►
From YouTube: IETF101-LSR-20180321-0930
Description
LSR meeting session at IETF101
2018/03/21 0930
https://datatracker.ietf.org/meeting/101/proceedings/
B
D
B
H
D
A
D
B
I
I
I
I
J
C
D
C
C
D
D
K
D
Get
in
you
Jen's
in
there
right
now
and
I'm
gonna
get
in
it.
You
can
get
to
it
through
the
agenda.
There's
a
little
button
under
the
right
I,
don't
know
if
we
have
a
jabber
scribe:
yeah
Mikkel,
hey
Jabbar,
Scott
and
Kelly
Grimson.
Thank
you
and
our
me
deco
support.
Well,
that's
us
now
this
used
to
be
a
slide.
They
made
us
read:
I,
don't
think
we
have
to
read
it
anymore
because
they
made
the
text
a
lot
longer,
but
this
is
our
note.
D
D
D
So
we
have
one
RFC
miss
rough.
It's
just
waiting
for
segment,
routing
extensions,
that's
LT,
bundles
the
reverse
metric!
It's
just
waiting
for
right
up!
I
was
contemplating
doing
having
a
separate
Shepherd
for
this.
So
if
you're
interested
in
shepherding
a
document
contact
me
the
MSD
we're
basically
just
waiting
for
the
OSPF
is
almost
done
with
this
as
well
would
be
nice
to
just
do
them
together.
So
there's
no
difference
make
sure,
there's
no
difference.
D
So
the
routing
extension
segment,
routing
extensions,
we're
on
version
15.
Now
it's
it's
been
in
last
call
since
April,
2017,
I'm
gonna,
pull
this
out
of
last
call
because
it's
not
really
in
last
call
right,
it's
just
being
developed
and
we
can
I
mean
if
it's
done
now,
then
we
can
last
call
it
again
right.
So
unless.
L
D
L
D
L
D
L
C
D
D
D
We
have
a
new
just
adopted
the
dust
source
routing,
so
that's
new
work
not
being
presented.
Today,
though
we,
the
bold
there,
is
being
presented
today,
the
yang
that
that's
a
segment
routing,
hang
I
sort
of
covered.
The
yang
together.
I
want
to
talk
about
the
other
yang
document,
the
MSD,
or
you
mentioned
these-
are
all
being
presented
their
individual
documents
and
then
well
have
those
dates
still
sitting
there
anyway.
D
So
right
so
I
yeah,
the
bold
ones
are
being
presented.
So
I
noticed
that
etz
got
updated
here,
I.
So
what
changed
since
2015,
when
we
decided
that
this
wasn't
work,
we
were
gonna,
carry
forward
with
I
mean
I,
don't
know
why
it's
being
updated!
It's
unless
the
the
comments
that
were
put
on
so
during
the
meeting
in
2015
were
addressed.
I,
don't
know
why
I
would
keep
getting
updated
and
also
the
OSPF.
D
It
did
go
into
experimental
with
so
I
mean
if
there's
something
to
share
there,
like
with
some
new
operational
experience,
since
it
was
accepted
as
experimental.
If
it's
been
implemented.
If
it's
been
deployed,
if
people
are
using
it,
that
would
be
good
to
know
before
we
start
working
on
in
eius,
eius
extension.
D
Geo-Coordinates,
so
this
I
wanted
to
mention
this
because
I
sort
of
got
pushed
to
the
side.
Peter
Lawford
brought
up
that
the
geo
coordinates
were
not
expansive
enough.
So
we
need
you
know.
We
need
to
be
able
to
identify
things
in
space
right
so
that
got
put
to
the
side.
I've
actually
got
some
a
model
design
that
makes
Peter
happy
so
I
could
introduce
that,
and
maybe
we
could
restart
this
work.
Yeah.
C
Yeah
I'm
an
offer
on
all
these
documents
and
that
kind
of
the
open-ended
comment
we
got
from
Peter
kind
of
put
us
on.
We
said:
well
we're
not
gonna
ask
for
adoption
of
these
and
other
things
were.
You
know
you
know
it
became,
went
from
low
hanging
fruit
to
very
high
fruit
to
reach,
so
I
think
this
is
well
I'll
get
with
my
off
co-authors
and
we'll
look
at
Peters
what
what
Chris
has
done
to
encode
the
geo
coordinates
that
satisfies
all
the
D
team
requirements.
F
M
C
I'm
gonna
go
real
quick.
Today
we
have
all
these
documents
sitting
on
the
RFC
editor
queue.
The
only
one
I'm
going
to
say
something
about.
Is
the
segment
routing
one
we're
gonna
change
this
on
the
queue
only
that
ref
and
I.
Don't
think
that
should
be
a
problem.
I
hope
you
know,
because
it's
now
not
the
conflict
draft
as
if
you've
been
following
spring,
that
draft
is
has
expired.
C
The
conflicts
are
now
handled
in
MPLS,
signaling
extensions.
You
can
see
that
tunnel.
End
caps
is
waiting
on
this
IDR
Tunnel
end
caps,
I
hope
that
will
go
forward
in
the
next
two
IETF,
the
other
to
the
other.
To
the
two
at
the
top
link
overload
and
extended
Ellis
ace,
extended,
ell
essays
is
actually
in
off
48
hours,
so
that
should
be
pretty
quick
link,
overload,
I'm
hoping
that'll
get
to
the
there's,
nothing
stopping
it's.
Just
the
RFC
editor
workload.
C
These
these
three
documents
I
want
to
work
in
group
last
call
this
may
be
a
bit
ambitious
because
there's
three
of
them,
but
to
do
them
all
before
the
next
IETF
there's
nothing
other
than
ELC
were
I'm
thinking.
We
should
wait
on
MPLS
I'm
gonna
go
to
MPLS
tomorrow,
see
where
that
one's
standing,
but
otherwise
there's
there's
technically
there's
nothing
that
should
keep
us
from
working
group.
Last
calling
these
three
and
all
the
other
drafts.
All
the
other
working
group
drafts,
not
individual
graphs
are
presented
today.
So
I'm
done.
D
H
H
So
this
is
the
changes
since
last
time,
so
with
the
first
one
is
because
of
the
PFT
model
is
finishing,
so
we
finally
knock
down
how
we
interoperate
with
the
FD
model.
That's
the
change.
We
did
for
that
part
first
bullet,
and
then
we
had
a
young
doctor
review
of
the
whole
data
model
and
they
thought
the
detail
changes.
We
did
I
think
pretty
much.
All
the
major
issues
addressed
still
a
cup
of
cosmetic
issues.
We
will
to
finish
it
in
the
another
update
and
then
we
will
started
last
call.
H
So
the
real
major
changes
is
this
part.
So
we
used
to
have
a
USP
of
the
instance
list
and
instead
is
install
SPF
instance.
We
have
a
list
of
address
families.
Now
we
remove
that
instance
list
also
the
address
family
list
so
right
now
we
only
under
nothing
config
mode,
also
and
the
control
protocols.
The
Lisa
is
over
there.
So
only
one
OSPF
instance
and
each
instance
has
only
one
address
family
defined.
H
C
H
C
D
H
D
H
D
H
N
N
So
it's
very
similar
to
the
ISS
I
says
draft
in
that
sense,
we'll
talk
about
that,
so
the
router
r1
new
one
uses
the
reverse
metric
signaling
towards
its
neighbor
r2,
and
this
is
done
on
a
link-local
mechanism
in
response
to
this
r2
modifies
it
metric
in
the
router
LSA
for
its
link
towards
r1,
and
this
is
a
mechanism
which
needs
which
needs
only
support
on
by
r1
and
r2.
Other
routers
in
the
network
are
really
not
involved
in
this
and
they
just
use
the
updated
metric
metrics
in
the
router
analysis
for
their
SPF.
N
A
couple
of
use
cases
which
wanted
to
capture
one
of
them.
The
first
one
is
many
OSPF
deployments
use
this
notion
of
an
auto
cost
used
on
based
on
some
reference
bandwidth,
which
means
that
the
metric
of
the
link
is
determined
based
on
the
bandwidth
of
the
link,
and
this
is
in
many
some
deployments.
This
is
symmetrical
in
both
direction.
Now,
as
some
of
these
networks
are
evolving
to
you
know
more
complex
there
to
aggregation
with
Ethernet
the
physical
topology
and
the
layer,
three
topology
are
different.
N
N
What
does
reverse
metric
mechanism
would
allow
in
this
case,
is
that
the
router,
which
detects
one
of
its
link
capacity
going
down-
let's
say
one
of
the
links
going
down-
can
signal
to
its
neighbor
to
reduce,
or
you
know,
increase
the
metric,
rather
in
the
other
reverse
direction
as
well,
so
it
still
symmetrical,
and
there
is
not
going
to
be
congestion
in
one
direction.
On
this
link,
the
second
use
case
is
when
there
is.
N
There
are,
let's
say,
a
couple
of
aggregating
routers,
which
are
you
know
connected
to
a
bunch
of
you
know,
access
routers
and
due
to
some
capacity
degradation
or
congestion
on
the
towards
the
core
network,
one
of
the
routers.
Let's
say
in
this
example,
the
aggregation
router
one
wants
to
shift
or
offload
some
of
the
traffic
coming
towards
it
to
another
router,
and
this
can
be
signaled
by
the
AGG
rr1.
Why
are
the
reverse
symmetric
mechanism?
N
D
N
D
Use
case
1
can
be
much
simpler.
Right,
I
mean
so
take
the
case
where
you
have
a
fiber
optic,
point-to-point
connection
right
and
you're
on
the
receiving
end,
and
you
see
that
the
signal
sucks
and
you
know
it's
got
a
lot
of
errors.
