►
From YouTube: IETF103-RIFT-20181107-1540
Description
RIFT meeting session at IETF103
2018/11/07 1540
https://datatracker.ietf.org/meeting/103/proceedings/
B
B
D
D
D
And,
and
about
their
supporting
documents,
we
have
the
young
specification
going
on
the
applicability
statement
and
stress
Inessa
documents
which
were
required
in
our
shuttle.
They
need
to
be
written,
so
if
anyone
can
help
with
appreciate
and
the
PGP
and
SR
graphs
that
were
separate
outs,
they
are
on
the
side
burner
now,
but
we
do
have
that
skeleton
draft.
Their
comments
and
contributions
are
welcome.
I
think
that's
the
admin
part.
B
E
All
right
so
now
in
the
middle
of
the
road,
the
crowd
is
getting
smaller,
still
bigger
than
you
know,
early
OSPF
sessions.
I
remember
so
that's
the
update
on
the
0-3
draft,
which
is
pretty
much
all
the
work
we
did
between
0
2,
&,
0
3,
so
that
last
IDF's
tons
of
authors
got
added
so
I
just
push
it
all
on
a
separate
page.
We
call
this
whole
thing,
the
rift
authors
and
they
all
fight.
You
know
whatever
the
publication
process
will
be
all
right.
E
So
the
last
one
we've
shown
was
the
o2
3
out
for
evolving
tons
of
specification.
I
give
you
some
stats
just
to
emphasize
the
volume
of
work.
Not
much
was
happening
on
the
big
list
which
I'll
talk
about
quickly
terms
of
weapon
source
code
is
rewritten.
We
under
interrupt
the
staff
learned
lesson
so
I'll
talk
about
it
stuff
extensively,
some
of
them.
You
know
entertaining
some
insightful,
some
plain
stupidity.
You
know,
as
it
goes
with
software,
and
we
had
these
ones
on
twice
weekly
meetings
which
were
partially
recorded.
E
E
So
now
that's
that's
a
marketing
geek
right.
So
what?
Where
were
we
in
the
to
you
know
all
these
colors
look
is
wrong.
Look
there!
That's
the
colors!
Okay!
This
thing
is
all
wrong,
so
we
were
pretty
much
nowhere
with
the
security
envelope
and
we,
the
flooding
if
sm,
wasn't
well
specified
it
was
like
wave
wave
is,
is
we
didn't
have
anything
on
the
multiplying
spine,
which
is
a
difficult
problem,
but
a
real
problem?
And
that
is
pretty
much.
E
You
know
a
very
big
addition
on
the
zero
three
and
Pascal
will
talk
about
the
stuff
extensively.
We
choose
for
a
couple
of
design
options
that
is
actually
extremely
interesting
corner
of
the
world.
But
again
you
know
most
of
the
customs
run
the
single
plane
so
somewhat
FM
arrow.
What
else
was
missing?
E
I
tried
to
read
the
stuff
yeah,
so
we
talked
about
moving
from
the
how
concept
like
best
available
level
that
died
on
the
wine,
it's
actually
extremely
simple
to
in
our
loop
things
like
ztp
direct
implementation
experience,
so
that
didn't
happen
right
and
the
rest
was
in
the
fairly
good
shape.
So
where
did
we
end
up
zero?
Three.
E
More
green
greens
good
right,
so
the
security
envelope
is
still
being
chewed
and
I.
Think
I'll
call
for
a
meeting
with
possibly
Alvaro
and
AC,
because
it's
not
entirely
clear
what
should
be
the
security
model.
We
can
go
to
a
private
key,
no
private
public
key.
We
can
go
to
a
shared
secret
opinions.
Very
people
have
different
ideas
or
security
on
IP
fabric
means,
so
that's
still
under
work,
but
the
envelope
starts
to
gel
because
we
talked
about
things
like
you
know:
nonce
exchanges,
so
to
secure
the
stuff
properly
compared
to
today's
routing
protocols.
E
All
this
stuff
is
kind
of
green.
By
now,
so
we
have
to
fool
multi-plane
spine.
We
have
the
negative
transitive,
disaggregation,
positive
dislocation.
We
have
to
fools
learning
FSM
in
excruciating
detail.
One
thing
I'll
talk
about
which
is
I'll,
probably
not
explaining
a
lot
of
data
because
that's
ongoing
work.
E
We
found
something
interesting
on
the
line
if
s
M,
which
we
still
have
to
add
it's
nothing
particular
well-understood-
has
to
be
done.
Gtp
was
all
stabilized.
Mobility
be
ft
bandwidth,
balancing
optimal
flooding
reduction.
It
was
more
well
stabilized.
Actually,
there
are
better
ideas.
Now,
flooding
reduction
and
I
refused
to
take
them
on
Pascal
always
improves
on
himself
right,
but
we
have,
to,
at
certain
point
in
time,
gets
finished
all
right,
so
tons
of
stuff
green,
really
secure.
E
It
is
the
only
interesting
stuff
with
chewing
like
I
said,
I'll
be
point
AC,
probably
and
Alvaro
in
because
it
is
more
of
an
architectural
discussion.
What
makes
sense
rather
than
you
know
how
to
do
it.
How
to
do
is
the
thing
is
fairly
well
understood,
good,
so
rough
statistics,
because
people
may
have
perceived
it
slightly
quiet.
So
most
of
the
discussion
happened
between
like
10
people
on
like
a
core
mailing
list
where
things
percolate
in
a
very
tight
cycles
and
on
the
weekly
meetings.
E
So
I
just
look
at
how
many
emails
flew
between
the
last
night
Eve
and
these
IDF
summer
to
the
order
of
300
400
emails
and
pretty
much
weekly,
sometimes
bi-weekly,
when
it
got
really
hot
right
and
people
got
really
interesting
and
we
pushed
a
lot
of
envelopes.
Nothing
has
been
done
before
so.
A
lot
of
ideas
have
been
scrapped
and
reinvented,
and
so
on.
Right
we
had
about
255
commits,
is
an
open-source
base
and
that's
without
the
merges.
E
E
If
you
look
at
the
death,
so
lots
of
work
has
been
done
there
as
well,
mainly
flooding
procedures,
multiplying
fabric
and
tons
of
small-fry,
because
we
were
operating
the
code,
I
mean
no
the
weekly
basis,
but
every
two
three
weeks
we
know
we
were
Bank
the
code
against
each
other,
which
I
know
I,
won't
even
talk.
Bruno's
framework,
open
source
framing
came
up
with
it's
one
of
their,
like
Colinas
frameworks,
actually
to
teach
a
protocol
both
building.
How
do
you
analyze
the
protocol?
Behavior
how
you
debug
it?
So
it's
very
easy
to
pull
in.
E
You
know,
be
unnamed
vendors
code
and
Bruno's
code
and
run
the
stuff
just
up
the
hill.
With
a
couple
of
meanings,
we
get
like
a
fool,
interrupts.
There's
a
lot
of
you
know
all
the
logs
graphs
and
so
on,
so
lots
of
stuff
can
be
shaken
up.
I'll
talk
a
little
bit
harder.
A
little
farther
down
the
road
I,
give
you
an
update.
E
What
Bruno
did
because
he's
on
his
walkabout
for,
like
two
weeks
now
crawling
up
some
mountain
trying
to
get
eaten
by
a
bear
or
something
and
then
he'll
be
back
and
writing
code
day
at
night.
That's
how
it
goes.
On
the
other
hand,
the
interesting
thing
is
that
the
model
has
stabilized
largely.
So
we
only
change
like
seven
objects
on
the
model
to
give
you
like
I
feel
right,
so
the
encoding
isn't
changing
much.
E
We
exchange
pretty
much
all
the
information
that
we
need
to
get
the
stuff
done.
It
is
more
like
how
do
we
chew
the
information
properly
ideas
discussing,
scrap,
that's
the
most
fun
part
right,
we'd,
never
get
boring.
Do
we
have
a
boring
meeting?
I,
don't
think
so
right
yeah,
so
that
what
sucked
people
back
in
because
know
tons
of
interesting
stuff?
We
were
really
pushing
the
curve
in
many
respects,
so
we
remove
tons
of
stuff
right.
E
This
is
the
7000
lines
good
chunk
is
removed,
and
that
comes
from
like
customer
iteration
looking
what
would
be
really
useful,
which
mechanism
are
old-style,
not
even
needed
anymore,
so
we
checked
out
the
whole
PGP
that
doesn't
seem
to
be
any
demand
or
any
realistic
way
to
do.
Traffic
engineering
on
fabrics
don't
as
much
as
my
gut
feeling,
so
the
PGP
went
in
a
separate
draft.
We
preserve
it,
we
drag
it
along
may
prove
useful.
E
It's
nice
mechanism,
the
segment
routing
stuff,
could
checked
out
into
separate
draft,
because
this
multiple
model,
how
you
do
segment
routing
on
the
fabric.
If
you
choose
to
write
like
this
using
the
labels
and
what
you
can
do
with
the
stuff
and
so
on,
it
just
didn't
seem
to
be
that
base
line
of
the
spec.
The
information
is
carried
on
the
encoding
model,
but
the
procedures
and
how
you
you
know,
get
all
these
labels.
What
you
do
with
them
didn't
seem
like.
E
You
know
that
the
right
he
just
diluted,
the
spec,
the
key
value
store,
will
go
into
its
own
draft.
It's
actually
looks
at
the
key
value.
Store
will
probably
need
some
logon
registry,
because
there
will
be
some
well-known
keys.
There
will
be
some
vendor
keys,
this
kind
of
stuff,
so
that
should
probably
go
off
into
its
own
draft
liked.
Also,
the
note
well-known
uses
for
a
key
value
store
and
some
kind
of
a
registry.
So
people,
don't
you
know,
do
the
same
thing
in
different
forms.
