328-1610
Description
LSR meeting session at IETF104
2019/03/28 1610
https://datatracker.ietf.org/meeting/104/proceedings/
A
D
There's
been
a
lot
of
routing
meetings
today,
a
lot
of
a
lot
of
a
lot
of
key
working
groups
in
routing
and
I
bet.
Everybody
is
a
little
bit
fatigued,
so
I'm,
not
I'm,
not
one
of
those
guys.
Gonna
have
you
stand
up
and
jump
around,
but
if
you
have
to
I'll
know
the
reason
anyway,
just
a
couple
one
working
group
thing
before
we
go
before
the
next
one:
please
try
and
get
your
your
requests
in
early
as
you
can
see,
we
probably
won't
use
the
whole
two
sessions.
D
D
I,
don't
know
how
many
people
are
going
to
have
proposals
for
flooding,
algorithms
that
fit
into
that
infrastructure,
and
we
followed
that
oh
yeah,
we
have
the
note
well
again
today,
I'm
sure,
you've
you've
seen
it
and
is
if,
if,
if
Gunter
says,
if
you
haven't
seen
this
already
this
week
and
in
the
last
week
in
group,
you've
probably
been
having
too
much
fun
and
not
attending
enough
IETF
meetings.
Is
he
in
here?
Yeah
I,
don't
see
him
anyway,
so
we
we'll
get
started
with
the
presentations
we
have
today
and
I.
Think
I'm.
E
F
E
D
Okay,
now
that
we
have
the
base
gang
model
is
with
Alvaro
and
hopefully
today,
IES
be
in
in
a
month
or
so
and
hopefully
published.
We
left
the
OSPF
be
three
extended
LSA
out
of
the
base
model,
because
at
that
time
we
didn't
have
any
implementations
already.
It
was
a
big
piece
of
work
in
order
to
firm,
ready
to
review,
and
we
just
published
this
last
year,
the
actual
functional
part
of
it
had
got
some
implementations.
D
If
you
remember
the
history
of
extended
Ellis
says
at
one
time
we
had
all
these
compatibility
modes
and
everything
and
everybody
thought
those
are
really
ugly
and
I
thought
it
was
limiting
the
implementation.
So
we
made
it
much
simpler.
Now,
here's
the
operational
state
in
all
the
in
the
link
state
databases.
D
We
have
the
state
data
for
for
the
LSD
B's
and
we
have
all
the
encoding.
So,
for
instance,
this
is
the
extended
router
LSA.
We
have
all
the
TL
B's,
you
see,
we
have,
you,
know
the
the
TLB
type
and
then
we
have
the
parts
that
are
the
base
type
for
for
all
the
different
types
of
links
and
the
rest
are
unknowns,
but
the
other
drafts
will
augment
this.
You
know
like,
if
you
add
a
link,
attribute
you'll
augment
this
tree
and
put
the
link
TLB
in
and
it
will.
D
It
will
no
longer
be
a
unknown
subtlety
of
the
link,
TLD
and
then
I
just
put
the
network
as
another
example,
we
had
the
attached,
router
TLB
one
thing
I
did
is
I,
didn't
do
a
per
neighbor
in
the
router,
let's
say
because
the
router
LSA
hasn't
this.
This
is
more
an
83
62
consideration.
The
router
I
mean
the
network.
Lsa
has
never
had
/
neighbor
stuff.
It
was
just
like
a
pseudo
note
that
had
the
connectivity
all
of
them,
so
you
don't.
D
D
The
one
thing
we're
not
doing
is
were
not
conditioning
LSA
types
by
the
scope
of
the
LSA.
For
instance,
we
don't
limit
the
link.
Ls
ADA
I
mean
link
the
link
extended,
link
LSA
to
the
link
database
and
I.
Think
I
could
do
that.
The
only
thing
is
if
we,
if
I,
do
that,
it's
it
wasn't
done
for
the
base.
L
essays
in
that
in
the
link
state
database,
but
I'm.
D
That
pretty
quickly,
if
I,
want
to
change,
put
this
constraint
on
it.
The
thing
about
configuration,
the
constraints
are
really
important
for
operational
data.
The
constraints,
are
you
know
it's
it's
you
know
what
you
see
is
what
you
get
really
putting
constraints
in
the
model,
isn't
isn't
isn't
as
important.
I
think
I
might
ask
some
of
the
other
yang
doctors
what
they
think
about
this
and
that's
it
so.
Eighty.
D
E
C
D
E
G
E
E
E
D
D
E
D
I
H
J
J
I
E
H
Yeah
hf
faso,
your
first
round
of
things
you
should
just
pick
what
the
core
items
are
and
put
them
in
there
that
this
is
the
same
thing
that
we
have
in
the
BGP
model.
A
BGP
base.
Rfc
is
very
tiny,
very
easy
to
manage,
and
the
yang
model
that
we
have
is
covering
a
dozen
technically
optional
features
and
they're
not
protected
by
a
feature
statements
or
anything
else
like
that.
It's
perfectly
reasonable
to
say
these
are
and
have
to
figure
out
what
that
from
a
working
group.