That's
the
you,
don't
need
all
the
aggravation
sweats
right,
so
one
neighbor,
the
neighbor
knows
that
it's
get
crap,
but
so
it
tells
its
its
neighbor
hey,
raise
the
metric
yeah.
N
So
how
is
this
gun?
This
is
new.
The
proposal
is
for
a
new
reverse
metric
TLV,
which
is
used
for
link-local
signaling,
and
it
really
has
the
reverse
metric
value,
which
is
advertised
there.
There
are
two
ways
in
which
this
can
be
done.
There
is
the
orbit
or
the
offset
bit,
which
is
what
it
indicates
is
that
the
value
advertised
there
is
a
metric
that
needs
to
be
added
by
the
route
by
the
neighbor
to
its
existing.
N
N
N
It's
coming
in
the
next.
Oh
sorry,
so
so
this
draft
applies
to
point
to
point
and
point
to
multi-point
links
only
we
already
have
in
OSPF
the
two
part
metric
and
that
mechanism
can
be
used
for
broadcast
and
NB
ma
networks.
So
this
thing
the
one
difference
in
this
draft,
as
opposed
to
the
two
part
metric
is
this
is
link-local
and
there
is
no
backward
compatibility
challenges
and
you
know
the
area
wide
capability,
things
which
come
with
the
two
part
metal.
N
Now
the
exact
mechanism
for
triggering
this
reverse
metric.
What
is
the
value
to
be
used?
You
know
whether
it's
something
done
by
a
management
station
or
it
is
done
by
the
router.
Those
are
implementation
aspects
outside
the
scope
of
the
draft.
The
draft
really
focuses
on
the
signaling
part
of
it,
since
we
also
have
the
ISS
reverse
metric
wanted
to
quickly
touch
upon.
What's
the
differences
so.
O
N
So
the
difference
really
is
that
the
ISS
draft
covers
a
variety
of
you
know
use
cases,
but
we
already
have
different
mechanisms
to
do
that
in
OSPF,
so
the
two-part
metric
handles
the
signaling
in
land
cases.
We
all
already
have
the
link
overload,
which
is
going
to
be
our
FC,
which
had
handles
the
link
overload
signaling,
this
particular
draft
know,
let's
say
handles
a
whole
or
missing
case
where
we
wanted
to
do
it
for
these
other
generic
use
cases
as
well.
N
The
other
difference
is
that
ISS
draft
only
allows
really
offset
mainly
in
addition
to
the
overload
scenarios,
but
here
because
of
this
deployment
use
cases
where
we
have
this
auto
cost
kind
of
a
thing
reference
bandwidth
deployments.
The
OSP
aircraft
also
allows
for
signaling
now
for
absolute
value.
P
L
Q
L
Okay,
I'm
les
Ginsberg.
This
is
a
presentation
of
an
update
to
the
SSS,
our
v6
extensions.
This
version,
2,
was
published
just
a
couple
of
days
ago
apologize
for
that
we
didn't
quite
make
the
deadline
before
the
IETF.
This
presentation
is
based
on
that.
There
are
two
key
documents
that
you
need
to
read
in
order
to.
You
know
fully
comprehend.
L
What's
going
on
with
the
IGP
drafts,
I'm
not
going
to
talk
about
a
lot
of
the
basic
concepts,
but
the
network
programming
is
sort
of
the
architecture
document
for
sr
v6
and
the
six-man
document
documents
the
segment
routing
header
for
ipv6.
So
those
are
key
background
documents
that
you
need
to
have
looked
at
I
did
want
to
mention.
There
is
an
equivalent
OSPF,
b3
draft
I
think
we
didn't
have
time
on
the
agenda
to
present
that,
but
it's
it's
basically
aligned
with
the
ice
ice
draft
in
terms
of
what
ad
amortizes
fast.
L
So
what
have
we
changed
since
v1?
A
lot
of
this
is
editorial:
we've
aligned
with
the
latest
version
of
the
network
programming
draft,
which
introduced
some
some
new
names
and
particularly
introduced
a
new
registry
for
the
functions
or
the
endpoint
types,
as
they're
now
called,
and
we
now
refer
to
that
registry.
Previously,
we
had
to
find
values
for
them
in
this
draft
itself,
but
now
we
have
one
common
registry
for
that.
We've
added
support
for
OAM
in
terms
of
what
the
IGP
can
advertise.
L
We
have
a
sub
TLV
for
the
the
function
descriptor,
because
originally
we
had
some
thoughts
that
we
might
want
to
associate
multiple
functions
with
a
single
Sid.
We've
now
decided
that's
not
needed,
and
so
we've
gotten
rid
of
the
sub
TLV
and
we're
now
encode
the
function
directly
in
the
city
LV,
which
I'll
show
you
and
we've
done
so
many
toriel
changes
in
terms
of
naming
that
we
thought
was
a
little
more
intuitive.
L
So
what
does
this
introduce?
It
introduces
a
new
SR
v6
capabilities,
which
is
a
sub
T
of
you
to
the
router
capabilities.
The
the
SR
capabilities
that
we
all
know
and
love
is
for
SR
MPLS.
So
we
need
a
new
one
for
SR
v6.
We
introduced
a
new
top-level
TLV
to
advertise
SR
v6,
note
SIDS,
we
have
endpoint
function,
types
associated
with
the
SR
v6
notes.
It's
and
we'll
show
you
that,
and
we
also
have
SIDS
associated
with
a
neighbor
sort
of
the
logical
equivalent
to
adjacency
SIDS
and
SRM
POS.
L
We
have
flags.
There
are
two
flags
defined
currently
one
of
them
to
indicate
whether
you
support
the
TN
cap
function
and
if
you
want
to
know
more
about
TN
cap
and
all
the
other
functions
that
are
mentioned
here.
It's
the
network,
programming
draft
that
that's
your
guide
there,
and
it
also
indicates
whether
the
Box
supports
the
o
flag
for
OEM
support,
and
then
there
are
optional
sub
sub
tlvs.
L
These
are
the
sub
sub
deal
these
that
are
currently
defined,
they're,
basically,
a
collection
of
things
which
indicate
should
we
say
the
the
maximum
imposition
of
various
types
of
of
SIDS
in
the
SR,
v6,
environment
and
again,
a
network
pro
mean
game
programming
guide.
Is
here
your
best
resource
to
get
all
the
details
on
this?
As
far
as
what
these
functions
actually
do,
there
is
a
general
descriptor
for
the
endpoint
function
and
it
looks
like
this.
We
have
8
bits
of
flags,
currently
there's
no
Flags
defined
and
the
actual
function
value.
E
L
So
we
have
the
node
C
TLV,
which
is
a
top-level
TLV
if
you're
not
SR,
v6
capable,
then
you
just
ignore
this.
It's
important
to
note
that
a
nodes
hid.
This
is
not
a
prefix.
This
is
not
something
that's
going
to
be
installed
in
the
forwarding
plane
and
we
one
of
the
reasons
that
we
chose
to
introduce
a
new
TLV.
L
One
could
have
imagined
that
we
could
have
done
sub
t
Ovie's
of
the
ipv6
reach
ability
TLDs,
but
in
particular
we
didn't
want
to
have
this
impact
on
the
nodes
that
are
not
SR,
v6
capable
so
by
introducing
a
new
TLV,
the
nodes
that
don't
support
us.
Every
six
simply
ignore
this.
They
they
don't
know
what
it
is.
L
L
It
does
share
the
same
sub
TLV
space
with
the
reach
ability,
TL
V's,
because
we
do
envision
some
use
cases
for
some
of
those
sub
tlvs
and
it
can
be
leaked
between
levels
similar
to
the
way
a
prefix
reach
ability
advertisement
is
leaked
and
just
it's
it's
possible
that
they
see
it,
which
in
SR
v6
is
nests.
Every
six
address
isn't
necessarily
covered
by
prefix
that
is
advertised
by
a
size.
It
could
be
advertised
by
another
protocol.
It
could
be
a
static
route.
L
So
this
is
what
the
TLV
looks
like.
We
have
Flags,
there's
one
flag
bit,
which
is
the
D
bit,
which
is
used
in
the
same
way
that
it's
used
in
the
prefix
reach
ability.
So
if
you
link
it
downward,
you
set
that
bit
the
function.
Flags
in
the
end
point
function.
Value
is
what
I
described
on
the
previous
slide
and
the
the
function
variants
that
can
be
advertised
in
this
TLV
are
the
engine
en
dot
OTP,
which
is
an
OEM
related
function.
L
And
then
we
have
the
the
the
actual
cid
itself.
We've
encoded
this
very
much
like
a
prefix.
In
most
cases,
it'll
be
a
slash
128,
but
we've
allowed
for
the
possibility,
if
it
isn't
that,
just
like
we
advertise
the
prefixes
today,
we
only
have
to
advertise
the
significant
bits
and
we've
allowed
for
sub
tlvs,
but
there
are
no
sub
deal
these
defined
presently.
L
Similarly,
we
have
an
adjacency
CID
sub
TLV.
This
is
advertised
in
the
is
neighbors
advertisement.
It
has
the
same
structure
as
the
note
cid.
So
again
you
have
the
flags
and
the
functions
and
the
actual
city
itself.
In
this
case,
the
function
variants
are
the
end,
X
and
n
dot,
DX
flavors,
which
are
sort
of
functionally
similar
to
adjacency
seeds,
as
you
know
them
in
SR
mpls,
there's
a
similar
advertisement
for
a
LAN.
D
M
C
C
Other
thing
is,
you
said
that
this
isn't
installed
in
the
routing
table,
but
it
is
in
the
same
sense,
like
the
MPLS
said
it
would
be
in
the
sr
v6
header,
as
as
for
individual
reach,
reach
ability
to
individual
effects
in
the
like
a
prefix
or
something
else
in
it.