E
We
got
it
some
interoperable,
well-known
stuff
and,
on
the
other
hand,
the
vendors
have
basically
a
nice
playing
field.
Some,
oh
you
I
kind
of
stuff,
alright.
So
what
is
that
we
do
except
removing
stuff
and
having
fun?
First,
of
course
we
have
to
change
the
language.
What
fun
is
a
spec,
where
the
glossary
didn't
change.
E
That's
where
the
teapot
stops
and
you
start
to
go
into
basically
building
your
spine
and
we
talked
about
the
radix,
because
that
is
a
very
central
concept
when
you
start
to
build
multi-plane
fabrics
because
multiplying
fabrics
happened
when
you
basically
run
out
of
radix,
so
you
have
to
break
up
your
fabric
into
multiple
crossbars
that
are
hooked
up
bar
cross
bars,
which
are
not
all
really
one
crossbar.
So
we
have
like
a
whole
pictorial.
E
One
of
the
really
fun
part
was
to
squeeze
a
picture
of
smaller
dimensional
fabrics
into
some
kind
of
a
two-dimensional
representation
which
actually
clarified
a
lot
of
thinking
right.
You
know
like
the
mortal
danger
and
ASCII
format
beautifully.
You
know,
concentrate
your
mind
now.
This
is
interesting,
so
Bruno
implemented
the
whole
flooding
and
he
implemented.
He
called
will
like
it.
E
That
way,
you
know
cleanroom
kind
of
stuff
and
most
of
the
stuff
he
could
derive
just
by
common
sense,
looking
maybe
than
III
as
I
expect,
once
or
twice
and
probably
just
being
good
at
it.
So
the
flooding
plantation
has
redone
clean,
clean
room,
wouldn't
talk
upfront
right
and
he
started
to
come
up
with
clarifying
questions
and
his
answers
were
mostly
right,
but
the
fact
that
he
had
to
ask
the
question
indicated
the
spec
wasn't
tight
enough
right,
so
the
scope
table
is
now
in
a
sense.
E
So
if
something
cool,
so
this
became
far
more
clear,
we
actually
spend
good
amount
of
time
of
eating
up
the
language
and
one
of
the
sloppy
language
cases
was
leading
to
a
flat
oscillation,
nothing
major,
but
something
that
never
stopped
the
flowering
converts
to
flying
okay
and
because
we
understood
that
the
top
of
fabric
is
actually
an
utterly
different
beast.
If
you
deal
with
multiplying
fabrics,
the
flooding
scope
also
broke
up
into
more
elements.
So
there
is
a
difference
when
you
have
a
tires.
E
The
request
is
the
tires
acts
in
the
scoping
right,
the
most
interesting
part,
probably
that
the
east-west
scoping
became
not
like.
Oh,
it's
just
like
any
North
link
when
you
low
in
the
fabric,
you
want
to
use
east-west
links
as
kind
of
the
last
resort
to
get
to
the
top
of
the
fabric
once
you're
at
the
top
of
the
fabric.
The
whole
game
actually
reverses.
E
So
you,
east-west
link
becomes
South
links.
That's
how
you
get
down
to
the
through
the
fabric.
If
you
can
go
directly
down,
you
just
push
it
east-west
to
your
neighbor,
who
can
actually
get
you
down
and
that
reflects
in
the
flooding.
Scopes
and
I
will
not
chew
that
stuff
to
pieces.
But
it's
more
detailed,
tighter
a
couple
of
things
became
much
clearer
after
bruno
implemented
the
stuff
and
he
interrupted
right.
E
I
mean
I,
knew
what
I
was
doing
right,
no
hands
always
root
password
right
and
blue,
though
an
idea
of
that
he
found
these
one
or
two
things
were.
We
were
just
you
know
Lucy,
so
that
was
the
second
one.
I
thought:
oh
I'm,
going
like
a
steam
train.
Okay.
So
what
did
we
do?
Third,
we
brought
all
the
flooding
rules
and
excruciating
details
and
shoot
them
through
in
the
appendix
p3
about
some
people
complain.
It's
way
too
complex
who
didn't
have
any
beef
with
implementation?
Actually,
you
ask
them
what
they
even
read:
the
stuff.
E
Okay,
I
think
this
is
the
clearest
write
down
of
flooding.
You
can
do
but
of
course,
and
John
Moyers
already
high
bar
to
pass,
and
he
basically
breaks
down
to
a
couple
of
cue
manipulation.
You
have
four
cues,
which
is
ties,
transmitters,
retransmit,
requests
a
neck
and
then
what
fell
out
when
we
started
to
write
down
these
rules
and
bro
start
to
look
at
my
implementation.
So
the
tie
generation
CSP
our
translate.
Real-Time
OSPF
is
none.
E
You
basically
priority
dump
out
your
database
break
our
description
of
your
databases,
header
and
you
dump
it
out,
and
when
we
found
out
that
there
is
something
called
the
mean
and
the
max
at
the
end,
you
indicate
I'm,
starving
and
I'm
finished
and
it's
the
last
one
and
it
wasn't
tight
enough.
It
was
inspected
out,
so
he
was
sending
a
different
mean.
I
was
sending
a
different
mean
it
just
worked,
but
we
both
got
confused.
E
He
actually
managed
my
application
to
go
like
what
the
hell
is,
this
guy
standing,
because
he
was
sending
something
that
looked
like
a
reasonable
mean,
tie
and
I
expected
something
which
is
Rachel.
Really
not
that
valid
mean
tie
they
included
lifetime,
wasn't
specified
tight
enough.
That
is
zero,
so
he
went
to
the
I
as
I
expect
and
you
need
the
right
thing
and
it
worked.
We
were
lucky,
so
a
couple
of
things
would
fix
by
looking
slavishly
at
the
is
I
expect,
and
then
it
was
very
interesting.
E
Bruno
was
kind
like
why
the
hell
is
flooding
done
that
way.
So
why
would
you
sort
the
headers
and
he
had
a
bunch
of
ideas
which
would
through
throw
them
all
away?
I
his
eyes
is
very
hard
to
improve
on
in
this
respect,
the
processing.
What
did
we
find?
He
found
a
major
bug
which
was
the
17
Clause
of
the
fourth
sub
Clause
of
the
like
17
thing
in
eyes:
eyes
I
for
God
they
bolted
on.
E
So
when
there
was
a
loop
in
the
you
know,
was
a
hole
in
the
tight
I
didn't
know,
check
whether
there's
a
hole
and
I
have
actually
to
send
those
things
request
them.
So
I
did
so.
The
flooding
is
so
positive,
Leary
bust,
especially
I
as
I
as
this
whole
thing
was
working
which
would
Lodge
it
by
accident
belts
in
suspend
her.
E
So
we
found
that
excellent
outcome
and
it
was
a
super
delicate
buck
where
I
have
wrote
into
spec
larger
equal
and
it
was
really
a
large
which
was
suck
showing
in
a
corner
case
tires.
So
on
a
generic
basis.
You
just
take
all
the
requests
in
the
X
that
you
have
on
your
Q's
and
you
just
push
them
out
didn't
found
any
issues
on
the
tie.
E
All
the
time
processing
we
didn't
find
anything
either
all
right
and
now
what
did
we
do?
Then?
We
actually
went
into
the
whole.
Multiplying
Farage
is
a
fascinating
topic
and
Pascal
will
be
chewing
the
stuff
to
pieces
because
he
came
up
with
a
lot
of
his
methodology,
pictures
and
so
on.
But
of
course
I
have
to
give
you
like
a
retro
picture
right
and
the
red
rings
will
become
clear
and
the
whole
you
know.
E
E
E
And
we're
working
on
the
secure
envelope,
lots
of
interesting
stuff
there,
okay,
so
there
link
state
protocols,
kind
of
like
bolted
on
the
security
and
they
found
all
kind
of
interesting
problems
like
and
and
they
kind
of
walked
around
couple
of
problems.
So
you
do
all
kind
of
checksumming,
but
you
live
for
certain
feels
like
lifetime,
which
leaves
you
open
to
attacks
which
led
to
another
draft.
E
We
were
also
looking
for
an
envelope,
so
once
you
start
to
work
with
the
model,
you
find
out
that
you
don't
really
want
to
deserialize
a
model.
You
want
to
treat
it
as
a
blob,
because
you
can't
just
go
into
a
field
you
just
muck
around
with
it.
You
don't
know
what
the
offset
is.
The
model
will
not
tell
you
and
you
go
deserialize.
It's
your
lives.
You
may
end
up
with
a
semantically
same
packet,
but
with
a
completely
different
binary
representation.
E
A
lot
of
cases
like
that
right.
So
if
you
want
to
authenticate
this
stuff
as
a
binary
blob,
you
better,
never
deserialize
your
lies
again,
because
we
never
what
you
end
up
with
so
the
stuff
you
can
muck
around
with
it.
You
really
want
to
pull
out
a
packet
and
then
there
is
a
bunch
of
things
which
we
don't
do
particularly
well
today,
which
is
something
like
nonces.
So
when
we
establish
an
adjacency,
we
really
build
security
association
which
gives
us
integrity,
but
it
does
not
give
us
privacy.
So
that's
kind
of
particular
thing
right.
E
It
leads
to
the
discussion.
Should
you
even
go
into
encryption
on
link
on
routing
Association,
but
even
without
that,
once
you
get
that
you
can
protect
against
flooding,
replay
attacks
and
all
kind
of
stuff,
so
SPF
has
some
of
the
stuff,
but
we
try
to
make
it
somewhat
cleaner
and
better
less
you're
welcome
fascinating
discussion.
Greenfield
we
can
do
what
we
want.
We
don't
have
to
cut
on
the
bytes.
Okay,
all
right
and
and
really
the
big
discussion.