D
H
H
E
D
A
About
yeah,
it's.
D
K
Yeah
good
afternoon,
everyone
I'm
Dan
from
ZTE
today
I
will
talk
about
packet,
Network
slice
and
you'll
insect
Minnesotan,
yes,
problem
statement,
draft
she's,
a
hosta
peon
specifies
the
layered
act
area
of
enhanced
with
yen
paste
the
layered
activator
of
enhanced
with
he
earn.
This
document
specifies
the
solution
to
Crete,
which
one
network
in
a
packet
Network.
Yes,.
K
Introduction
explicit
with
to
a
network
edification,
the
maintenance
treaty
instance
admire
aii
to
meet
the
different
Zoe's
entire
network,
which
is
not
only
itg
domain,
but
also
in
cross
domain
peace
yeah.
It's
one
of
the
constitute
cry
tier
of
the
color
complete,
which
is
defined
in
the
truck
of
Sprint
segment.
Eroding
polish
analog
template
with
a
provides
a
more
flexible
control
which
will
network
based,
I'm
he
to
P
as
a
PE,
equal
best
efforts,
OS
and
a
p2p
I
thought
he
for
traffic
and
near
so
is
respectively.
K
Uniform
color,
translate,
include
secessionist
and
the
distribute
intro
intro
domain,
for
always
so
is
matching
to
underlays
results.
There
is
no
modification
to
the
folding
table
in
order
to
distinguish
forwarding
to
follow
team.
We
have
behaviors,
we
we
prefix,
the
prefix
situ
will
be
allocated
for
for
I.
Yes,
this
is
the
solution.
There
are
two
which
one
network
a1
and
a2.
K
This
is,
as
example,
based
on
the
color
template
with
AII
in
the
candy
calculate
two
three
three
paths
that
is
I
thought
he
won.
I
thought
he
to
and
I
shall
be
in
the
AIA
one.
The
resource
allocation
sauce
allocations
based
on
the
link
yeah
this
not
this-
is
only
the
solution
to
creat
which
will
network
in
a
packet
Network.
K
Yes,
this
is
matadormon
the
plot,
a
deployment
we
support.
I
think
it
has
really
to
provide
the
end
to
end
to
end
to
end
in
the
domain
in
the
domain.
Peace.
Yes,
in
some
disappointment
operator,
adopt
PPI
Oh
for
setup
pru.l
sushi,
always
always
we
are
directly
over
the.
If
he
are
you
out
see
if
it's
always
always
have
the
requirement
of
food
tea
which
defendants
at
the
color,
the
PPR
euro
will
have
also
supported
the
color
PPR.
Your
labels
will
be
allocated
for
copper,
color.
K
K
K
Yes
combined
with
essa
flex.
Algorithm
color
tablet
with
I
could
could
make
to
the
FAI
de
Filippo
stack
of
the
line
of
of
the
medicine
when
the
winter
algorithm.
While
you
know
the
world
has,
is
a
why
Leo
the
prefix
si
T
is
associated
with
the
path
calculated
using
the
F
way
as
algorithm
in
the
Associated
popularity
AI
specific,
yes,
next
step
comments.
Welcome.
Thank
you.
A.
D
Ceiling,
I
I
just
we
went
back
and
forth.
I
read
this
on
the
plane
and
I
had
a
lot
and
I
said
well.
This
has
a
lot
of
touch
points
with
with
a
lot
of
different
working
groups
and
I
thought.
Maybe
it
should
go
to
te.
The
te
chairs
I
talked
to
Lou
and
Pavan
and
well
at
least
to
Lou.
Personally
and
Pavan
didn't
disagree
that
it
probably
we
should
look
at
that.
The
touch
point
it
has
here
is
it's
leveraging
with.
D
In
the
intra-domain
case,
it's
leveraging
the
flex
algorithm
the
difference
between
this
one
and
the
enhanced
VPN
is.
It
does
not
use
a
VPN
or
a
verb.
It
just
does
data
lane
data
plane,
reservation
usage
by
using
different
algorithms
and
mapping
them
to
different
colors.
So
it's
a
lower
overhead
than
having
a
separate
birth
for
these.
These
VMs
right
that
we
talked
about
that.
A
D
K
K
D
I
D
K
No
in
our
chapter,
there
is
nothing
extensions
right,
I
think
it
is
more
general.
A
include
just
just
like
here
says
it's
more
general
and
requirements.
Okay
problem,
it's
the
problem
prop
sorry,
so
I
think
I
will
I,
will
tutor
presentation
under
cheese,
okay,
first
under
they,
cracked
the
feeder
pipe.
Okay,.
L
K
Sorry,
food,
just
the
Scylla
question,
you
say
yes,
I
know:
y'all,
we
have
discussed,
I,
know
your
job
and
we
are
trapped.
Your
solicit
you
use
Mattie,
but
ecology,
yes,
but
I
think
it
is
conflict
to
to
in
certain
in
the
domain
so
like
in
human.
We
show
Network,
but
so
we
defined
oh
yeah
identifier
of
foods,
always
connect
with
which
one
which
one
had
to
yes.