That
is
in
the
14
table.
Correct.
Yes,
so
you
certainly
you.
L
L
L
L
So
I'm
going
back
to
IETF
s,
because
the
two
documents
were
updated
at
different
times
along
the
way,
but
the
two
documents
are
now
fully
aligned.
So
they've
been
adopted
as
working
group
documents,
we
have
introduced
some
additional
standard
applications
since
the
earlier
drafts,
we've
clarified
the
relationship
between
attribute
advertisement
and
whether
that
indicates
an
application
is
actually
enabled
on
an
interface.
L
L
Okay
with
so,
we
now
have
the
following
standard
applications.
We
have
rsvp-te,
we
have
a
segment
routing
srte,
we
have
LFA
and
we've
now
incorporated
the
Flex
I'll
go
they'll,
be
presentation
on
flex
I'll
go
later
today
for
application
enablement.
We've
now
clarified
in
the
case
of
our
CPT
II,
that
if
you
advertise
link
attributes-
and
you
specify
the
rsvp-te
bit,
then
that
does
indicate
that
our
CP
is
enabled
on
the
link
for
the
other
application
types.
That's
explicitly
stated
as
not
the
case
that
it
does
not
indicate
enablement
of
the
application
on
that
interface.
L
So
this
maintains
I
think
the
the
existing
paradigm
that
rsvp-te
is
explicitly
enabled
on
a
link
and
its
associated
with
the
link
attribute
advertisements,
and
this
addresses
an
issue
that
had
been
brought
up
in
a
draft
from
Chris
powers
and
Shradha
wherein
in
existing
implementations.
Not
everybody
agrees
on
exactly
which
link
attribute
indicates.
L
So
these
are
the
the
statements
about
each
of
the
applications
as
to
whether
it
indicates
that
they're
enabled
on
a
link
or
not
enabled
on
a
link
and
we've
specified
that
if
you
introduce
new
applications
in
the
future
that
you
are
required
to
state
whether
the
in
the
advertisement
of
link
attributes
to
be
used
by
that
application
indicate
whether
that
application
is
enabled-
or
it
does
not
indicate
that
it's
enabled
that's
that's
a
requirement
of
any
application.
That's
incorporated
into
this
functionality.
L
So
here's
the
list
of
supported
link
attributes
and
we
made
some
clarifications
on
particular
link
attributes
in
the
case
of
link
bandwidth.
This
is
clearly
an
application
independent
attribute.
So
when
it
is
advertised,
if
it's
advertised
associated
with
multiple
applications
in
different
sub
TVs,
then
the
value
value
must
be
the
same.
In.
L
The
case
of
maximum
reservable
we're
allowing
for
the
possibility
that
there
are
application
use
cases
I,
think
it's
in
in
the
most
common
use
case.
This
will
be
shared
among
all
the
applications
that
are
using
a
particular
link,
but
the
infrastructure
that
we
have
allows
for
the
possibility
that
you
could
have
different
values
for
different
applications.
L
L
L
D
C
K
H
Q
K
D
G
G
G
K
L
Okay,
Tony
I
think
you're,
probably
in
a
time
warp
now,
but
so
there's
an
existing
RFC
for
graceful,
restart
and
that's
about
15
years
old.
But
there's
an
update
that
we
are
proposing.
E
L
So
how
does
this
work
today
today
in
hellos?
If
you
have
a
router,
that's
undergoing
a
control,
plane,
restart
and
the
forwarding
plane
is
maintained.
We
want
to
be
able
to
support
a
hit
list
transition,
and
so
we
have
this
TLV
that
we
insert
in
the
hellos
and
the
restarting
router
sets
the
our
Arbit
to
say:
hey
I'm,
restarting
on
the
neighbor.
L
L
It
can
be
useful
where
you
have
redundant
control
planes,
but
for
whatever
reason
you
haven't
implemented
any
local
checkpointing.
So
you
need
to
reacquire
the
link
state
database
from
your
neighbors,
but
it
does
not
support
a
reload
of
a
control
plane
that
takes
an
extended
period
of
time.
Such
as
two
three
minutes,
you
can
hack
this
by
sending
a
halt
before
you.
You
start
your
plan
to
reload
and
extend
the
whole
time,
but
the
risk
associated
with
this
is
your
neighbor
has
no
clue
that
you're
actually
restarting
he
just
sees.
L
Oh,
you
want
me
to
have
a
longer
whole
time
and
if
there
are
topology
changes
which
occur
during
this
extended
period
when
you're
restarting,
the
neighbor
has
no
reason
to
think
that
you're
not
receiving
the
updates
and-
and
so
this
can
lead
to
traffic
loss.
So
what
have
we
introduced?
We've
introduced
two
new
bits,
so
we
use
the
same
TLV
in
the
hellos,
but
we've
introduced
the
PR
bit
and
a
PA
bit
PR
bit
signals
to
your
neighbor.
L
I
am
going
to
to
do
a
planned
restart,
but,
unlike
the
RR
bit,
it
says,
I'm
not
actually
I
haven't
finished
my
reload
or
restart
yet
so
don't
start
sending
in
the
link
state
database
I'm
just
telling
you
that
I'm
about
to
do
this
so
that
you're
aware
and
it
provides
it
still
provides.
The
existing
mechanism,
which
says
I,
can
extend
the
whole
time,
but
the
difference
is
now.
The
neighbor
is
aware
that
you're
actually
restarting
it,
can
take
this
into
account.
Should
there
be
additional
changes?
L
So
what
does
the
neighbor
do
whenever
he
sees
Br
it
keeps
the
adjacency
up,
assuming
the
adjacency
was
in
the
up
States
already
it
marks
as
an
internal
state
that
paid
my
neighbors
in
planned.
Restart,
did
state
updates
the
whole
time,
but
updates
at
only
once
so
if
he
gets
another
hello
with
the
the
PR
bit
said,
it
says:
hey
I'm
already
in
play
and
restart
state
I'm,
just
not
going
to
update
the
whole
time,
and
it
sends
that
acknowledgement.
L
D
L
I
think
the
the
danger
here
is
that
you
know
something
goes
wrong
and
you
just
keep
sending
this
and
it
keeps
maintaining
this
adjacency
forever.
There's.
What
I
would
expect
people
to
do
is
to
send
the
PR
and
go
okay
based
on
my
platform.
I
need
three
minutes
for
this,
so
they'll
set
the
whole
time
to
five
minutes
or
something
generous,
so
I,
don't
think,
there's
any
any
need
to
keep
updating
this.
This
is
supposed
to
be
a
limited
time
period,
not
I'm
not
supposed
to
do
this
forever.
L
So
what
does
the
neighbor
do?
And
this
is
the
value
add
here
since
the
neighbor
now
knows
that?
Okay,
your
are
restarting
your
forwarding
plane
is
maintained,
but
it's
not
getting
updated
because
the
control
plane
is
is
busy
you're
going
to
do
in
the
reload.
If
topology
changes
occur,
then
the
neighbor-
and
we
introduced
this
as
a
may,
because
you
could
put
some
intelligence
around
depending
upon
the
significance
of
the
topology
change.
L
Maybe
a
topology
change
occurs
which
doesn't
affect
any
traffic
transiting
at
the
restarting
node,
but
the
neighbor
now
has
the
choice
to
bring
the
adjacency
down
saying.
There
are
apology,
changes,
there's
nobody
home
and
the
control
plane
on
my
neighbor
he's
not
going
to
if
I
can
send
him
the
LSB
updates,
but
he's
not
going
to
process
them
if
the
restarting
router
is
the
D
is
on
a
LAN.
D
is
is
responsible
for
driving
the
the
update
process
on
a
land.
T
E
L
T
D
D
L
D
L
That's
so,
if
you're,
the
restarting
router
your
plan
recommend
that
and
we're
now
getting
into
implementation
details,
I
wouldn't
send
this
just
once
I'd
send
it
several
times
before.
I
actually
go
down
because
house
are
unreliable
and
you
want
to
make
sure
the
as
sure
as
you
can
that
the
neighbor
has
received
it.
But
of
course
there
is
the
acknowledgement
mechanism.
So
once
you
get
the
announcement,
you.
D
C
L
L
C
C
The
neighbor
yeah,
but
but
there
is
one
good
good
point
you
made
and
we
are
doing
some
some
extensions,
I'm
working,
I
think
from
the
people
from
brocade
or
someplace
on
a
draft
for
the
grace
where
we
started
extensions
and
I.
Think
I
should
put
this
piece
in
about
the
handling
the
DR,
because
right
now
it
assumes
that
you
terminate
helper
mode
and
what
we
should.
What
what
what
a
dr
should
do
or
diz
should
do
is
advocate
the
digital.
That
would
be
one
other
thing
you
could
do
as
opposed
to
terminating
graceful.
C
L
D
U
I'm
Peter
shinnok
francisco
I'm
going
to
cover
the
segment
routing
expansion
as
a
first
item,
so
there
has
been
a
significant
rewrite
updates
companies
jobs.
Basically,
we
put
it
in
a
parity
with
OSPF
v2s,
our
graph.
So
one
thing
that
we
removed
from
the
draft
was
the
asari
6,
because
the
SIV
6
extensions
are
now
in
a
separate
document.
So
the
whole
section,
3
5,
was
removed.
U
The
seed
label
Sub
T
L,
be
the
binding
sub
T,
the
syllables
of
GL.
We
caught
bone
changes
from
3
to
7,
because
the
C
has
been
used
for
the
tags
in
the
OS
PVC
LSX.