E
Should
we
go
to
a
private
public
key
model,
or
should
we
keep
at
a
shared
key
model
or
shared
keeper
interface
model
that
actually
has
a
lot
of
impact.
How
you
build
this
stuff
and
I-
don't
have
a
clear
answer:
I
see,
people
being
all
the
way
from
paranoia
to
like
who
cares?
Okay,
so
we
have
to
pick
some
point
which
you
think
is
as
much
feasibility
design
simplicity
driven
as
you
know,
how
far
in
the
paranoia
scale
do
we
want
to
get
all
right?
E
What
still
out
is
last
one
of
this
one
which
I
almost
didn't
over
run
a
lot
of
explanation
section,
so
the
desire
came
to
really
explain
how
much
of
the
specter
you
have
to
implement
on
the
leaf
on
the
spine
or
at
the
very
top
of
the
fabric.
So
we
have
to
write
that
right
uses
to
tell
an
implementer.
You
know
this
is
the
slice,
and
you
know
the
complexity
raising
with
the
higher
you
get
up
in
the
fabric,
which
is
exactly
the
desired.
E
You
know
outcome,
then
the
east/west
sub
super
spine,
like
I,
already
Alert,
where
the
direction
of
the
east-west
link
actually
changes
to
run
the
fabric,
which
is
mildly
confusing.
Originally,
that
needs
to
be
more
explained.
We
have
to
spec,
but
the
spec
is
just
a
spec.
Tells
you
what
to
do,
but
not
necessarily
why
to
do
that
security
envelope,
soft
ocher
generation.
E
All
my
stuff
was
interested
in
that
join
the
weekly
grab-ass,
fascinating
discussions
greenfield
right,
I'd,
you
don't
have
to
take
you
mom
you're,
actually
involved
in
Riften,
okay,
I'll
talk
about
the
interesting
oscillation
we
found
and
that's
a
discussion.
Yeah
I'll
explain
the
oscillation
we
found
on
the
lies.
Actually,
the
Bruno
by
sheer
miss
configuration
file
which
is
exactly
intended,
and
then
there
is
a
delicate
case
on
the
tight
reception
join
the
weekly
that
is
caused
by
very
sophisticated
sequence
of
reboots
of
nodes
in
the
fabric.
E
People
have
to
reboot
in
just
the
right
sequence,
and
then
you
may
end
up
stuck
with
the
north
in
the
north,
with
the
tie,
which
can
be
relative.
We
have
a
good
proposal
how
to
clean
it
up,
but
that's
basically
still
needs
to
be
SPECT
out
which
will
give
actually
in
the
flooding
rules.
The
only
case
where
you
have
to
understand.
Where
are
you
going
north
or
south
beside
the
Scopes
negative
disaggregation?
Moscow
will
talk
about
it.
We
need
a
section
on
that
and
that's
just
information
rules.
E
G
H
I'm
very
happy
that
you
guys
are
making
progress.
I'm
really
happy
that
you're
picking
up
interest
from
other
people
that
were
not
originally
working
on
this
with
you
I'm
also
happy
that
you
guys
are
recording
the
meetings
that
you
said
several
times
just
in
last.
You
know
other
people
could
join
them.
Yeah.
H
H
G
H
E
Bullets
yeah
easy
because
mostly
it's
a
quite
a
linear
progress.
Okay,
so
we
were
in
like
jumping.
Maybe
there
will
be
two
topics
coming
up
max
right
and
yeah.
That's
actually
easily
done
so
people
can
track
progress
without
actually
even
listening.
Yeah,
that's
true
yeah
good
suggestion.
All
right.
E
Right
so
no
problem,
almost
a
movie
all
right,
so
I
think
this
is
the
place
where
he
starts,
and
this
is
the
mountain
he
aspires
to
fall
down
from
that's
Bruno's,
update
right,
so
I
think
it's
one
of
these
things
he
goes
about.
You
know,
look
for
trouble,
so
Bruno
isn't
here.
He
gave
me
this
mount
at
this
stuff
and
this
is
pretty
much
the
output
of
the
open
source
work,
which
he
frames
in
his
own
way,
alright
good.
E
So
the
usual
marketing
blurb,
you
know
the
stuff
is
so
originally
we
wanted
just
to
test
Alya
for
Sam
yeah.
It's
you
know
things
too
take
their
own
dynamics.
It
looks
like
I
hope,
we'll
have
basically
a
server
implementation
out
there
for
free
good.
So
basically,
Bruno
is
really
kicking.
My
rear
behind
to
get
is
now
all
this
back
to
his
standards,
which
are
excruciating
as
they
should,
and
it
should
be
the
reference
and
you
release
and
actually
I'm
using
it
even
farther
and
telling
people.
If
you
want
to
learn
how
the
link
state
works.
E
This
is
probably
the
place.
Did
you
want
to
go
and
I
show
you
some
of
the
output,
which
otherwise
are
very
very
hard
to
get
except
no
proprietor
of
development
tools
with
within
corporations
that
you
know
are
not
too
interest
in
exposing
any
of
the
stuff
for
obvious
reason,
all
right
so
he's
building
a
framework
which
is
very
heavy
on
testing
the
but
ability
visualization.
This
kind
of
stuff
performance
will
be
lousy,
but
it
literally
is
irrelevant.
E
When
you
look
down
that
you
know
at
the
requirements
on
the
host,
it's
just
different
for
all
practical
purposes:
Python
he's
you
know
very
which
call
it
he
likes
his
documentation,
as
you
should
know
this
shape
with
any
vendor,
maybe
yet
all
right.
So
here's
how
you
get
into
that
stuff.
Here's
a
nice
no
page-
and
you
can
install
this
now
and
start
up.
So
you
basically
pull
up
some
Python
thing
at
a
virtual
environment.
E
You
know
you
should
PI's
and
stuff
three
four
libraries
which
you
just
pulled
automatically
and
you
in
the
business
all
right
worst
stuff.
So
a
Jason
sees
seventy
five
years,
something
like
that.
Eighty
I
don't
even
know
what
he's
missing
so
there.
The
hello
has
been
banked
on
for
quite
a
long
time.
The
CTP
has
been
pretty
much
completely
banked
out.
He
he
was
a
disbeliever
as
he
should
so.
He
built
all
kind
of
weird
nice,
wiring
cases
and
so
on
all
seems
to
come
up
as
intended.
E
No
routing
computation
has
been
done.
Yet
if
it's
a
host
versus
trivial,
it's
basically
go
on
level
up
and
grab
your
tours
pretty
much.
Unless
you
do
bandwidth
adjusted,
which
is
a
little
bit
more
work
and
management
interface
I,
don't
know
what
he
means
by
about
half
done
by
show
you
what
he
has
development
tool
chain,
not
even
clear
where
there
is
no
automated
testing
continuous
integration,
all
the
stuff
in
place
since
a
long
time,
all
right
so
adjacencies,
oh
I,
see
why
he
is
unhappy.
So
we
have
all
the
live,
packets,
fine
state
machine.
E
We
interrupt
the
staff
through
a
whole
slew
of
like
you
know,
you
send
me
weird
stuff.
We
went
through
the
whole
thing
really
80
to
90%
of
the
heavy
lifting
like
nice.
Formatted
package
is
not
there
right,
because
then
you
can
even
deter
the
model.
So
that
says
all
the
serialization
deserialization
there's
no
code,
it's
so
generated.
So
at
the
end
you
just
look
at
did
I
get
information
that
makes
sense
nothing
else.
E
What
he
doesn't
have
is
the
ipv6
adjacencies.
As
everybody
knows,
ipv6
is
a
hood
to
work
with,
especially
when
you
get
into
multicast,
so
can
interfaces
as
I
warned
him,
so
he
didn't
get
to
that.
Yet
options
are
around,
but
it
works
utterly
different
as
no
than
ipv4,
multicast
and
sockets.
Just
because
we
will
talk
about
the
Malta
neighbors
that
he
doesn't
have
any
interactions
with
BFD
which,
on
the
server
side,
is
an
interesting
kind
of
sidetrack
discussion.
E
I
should
don't
think
the
server
should
be
doing
any
BFT
should
be
just
used
in
a
mirror
mode.
When
you
bounce
it
off
of
PFD
join
some
deep
call,
we'll
talk
about
the
stuff
at
security,
because
that's
what
we
chew
on
right,
a
security
envelope,
the
knowns
exchange.
What
we
shall
do
we
do
shared
key
public
probe
key,
will
all
shake
out
zero
touch,
pretty
much
everything
done.
We
interrupt
weird
stuff,
all
done
flooding,
so
the
flooding
is
about
I'd,
say
halfway
there.
E
No
flooding
is
actually
like
80%
there,
so
we
do
all
this
packets
different
times
and
no
notes.
Prefixes
databases,
all
the
cues
procedures,
whatever
we
interrupt
the
staff
finding
all
this
interesting
stuff,
I'll
be
talking
about
some
of
that
code.
Of
course
sucks
right.
So
it's
always
deserters
and
so
on
decides
we're
doing
the
stuff
server
implementation.
Pretty
much
of
no
relevance.
E
He
doesn't
do
positive
distillation
ties,
which
means
nothing,
because
those
are
basically
normal
prefixes
in
a
different
tie,
the
negatives
we
don't
have
SPECT
out
yet
well,
we
have
to
inspect
out,
but
we
know
we
don't
have
to
implementation
either,
because
it's
all
on
discussion,
that's
our
talk.
He
doesn't
do
key
values.
We
have
external
prefixes,
doesn't
do
that
stuff,
PGP
overload
and
the
clock
comparisons,
cool
thing:
it's
not
much
work,
but
it's
pretty
cool,
because
if
you
have
a
clock
on
the
fabric,
it
will
give
you
perfect
protection
against
people.