D
I
Maybe
last
one
is
as
I
think
you
mentioned
the
used
a
is
constraining
the
Assad
policy
right
and
you
and
I
supposed
to
be
usually
being
translated
on
the
head
end.
Do
you
also
consider
those
supported,
like
those
paths
computation
by
the
trends?
In
knows,
in
that
case
of
how
do
you
build
that
information
to
the
transit?
No
Henderson,
maybe
not
that
clear
in
the
current
draft.
Yes,.
D
N
O
Pollen
bit
on
t-shirt-
you
have-
you-
have
already
met
this
in
this
session,
so
you
don't
need
to
wait
till
the
next
session
to
present
it.
So
you
can
start
that's
right
to
initiate
discussion.
All
right.
I
mean
at
this
point
anything
to
do
with
topology
slicing
reservation
within
a
slice,
slice
identifier,
steering
into
slice.
We
would
like
to
discuss
that
at
T's
first
and
then,
based
on
the
solutions
we
can
take.
It
thanks.
K
Senator
that
he
just
some
clarification
and
this
the
AIT
funding.
This
document
is
not
a
conflict
with
the
multi
domain.
I
didn't
far,
so
it's
two
dimension
to
use
the
for
the
past
calculation
and
we
think
that's
the
yeah,
it's
easy
to
mapping
for
internal
main
use
case.
So
we
we
don't
think
they.
There
are
any
complexion
yet
so
you
can
find
that
they
can
work
together
and
makes
the
virtual
network
er
easy
to
deploy
yeah.
Thank
you,
okay.
Thank
you.
Thanks.
E
F
H
E
G
E
F
F
Then
I'm
sure
you
guys
will
help
me
out.
Okay,
this
is
about
looking
at
fast.
Convergence
for
IP
ran
grid
ring
topologies.
There
are
many
types
of
IP
ran
topology,
so
I
noted
that
grid
ring
topologies.
Now
this
is
basically
research.
This
is
not
a
deployed
Network.
This
is
just
looking
at
a
problem
that
was
reported
in
various
places
to
the
company,
and,
and
so
they
asked
me
to
look
at
it.
F
Theoretically
and
another
set
of
people
were
doing
a
fast
SPF
calculation
and
we
were
trying
to
put
initial
data
together
to
just
look
at
the
look
at
the
problems
that
were
reported
and
the
actual
theoretical
piece
I
am
NOT
presented.
The
theoretics
are
based
on
work
coming
from
Genghis
I
loose
enough
and
people
at
packet
design,
starting
in
1995.
So
this
is
old
equations,
but
we've
I've
dusted
them
off.
F
Now
that
Tony
made
a
good
point
on
the
draft
in
that
I'm,
looking
always
in
in
worst
case,
because
this
is
a
back-end
for
a
phone
network
and
you
are
looking
for
worst-case
convergence
time
so
that
if
the
phone
is
gonna
drop,
we'll
know.
Okay,
now
you
can
go
on
so
what
makes
IP
ran,
taking
apologies
and
the
ring
hub
unique.
This
is
a
sort
of
global
design.
Please
look
at
the
I
guess
it
didn't
turn
out
so
nice
on
the
screen.
Please
look
at
the
slides.
Essentially,
what
I
have
is
a
very
small
grid.
K
F
F
Policy
I
should
have
made
this
again.
The
ring
down
here
is
a
pretty
standard.
Sonet
ring.
It
connects
in
two
places.
So
that's
doubly
home,
but
summit
rings
have
pretty
normal
failure
characteristics.
You
can
break
in
one
place
and
maybe
you
can
use
rotating
rings
and
we
go
back,
but
it
has
a
limited
set
of
failure.
F
Techniques,
it'll
break
it'll
break
in
half,
it
won't
return,
but
because
it
has
two
endpoints
now
in
most
of
the
study,
since
the
ring
convergence
time
is
based
on
these
simple
breaking
patterns
that
didn't
become
as
interesting
as
some
of
the
rest
site.
I
go
into
that,
but
I've
got
limited
time.
The
interesting
piece
in
the
research
or
you
have
to
go
forward
now
is
to
look
at
the
graph.
Why
is
that
interesting?
F
Well,
the
ran
technology
that
I
was
asked
to
look
at,
has
power
back
up,
so
it's
it's
in
many
strange
locations,
maybe
northern
Michigan,
where
we
lose
power
or
other
places
in
the
country,
and
it
has
power
for
the
r
and
radios,
but
it
may
get
to
these
back
rings
that
are
connect.
Then
you
may
have
rolling
power
blackouts.
Why
is
this
different
than
data
centers
or
current
provider
networks?
F
You
don't
usually
use
multiple
links,
although
your
fast
reroute
calculations
and
I
talked
to
alia
the
basic,
fast
reroute,
and
maybe
the
the
multi-link
fast
reroute
only
look
at
one
or
two
when
you
get
to
three
or
four
in
this
grid,
you're
pretty
distracted
and
if
it
comes
and
goes
it's
even
worse.