Indeed,
we
changed
some
of
the
code
points
for
a
s,
I'll
be
an
SMS
reference
sub
theories
because
again
these
are
shared
with
the
OSPF
v2
as
a
router
information
sub
theories
and
some
of
these
values
has
been
used
for
other
expansions.
U
So
the
binding
TLB
has
been
removed.
It
has
been
removed
from
the
OSB
v2
as
well.
So
it's
the
same
change
there.
We
added
the
support
in
the
range
TLV
for
the
ipv4
address
family
because
we
can
run
OSPF
v3
for
both
ipv4
and
ipv6.
So
yet
it
there,
we
changed
the
values
to
be.
You
know
the
same
for
the
ipv4,
so
we
used
the
value
1
for
before
and
value
a
value
0
for
before
and
value
146.
So
we
are
in
Singh
there
there
has
been
a
security
consideration
area.
U
References
has
been
updated
and
lot
of
comments
that
went
into
the
OSPF
v2
assad
wrapped.
As
a
part
of
the
you
know,
various
reviews
in
the
process
has
been
incorporated
here
as
well.
So
basically,
this
document
now
has
all
the
review
comments.
We
got
through
the
OSP,
a
v2
review,
so
I
guess
this
document
can
be
last
call
I,
guess
because
it
has
everything
it
needs
at
this
point.
So
did.
C
We
Acey
Linda
ma.
Did
we
ask
for
early
allocation
on
these
point
points?
Yet
I
know?
Yes,
we
did
I
know
nokia
was
implementing
it.
So
I
believe.
C
U
C
U
U
C
Is
there?
Anybody
who
doesn't
think
this
is
ready
for
working
grew
last
call
we'll
take
it
to
the
list,
of
course.
Okay.
That
would
be
this.
Would
this
much
I
think
I
put
this
ahead
of
OSPF
III
segment
routing,
because
that
needs
much
more
review
than
this
one?
Would
it's
funny
simpler
for
sure?
Yes,
I
put
it
on
probably
on
the
list
of
those
four
that
I'm
fit
those
longtime
documents
that
we'd
like
to
finish.
I
U
U
So
I
mean
this
has
been
presented
before
in
the
last
fight
here.
I
just
go
quickly
and
describe
what
the
flags
always
I
mean.
We
have
the
the
algorithm
types
registry
which
has
the
values
from
0
to
255.
Basically,
what
we
want
to
do.
We
want
to
use
half
of
the
values
for
a
user-defined
values
and
what
basically,
what
it
is
we
call
it
a
Flags
algo,
because
it's
defined
it
has
the
flexibility
for
a
user
to
define
it.
U
It
says:
flex,
Agha
represents
the
set
of
constraints
which
is
affinity,
exclude
include
whatever
we
can,
you
know
introduce
in
the
future.
It
has
a
metric
type
associated,
so
it
can
be
either
a
GP
metric
can
be
relay
can
be
teammates.
It
can
be
whatever
exotic
stuff.
We
can
add.
If
we
see
a
need,
it
also
is
an
algorithm
type,
which
means
what
type
of
computation
do
we
want
to
use.
So
basically
it
points
to
one
of
the
standardized
III
GP
algorithms.
So
we
can
use
SPF,
that's
what
our
app
uses
today.
U
We
can
use
something
else
if
you
need
to
in
the
future,
so
it
is
user
defined
user
has
the
flexibility
to
associate
the
value
with
this
set
of
metric
controller
constraints,
algorithm
types,
etc.
Now
it
allows
the
IGP
to
compute
the
pass
for
the
destination
using.
You
know
these
constraints.
On
top
of
the
topology
that
we
have.
We
are
not
doing
any
market
apology
here.
We
are
not
defining
new
topology.
We
are
doing
this
on
top
of
the
existing
topologies.
U
The
way
we
forward
a
traffic
for
the
Flex
algo
is
use
the
algo
specifics,
prefix
seats
and
they.
Basically
this
is
a
distributed
computation.
Everybody
uses
just
the
same
flex,
algaas
it
and
we
use
that
set
to
format
the
traffic
using
this.
These
flex
I'll
go
pass.
So
that's
what
the
job
is
talking
about.
U
Okay,
so
we
used
to
I
mean
most
of
the
discussion
in
the
draft
talks
about
how
do
we
make
sure
that
the
router
is
in
the
network
agree
what
the
definition
of
the
particular
algorithm
is,
and
we
have
two
ways
to
advertise:
it
we
simplified
it.
We
only
advertise
it
in
the
router
capability
TLV.
We
have
this,
what
we
call
a
flex,
algorithm
definition,
sub
TLV,
which
basically
tell
us
what
the
what
the
algorithm
definition
means.
So
everybody
agrees
and
can
do
the
consistent
computation.
U
The
original
version
talked
about
conflict,
so
what
the
conflict
was
detected.
Basically,
the
the
outcome
was
that
we
stopped
the
computation
removed
all
the
forwarding
associated
with
it.
We
changed
this
and
basically
we
we
change
it
to
deterministic
selection,
where
we
put
a
priority
field
in
the
in
the
sub
TLB,
and
then
we
also
use
the
router
ID
to
select.
If
you
have
a
conflicting
definition
coming
from
different
routers,
we
edit
the
algorithm
type.
So
we
can
support
non
SPF
computations
if
needed,.
U
Yeah
that
one,
so
we
added
so
originally
we
had
exclude
affinity
as
a
constraint.
We
added
include
any
include
all
affinities,
so
we
have
more
flexibility
and
we
also
updated
the
section
which
talks
about
how
to
be
compute.
So
there's
a
whole
section
talking
about
what
do
you
do?
What
are
the
steps?
What
is
the
order?
So
you
know
the
the
computation
is
consistent
as
being
about
what
of
editorial
changes
to
make,
make
it
clear
and
easier
to
read.
Also.
U
U
So
is
at
least
one
implementation,
maybe
more
I,
don't
know
I,
don't
know.
I
I
know
about
one.
We
would
like
to
make
this
a
working
group
document
and
allocate
the
code
points.
So
we
can,
you
know,
use
it
in
the
implementation
and
one
proposal.
I
don't
know.
I
would
be
interesting
to
see
the
feedback.
Maybe.
E
U
G
W
Organa
Nokia,
first
of
all,
thanks
for
bringing
this
app,
it's
definitely
interesting
work.
Few
kind
of
comments
and
I,
probably
extremely
misleading,
calling
this
segment
routing
flex
I'll
go
because
it's
not
it's
an
AGP
and
not
segmental
everything,
so
I,
work
on
and
then
I
already.
A
use
case
is
talking
beyond
segment
routing.
So
I
would
like
others
to
kind
of
change
that
and
and
focus
our
AGP
will
using
a
GP
registry,
we're
moving
it
to
a
GP
and
I
would
like
also
to
know
are.
U
So
so
we
are
using
signal
routing
as
a
forwarding
plane.
You
can
use
different
one,
so
I
agree.
We
can
make
it
more
independent,
but
the
the
reason
why
we
brought
it
up
with
the
segment
routing,
because
we
have
a
very
easy
way
to
format
using
the
seats
which
I'll
go
specific
sure,
but
we're
not
computing.
U
Sure
I
think
what
we
want
to
make
sure,
though,
that
a
JDI
base
is
IGP
computation
and
we
can
use
different
control
planes.
So
I
got
the
point
regarding
the
the
work
I
think
it
belongs
here,
because
this
is
all
about
igb
and
distributed
computation.
So
I
think
this
is
the
right
working
group
I,
don't
know,
maybe
chair
skin,
no.
X
X
X
T
U
W
W
A
E
Z
Other
than
a
future
rear
of
all
the
drafts
that
come
out
of
here
so
obviously
because
I
have
to
read
everything:
I
want
less
documents.
That
would
be
better,
especially
if
the
text
is
the
same.
I
don't
want
to
say
well,
this
wrap
says
that,
but
there's
other
dresses.
So
the
thing
right
now.
Having
said
that,
you
know
I,
think
I,
don't
think
it's
a
good
idea
to
make
a
blanket
policy
that
covers
everything,
because
you
know
there'll
be
exceptions.
Z
There's
going
to
be
cases
where
it
did
I
don't
know,
one
per
girl
has
to
be
has
to
be
has
to
require
more
work
than
the
other
protocol
for
the
same
type
of
thing,
and
it
may
not
make
sense.
I
think
this
is
something
we
need
to
look
at,
maybe
on
a
case-by-case
basis.
I
think
the
chairs
are
probably
better
situations
and
look
at
that.
Z
The
other
thing
that
that
care
about
up
in
the
back
is
conformance
as
we
look
forward
and
vendors
have
to
be
conformed
with
an
RFC,
and
they
only
implemented
one
of
the
protocols.
Why
I
don't
know,
but
the
only
permit
the
one
their
protocols
are
there
issues
there?
So
those
are
things
that
maybe
we
need
to
consider.
Maybe
there
are
other
things
I
said
well,
it
makes
sense.
Yes,
I
would
like
to
see
less
drafts,
but
I
understand
why
we
could
have
you
know
two
drafts
for
the
same
side
resolution
right.
D
X
So
experience
something
related,
but
not
exactly
matching.
You
can
very
well
set
up
something
like
a
framework
document
explaining
technology
setting
up
the
architecture,
and
then
you
have
per
protocol
specific
implementation.
Okay,
so
yeah
you
have
three,
but
that's
the
clean
framework
to
move
forward
to
make
one
out
of
two
making.
E
X
X
D
D
C
K
C
A
AA
Michael
Lorenzen,
so
I
can
tell
you.
I
once
asked
a
vendor
to
document
one
RC
every
should
that
they
did
not
follow
and
they
did
not
have
that
documentation,
and
this
was
one
of
the
major
vendors.