E
Actually
mucking
around
compromising
source
is
reassuring,
stuff
or
stuff
that
is
still
in
the
fabric,
all
right
route.
Calculation.
You
do
anything
yet
for
the
server
we'll
need
to
do
the
northbound
and
yes
for
for
for
the
leaf.
We
don't
need
any
of
that
stuff.
So
we
see
where
to
know
what
the
trebles
take
us.
We
start
probably
from
the
host
from
the
server,
which
is
kind
of
a
natural
growth
up
the
fabric
management,
CLI,
boring
stuff.
Sorry,
so
actually
we
aligned
on
the
yellow
format
how
we
describe
topologies
to
run
tests
very
quickly.
G
E
Take
my
stuff
I
can
take
his
stuff
and
we
can
run
the
same
file
and
half
of
the
stuff
is
his.
How
about
the
stuff
is
mine
or
whatever
we
mix
and
match
implementations,
and
then
we
can
look
at
the
stuff
three
kind
of
a
standardized
interface.
Do
we
get
what
we
need
configuration
common
history
stuff
like
that?
He
wants
an
SSH
client.
We
don't
have
the
yank
data
models
or
kind
of
early
development
tool
chain
yeah.
Ninety
percent
code
coverage
he's
a
drama
and
rule
of
things
about
writing
a
Wireshark
dissector
for
rift.
E
It's
kind
of
a
natural
I
liked
it
and
actually
Wireshark
supports
now
a
very
sophisticated
plug-in
interface.
So
you
don't
actually
even
have
to
like
heck
Wireshark.
You
can
write
plugins.
This
is
that
something
that's
on
his
mind
because
right
now
you
know
in
flight.
We
don't
decode
back.
So
just
look
on
both
ends,
so
we
get
what
we
need.
E
Everything
is
fully
automated
and
then
some
trayvis
Pilon
think
so
very
nice,
finite
state
machine
framework
and
visualization
tools
we'll
see
some
of
that.
Okay.
So
what
did
we
find?
I
know
I'm.
You
know
I
still
have
tons
of
time.
We
found
the
Malta
neighbor
oscillation.
So
when
you
end
up,
we
say
it's
a
point-to-point
link
right
and
it's
about
it.
Now,
if
you
miss
configuring
bunch,
three
or
four
people,
funky
stuff
happens,
and
we
fell
like
a
an
Omega.
E
Do
turn
out
that,
because
of
all
these
cool
optimizations
to
bring
up
adjacency
real
fast
bad
stuff
was
happening,
so
we
had
to
back
off
so
that
is
caused
by
certain
what
he
calls
triggered
loop,
so
the
packet
formats
can
actually
the
packet
sending
are
basically
building
up
the
oscillation.
You
don't
see
the
graphs
and
we
have
to
introduce
a
new
multi
neighbor
stage,
which
is
not
much
of
work
on
fatik
was
starting
on
it.
We
basically
have
to
cool
down
when
we
see
multiple
nodes.
E
You
know
on
the
same
link
more
than
two
and
give
some
alarm
or
something
or
we
extend
the
protocol
to
eaters.
But
since
there's
no
demand,
it's
kind
of
you
know
futile
at
the
moment:
flooding
constellation
okay,
so
we
found
flooding
constellation,
which
is
nothing
major
but
just
know
and
knowing
and
really
poke
the
finger
into
the
loose
language
it
wasn't.
Even
the
body
was
really
loose
language
and
okay
I
show
you
the
staff,
so
we
had
to
tighten
up
the
language
of
the
flooding
scope.
It
wasn't
clear
who
was
filtering?
E
Can
you
send
what
you're
not
supposed
the
other
guy
should
suppress
it,
or
you
always
must
filter
and
I
was
sloppy
as
I
am
I
was
sending
will
be
too
much
like
can't
hurt
right
and
he
wasn't
filtering
and
the
next
thing
we
had
an
oscillation
all
right,
minor
issues,
all
right
so
about
the
neighbor
scenario
is
simple:
we're
supposed
to
be
a
point-to-point
as
all
day.
You
know
respecting
fabrics,
artist,
a
model
or
maybe
some
you
know,
discussions
I'm
having
with
people,
but
then
the
whole
bandwidth
management
becomes
impossible.
E
Well,
doesn't
mean
I,
put
you
on
an
ether
and
there's
20
people
on
it.
Who
gets
what
right
so
could
happen
by
accident.
So
what
happens
to
the
protocol?
Well,
that's
what
happens
to
the
protocol.
He
looked
at
the
Wireshark
and
Bank,
which
is
filled
the
wire
for
every
those
people
who
didn't
get
it
not
a
desirable
outcome.
Okay,
it
was
fun,
so
Bruno
accused
me
of
a
back
I
accused
him
of
a
back
fantastic.
We
found
that
actually,
this
fully
intended
behavior.
E
Okay
right
all
right-
and
here
you
see
one
of
his
first
tools,
so
he's
generating
all
this
beautiful
SVG
that
you
can
visualize
in
time
and
you
can
flip
on
things
and
just
focus
on
a
certain
flow
and
so
on,
just
because
he
got
desperate,
which
a
thing
is
a
great.
You
know
motivation.
So
that's
what
it
look
like
right.
People
started
to
send
what
those
are
the
three
people
and
they
start
to
send
to
each
other
hellos
and
this
thing
starts
to
trigger
and
they
get
a
beautiful
bifurcation.
E
If
you
know
anything
about
chaos,
theory,
you
could
teach
it
right
here.
So
explanation
was
like
that,
if
you're
in
neighbor
and
you
are
into
a
which
means
you
heard
from
someone-
don't
you
tell
them-
you
were
hit
by
two
people
and
Y
and
the
Z
is
an
axe
who
told
you
you
know
okay
I'm
into
way
now
the
two-way
was
really
eager,
like
I
want
a
three
way
from
you,
which
was
immediately
pushing
out
the
hello
yeah
I
know,
but
it
helps
so.
E
It
definitely
created
two
packets
on
a
multicast
to
those
two
people
who
all
received
them
were
very
eager
to
become
immediately
three-way
with
you.
So
you
know,
multiplying
by
two
is
a
pretty
powerful
tool,
all
the
other
stuff,
not
that
relevant.
So
we
ended
up
with
exponential
growth
and
number
of
live
messages
and
the
FSM
was
oscillating
a
specific
code
right
just
everybody
was
doubling
the
Packers.
Everybody
was
very
eager
to
send
out
two
packets
to
get
two
to
three
way
to
these
people.
Well,
that's
how
things
goes
so
we
have.
E
We
need
to
basically
new
multi
neighbor
state,
which
means
that,
if
I'm
in
the
two-way
and
I'm
getting
hit
by
some,
so
if
I'm,
an
axe
and
I
get
something
from
Y
and
I
get
into
two-way
and
I
push
thing
out
towards
him
like
hey
here's,
the
three-way
I
reflect
you,
but
then
I
get
hit
by
something
from
ze,
then
I
have
to
say
I'm
in
a
multi
neighbor
state
and
we
basically
cool
down
for
whatever
it
is.
Second
all
right
and
probably
have
to
raise
an
alarm
saying
like
misconfigured.
E
Don't
do
that
next
one
was
a
flooding,
oscillation,
mildly,
more
interesting,
nothing
amplifying.
So
this
was
an
amplifying
oscillation
caused
by
the
fact
that
we
were
very
eager
by
arrival
of
packets
to
trigger
events
of
the
FSM
instead
of
just
waiting
for
time
or
like
I
heard
from
you
in
a
second
you're
here.
For
me,
we
just
wanted
to
be
really
fast.
So
this
is
far
less
dramatic.
E
We
were
seeing
a
tide
which
caused
a
tie
so
csmp
sending
an
LSP
and
PSP.
For
you
know,
people
defective
in
the
language
talking
like
is
is
East,
Wish,
20
years
old,
and
that
was
going
on
on
every
time.
We
saw
this
tie
and
a
neck
and
then
against
a
time,
and
the
guy
asked
you
know
was
sending
the
same
time
so
like
nothing,
dramatic
but
a
glitch.
E
So
we
went
looking
and
he
was
again
caused
by
Luke
my
loose
language,
I
didn't
say
who
is
supposed
to
filter
and
do
you
send
only
this
much
information
or
tight,
or
can
you
send
more
and
if
I
send
more?
Is
the
receiver
supposed
to
filter?
Or
is
she
supposed
to
send
right?
I
was
just
like
sloppy
sloppy
I
as
I
as
flooding
is
so
stable.
You
really
have
to
do
stupid
stuff.
So
in
more
detail,
he
was
going
like
that
the
tide
was
pushing
something.
E
Was
that
Oh,
so
the
tide
was
pushing
something
which
is
really
out
of
this
flooding
scope,
and
this
guy
would
be
pushing
the
tie
because
he's
so
like.
Oh,
you
know,
you
request
that
stuff
and
this
guy
will
be
acknowledging
that,
but
technology
would
have
been
dropped
as
far
as
I.
Remember
it
because
would
have
filtered
so
next
time.
This
thing,
so
this
tie
was
actually
not
accepted
because
we
were
like
eh.
You
are
giving
me
something
out
the
flooding
scope.
He
described
something
out
of
flooding,
scope.
E
Guy
was
pushing
this
thing
out
the
flooding
scope.
He
got
an
AK,
but
this
thing
never
got
accepted
in
database
because
without
the
flooding
scope
yeah
so
like
we
have
to
specify
like
who
is
filtering,
do
you
double
filter?
Are
you
allowed
to
send
more?
Those
are
the
weekly
calls
which
are
fascinating
because
a
lot
of
the
stuff
burns
down
to
like
what
is
the
data
structure?
What's
efficient
right,
I
mean
when
I
build
those
ties
tights
that
I
can
I
catch
them,
do
a
filter
dam?