So
let's
look
at
a
nine
by
twelve
grid
to
sort
of
do
that,
I
need
to
go
back.
One
slide,
please.
F
F
So
what's
the
problems
as
I
mentioned
the
summit,
the
ring
the
rings
were,
the
routers,
usually
SONET
rings,
connect
to
two
good
routers
and
the
mobile
ran
nodes
handoff,
the
routers,
the
grid
size
may
vary.
It
may
be
20
by
20
in
some
places,
10
K
by
10,
K
or
even
they'd
love
to
have
it
bigger
in
in
5g
networks.
You
know,
let's
put
the
grid
over
the
top
of
all
of
northern
Michigan
and
life
is
good
or
remote
Britain
or
some
other
remote
places.
F
F
My
terminology
might
be
centered
and
I'm
sure
Genghis
Khan
sends
you
pointers
to
his
Richmond.
The
rolling
power
outage
has
the
problem.
I
think
I've
gone
through
that:
okay,
what's
the
desired,
while
you'd
like
zero
configuration
and
Yaakov
Heights,
is
talk
about
that
for
data
centers,
but
this
is
a
different
type
of
environment,
same
desire,
you
want
to
fast
I
GP,
and
you
want
the
IGP
to
be
one
of
the
fast
algorithms
where
it
is
used
only
after
initial
IGP
convergence,
and
it
must
handle
a
fast.
F
So
this
really
provides
a
difference
either
through
paula,
geez
I
totally
kindly
ran
his
algorithm
to
see
what
improvement
on
this
type
of
anthropology
and
it
it's
only
70
percent.
Seventeen
percent-
that's
probably
minimal.
It's
not
built
for
this.
It's
not
bent
this
is
so
what
the
purpose
of
all
of
this
that
I
get
is
I
just
provided
the
equation.
I
wanted
to
talk
about
the
problem
and
I
wanted
to
know.
If
the
group
is
willing
to
look
at
these
type
of
topologies,
they
are
not
fast
fast.
P
F
Know
I
tried
to
say
I
peer
and
grid
and
hub
and
I
should
have
said,
not
5g
right.
These
are
existing
older
networks,
it
messed
me
up
was
you
talked
about
a
ring
and
then
you
showed
a
grid.
Oh
I'm,
sorry
I
talked
about
a
ring
if
to
go
back
to
the
beginning.
The
the
topology
is
both
ring
and
is
rings
attached
to
a
grid,
so
all
of
those
rings
will
take.
Then
your
well
hang
off
a
grip
and
I
put
it
in
my
slides,
a
shape
in.
P
P
F
Welcome
I
I'm
raising
the
topic
because
I
asked
the
question
on
the
list.
You
know.
Maybe
you
can
help
me
Confucius,
better
layer,
two
or
work
with
me
on
the
draft,
so
we
can
tell
people
about
virgins.
P
F
F
I
gotta
guess
blackout
case.
Ok,
so
let
me
let
me
just
go
it
in
several
cases.
You
want
to
get
any
type
of
algorithm
without
fast
reroute
that
can
improve
the
convergence
in
general.
Then
you
might
want
to
add
fast
reroute
and
only
when
you
get
all
of
that
done.
Would
you
want
to
do
the
multiple
link,
because
it's
tough
stuff
that
you
have
to
have
the
other
stuff,
then
ready
for
any
questions
on
the
problem?
P
D
Others
say
speaking
as
a
working
group:
member
I,
don't
think
it
would
I
would
I,
don't
think
it.
We
bet
it's
not
like.
Our
list
has
so
much
traffic
to
start
an
email
thread
on
it
and
let
people
brainstorm
I.
Don't
want
to
do
it
here
right
in
this
meeting,
but
we
we
could
because
because,
as
you're
saying
that
I
have
all
sorts
of
questions
on
it,
I
say
no
say
anything
more
I
started
to
copy.
D
Q
Q
Yeah
no
changes
since
Bangkok.
Okay,
the
point
of
this
I'm
not
gonna,
go
through
all
of
these
unless
people
really
want
to.
But
the
basic
idea
here
is
is
today
has
two
levels:
we
could
have
many
more.
Very
simply,
it's
a
lot
of
code
point
allocation,
but
there's
nothing
really
complicated
about
it.
We
can
just
basically
dial
it
up
and
go
all
the
way
to
eight
levels.
Without
a
problem
now
I
did
get
a
bug
or
set
of
bug.
Reports
less
Ginsburg
and
Paul
wells
have
found
lots
of
mistakes
that
I
made
in
the
details.
Q
E
I
think
I've
asked
before,
but
I
didn't
I
scan
the
the
the
draft.
I
don't
see
an
answer.
The
area
is
how
I
think
you
have
to
specify
how
at
least
talk
about
how
you
do
like
you
know
now
that
level
2
will
have
an
area
that
has
to
subsume
the
level
1
areas.
You
know,
how
does
that
interact?
How
do
you
divide
the
any
tea
up
you
know
and
so
forth
and
so
on,
like
a
whole
section
on?
That
would
be
good.
Thank.