So
this
is
just
not
done
if
we're
talking
about
that
I
think
it
would
be
a
major
undertaking.
I
would
like
it
to
be
done
because
it
would
be
an
easy
way
to
say
like
where
did
they
deviate
from
the
shoulds?
But
nobody
does
this.
As
far
as
I
know,.
S
C
One
more
thing:
I'd
encourage
everybody.
Look
at
this
one
early.
You
know
that
we're
asking
you
know,
you
know
I
know
we
haven't
asked
for
adoption
or
anything
because
there's
a
lot
of
flexibility
here,
just
make
sure
everything
I
don't
know
if
you
followed
the
the
beer
algorithm
discussion,
but
it
got
almost
so.
The
document
was
almost
to
the
iesg
and
then
the
encoding
change.
So
let's
get
this
make
sure
everybody's
happy,
or
at
least
we've
reached
consensus
on
what
we
have
here
for
is
is
no
SPF,
so.
M
Just
a
comment
on
in
the
word
stood
it
tends
to
get
used
in
two
different
ways
of
one
is
that
this
is
very
good
practice
to
do
this.
The
other
is
that
this
is
what
you
should
do
99.9%
of
the
time.
Well,
we
know
that
there's
a
couple
corner
cases
where
it
would
be
the
wrong
thing
to
do
so
that
you
know
you
can't
quite
say
I
follow
this
should
or
not
yeah.
T
Follow
everybody.
My
name
is
Uma
myself.
Jeff
fundings
input
this
document.
This
is
any
speed
of
mentis
I'll
define
what
is
n,
SP
and
SP
can
be
it's
it's
non
shortest
path.
It
could
be
a
te
path
or
it
could
be
some
FRR
path
or
it
could
be
some
segmented
path
in
some
deployment
use.
Cases
like
you
know
where
you
need
to
change
couple
of
things:
simple,
SR
Network.
It
shows
router
a1
to
a7
b1
to
b7
and
all
the
six
are
marked
wherever
the
number
is
not
there.
The
metric
is
one
bi-directional
metric.
T
T
As
a
couple
of
four
edges,
NC
say
it's
and
knodel's
it's
together
to
represent
that
if
using
SR
MPLS
it
takes
8
labels
and
if
you
do
SRH,
it
takes
sr
8
sr
v6
its
and
the
labels
in
MPLS
case
can
be
more
if
you
are
using
entropy
based
on
the
our
LD
capabilities
unit,
insert
ela
al
al
al
pairs,
so
the
number
could
be
more.
So
what
are
the
issues
there
might
not
be
any
issue
for
large
label
stack,
so,
okay,
some
good
hard
ways.
T
If
you
deployed
with
capable
hard
ways,
it's
ok,
but
there
are
three
problems
we
identified.
One
is
the
fragmentation
issue:
MTU
and
fragmentation
choose.
Some
diplomats
is
very
sensitive
with
large
label.
Stats
are
with
large
SRH
header
and
the
second
one
is
hardware
and
performance
related
issues.
Mst
definitely
helps
to
mitigate
if
the
controller
is
competing
in
a
surpass
and
some
of
the
nodes,
if
you
are
using
Broadcom,
it
can
support
only
four
levels.
For
example,
chemisty
will
help
you
to
mitigate
this
by
choosing
an
alternate
path.
T
Controller
can
compute
and
Pt
PC
initiated
LS
B's,
updated
LSP
on
the
ingress
node.
But
if
you
don't
have
any
other
path,
it's
a
problem,
so
there
that's
where
it
helps
here
and
the
other
one
is.
There
are
deployment
scenarios
where
small
packets
are
extremely
like.
You
know
this
the
whole
whole
slices.
For
example,
there
are
some
slices
where
only
the
packet
sizes
are
only
4200
bytes.
If
you
take
it
to
account
the
transport
and
network
and
path
overhead,
it's
almost
300
to
400
percent
overhead.
T
So
what
is
we
are
proposing?
There
is
n
SPF
I
DT
L.
We
are
talking
about
it's
an
empty,
empty,
capable
prefix
advertises
this.
It
could
be
any
node,
it
advertises
the
fake
fact
prefix
and
also
it
is
the
ordered
segment
list.
It
could
be
n
spi,
for
example,
any
spare
PT
type
is
MPLS.
The
audit
list
is
combination
of
segments
or
no
decency
are
terrific
saves
and
there
are
a
couple
of
others.
Tlv
suede
that
can
be
used
for
other
purposes.
I'll
talk
about
it
later.
D
T
So
the
computation-
how
do
we
do
this
once
we
advertise
this?
For
example,
this
in
example
r1
to
r4,
it's
an
SR
path
once
it
is
received
or
SP
regular
SPF
computation
for
each
topology
at
the
end
of
SPF
computation.
You
just
have
to
look
into
in
the
path
NSP.
If
you
are
out
of
is
in
your
path.
If
you're
out
raised
in
your
path,
your
next
stop
is
computed
to
the
eventual
destination.
For
example,
r9
take
turn
in
case
our
name.
T
If
r4
is
the
prefix,
it
isn't
raising
the
shortest
path
or
prefix
NH
computed
on
SPF
is
11,
so
instead
the
next
segment
is
our
10.
The
next
stop
is
changed
to
our
10,
so
the
same
thing
will
happen
in
our
8
in
our
1.
So
in
summary,
what
we
are
talking
about
is
a
small
change
after
SPF
computation
adjust
your
next
stop
to
the
next
segment
next
top
of
the
next
segment
instead
of
the
eventual
destination.
That's
all
after
that
you
have
to
come
program
the
data
plane
with
the
received
n
SPF
ID.
T
T
So
what's
the
relation
we
two
bindings
it
it's
not
related.
There
are
a
later
version,
Zinna
solace
in
expertise.
A
CR
was
binding.
Cities
basically
used
for
advertising.
The
six
fur
on
behalf
of
one
node
or
multiple
nodes,
laugh
SR
or
not.
So
there
are
two
things
we
are
thinking
to.
Update
I've
got
some
offline
feedback
on
this.
One
is
way
to
advertise
this
NS
piece
from
today,
SR,
if
PC
initiated
LS
piece
are
programmed
on
the
ingress
node.
You
can
under
it
s
from
the
ingress
node.
T
But
the
thing
is
your
prefix
is
not
from
your
ingress
node.
Your
address,
node
is
the
owner
of
the
prefix,
so
binding
cities
does
differently
any
node
candidate
as
the
binding
said,
but
we
were
thinking
if
address
nodes.
If
the
owner
of
the
prefix
advertises
this
path,
the
advantage
is
any
caste
is
easy.
We
can
solve
this.
Any
node
can
give
the
flexibility
any
node
can
advertise
this,
but
you
need
to
put
the
system
ID
to
it.
Disambiguate
the
any
cash
prefixes
and
the
other
TL
is
I
was
talking
about
the
traffic
accounting.
T
You
don't
have
to
do
advertise
the
traffic
accounting
information
in
the
IG
piece,
your
controller
programs.
This
and
controller
could
fix
this
up.
The
traffic
information
on
all
the
nodes.
Only
thing
is
the
NSP
of
ID
will
allow
you
to
control
or
manage
this
traffic's
cleanly
to
the
each
paths.
Another
thing
is,
you
can
advertise?
Let's
say:
I
need
this
traffic
accounting,
one
to
you
can
advertise
along
this
nsv
path.
Let's
say
four
out
of
my
ten
pass.
These
two
new
paths
need
my
traffic
atlantic.
It
can
be
done.
T
D
T
C
E
T
Q
T
C
T
W
W
This
can
be
solved
by
creating
labels
and
like
binding
type
of
set
labels,
but
pushing
the
like.
What
are
we
gonna
draw
the
line
because
you
could
say
they
will
do
the
same
sort
of
binding
sites
with
like
we
can
start
advertising
stuff.
You
know
SPF
in
ISAs
and
come
up
with
solution
for
that.
So
once
we
make
an
exception
it
we
may
open
the
Pandora
Boxx.
T
W
W
If
I
exclude
the
links
that
you,
if
you
go
to
the
apicture
with
links
and
exclude
certain
links
through
whatever,
through
whatever
come,
like
something
dictates
you
to
take
a
path
from
A
to
B
and
immediate
the
links
that
you
are
meeting
from
shortest
path.
So
if
you,
if
you
do
this
as
a
input
to
your
shortest
path,
calculation,
that
you
the
fact
of
computing
shortest
paths
with
a
different
links,
this
is
no
different
than
then
computing
shortest
paths
that
that
has
total.
W
You
know
redundant
shortest
paths
that
that
has
zero
correlation
between
the
primary
you.
You
still
computing
shortest
paths,
avoiding
every
link
from
the
other
path,
and
so-
and
we
don't
signal
that
stuff.
So
why
would
we
do
the
exception
here?
There
is
a
real
problem:
I'm,
not
arguing
that
there
is
a
real
problem.
There
is
some
technologies
that
avoid
talking
about
how
many
labels
we
adding
that's.
Okay,
that's
also
a
problem.
AD
T
AE
AE
T
AE
AE
But
I
think
to
address
the
scaling.
You
need
to
somehow
limit
how
many
of
these
you
are
gonna
have
and
then,
as
the
links
fail
and
stuff,
because
this
is
a
n
square,
computation,
theoretically
right
and
how
many
of
T's
is
going
to
change
and
then
what
does
it
mean
for
iGPS
flooding?
You
have
to
be
to
address
these
issues.
If
you
want
this
to
be
operationally
practical
in
a
single
nsv,
this
looks
beautiful,
but
it's
not
we're
not
talking
up.
You
are
talking
about
n
square
problem
here.