What's
the
cost
of
filtering
them?
E
Who
can
do
cheaper
filtering?
This
is
how
you
build
those
beasts
right,
two
minutes,
if
perfect
and
the
flooding
oscillation
I
absolutely
do
not
remember
what
it
I
just
looked
at
it.
What
did
he
cook
up
there?
Okay,
so
we
sent
it
tight.
We
announced
a
tie.
Header
north,
you
know
one
is
probably
no
tie
which
no
note
tie
and
some
no
sequence
numbers-
and
this
note
was
sending
tire
or
because
it
didn't
have
this
tied,
oh
yeah,
that
was
the
other
installation.
E
So
now
we
are
selling
this
stuff
and
these
guy
requests
it,
but
we
are
filtering
on
the
output
like
you
are
not
supposed
to
ask
me
for
this.
Tie,
yeah
I,
send
you
this
information,
but
you're
not
supposed
to
ask
me
so
I
won't
send
you
that,
so
he
keeps
on
asking
it's
a
who's.
Filtering
am
I
allowed
to
ask
then
or
you
shouldn't
have
sent
it
in
the
first
place
because
then
I
keep
on
asking
or
yeah
I
can
send
you,
but
you're
not
supposed
to
ask.
E
So
these
are
like
no
tied
language
and
descriptions
all
right.
So
the
flooding
scope,
rules
of
lessons
learned,
that's
the
interesting
stuff
and
I
still
have
15
minutes.
So
they're
flooding,
scope
rules
are
obvious
sensitive.
We
all
know
flooding
East,
delicate
right.
That's
what
people
have
BGP
just
less
potential
to
hang
yourself
up
when
flooding
works.
It's
like
the
most
beautiful
distribution.
You
know
lose
synchronization
pass
in
the
world,
but
it
is
delicate
and
we
have
flooding
scopes.
As
do
the
traditional
links
that
protocols
grow.
Right.
I
mean
we're
talking
here.
E
All
northbound
flooding,
saldaña
learning
is
tribulations.
One
Habra
would
defer
out
pretty
much
so
tiny.
Changing
rules
can
lead
to
oscillations,
nothing
amplifying
it
looks,
but
you
know
you
lost
light
persistently,
not
elegant,
but
the
positive
stability
is
intoxicating.
We
were
playing
it
loose
and
we
never
melted
it.
E
So
the
rules
for
the
flood.
What
you
plan
out,
what
you
show
and
what
you
ask.
They
must
be
consistent:
okay,
which
is
not
trivial
because
there
are,
they
are
not
orthogonal.
There
are
basically
joined
at
the
heat,
so
Bruno
had
some
ideas:
how
to
maybe
write
and
represent
the
whole
flooding
scopes
differently.
First
iteration
didn't
leak
too
much,
but
this
is
not
a
simple
problem:
some
kind
of
representation
where
this
at
hippies
jointness
will
become
welded,
maybe
some
model
or
otherwise
you
know
very
hard
to
fracture
in
a
sense
yeah.
E
So
he
was
thinking
about
maybe
putting
the
flooding
scope
in
the
tie,
header,
which
I
think
is
a
solution
from
hell,
because
if
I'm
getting
his
tie
and
the
flooding
skull,
my
supposed
to
or
maybe
choke
me
flooding
scope.
This
thing
is
never
supposed
to
show
up
on.
You
have
like
the
presence
of
these
things,
bellies
its
scope.
Well,
what
it
tells
you
about
it.
Scope
contradicts
where
it
is
so.
Would
you
do
that
more
details?
There's
a
some
press,
oh
and
some
graphs,
and
so
on.
E
You
know,
I
should
document
it
cool,
so
we're
and,
like
I,
said
we're
on
in
a
the
vendor
staff
one
process
Python
another
stop
process,
we'll
pull
in
one
topology
file,
which
describes
the
topology
and
by
very
simple
mechanism
you
decide
who
gets
what
all
fully
automated
you
run.
This
thing
pull
out
state
from
both
through
different
means
and
see
whether
you
got
what
you
needed.
E
Flooding
is
not
automated
yet
and
the
rest
is
all
fully
automated.
Alright
yeah,
it
helped
lots
of
editorial
small
things.
Protocol
do
interoperability
tested
an
early
stage
which
means
implement
in
the
early
stage.
We
knew
that
since
25
years,
we're
forgetting
that
the
duration
yeah,
but,
of
course
it
takes
time
to
build
as
eg
helps.
You
know
this,
which
is
a
great
thing.
It
saves.
You
tons
of
you
know
like
building
the
visualization
out
of
the
format.
E
Yeah
the
code
calls
are
cool
one
hour
a
week.
You
know
you
learn
by
osmosis
and
you
can
you
know
everyone's
welcome.
Oh
yeah,
so
more
people
to
hack
the
stuff
is
fine.
Bruno
has
no
levee
walls.
You
won't
even
be
able
to
ask
him
for
a
pool
up
unless
you
meet
the
stringent.
You
know
coverage
and
whatever
not
testing
requirements
which
is
cool.
You
know
allows
us
to
keep
people
out
who
will
break
stuff,
I.
Think.
D
C
C
So
we
have
a
DK
text.
That
explains
you
know
when
you
get
this
breakage,
do
you
need
to
desegregate
when
you've
got
this
breakage?
Do
you
need
to
desegregate
it's
more
like
also,
you
understand
how
we
get
the
fabric
and
you
understand
what
are
the
consequences
of
losing
this
thing
versus
losing
that
thing
and
the
disaggregation
is
read
the
process
that
you
have
to
do
from
the
top
to
actually
say
I.
C
C
So
the
positive
disaggregation
is
flooded
by
all
the
guys
who
can
still
reach
the
leaf
so
as
to
install
a
rod
via
them
as
a
more
specific
route
to
the
loss,
prefix,
the
primary
that
it's
a
lot
of
flooding
because
it
comes
from
all
those
super
spy
mode
which
can
steal
which
reach
the
guy
and
the
idea
of
flooding
that
recursively
down
the
graph
is
kind
of
ooh
big.
So
so
we
looked
at
it,
we
looked
at
the
color
cases.
C
We
went
through
a
lot
of
exercises
with
Toni
and
a
lot
of
possible
weird
topologies,
and
we
understand
a
let's
do
what
the
draft
always
said,
which
is
it's
not
transitive.
It
follows
the
source,
flooding
scope,
which
is
one
hub,
and
this
also
has
consequences.
So
we
went
through
those
discussions
and
then
we
said
hey.
If
we
want
to
go
transitively,
there
are
rare
cases
like
Tony
said,
which
is
probably
not
the
case
when
you
don't
partition
a
super
spine.
C
But
if
you
do
partition
the
super
spine,
like
I,
did
here
top
of
fabric,
then
you
will
end
up
in
cases
where
you
have
to
transitively
go
down
all
the
way
down
the
leaves,
so
the
leaves
here
are
represented.
That
are
these
guys
and
you
see
I
colored,
the
two
topologies,
the
two
planes
and
you
realize
that
the
leaf
is
the
guy.
C
So
this
is
this.
This
partition's
super
spine
is
the
case
where,
what's
described
in
in
the
document
as
positive
disaggregation
cannot
work
because
the
set
of
spine
in
different
planes
don't
see
one
another,
and
so
you
have
to
go
down
all
the
way
to
the
leaves
to
tell
them
a
don't
use
plan
right
or
don't
use
plain
blue,
it's
pretty
much
done,
and
so
we
figured
that
the
positive
was
just
too
expensive
to
do
recursively
and
flood
all
the
way
down.
C
You
will
need
to
this
aggregator.
You
want
it
to
the
aggregate
and
I
think
this
text
is
useful
as
it's
time
to
understand
the
consequences
of
things.
Will
that
text
stay
in
the
final
RFC
I,
don't
know
your
decide,
it
doesn't
give
you
anything
that
you
have
to
do
it
helps
you
understand
why
you
have
to
do
it.
So
basically
what
this
is
and
they
put
the
text
in
line
pretty
much
wrote
in
this
slide.
C
What
I
wrote
it
in
the
spec
so
that
you
can
see
what
deals
about
and
basically
what
that
tells
you
is
in
any
of
those
breakages.
You
know
this
link
down
below
the
leaf
this
leak.
There
is
this
leaf
node
or
the
surface
by
node
when
they
are
broken
down
broken
for
everybody,
so
they
don't
create
a
special
case
in
one
plane.
So
there
is
nothing
special
to
disaggregate,
a
normal
writing
that
the
stuff
for
you,
for
instance,
if
this
guy
sure
dies,
nobody
will
use
him.
So
it's
not
a
disaggregation
problem
same
thing.
C
If
this
leaf
goes
away,
it
goes
away
in
every
plane.
So
there
is
nothing
special
to
doing
one
plane
like
a
disaggregation
right.
So
all
those
classes
Prime's
don't
need
this
aggregation
the
text
once
a
issue
in
this
case
you
don't
disagree,
gate
right!
You
can't
you
can't
miss
something
like
that,
just
to
explain
you
in
which
the
constants
it
depends
on
which
it
does
not.
C
Okay,
so
the
I
would
say
the
normal
case,
which
is
always
the
case
when
you
don't
partition.
The
circle
spine
is
that
a
breakage
which
happens
in
your
plane
can
be
detected,
the
South
and
reflection
within
the
same
plane
by
the
guys,
who
had
just
above
the
breakage
right.
So
you,
the
breakage,
happens
here.
The
guys
above
the
breakage
see
one
another
inside
the
plane
they
can.
They
can
do
the
reflection
or
as
four
figures
that
is
two
can
not
reach
the
red
plane.
C
C
Okay,
so
and
and
that
again
it's
the
normal
case
when
you'd
only
have
got
red
guys,
whichever
hallucis
is
linked
here
completely
solved
within
the
plane.