Q
The
obvious
thing
you
could
do
is
to
have
that
be
a
per
implementation
configuration,
and
you
have
to
agree
on
that
configuration
between
two
two
systems
before
they
actually
would
join.
Now.
One
of
the
comments
that
Les
and
Paul
made
was
that
that's
really
really
hard,
given
that
almost
nobody
in
the
room
knows
what
an
in
SAP
is.
Q
R
S
Q
Bgp,
at
least
as
it
was
originally
intended,
not
how
it's
been
abused
was
meant
to
interconnect
domains,
okay,
separate
management.
What
we
are
trying
to
do
is
to
scale
up
is,
is
so
that
we
can
build
ever
larger
networks
under
the
same
management
and
clumpy
and
policy
should
not
be
necessary
within
a
single
administrative
domain.
So.
D
Other
questions,
hey
Tony,
I,
think
what
may
be
obvious
should
be
stated
in
the
backward
compatibility.
So
you
really
four
areas
beyond
how
do
you
level?
Two
is
the
place
where
it
gets
a
little
confusing,
because
a
router
that
doesn't
support
multi
levels
could
support
could
be
in
level
two
and
I
know
above
level.
Two
because
there's
new
PDUs
it
doesn't
you,
you
know
you
don't
have
4000
backward
compatibility
today,
supported
or
done.
Is
there
are
there
any
considerations
that
need
to
be
put
in
there?
So
if.
Q
We've
done
things
right:
there
should
not
be
any
interactions
with
existing
l1
and
l2
routers
that
cause
any
interoperability
issues.
Okay,
one
of
the
things
that
is
less
pointed
out
was
that
there
were
some
bugs
in
that
area
and
after
we
fixed
them
the
way
the
less
proposes
I
believe
we
will
be
in
a
situation
where
we
have
no
intersecting
PDUs.
Q
T
T
D
E
Q
A
T
U
Tajima
sorry
just
had
the
pain
and
pleasure
of
architecting
some
of
the
large-scale
networks
on
the
planet.
You
know
the
definition
of
large
scale
is
more
than
hundred
thousand
routers.
U
So
when
I
think
about
the
problem
that
I
ran
into
this
rings
bill,
because
the
same
approaches
that
get
used
to
construct
that
bill,
the
network
is
a
lot
of
hierarchy
and
when
we
think
from
hundreds
of
thousands
of
routers
perspective,
you've
got
level
two.
You
got
lots
of
level
one
to
delineate,
but
after
point
scale
becomes
in
convergence.
That
goes
with
it
become
the
most
foundational
roadblocks
to
extend
level
one.
U
Is
the
notion
here
that
we
can,
you
know,
go
a
greater
scale
with
another
level.
Q
Well
right
now
you
there
is
some
upper
bound
on
what
we
can
manage
to
fit
in
side
of
one
area
in
the
early
90s.
The
conventional
rule
of
thumb
was
about
1,000,
routers
I,
don't
want
to
mumble
something
because
as
soon
as
somebody
there's
a
number,
they
will
go
off
and
sue
me
if
I
I
got
if
their
implementation
can't
handle
a
number
but
you're
a
number,
but
even
using
that
1,000
routers.
Q
Okay,
if
we
have
a
eight
level
hierarchy,
then
that's
ten
to
the
third
to
the
eighth,
which
is
ten
to
the
24th,
which
is
a
whole
lot
of
routers.
Okay,
so
yeah
I
mean
the
idea
is
to
be
able
to
scale
right.
That's
the
whole
point
of
this
is
to
we.
We
are
up
against
the
point
where
some
people,
if
they
were
using,
is
is,
would
be
bumping
up
on
today's
existing
limitations
and
I
want
to
relieve
that
before
they
get
to
those
limits
in
practicality
that
that.
U
V
E
E
Q
A
Q
E
Q
E
Q
Q
W
Q
Right
so
the
next
talk
is
about
area
abstraction.
This
is
a
talk
that
I
tried
to
give
in
Bangkok
and
I.
Think
I
kind
of
buffed
it
and
I
would
like
to
try
again.
So
this
is
a
new
slide
deck
all
the
same
concepts,
all
right
so
suppose
we
have
this
large
scalable
is
is
network,
and
this
is
only
60
routers.
Q
Q
Q
If
we
have
an
is,
is
level
one
area
today
and
we
want
to
have
transit
of
that
level.
One
area
we
have
a
small
problem
take
a
look
at
the
two
yellow
paths
highlighted
here.
If
we
want
level
two
traffic
to
transit
that
level
one
area,
then
all
the
systems
that
we
touch
with
those
paths
have
to
be
l1
l2
systems.
Q
Q
Okay?
Well,
there
are
some
simple
things
we
have
to
do.
A
level
two
router
has
to
send
out
a
system
ID
in
its
IHS,
and
we're
gonna
have
to
send
out
a
single
system
ID
from
all
of
our
level.
Two
interfaces
now
nevermind
the
fact
that
we're
sending
it
out
from
different
routers
well
group
that
this
is
kind
of
what
we
had
do
to
fake
this
out.