T
U
T
T
Q
AB
T
There
is
somebody
some
of
you
had
attended
yesterday's
DMM
working
group
meeting.
There
is
a
3gpp
si
for
a9
interface,
the
new
interface
for
Phi
G,
where
they're
looking
for
idea
of
technologies
to
include
this
for
the
data
plane,
SR
v6
is
one
of
the
proponent
there
we
have
I
am
working
with
Dino
on
the
Lisp
lisp
advantage
of
the
Lisp.
Is
it
sauce
the
mobility
plus
its
agnostic
to
underlay?
It
doesn't
propose
that
it
will
not
put
SR
v6
or
ipv6
and
everybody
people.
T
If
you
want
to
serve
a
six,
they
can
use
it
and
the
advantage
of
Lisp
is
it.
Is
agnostic
to
underlaid
can
be
sr
MPLS
or
sr
ipv6,
but
the
problem
is,
there
are
slices
mm
TC
I
use
mm
TC
slice
and
URLs
is
slice.
The
short
packets
are
there
all
the
all,
the
all
the
all
the
data,
all
the
data
is
constituted
of
short
packets,
so
they
cannot
take
this
warhead.
W
For
the
yesterday's
DMM
working
go
okay,
so
now
we
talking
cuz
that
will
be
six
okay,
thank
you
so
because
it's
a
lot
of
our
hat
and
hardware
doesn't
support
it.
This
is
not
the
reason
to
hack
a
GP,
and
this
is
not
the
reason
to
say
that
not
to
use
binding
zips,
and
this
is
not
the
reason
to
say
non
shortest
path
to
something
that
is
sure.
This
path
again,
I
can
have
a
discussion
of
line.
V
C
AA
I
known
that
topic
of
jabber
scribe,
typically
six
or
eight
sessions
per
IETF,
this
is
the
worst
route
with
people
wearing
badges.
So
on
algebra
here
you
get
the
slides.
You
get
the
voice,
recording
and
I'm
typing
down
the
name
as
you
walk
up
to
the
mic,
but
if
you
don't,
if
you're,
not
wearing
your
badge
and
you're
saying
your
name
well,
even
if
you
tell
me
your
name
are
gonna
get
it
wrong.
So
please
wear
your
badge
and
this
will
be
properly
recorded
and
the
minute
taker.
AA
D
AA
D
E
AB
Morning
they
say
situation
from
Maui
and
the
name
of
the
trap
saying
is
avoiding
traffic
black
hosts
for
joke
aggregation
in
ISS,
and
this
is
version
2
drugs,
and
this
is
the
change
things
were
in
their
own.
We
have
added
the
team
Chun
as
a
closer
and
in
the
previous
draft
we
extended
the
protocol
and
use
a
new
ice
ice
theory
to
carry
and
reachable
IP
prefix,
but
in
the
latest
draft
we
introduced
the
infinite
cost,
such
as
to
reusing
existing
theory.
AB
Instead
of
creating
a
new,
and
they
says
the
problem
we
are
going
to
solve
and
when
I
sized
running
in
DC
topology
such
as
benefit
apology,
the
leaf
nodes,
often
configured
to
now
to
receive
expressing
notes
from
other
leaf
nodes,
they
only
receive
a
Depot
Road
and
the
sets
the
next
hope
to
be
outer
spaniel.
So
in
this
example,
the
traffic
form
from
leaf
1
to
live
war
will
be
easy,
impede
among
all
these
Spanos.
AB
However,
the
link
fader
in
the
network
will
cows
like
holding
problem
problems.
We
use
this
example
to
illustrate
this.
If
the
link
between
spine
3
and
leave
for
goes
down,
if
one
will
not
know
it
and
still
send
traffic
to
span
three,
so
the
traffic
will
be
discarded
as
bansuri
such
causing
black
holing
problem.
So
this
is
the
big
picture
of
our
solution.
AB
We
also
use
this
example
to
illustrate
this
after
their
link,
filler
sponsoring
should
advertise
prefix
a
and
pricks
the
perfect
speed
to
leave
one
two
and
three
and
set
the
metric
to
be
infinite
and
after
receiving
the
prefix
leaf,
one
should
ID
prefix
a
and
privy
to
his
routing
table
and
the
stats.
The
next
hop
will
be
spam,
Wong
and
the
span
to
such
avoiding
the
black
holding.
AB
These
are
some
details.
When
link
failure
happens
between
a
Spaniard
and
a
leaf
node,
the
Spaniard
should
include
our
prefix
attached
to
the
leaf
node
into
the
IP
edge
usability
theory
and
cites
the
cost
of
the
prefix
to
be
infinite.
The
value
of
infinity
is
built
to
be
determined
a
pet
and
appended
the
IP
reachability
to
our
way
into
the
ICS
RSP
and
descended
RSP
to
every
other
leaf
node
in
connect
to
the
leaf.
AB
W
Android
Organon,
okay,
I'm,
trying
to
understand:
what's
the
intent
of
the
draft
telling
people
who
know
how
to
do
protocols
and
advertisement
how
they
should
be
doing
this
showing
one
of
the
options
there
are
multiple
options
to
solve
your
problem.
This
could
be,
one
of
them
doesn't
need
to
be
standardized.
Anybody
who
does
routing
protocols
probably
knows
how
to
do
an
advertisement
when
this
happens,
so
you
don't
have
black
hole
with
existing
mechanisms.
AB
L
AB
D
So
so,
in
any
case,
this
is
this
is
a
work
building
and
Hoppus
fine
leaf
right
I
mean
it's
related
to
this
fine
leaf
work
by
naming
yeah
yeah
so
white.
If
it's
a
good
idea,
then
I
think
you
guys
it
should
not
be
done
separately
right,
I
mean
I.
Think
you
need
to
convince
the
naming
and
you
guys
if,
if
it's
needed,
put
the
work
together,
okay,
okay,.
D
H
D
AB
O
S
AF
So
for
large
and
the
Vince
network,
so
the
ITB
flooding
were
flooded
to
every
link
and
the
foursome
link
multiple
being
stay.
They
were
flooding
through
those
links,
so
this
will
raise
a
lot
of
issues
such
as
reduced,
secure
mobility
and
also
slowed
down
on
the
convergence.
I
heard
that
for
some
people
at
work
that
converging
time,
maybe
a
second
200
seconds.
So
here
we
proposed
a
solution
to
resolve
these
issues
through
flooding
to
some
small
portion
of
links,
and
also
we
only
flood
and
say
link
was
not
a
multiple
times.
AF
D
AF
AF
Also,
it
must
flood
the
link
States
to
Evernote
in
any
conditions
such
as
an
ingot
hang
loaded
on
the
link
hub
and
also
which
we
should
or
must
reduce
the
flooding.
The
past
cases
are
almost
minimized
flooding,
so
we
don't
wanted
to
introduce
more
flooding.
So
that's
I
think
that's
the
most
of
the
requirement.
AF
So
for
loving
our
first,
we
needed
building
a
fight
anthropology
so
that
there
are
many
ways
to
build
a
flat
entomology.
So
here
I
just
present
an
example
way
to
building
a
flat
entomology.
In
fact,
you
didn't
follow
entomology.
There
is
a
symbol,
so
here
we
can
just
give
three
steps
to
building
a
flap
in
topology.
We
maybe
I
just
need
two
steps.
So
first
we
select
a
node
in
the
network.
You
know
in
some
ways,
for
example,
we
can
select
a
the
largest
and
the
node.
AF
AF
AF
So
this
is
just
to
give
more
details
about
about
a
building
for
a
flatte
entomology,
so
we're
building
for
lighting
for
money.
We
can
have
two
two
types
of
algorithms,
so
one
one
type
is
that
we
don't
consider
which
in
OT
just
a
portal
with
which
you
know
doesn't
support
for
lighting
reduction.
So
these
are
assume
we
assume
that
Evernote
in
network
support
flooding
reduction.
That's
a
one
type
of
algorithm,
so
announce
table
is
that
we
should
consider
whether
a
node
supporting
flat
for
lighting
reduction.
AF
So
in
order
to
have
this
type
of
a
chrism,
so
we
just
introduced
an
a.1
feed
flag
F,
which
indicates
that
whether
a
node
supports
flatting
reduction.
So
if
we
set
to
1-
and
then
that
means
this
notice
supports
14
reduction
so
with
this
flag,
and
we
can
also
consider
the
algorithm.
So
when
we
build
a
tree,
we
want
those
those
supporting
flattened
reduction,
maybe
just
connect
together
continuously,
just
chief
a
patter
result.
Of
course
it
you
can
just
leave
it
all.
This
kind
of
connecting
those
nodes
supporting
the
adoption
together
as
given
are
you.
AF
Just
those
are
for
encoding
Tony
great,
so
for
this
flag
we
can
chasity
at
one
beat
in
the
different
protocols.
You
know
spear
for
v2,
it's
one
bit,
you
know
speed,
V
3
is
also
one
bit
in
the
rotors
days,
so
you
ISS
it's
also
a
one
bit,
so
these
are
in
the
different
data
structures
in
the
ISS,
the
rotor
ability
RV
at
one
people.
So
this
is
a
extension
is
a
very
simple
and
then
we
let's
look
at
the
revised
the
flooding
procedure.
AF
So
this
procedure,
the
check,
is
very
simple
so
a
week,
so
we
consider
different
case.
One
case
is
that
for
receiving
an
Inc
state,
so
in
this
case
we
just
flood,
that
is
a
L
say
according
to
the
flooding
or
G,
for
example.
For
this
note,
a
we
load
a
receives
that
in
state
from
node,
R
1
and
then
not
a
willful
oddities
as
a
to
the
other
link
on
the
flood
entomology.