So
that's
kind
of
an
easy
case.
Well,
the
spec,
without
the
negative
disaggregation,
already
solve
the
situation
now.
Sadly,
there
are
other
cases
and
one
of
the
other
cases
if
the
break
agent
in
this
link,
as
opposed
to
this
link
and
this
link,
the
solve
and
reflection,
will
happen
between
those
two
guys.
So
the
situation
is
locally
solved
by
the
non-transitive
positivity
segregation
within
this
particular
pod.
C
You
could
observe
it
on
on
the
parent
on
both
sides
right,
but
the
this
it
cannot
be
observed
by
the
sound
on
reflection
of
just
the
share
is
here
and
well.
It
could,
but
then
those
guys
would
not
be
able
to
go
all
the
way
down
the
leaves
to
tell
them
hey.
You
need
to
pick
the
other
plane,
because
the
positive
this
aggregation
is
not
transitive
right.
So
there's
no
way
with
the
positive
disaggregation
to
discover
something
here
and
then
go
way
down
to
inform
the
Leafs
a
don't
use
flame
right
friend
right.
C
When
we
do
they
get
here,
the
parents
here
above
the
breakage,
so
that
would
be
this
guy-
could
do
a
negative
to
all
the
red
plane
and
say:
don't
use
me
the
primaries
if
it
never
so
what's
behind
l2,
it
doesn't
know
which
prefix
to
disaggregate
okay.
So
we
can
know
there
is
a
problem,
but
it
doesn't.
It
cannot
know
for
sure
what
the
promise
cannot
the
us.
C
So
if
the
fabric
came
up
and
all
the
prefixes
were
advertised,
etc,
etcetera,
many
many
fish
in
the
world,
then
you
would
know
which
free
fixes
were
reachable
through
this
link
and
it
would
be
able
to
disaggregate
them.
But
if,
for
some
reason
it
was
never
aware
that
something
came
up
here
or
if
something
comes
after
the
breakage,
he
cannot
know
all
the
blue
guys
will
know,
but
the
red
guys
won't
because
the
flooding
is
is
cut.
C
So
you
don't
know,
what's
behind
so
you're
in
this
in
this
situation,
where
the
red
guys
know
that
there
is
a
loss,
but
they
don't
know
what
it
is.
The
blue
guys
know
what
prefix
is
associated
to
this
lost
link,
but
they
can't.
They
can't
tell
the
red
guys
that's
why
we
will
go
through
those
rings
that
10
years
as
we
presented,
you
know
there
is
a
more
broken
situation,
but
it's
pretty
much
with
the
same
logic.
In
the
end,
if
you
lose
this,
we
call
it
spine
or
top
of
that
in
this
guy's
node.
C
Then,
yes,
you
cannot
go
to
any
of
those
leaves
in
the
red
plane
and
the
same
thing
applies.
This
guy
knows
there
is
a
breakage
doesn't
know
which
prefixes
are
affected.
These
guys
know
that
something
is
affected,
but
they
can
tell
exactly
what
it
is
to
those
guys,
because
there
is
no
reflection
between
those
two
planes.
C
So
that's
where
we
are
so
to
solve
this
problem.
We
what
one
suggestion
we
have
in
the
draft
is
actually
to
to
enable
to
ring
2
by
2.
Actually,
the
the
peers
can
be
many
by
many
if
in
a
representation
12
in
the
draft,
it's
actually
maybe
here
or
this
way,
so
you
actually
ring
all
the
guys
in
kind
of
the
same
line
bringing
them
together.
C
You
recreate
a
bigger
switch
in
a
non-partisan
world,
see
if
you,
if
those
guys
were
a
single
guy
and
those
guys
were
a
single
guy,
you
would
have
a
fabric
that
is
not
partitioned.
So
we
recreate
that
and
that
enables
these
guys,
the
red
guy,
who
discovers
the
breakage
to
get
the
list
of
prefix
that
that
are
broken
from
the
other
one
I
know
there
is
a
breakage
behind
me.
I,
don't
know
what
prefixes
are
lost.
I
know
what
prefixes
that
mean.
I
can
tell
so
on
this
link
here.
C
Actually
we
will
advertise
prefix
types
prefix.
Yes,
so
we
know
if
you
have
a
breakage
what
that
means.
Okay,
so
let's
pretty
much
how
we
we
can.
We
can
do
this
game.
One
could
also
design
a
super
super
spine
up
there,
so
that
this
information
is
actually
shared
by
above.
That
would
be
an
ultimate
way
of
doing
it,
and,
with
this
we'll
go
into
the
details
on
how
you
do
the
negatives
aggregation
so
remember,
we
are
in
distance
vector
space.
So
it's
a
bit
different
from
what
you
may
see.
C
Some
people
are
doing
that
you
know
naming
is
doing
that
in
I
guess.
Well,
there
is
a
draft
which
attempts
to
do
that.
Yeah,
yes,
I
spiked,
so
space,
but
here
we
are
in
distance,
vector
world
and
you'll
see
that
makes
actually
life
much
much
easier
than
it
would
be
to
do
any
form
of
flooding,
because
it's
a
one
hop
oops.
The
flooding
scope
is
basically
one
heart,
so
you
can
really
make
a
decision
that
each
heart
about
what
you
already
got
and
we'll
see
how
that
happens.
C
The
whole
principle
of
negative
is
that
if
there
is
a
breakage
like
this
is
a
foreign
leaf,
I
can't
I
don't
need
to
tell
you
what's
behind,
but
we
lost
the
connectivity
between
this
particular
leaf.
If
you
like,
and
this
particular
top
of
fabric,
it
could
be
just
a
link
or
could
be
just
the
logical
visualization
that
the
admin
very
first
slide,
whatever
connectivity,
the
creativity
between
this
guy
and
this
guy.
This
finally
is
lost.
C
Okay
and
if
you're
in
this
situation
with
the
negative,
it's
going
to
be
s
1,
which
lost
the
connectivity
to
prefix
a
that
will
inject
to
all
its
remaining
children
in
its
own
plane
that
it
lost
the
connectivity
twice.
So
it
kind
of
injects
a
rot
non,
a
right!
That's
how
we
presented
you
and
the
reaction
by
any
of
these
children,
and
here
I
picked
I'm
one,
but
it's
gonna
be
the
same
for
M
2,
M
3
is
that
they
will
in
the
in
the
forwarding
trotting
table.
So
we
don't
respect.
C
C
In
the
forwarding
plane,
what
you
really
end
up
doing
is,
if
you
receive
an
on
a
from
one
parent,
one
of
the
guys
at
the
level
above
you
really
install
a
route
to
a
via
all
the
other
parents.
So
basically,
this
one
message
is
equivalent
to
getting
at
this
very
same
time
and
that's
kind
of
interesting
at
the
very
same
time
as
many
messages
as
you
have
other
parts.
C
So
when
you
apply
this,
then
the
load,
balancing
that
you
could
have
with
all
the
other
parent
is
still
maintained.
It's
kind
of
an
interesting
property.
If
you
got
a
single
positive
from
one
of
them
and
then
later
a
positive
from
him
and
then
later
a
positive
from
him.
The
result
is
during
a
small
period
of
time.
C
All
the
traffic
to
prefix
is
funneled
through
the
first
parent,
which
sends
you
the
positivity
segregation,
something
that's
kind
of
so
do
you
have
two
strands
this
transient
state
with
the
positive
that
makes
it
so
that
you
get
an
in
cast
in
this
particular
mode,
which
is
good
to
remember.
Okay,
so
with
the
negative,
you
don't
have
this
particular
line
cast.
You
just
divide
the
traffic
with
all
the
other
parts.
C
Now,
if
we
continue
on
this,
if
actually
the
breakage
is
more
extensive
than
then
what
you
thought
right
this
other
parent
I
cannot
see
the
same
prefix
one,
since
these
are
those
two
links
which
are
broken
but
could
be
any
sort
of
situation.
It
will
also
send
not
a
right
and
the
reaction
of
the
second
non,
a
is
to
go
through
the
state
that
you
already
have
in
your
phone
in
table
and
say:
oh
I
have
to
remove
this
one,
because
now
it's
no
more
valid
and
that's
it.
That's
all
you
have
to
do
so.
C
One
cause
one
action,
I'm
done,
which
is
pretty
cool.
If
you
are
looking
at
doing
that
and
positive
world
you'll
see
it's
much
more
complex
and
you
would
need
actually
information
that
you
don't
necessary
to
know.
If,
if
you
need
to
do
something
about
because
you
would
depend
on
what
your
neighbors
do,
it's
real,
so
there
we
go,
then
the
second
negative
advertisement
from
from
a
second
parent
I,
just
remove
an
untreated
I
just
created,
and
we
continue.
C
Obviously,
so
you
have
you
end
up
with
the
knowledge
that
you
don't
have
any
right
to
I,
which
boils
down
to
the
fact
that
you
need
to
tell
your
children
all
of
your
children
recursively.
That's
when
the
process
recursos,
you
tell
brutally
all
your
children,
I,
don't
use
me
and
then
we
start
using
all
their
other
possible
parents
see.
So
so
you
get
to
negative.
C
Same
thing
happens
the
other
way
around
right.
If,
if,
if
the
the
connectivity
is
restored,
I
wrote
it
receive
a,
but
it's
really
an
ammonite
right.
You
receive
the
the
undoing
of
the
non
a
from
a
first
parent.
As
soon
as
you
receive
the
undoing
from
the
non
off
of
the
non
a
from
a
first
parent,
it
means
a
you
can
which
a
via
this
pallet
right.
So
it
is
the
time
while
you
tell
on
your
channel
a
I'm
back.