Q
We
have
to
generate
the
level
two
LSP
that
looks
like
we
have
all
of
these
adjacencies
and
for
scalability
purposes.
We
don't
want
to
inject
anything
into
level
two
unless
we
have
to
so
really.
It
should
just
be
this
one
LSD
and
oh
by
the
way.
Yes,
we
have
to
forward
traffic,
so
the
level
two
transit
traffic
somehow
has
transit
or
a
level
one
area
and
some
of
the
things
inside
that
level
one
area
may
only
be
l1
routers.
So
how
do
they
handle
the
traffic
that
they
may
not
have?
A
prefix
for.
Q
Okay,
well
to
attack
the
first
problem.
We're
proposing
is
that
we're
gonna,
add
a
proxy
system
ID
all
right
when
you're
busy
configuring
your
system
and
you're
configuring,
your
network,
and
you
are
assigning
system
IDs
to
everything.
You
now
just
add
one
system
ID
for
your
level,
one
area
this
gets
advertised
by
the
area,
leader
and
all
of
our
level.
One
edge
routers
are
going
to
see
this.
Q
Q
I
D
L
L
A
X
Q
Hold
that
for
about
three
slides,
okay,
okay,
all
right!
So
now
we
need
to
generate
an
LSP
that
represents
our
level
one
area
into
level
two.
So
the
way
this
happens
is
pretty
straightforward.
Our
area
leader
now
has
a
working
level,
one
link
state
database
and
a
working
level,
two
link
state
database
and
from
the
level
two
link
state
database.
Our
area
leader
can
see
all
of
the
external
adjacencies
that
the
l1
area
edge.
B
B
Q
Q
Q
Q
All
right
now,
the
forwarding
part
suppose
we've
got
transit
traffic
that
arrives
at
our
l1
l2
edge,
router
and
that
traffic
is
going
to
match
some
l2
prefix.
So
we
know,
we've
got
a
forward
in
one
of
the
things
that
we're
gonna
have
to
do
is
to
figure
out
a
path
that
transits
our
level
one
area,
and
it
would
make
sense
that
it
would
transit
directly
to
an
exit
egged
router.
Q
Now,
because
we're
transiting
level
one
area
level,
one
routers
that
don't
have
a
level
two
link
state
database,
they
don't
have
forwarding
entries.
So,
of
course,
we
have
to
tunnel
across
them
I'm,
specifically
not
trying
to
specify
what
the
tunneling
technology
should
be
because
I
know
lots
of
people
will
choose
different
things.
But
the
point
is
that
when
we're
doing
the
level
two
SPF,
we
should
insert
forwarding
table
entries
that
point
tunnels
to
those
exit,
routers
and
yes,
vendors
I-
do
expect
that
people
will
make
automatic
towels
for
doing
this.
Q
Q
Q
We
have
to
be
careful
about
intra
area
metrics,
and
we
can
only
use
them
for
path
differentiation
when
we
have
inter
area
metrics
to
the
destination
that
are
identical,
and
the
reason
for
this
is
you
don't
ever
want
to
accidentally.
Have
your
metrics
inside
of
your
area
force
you
to
choose
a
path
going
back
out
of
the
incoming
side
of
your
area.
You
end
up
with
in
creating
a
forwarding
loop
instantly.
Q
W
Questions
Tony,
actually
I,
didn't
think
your
Bangkok
talk
was
not
at
all,
but
actually
pretty
clear.
The
first
of
all
the
first
one
is:
are
these
topologies?
Where
do
you
see
them
because
any
service
providers,
usually
if
there
is
the
band
router
success,
routers
aggregation
that
there's
a
clear
hierarchy?
You
don't
come
and
then
go
to
the
other
layers,
and
the
data
center
is
also
pretty
hierarchical.
You
know,
Leafs
point
is
that
I
mean
if
it
is
very
common.
W
This
is
a
lot
of
stuff
to
do
right
if
it
is
very
common,
this
kind
of
topology
is
absolutely,
but
if
it
is
like
one-off
or
something
that's
my
first
question
mark,
you
see
them
powerful,
you
see
them
and
so
forth,
and
the
second
question
is
all
of
your
proposals
are
kind
addressing
the
same
problem
scale
and
I.
Definitely
like
the
first
one
I'm
scared
of
this
one.
It
definitely
achievable,
but
you
know
it's
more
scary.
W
Q
Both
are
useful:
okay,
yes,
they
are
both
attacking
scale,
but
in
at
the
end
of
the
day,
I
think
we
can
use
both
and
what
I'm
trying
to
do
is
to
give
us
tools.
My
history
of
attacking
scale,
you
use
every
single
tool
at
your
disposal,
because
every
single
tool
might
give
you
an
order
of
magnitude
improvement.
Yes,.
W
I
agree
with
that,
however,
maybe
first
order
of
magnitude
improvement,
you
get
from
your
first
proposal
may
be
sufficient
for
the
next
20
years
or
whatever
right,
I'm,
just
kind
of
sometimes
it's
better
to
staged
either
supposed
to
go
advanced
about.