AF
So
you
know
another
suppressor
case,
so
no,
they
may
receive
LSA
from
down
below
this
note,
and
then
this
thing
is
not
on
the
flood
entomology.
So,
in
this
case,
nota
a
we
are
flat.
This
L
say
to
all
the
links
on
the
flooding
for
G.
So
in
this
case
not
a
will
flood
this
sa
to
this
one
and
that
to
this
one
on
the
flood,
mythology
and
entities,
one
on
a
flood
of
authority,
so
this
is
for
the
LSA
receiving
FSA,
so
a
magic
aces
that
a
node
may
originate
NSA.
AF
So
in
this
case
we
is
divided
into
different
sub
cases.
One
case
is
that
this
as
a
without
any
change
just
to
refresh
so
in
this
case
with
flood
that
deserves
a
according
to
the
valerian
topology.
So
in
another
case,
is
that
this
and
say
with
changes.
So,
for
example,
this
lot
is
done
and
then
the
Earth
Day
will
generate
this
is
where
generator
assayed
so
with
with
changed
with
changes.
So
in
this
case
we
flood
this
L
say
to
every
adjacencies.
AF
C
AF
So
also
we
consider
load
up
so
in
this
case
a
new,
a
new
notice
is
up
in
this
case.
We
do
some
special
process
considerations,
so
in
this
case
we
assume
only
this
old
assumed
that
this
new
adjacency
is
under
flood
of
ammaji.
So
that's
all
so.
With
this
revised,
the
procedure
looks
like
we
can
cover
all
cases,
which
means
that
in
any
case,
we
can
guarantee
the
Dean
state
will
flood
into
Evernote
in
the
Wemo.
F
AF
F
C
AF
AG
O
AC
AD
C
AD
AD
S
AF
For
those
case,
I
think
in
those
a
transition
stage,
I
think
we
consider
that
one
is
not
on.
That
seems
not
no
serious
issues
and
then
I
think
we're
for
each
case
Eve
over
someone
paper
real
scenario,
and
then
we
can
pick
out
whether
we
cover
that
one
yeah,
a
single,
also
some
people.
So
this
is
also
some
kind
of
mathematical
problems
and
there's
some
people
may
be
interested
in
and
then
give
some
proof.
It's
covered.
AF
Summary
presented
that
topology,
so
you
should
also
wear
generator
flagging
Havarti,
so
we
don't
need
a
flood
is
a
lot
in
proportion
to
the
network
and
then
I
think
it
will
go
to
procedural
solution.
We
propose
I
think
we
at
the
first.
A
short
we
think
we
might
add,
is
requirement,
and
so
we
welcome
comments.
My.
AE
Name
is
Alicia
who,
from
Alibaba
I
query
with
ASUS
and
observation,
it's
same
that
it
has
angry
chicken
issue
for
just
add
a
summation
if
the
link
on
the
flagging
table
gh
and
you
need
the
dependency
on
the
original
flooding
mechanism
right.
In
that
case,
the
link
tongue
event
had
been
this
flooded
to
use.
Do
you
need
another
flood
on
the
reconverted
flooding
topology.
AF
We
just
the
so
people
says
that
they
don't
really
issue
she
can't
issue,
so
we
just
maybe
just
go
through
one
real
case
to
see
whether
we
really
have
an
issue.
Okay,
so
here
we
suppose
D
something
is
done,
okay,
so
in
this
case
this
and
English
is
done.
The
procedure
like
this
one
so
note
a
willful
at
F
say
to
everyone
to
this
one
to
this
one
to
this
one
to
this
one.
So
no
two
are
for
way
to
the
same
machine.
AF
So
lot
because
note
our
forward
practice
lingered
on
LS,
a
change
right
according
to
the
revised
procedure.
These
are
all
the
way
flat
those
changes
way
to
everyone
and
then
everyone.
So
in
this
case,
where
is
the
corner?
You
can
indicate
where
its
corner,
so
this
L
say
change.
They
were
flogged
to
every
one
ever
knows.
AF
AF
AC
W
Just
a
quick
one
end
organ
okay,
you
really
need
to
do
a
better
to
justify
this.
This
is
basic
mechanisms
you're
trying
to
optimize
making
them
complex,
aka,
making
them
wrong.
And
if
you
have
flooding
issue
induct
apology,
then
you
change
whoever
software
using
because,
like
this
is
a
simple
topology
and
people
are
poking
issues
in
your
in
your
solution.
So
you.
AF
R
W
C
I,
have
one
thing:
I
mean
I
mean
there's
there's
you
know,
there's
a
lot
of
ways
that
people
are
looking
at
this
data
center
flooding
problem
and
we're
obviously
not
going
to
standardize
all
of
them,
and
so
you
know
it's
going
to
be
sort
of
a
dependent
on
it.
I
think
a
lot
of
its
gonna
be
dependent
on
implementations
and
what
works
you
know,
and
it
was
simplicity,
so
yeah
yeah,
you
can
initiate
more
discussion
on
the
list.
Thank
you.
Thank
you.
C
AH
J
AH
AH
Things
touching
so
we've
seen
this
problem
before
long
ago.
In
a
galaxy
not
so
far
away.
We
had
a
network
built
on
top
of
frame
relay.
It
only
had
20
nodes,
it
had
a
full
mesh
of
VCS
and
it
collapsed,
and
we
scratched
our
heads
and
looked
at
it
and
played
with
it
and
said:
oh
there's
too
much
flooding.
Now
that
was
a
full
mesh
network
and
as
soon
as
you
have
a
full
mesh,
you've
got
tons
and
tons
of
flooding
going
on,
and
the
solution
that
we
had
at
the
time
was
mesh
groups.
AH
This
this
slide
took
me
about
one
minute
to
make,
because
that
little
pretty
picture
you
see
there
was
courtesy
of
Google.
I
did
not
draw
this
alright,
so
data
center
routing
is
actually
a
subset
of
this
problem
and
all
of
the
techniques
that
we
know
about
for
solving
the
complete
graph
actually
apply
on
the
data
center.
You
could
use
mesh
groups
in
the
data
center,
except
for
one
problem
configuration
that's
unacceptable.
Next
slide!
Okay!
AH
So
if
we're
not
going
to
do
mesh
groups,
we're
not
going
to
do
manual
configuration
how
about
we
just
compute
what
links
we
want
in
our
topology,
okay,
reasonable,
easy
enough
to
do
which
system
is
going
to
do
this
configuration
well?
How
about
we
elect
one?
Okay?
Well,
that's
not
too
hard!
We've
got
di
a
selection.
We
can
apply
the
same
thing
at
the
area
level
and
elect
somebody
at
the
area
level.
Do
you
sit
and
do
this
computation
compute
with
the
flooding
topology
should
look
like
and
then
flood
that
out
next
slide?
AH
Okay,
so
here's
a
really
straightforward
example!
Suppose
we've
got
a
small
data
center
k,
2220
spines
200
leaves
I'm,
told
that's
tiny
compared
to
what
many
people
are
doing.
That's
4,000
links
a
flooding
topology
on
that
we
can
just
build
the
cycle.
Doing
round-robin
selection
every
leaf
ends
up
with
only
two
links
on
the
flooding
topology,
where
we've
got
a
cycle,
so
we
have
no
problems
with
single
points
of
failure,
and
so
we
end
up
flooding
only
on
twenty
links,
20
spines
and
then
two
400
links.
AH
What
do
we
have
to
do
for
a
leader
election?
Well,
it's
very
straightforward.
We're
just
going
to
steal.
D
is
election
completely
advertised
one
TLV
into
the
area
to
indicate
a
preference
do
election
on
on
all
of
those
whoever
wins
the
election?
You
don't
have
to
signal
anything.
You
just
go
ahead
and
do
the
computation
gonna
get
you
back
up
once
yes,
sir
I
mean
IIIi,
see
that
90%
reduction,
but
it's
you
said
20
likes,
you
mean
400
likes
spines
end
up
flipping
on
20
links,
each
all.
C
AH
AH
The
topology
and
then
the
replication,
the
maximum
maximum
degree
of
the
node
okay,
okay
I
thought
you
meant
the
total
sorry
yeah.
Okay,
we
can
use
the
usual
tie
breakers
here.
This
seems
relatively
straightforward,
okay,
so
now,
how
do
we
flood
the
topology?
How
do
we
talk
about?
What's
going?
What
we're
gonna
do?
We
need
a
reference
mechanism
for
talking
about
the
nodes
in
the
topology
and,
unfortunately,
I.
Don't
can't
think
of
a
good
way
of
absolutely
describing
everything.
AH
That's
going
on
so
proposal
here,
let's
flood
a
list
of
the
systems
in
the
area
and
the
list
of
that.
The
point
of
this
is
just
to
give
us
a
set
of
indices
to
work
with.
If
we
have
indices,
then
we
can
describe
things
very
succinctly
rather
than
carrying
around
a
bunch
of
system
IDs
all
over
the
place,
and
all
of
this
is
just
to
get
encoding
down
to
something
smaller.
AH
Okay.
Now,
in
the
initial
version
of
this
draft
I
proposed
sending
out
an
adjacency
matrix,
my
boss
took
one
look
at
this.
He
mumbled
some
numbers
and
told
me
that
I'm
daft
to
use
the
British
ISM
he's
absolutely
right.
This
is
a
really
bad
way
of
doing
things.
I've
withdrawn
this
and
published
another
version
of
the
draft,
where
we
do
next
slide
a
flooding
path,
if
you
use
sparse
matrix
techniques
and
instead
just
list
the
indices
of
the
systems
that
you
want
in
your
flooding
topology.