C
C
That's
when
you
consider
removing
the
we'll
see
in
the
RHIB
what
happens,
but
the
entry
gets
away
from
the
rib
and,
and
you
can
remove
all
the
fields
altogether,
because
they
all
they
tell
you
that
the
rot
both,
which
is
the
aggregation,
is
actually
aggregating
everything
you
could
do
with
the
aggregated
rod.
So
that
the
time
where
you
get
from
the
last
parents,
they
came
when
you
consider
removing
everything.
C
Okay,
so
that's
basically
the
data
function,
but
the
operation
and
it's
as
simple
as
this
now
I
represented
here.
It
cannot
be
in
the
spec
because
we
don't
we
don't
talk
about
fib,
but
I
represented
a
kind
of
an
abstract
way
of
doing
the
red
and
then
kind
of
an
abstract
way
of
of
looking
at
the
field.
That
goes
with
that,
just
for
the
sake
of
visualizing,
what
happens
in
the
falling
plane?
C
So
if
we
have
this
default
route,
which
is
all
my
parents
as
well
as
two
or
three
or
four
and
now
I
get
one
parent
like
say
it's
a
swamp
which
tells
me
a
there.
Is
this
prefix,
which
is
like
aggregated
by
default
right
now,
2/16
and
I'm,
getting
in
on
a
four
via
a
swamp.
So
what
I
end
up
doing
is
I
had
a
fib
which
told
me
default
only.
It
was
like
that
now
I
create
this
fibrin
tree,
the
most
specific
to
this
/
16,
prefix
and
I.
C
Look
at
what
was
above
what
I
inherited
from
what
my
parents
would
you
what
I
could
do
without
this
non
ie?
So
what
is
above
in
this
table
here
and
I,
just
subtract
I
copy
it
and
I
subtract
what
I
cannot
reach
so.
The
result
is
this
that
gives
me
this
okay,
so
this
is.
This
is
what
I
inner
it
from.
This
is
what
there
is
the
result
of
the
inheritance
by
removing
this.
C
Let's
read
the
operation
that
you
have
to
do,
then,
if
there
is
another
prefix
which
also
inherits
from
default
right,
there
are
prefix
of
the
same
level
in
my
table,
and
this
one
says:
oh
it's
via
s
for
that
they
cannot
read
well
this
one.
This
one
inherited
from
this.
This
one
also
inherits
from
this
right.
So
in
these
guys,
I
subtract
S
4
from
what
I
inherit
from
which
is
the
in
this.
The
rib.
C
Okay,
so
so
far,
not
too
much
surprise
a
little
bit
more
surprises.
What
happens
is
in
default.
Iii
I
lose
as
to
global
as
to
his
dead
Manning
to
his
to
his
dead.
Well,
all
those
things
not
only
v
here
which
was
associated
to
this
ribbon
tree
here,
but
also
all
the
inherited
prefix
remember
they
were
inherited
from
him.
They
depend
on
him
if
he
goes,
those
can
stay.
So
you
have
to
look
at
the
negative
inherited
prefix
that
you
installed
because
of
this
inheritance
and
you
have
to
remove
them.
C
Okay,
it's
like
remember
a
negative
is
like
multiple
messages
in
one
and
now
we
look
at
of
may
be
more
curious
case
and
probably
less
realistic,
but
imagine
that
you
submit
and
subnet
and
you
get
negative
with
it
negative,
which
is
kind
of
weird.
If
you
ask
me
when
that
can
happen,
I
can
make
your
drawing,
but
it's
a
really
really
weird
right.
C
C
That
gives
me
this
now.
This
guy
is
the
part
of
this
guy
right.
In
this,
your
key
here,
this
one
is
a
subset
of
this,
so
so
the
the
the
capability
to
reach
10.0
the
10
before
I
got
this
negative
was
the
capability
to
reach
this,
which
means
that
it
was
through
those
guys.
That's
what
I
mean
everything
from
so
now,
if
I
get
an
honest
for
this
sub
prefix
of
this
guy
I
have
to
subtract
it
from
this,
not
from
this,
because
that
is
my
parent.
So
the
result
is
that.
C
There
same
story
again:
if
I
lose
this
guy,
from
which
all
this
is
narrated.
Well,
not
only
do
you
lose
it
from
here,
but
you
have
to
recursively
go
through
here
and
then
you
have
to
recursively
goes
from
here,
so
you
have
to
look
at
down
the
tree
of
inheritance,
which,
in
the
real
world,
is
not
expected
to
be
anything
more
than
one
one,
one
heart,
but
you
could
devise
web
games
where
you
would
actually
have
to
go
deeper
down
now.
There
is
this
rule
in
rift.
C
You
could
get
a
positive
route
at
the
negative
rod
for
the
same
prefix
from
different
people,
so
they
appear
in
that
table
in
the
Trib
here
like
in
the
same
place.
You've
got
different
types
of
rods,
like
would
have
a
seven
rod
and
the
northern
rods
and
you've
got
priority
rules
in
rift
which
tells
you
a
you,
always
save
our
route
going
south
over
a
road
going
north.
Even
if
it's
about
the
same
prefix
or
same
in
the
same
logic,
we
made
this
decision
and
it's
actually.
It
was
an
interesting
discussion.
C
There
are
pros
and
cons
here
so
that
the
positive
route
wins
over
the
negative
world,
but
the
consequence
of
that
is,
if
you
have
a
positive
here
somewhere
for
that
prefix
as
well.
Actually
all
those
negative
rods
go
away,
only
the
positive
space,
meaning
that
there
is
no
recursion,
because
those
negative
are
only
installed
if
there
is
no
positive
edges
for
the
same
prefix.
C
So
normally
when
you
have
the
same
level,
so
you're
one
hop
away
from
the
discovery
you
get,
the
certain
reflection
you
get
the
positive
well
and
if
the
fabric
is
not
broken,
that's
exactly
what
partition!
That's
exactly
what
you'll
get
you'll
get
a
positive
and
you
get
a
negative
you'll
prioritize
in
your
fav,
the
positive
meaning
that
again,
this
entry
will
not
exist.
So
we
did
not
consider
that
this
recursion,
which
could
in
theory,
be
a
V,
would
realistically
be
it
right.
C
So
the
the
archer
is
the
most
probable
cases
you
get
something
minus
something
and
that's
gonna,
be
it
around.
My
slide.
I
call
a
different
leader,
the
forwarding
table,
depending
on
where
what
type
of
ribbon
tree
was
used
to
build
it.
So
as
the
result,
if
you
look
at
this,
the
the
normal
flooding
gives
me
Selden
rods.
So
the
rod
to
a
is
installed
ya
s,
3,
S,
4
and
M
4
to
L
4.
So
that's
that's!
How
I
can
get
get
away
with
normal
flooding?
That's
basically
a
shadow
cone
of
a
in
that
case.
C
That's
what
I
can
see
when
I
flood,
knowing
that
this
link
here
is
expected
to
be
broken
now
because
of
the
southern
reflection
here.
Well,
actually,
because
this
is
broken,
which
we
happens
here
and
for
is
aware
that
a
is
broken
for
else
rewrite.
This
link
is
come
down,
so
m4
will
positively
disaggregate
one
hub.
So
this
right
here
is
the
result
of
a
positive
disaggregation
and
now
and
now
due
to
the
negative
disaggregation,
s1
and
s2,
because
now
you
see
s1
and
s2,
they
cannot
see
a
anymore.
They
will
do
this
recursive
operation.
C
It
will
disagree,
gate
negatively,
well
disaggregate
negatively,
meaning
that
he
world
is
aggregating
actively
same
thing
for
us
to
evil
desire,
get
negatively
evil
disaggregate
negatively.
So
but
here
since
there
is
still
well
here,
you
would
have.
You
would
also
have
a
green,
but
since
the
the
positive
wings
about
the
green,
that's
why
you
don't
see
the
green.
Otherwise
you
would
have
you
do
whatever
the
green
route
here
to
and
check
check
check
which
much
right.
C
And
I
said
that
this
one
is
also
installed.
I
think
I'll
rerun
that
in
my
mind,
oh
it's
because
those
are
the
ribbon
trees
and
these
are
the
five
entries.
That's
whether
to
reread
my
slides,
but
the
real
entry
follows
you
know
the
the
the
the
propagation
of
the
negative
the
Theban
tree
is
what's
left,
so
the
forwarding
is
really
along
the
red,
so
I
can
get
into
one
of
those
two
which
can
run
and
and
the
green
is
really
the
the
ribbon
trees,
not
the
five
entry.
C
So
that's
what's
interesting
right
as
long
as
you
do
positive
advertisement,
the
ribbon,
the
fifth.
They
tell
you
the
same
thing,
but
when
you
advertise
negatively,
which
is
the
the
the
other
way
of
the
green
arrows,
which
are
really
the
ribbon
trees,
the
Phebe's
whatever
is
left
so
the
freebies
this
guy
in
this
guy.
C
Well,
the
rib
is
this
guy,
this
guy,
this
guy,
this
guy,
so
in
fever,
read
out
of
many
states
which
is
interesting
to
read
to
remarked
my
sheep
is
very
minimum
and
it
points
exactly
I
wanna
go
so
you
follow
the
red
and
yellow
arrows,
and
this
one
as
well
and
you're,
always
good
way
and
you've
solved
your
watch
in
Prague
and
that's
pretty
much
it.
We
have
questions.
B
C
C
C
It
cetera
and
not
seeing
not
being
able
to
actually
go
deep
enough
is
because
those
guys
don't
see
one
another
and
they
can't
clear
I
call
it
a
ceiling
right,
but
it's
it
basically
might
mean
if
I
see
everybody
and
I
can
touch
all
the
level
below
then
I
can
install
something
T
online
at
the
level
below
which
passed
everything
wrong
like
a
ceiling.
So
nothing
can
happen
without
the
ceiling
intercepting
the
packet
and
sending
it
to
the
right
satisfied.