My
first
question:
where
do
you
see
this
kind
of
topology
is
where
you
must
actually
go
into
your
logo
layer
to
come
out
at
the
other
end,
this.
Y
Robert
Ski
juniper
is
there
any
reason
why
l1
is
is
is
l1
in
this
case?
Can
it
be
anything.
Q
Q
Y
H
H
It
was
something
that
was
BGP,
for
example,
what
happens
if
you
partitioned
your
internal
proxy
node?
It
was
like
you
got
this
l1
that
it's
messy
it's
all
connected.
If
you
cut
it
in
half,
no,
you
would
effectively
have
to
split
brains
that
really
aren't
connected
to
each
other.
I
thought.
I.
Gps
didn't
really
like
that.
You.
Q
Are
absolutely
correct,
my
GPS
do
not
like
it
when
an
area
partitions,
ok,
they
get
indigestion.
There
have
been
some
choice:
a
hacky
proposals
that
people
have
used
in
the
past
to
protect
against
that.
Basically,
you
tell
from
one
side
of
the
partition
together
and
tunnel
three
level
two
and
I'm
it.
S
D
S
So
I
know
that
the
idea
of
Oryx
is
what
I'm
trying
to
say,
but
I
do
have
at
least
two
questions.
The
first
one
is:
did
you
actually
look
at
how
much
you're
saving
in
open
flooding
overhead
by
having
this
special
role
of
the
one,
a
system
that
is
responsible
for
for
managing
the
LSPs,
because
I
think
you
can
do
the
entire
system
without
such
a
special
note
and
I'm,
not
convinced
that
the
reduce
flooding
overhead
is
worth
the
extra
hassle?
At
this
point,
you.
D
Know
another
thing
you
could
do
is
because
it's
somewhat
proven-
and
it
is
simpler-
you
know
most
of
the
bugs
with
the
virtual
links
are
to
to
these
having
AAS
externals
come
in
in
in
the
different
areas
or
having
to
look
at
for
summaries
in
different
areas.
You
could,
depending
on
how
much
it
is
I'm
not
trying
to
change
your
graph.
D
S
Q
S
S
Q
S
Okay,
so
the
other
question
I
have
is
in
this
topology
hiding
thing,
so
I
would
need
to
look
at
the
details,
but
this
is
a
pretty
pretty
large
change.
Wouldn't
you
want
to
make
this
something
to
be
considered
for
the
new
like
level
3
and
up
and
just
leave
layer
level,
1
and
2
as
they
are,
and
just
for
making
things
easier
in
terms
of
interrupts.
S
Q
Y
Q
It
is
not
a
common
failure
mode
for
a
router
to
fall
into
two
pieces
and
have
the
two
pieces
both
start
behaving
like
routers
and
behaving
like
themselves,
right
weaving
like
two
clones.
Yes,
so
and
I
I
was,
you
know,
waiting
to
hear
you
say:
oh
yeah,
I'm
gonna
fix
that
with
the
area
leader
and
1/2
us
can
I
have
the
area
leader
and
that'll,
be
the
router
and
1/2
isn't
and
that
one
will
just
shoot
itself.
But
you
didn't
say
that
so
I'm
giving
you
another
chance,
I
mean.
Q
R
Q
Would
you
like
bad
or
would
you
like
worse,
you
know,
or
you
like
one
one-half
to
rename
itself
as
a
different.
You
know
LT
router.
Actually,
what
I'd
like
to
do
is
adjudicate
it
and
say
the
areas
can
never
partition
and
we
don't
have
to
worry
about
it.
I,
don't
think
anybody's
going
to.
Let
me
get
away
with
that.
Q
The
second
thing
I'm,
the
one
I
got
up
for
to
begin
with
and
I
assume
that
you
know
either
I
just
missed
this
or
like
I,
took
my
eyes
off
the
the
coconut
shell
at
the
wrong
time.
I
get
how
you
originate
your
LSP
from
outside
of
one
of
these
things,
your
fake
LSP
I,
get
how
you
have
forwarding
connectivity
across
it,
that's
straightforward:
how
do
you
flood
across
it.
Q
Okay,
so
with
the
hack
here
is
basically
flood
everything
through
the
area,
leader
who's,
going
to
be
an
l1
l2.
Note:
okay,
and
the
way
to
do
that,
if
you
don't
have
a
direct
connection,
then
go
ahead
and
tunnel
across
to
the
area
leader,
so
I'm
modeling,
my
internal
stuff
as
l1
l2,
but
I'm,
just
not
I'm,
going
to
quietly
not
expose
that.
Yes,.
A
Q
Q
Q
Let's
see,
they've
got
a
smooth
migration
mechanism,
so
this
actually
can
turn
itself
on
and
off
gently,
and
that
may
that
seems
like
that
might
be
very
valuable.
So
so
I
was
going
to
propose
that
we
combine
drafts
and
since
you're
play
mode
since
your
draft
proceeds,
mine,
I
think
we
should
combine
into
yours.
So
we
should
talk
privately.