AH
AH
AH
You
could
do
things
that
are
higher
connected,
but
of
course,
if
you
do
that,
you
increase
your
amount
of
amplification
that
your
flooding
does
probably
want
to
think
very
hard
about
doing
that.
We
want
to
minimize
the
node
degree.
We
don't
want
one
node
that
has
to
slide
12,000
copies
of
things.
If
we
don't
have
to
probably
really
like
to
distribute
the
workload
as
much
as
possible,
we
want
to
minimize
the
diameter
of
the
flooding
topology.
AH
AH
AH
AH
So
next
slide
here's
another
example.
This
one
I
did
write
this.
This
one
took
a
long
time
to
draw.
It
was
a
boring
meeting.
If
anybody
is
from
this.
I
did
this
in
keynote.
If
anybody
from
Apple
is
here,
I
want
connectors,
connectors,
connectors
connectors,
please
I'm
the
graft
that
would
that
would
be
okay,
Donna!
Thank
you.
I
think
you
can
afford
it.
This.
AH
Is
K
for
17
it
was
got
68
links
to
begin
with,
via
very
small
topology.
If
we
look
at
the
minimal
topology
next
slide,
that
alone
drops
it
down
to
34
links,
that's
a
50%
reduction
already
on
a
very
tiny
topology
and
and
the
savings
go
up
just
as
you
increase
the
replication.
Okay,
now
I,
don't
think
I've
ever
seen.
Anybody
at
IETF
present
a
theorem
before.
J
AH
So
I
define
a
minimal,
minimal
flooding
topology
on
a
spine
leaf
as
by
connected
with
the
degree
of
the
leaf
being
okay
leaf
only
has
two
links
throughout
I
can
show
that
for
a
such
a
graph,
where
the
number
of
leaves
becomes
much
larger
than
the
number
of
spines
close
to
N,
squared
that
there's
a
minimal
flooding
topology
where
the
diameter
is
4.
This
is
a
fun
result,
because
this
what
this
night
means
is
as
we
grow
the
topology
as
we
add
racks,
the
flooding
topology
stays
fast
and
if
anything
gets
faster.
AH
F
AH
Now
another
person
brought
up
the
fact
that
we're
still
flooding
on
too
many
links,
and
they
were
of
the
opinion
that
we
didn't
need
to
be
by
connected.
So
what
happens
if
we
throw
out
the
requirement
for
full
blind
connectivity
if
you're
feeling
more
risky-
and
you
don't
care
about
losing
the
top
of
rack,
you
don't
care
about
losing
a
leaf.
Well,
what
happens
if
we
just
keep
our
spines
by
connected
and
leave,
make
the
leaves
some
of
the
leaves
singly
connected
okay?
Well,
we
can
reduce
the
topology
even
further
next
slide.
AD
Can
I
ask
you
a
question
about
the
sweeter
side,
sure
sure
you're
saying
that
as
the
topology
grows
this
previous
into
this,
the
the
problem
is
you
you
want
N
squared
leaves,
and
so
that
apology
is
not
growing
only
in
the
leaves.
It
also
is
growing
in
the
spine,
and
so
the
number
of
cases
that
you
actually
see
this
I
don't
know
data
center
networks.
AD
AH
B
AH
AH
So
so,
if
you
take
that
and
you're
willing
to
take
risk-
and
you
can
see
you
end
up
with
a
bunch
of
leaves
that
are
singly
connected
here-
you
can
cut
the
topology
even
further
you're
down
to
21
licks,
okay,
all
right
a
little
bit
of
wisdom
from
dealing
with
scalability
problems
for
my
entire
career.
Whenever
we
have
these
problems,
we
should
use
all
available
tools.
AH
H
AH
Just
one
more
tool
in
the
tool
chest
this:
this
solution
is
nice
because
it
is
really
strictly
not
for
data
centers,
but
it's
for
any
dense
topology.
We
designed
link-state
protocols
and
the
flooding
back
when
links
were
really
expensive.
You
know
64
kilobits,
synchronous,
modems
and
imps,
and
you
know
if
that
cost
millions
of
dollars,
and
we
no
longer
have
that
problem
now
a
link
is
$5.99
at
Fry's.
You
know
it's
just
not
worth
it,
so
we
really
have
a
different
design
point
than
where
we
started
building
link
state
protocols.
AH
It's
not
surprising
that
flooding
doesn't
work,
and
this
is
another
approach
for
rethinking
how
to
do
things.
In
the
general
case
where
we've
got
dense
topologies
the
solution
is
nice
because
we
don't
care
about
lateral
links.
You
can
add
whatever
you
like
to
your
topology.
If
you
decide,
you
have
to
have
a
link
between
two
racks.
Well,
you
know
this
will
adapt.
AH
External
links
are
not
a
problem
if
you
want
to
interact
and
expand
your
area
outside
of
your
data
center.
This
works.
We
can
also
combine
this
with
some
of
the
other
proposals.
Naming
and
less
has
an
eye.
I
have
a
nice
draft
if
you
apply
that
to
you,
have
some
restrictions,
but
you
can
get
even
better
scaling
because
you're
flooding
less
information.
So
all
right
last,
if
you
like
this,
if
you
like
this,
please
seek
out
this
lady,
her
name's
Artie
tell
her
you
like
it.
AE
AH
Much
the
only
drawback
that
I
see
is
that
we're
going
to
add
to
the
link-state
data
by
base
we're
actually
adding
more
information
that
has
to
be
flooded
around.
So,
if
you
take
a
look
and
say
well,
we're
decreasing
a
multiplier
on
this
side
of
the
equation,
but
we're
adding
to
the
value
on
this
side
of
the
equation
right.
So
it's
a
trade-off.
Yeah
I.
AH
AD
D
D
AH
AD
AH
AD
AD
AH
D
L
AH
Transition
cases
I,
don't
think,
are
problematic.
This
is
because
the
worst
problem
that
you
have
is
that
you
have
an
old
topology
and
a
new
topology
both
are
being
flooded
at
the
same
time.
If
that
does
happen,
you
don't
have
to
decide
you
just
flood
on
both
and
they're
very
likely
to
overlap
significantly
so
you're,
not
having
a
real
bad
time
anyway.
Also
do
the
other
transition.
That's
going
to
happen
is
if
you
actually
lose
wrong.
Answer
hang
on
actually.
L
AH
AB
AH
The
network
converges
quickly,
even
if
you
lose
the
area
leader
now,
when
you
elect
a
new
area
leader
assuming
the
first
area
leader
doesn't
come
up
quickly,
then
you're
gonna
want
to
have
that
new
area
leader
flood,
a
new
topology
and
and
a
smart
thing
that
you
might
want
to
do,
and
this
is
more
risky,
but
he
might
want
to
purge
the
original
area
leaders
topology
in
the
overlap.
We
have
this
dangerous
transition
where
you
might
have
to
flood
on
both
topologies
okay,
I.
C
Have
one
before
Alvaro
makes
a
question
when,
when
let's
say
you
have
by
connected,
are
you
accepting
the
fact
the
area
leader
doesn't
go
down,
but
you
have
some
information
that
is
flooded
in
between
computations
of
topologies?
How
do
you
sure
that's
reliable
if
you
lose
your
funding
topology
if
somebody's
completely
isolated?
C
AH
C
C
AH
Z
Donna,
why
we
speaking
as
a
working
member
so
I
don't
have
a
specific
comment
about
your
solution,
but
what
I
want
to
say
is
that
most
of
the
questions
that
have
come,
our
part
around
what
happens
is
this
fails?
What
happens
is
that
node
fails?
What
happens
when
the
link
goes
out?
People
for
finals
presentation
before
it's
20
said
this
is
not
a
new
problem
right.
This
is
an
old
problem
and
it
has
been
solved
before
for
mixing
protocols
as
well.
Z
Now,
mobile
networks
have
failures
all
the
time
right,
you're
moving
all
the
time
you
never
know
where
the
leader
is,
you
never
know
if
the
link
is
gonna
go
down,
you
know
what
second
from
now
you
never
know
what
happens
once
you
transmit
the
LSA
if
it
actually
got
to
the
other
side
or
not,
but
the
mechanisms
that
have
been
done
only
are
implemented
for
us.
We
have,
but
most
importantly,
for
all
this
R,
which
is
deployed,
and
it
works
pretty
well
the
mechanism
you
can
go
read
all
about
all
this
are
I.
AE
Need
more
follow-up
to
what
AC
said
and
I
think
you
answered
it,
but
I
just
wanna
double
checks,
more
of
a
clarification.
So
when
the
flooding
topology
changes
and
a
new
link
becomes
part
of
the
flooding
topology,
those
two
routers
and
CS
MPs
of
some
sort
to
synchronize
what
they
know
about
what
has
been
flooded.
While
the
topology
was
changing.
AH
AE
D
Z
AD
I
liked
the
idea
in
general,
but
the
idea
of
you
know
sending
out
the
having
one
person
one:
what
area
leader
decide
what
the
flooding
topology
is
good
and
bad.
It's
good,
because
you
know
you
don't
have
these
fun
I!
Think
in
the
previous
presentation
there
was
a
big
big
flaw.
On
the
other
hand,
if
you
could
have
everyone
determinacy
compute
the
same
thing,
then
you
wouldn't
have
to
actually
send
it
out,
and
that
might
be
better.
AD
AD
One
of
the
common
QD
again
is
if
something
along
the
flagging
to
potty
breaks
and
you
fall
back
and
I
mean
we
haven't
discussed
it.
I
haven't
seen
it,
but
if
you
fall
back
to
flooding
on
everything
because
you're
just
not
sure
about
things,
that'd
be
really
bad
because
you
get
into
the
problem
that
you're
trying
to
solve
right.