C
D
I
I
It
is
about
the
general
problem
that
was
discussed
in
draft
Lee
and
LSR
and
is
also
brought
forward
a
plethora
of
discussion
and
solutions
for
how
to
produce
a
constrained,
flooding
topology
in
a
dense
graph,
with
some
of
the
requirements
being
that
it
would
be
immune
to
single
failures.
But
the
key
thing
was
it
would
reduce
the
number
of
copies
of
an
LSA
that
would
be
continually
interrupting
the
control
plane
for
any
given
note
in
the
system,
so
it
this
is
a
proposal.
I
The
actual
approach
is
Tiye
to
your
starting
point
is
to
use
two
diverse
lis
routed
spanning
trees
constructed
such
that
each
node
in
the
dense
graph
ends
up
being
by
connected
to
the
flooding
topology
each
node
computes,
the
spanning
trees,
knowing
the
roots
and
then
determines
its
its
role
in
it
with
respect
to
how
it's
going
to
treat
LSAs,
and
this
is
based
on
information
obtained
from
the
IGP
and
the
flooding.
Topology
itself
is
the
sum
of
the
spanning
trees.
I
So
the
first
copy
of
an
LSA
received
by
a
node,
for
example,
is
the
one
that's
propagated
further
on
both
trees,
regardless
of
how
it
arrived.
So
it's
if
it's
a
first
batch
first
past
the
gate,
sort
of
solution,
the
actual
flooding
itself
is
split
horizon
between
the
upstream
and
the
downstream
from
Forel
essays
that
are
received
from
an
upstream
interface.
So,
if
I
received
one
coming
downstream
from
the
root
of
one
spanning
tree,
I'm
not
going
to
reflect
it
back
upstream,
that
would
effectively
defeat
about
the
closest
thing
to
a
loop
you
could
produce.
I
The
net
result
is
in
a
fault,
free,
Network,
all
nodes
that
participate
in
the
flooding
topology
will
receive
two
copies
of
a
flooded
LSA,
a
and
in
any
single
fault.
They
will
get
typically
to
one
or
two
of
them
may
only
get
one.
It
depends
on
whether
the
fault
actually
impacts
the
flooding
topology
itself.
I
I
Bridging
mac
mode,
the
actual
algorithm
and
an
example
of
using
it
with
is,
is,
is
documented
in
RFC,
63
29,
that's
also
available,
and
at
least
one
or
more
commercial
stacks
that
I
know
of
and
of
course,
802
dot
when
at
a
queue
itself
is
deployed
in
about
a
thousand
networks
that
I've
heard
about
the
actual
tiebreaking
algorithm
uses
a
lexicographically
sorted
list
of
the
node
IDs
to
tiebreak.
When
multiple
equal
cost
paths
are
found.
In
essence,
you
construct
the
path
ID
for
each
path,
sort
them,
and
then
you
can
rank
them
now.
I
One
of
the
interesting
things
you
can
do
with
this
is
prior
to
ranking
them.
You
can
also
use,
what's
known
as
an
algorithm
mask
if
you
want
to
produce
more
than
one
diverse
path
and
you
xor
the
node
ID
list
with
the
algorithm
mask
prior
to
ranking
so
and
if
you
use
an
algorithm
mask
of
zero
or
ffffff
effectively,
what
you
can
do
is
use
the
bookends
of
diversity,
so
each
tree
in
this
example
is
constructed
using
one
of
the
two
algorithm
masks
and
the
trees
themselves
are
both
lincoln
node
diverse.
I
The
net
result
of
this
is
is
for
each
set
of
nodes.
That's
equidistant
from
the
root
the
low
and
the
high
node
IDs
are
selected,
the
low
node
ID
for
one
tree
high
node
ID
for
the
other
tree,
and
this
is
how
you
actually
get
the
both
the
note
and
the
link
diversity,
so
the
trees
using
algorithm
mask
0.
Always
you
will
select
the
node
ID
the
trees
using
algorithm
mask
minus
1
will
always
select
the
high
node
ID.
The
other
nice
thing
about
this
algorithm
is
not
not
all
order
and
our
equal.
I
One
of
the
things
you
can
do
with
this
is
any
portion
of
the
shortest
path
is
also
the
shortest
path.
So,
while
traversing
the
graph,
you
can
incremental
eat
I
break
and
purge
an
awful
lot
of
state.
This
makes
it
quite
quick
and
tends
to
explain
my
shortest
path.
Bridging
networks
converge
very
quickly,
despite
the
fact
that
are
actually
doing
it
all
pairs
shortest
path
computation.
We
are
not
doing
all
pairs.
We're
we're
only
doing
two
Dijkstra's
in
this.
I
So
to
give
you
a
visual
representation
of
the
results.
I
have
an
example
network
here,
where
I
have
the
two
spanning
trees,
the
red
and
the
green.
The
red
is
routed
on
node
0
down
at
the
bottom
left.
The
green
is
routed
on
node
55
at
the
top
and
you'll
notice
that
all
the
transit
nodes
for
the
red
tree
are
the
low
node
ID
in
any
given
set
of
nodes
in
the
graph
and
for
the
green
it
tends
to
be
the
high
ID.
So
this
is
actually
how
the
diversity
is
achieved.
I
Like
I
said
you're
you're,
ensuring
you're
picking
the
book
ends
of
the
node
IDs
at
any
particular
point
in
the
graph.
Now
the
flooding
rules
themselves
are
quite
straightforward.
If
I've
received
an
LSA
from
an
upstream
adjacency
and
it's
one
I
haven't
seen
before,
of
course,
I
flood
on
the
downstream
member
adjacencies
and
any
non-participating
adjacencies.
One
of
the
things
Tony's
graphs
discussed
was
the
idea
that
you
could
have
a
flooding
topology.
I
And
what
I
mean
by
a
member
of
Jason
C
is
it's
an
adjacency
that's
been
identified
as
part
of
the
flooding
topology
now
I
also
use
the
color
of
the
red
and
green
trees
to
illustrate
the
sort
of
the
notion
that
ultimately,
I
really
don't
care.
What
the
tree
of
arrival
was
simply
whether
it
was
upstream
or
downstream.
It's
the
sum
of
the
trees.
If
we're
working
with.
I
I
That's
an
unsolved
problem
in
this
draft
I'm,
just
sort
of
throwing
the
basic
algorithm
out
and
seeing
where
things
go
from
there
once
the
community
takes
a
look
at
it,
so
I'm
not
documenting
any
changes
or
proposing
any
protocol
elements
I'm
just
identifying
the
requirements
at
this
point
and
we'll
see
where
it
goes
from
here
now,
one
of
the
things
is
is
the
number
of
copies
that
needs
to
be
generated
by
certain
nodes.
If
you
looked
at
the
example
is
not
constrained,
that
was
suggested
as
being
desirable.
This
draft
does
not
achieve
that.
I
I've
considered
possibilities
of
trying
to
partition
the
space
by
using
multiple
trees.
So
far,
I
don't
really
have
anything
satisfactory
at
the
moment,
so
very
quickly.
What's
in
the
draft
the
problem
space,
the
algorithm,
the
flooding
rules,
a
discussion
of
the
requirements
for
route
selection,
interaction
with
non
participants
which
is
pretty
brain-dead
when
you
think
about
it?
No
two
network
startup
some
thoughts
on
reoptimize
ations.
I
Of
course,
once
you
start
getting
failures,
something
may
end
up
as
incomplete,
although
the
some
of
the
graphs
may
be
complete,
but
you
don't
take
the
chance
of
subsequent
failures,
eventually
degrading
things,
and
it
also
suggests
and
strategies
for
dealing
with
catastrophic
multiple
failures
and,
as
I
mentioned,
I,
actually
don't
get
into
protocol
details.
Yet
so
two
very
quickly
to
summarize
the
characteristics
of
the
solution.
The
structure
is
an
interconnected
of
1+1
multi-point
to
multi-point
trees.
I
The
protocol
changes
required
are
some
means
of
advertising
or
communicating
information
to
allow
route
selection
advertisement
of
the
desire
to
participate
in
the
flooding
topology.
The
actual
computation
of
the
flooding
topology
is
the
order
of
two
Dijkstra
computations.
The
maximum
diameter
of
the
flooding
topology
in
the
fault
tree
worst
case
is
twice
the
distance
from
the
leaf
to
the
spine,
and
the
most
pathological
case
I
have
been
able
to
find
is
when
you
can,
a
link
between
the
routes
fails,
and
so
the
net
result
is,
is
the
distance
between
the
routes?
I
One
needs
to
get
factored
into
the
maximum
diameter
of
the
flooding
topology
and
the
typical
and
maximum
number
of
LSA
is
a
node
will
receive,
would
be
two
so
there's
a
couple
of
things
in
the
draft.
I
want
to
update
about
the
route
selection
criteria
for
the
next
version,
and
otherwise
I
just
want
to
collect
feedback
and
figure
it
out
from
here.
I
D
D
J
J
Okay:
okay:
this
data
model
is
defined
according
to
rift
protocol,
and
this
model
includes
protocol
configuration
state
and
a
notification,
and
some
features
are
added
to
enhance
the
protocol,
such
as
PP
and
a
crystal
and
at
the
other
future,
and
the
the
updater
of
this
version
is.
We
changes
the
community
from
R
to
list
and
the
reference
common
policy
defining
and
a
DWG
policy
model
structure.
J
So
it's
the
main
updater
of
this
version,
and
next
this
is
the
configuration
of
the
young
model,
and
so
we
know
that
the
configuration
includes
not
info
interface
policy
and
the
other
things
you
can
can
see
from
this.
Okay-
and
this
is
the
state
of
the
rift
protocol
include-
includes
neighbor
database.
Then
a
chemist
all.