T
N
T
U
Q
The
square
boxes
in
here
are
l1
only
routers
okay,
so
they
do
not
have
visibility
outside
of
their
area
or
communications,
and
obviously
there
could
be
a
lot
more
complicated
topology
there
than
what
I've
shown.
Q
U
Q
That
I
think
away
everything
has
any
any
reach
ability.
Okay,
reach
ability
is
still
there.
The
question
is:
what
is
in
the
link
state
database?
Okay,
the
square
boxes,
because
they
are
on
only
we're
going
to
see
a
area,
border,
routers
and
they're
going
to
pick
default,
pointing
outwards
and
those
l1
l2
boxes,
then,
because
they
have
l2
information,
presumably
you're
gonna
be
able
forward.
Q
T
You
they
get
fly
again.
So
if
you
start
to
think
about
exactly
this
manageability
aspect,
it
becomes
pretty
soon
what
I
see
said
and
I
kind
of
agree
with
right.
The
TTC
was
justified
because
they
really
wanted
to
carve
out
something
which
was
not
an
area
boundary
right
and
made
it
very
hard
to
do
anything
else.
But
if
you
leaked
the
prefixes
doesn't
work
with
the
default
right
like
with
the
attach
and
the
default
that
doesn't
work,
it
will
black
hole.
T
A
Z
Speaking
as
myself,
I'm
one
of
the
authors
of
the
TTC
work,
so
yes
you're
right,
you
know
we
have
similar
but
slightly
different
objectives.
I
think
that
were
especially
proud
of
that
migration
mechanism,
which
we
thought
was
absolutely
smooth,
so
I,
mean
obviously
I.
Think
the
right
thing
to
do
is
to
work
together
on
this.
There's
no
point
in
having
to
similar
solutions.
Z
N
Speed-Up
shinnok
Cisco,
so
it
has
been
mentioned
here
that
it's
somehow
this
resembles
the
virtual
links.
You
know
SPF
and
I.
Remember
that
the
virtual
links
in
OSPF
were
actually
only
the
one
use
case
to
fix
the
bad
design.
It
looks
to
me
they
are
trying
to
create
from
here
he
got
here.
He
hasn't
been
ghost
from
the
beginning,
I'm,
not
sure
this
is
what
we
wanted.
There.
Q
B
D
F
E
So,
okay,
so
from
a
train
this
again,
maybe
this
one
today
will
work
better
from
a
work
from
a
chair
standpoint.
We
decided
not
to
do
ttz.
This
was
before
we
merged
groups,
so
this
is
sort
of
an
interesting
thing,
because
OSPF
published
an
experimental,
RFC
and
eius
eius,
basically
looked
at
after
expire
I,
would
be
interested
in
hearing
what
changed.
E
Hey,
I'm,
not
saying
that
I
mean
there
definitely
could
be
things
right,
but
why
did
we
let
Y
as
a
working
group
prior?
Did
we
feel
that
ttz
wasn't
needed
and
why
do
we
need
it
now?
So
that's
just
something
to
consider
I'm,
not
actually
looking
for
an
answer
right
now,
but
I
think
it's
important
to
maybe
address
that.
Well,.
D
D
Q
W
Planet
actually,
my
we
didn't
change
is
the
same
and
basically
virtual
links.
Ospf
we
have
about
600
deployments,
I
have
not
run
into
it.
Yet
I
am
sure
is
out
there
someplace,
but
yours
providers
typically
do
not
like
complicated
stuff
I
like
the
multiple
layer
levels,
because
you
know
it's
the
same
thing,
but
we
know
it
generalizes
it.
It
doesn't
change.
Anything
is
the
same
rules.
It's
just.
You
know
multiple
layers,
this
one,
it's
complicated,
I'm,
pretty
sure
the
initial
implementations
will
be
buggy
eventually
we'll
get
it
right.
W
U
Q
AA
Hooke
so
I
have
a
question
here,
so
if
they
el
ones,
are
really
only
pointing
default
out
to
the
ABR's,
it's
works
duplicate
quite
well
when
the
l1
is
stopped
not
when
the
l1
is
transit.
So
don't
you
afraid
about
suboptimal
exits
being
taken
for
the
path
for
the
originated
traffic
within
the
L
ones.
Q
Q
U
U
Q
Okay,
I
have
one
more
quick
one
I
don't
see
any
reason
you
couldn't
apply
this
recursively
right
as
in
having
this
apply
layer,
three
over
layer
level,
3
over
level.
Two!
Yes,
of
course,
no,
as
in
take
that
whole
thing,
yeah
draw
a
big
circle
around
it
and
then
build
the
topology
out
of
those
yeah.
Yes,
the
point
is
for
these
two
to
work
together.
Thank
you.
E
Okay,
I
think
that
was
it.
Let
me
reiterate
what
AC
said
last
time,
which
is
we
really
don't
we
don't
wanna
waste
any
time.
So,
let's
we're
gonna
try
to
get
our
agenda
set
up
early
before
the
next
IETF,
so
send
your
requests
early,
but
as
in
Tony
piastre,
it
can
be
interesting.
Things
to
talk
about.
It
doesn't
just
have
to
be
silly.
Tl,
B's.