►
From YouTube: IETF115-RIFT-20221108-1500
Description
RIFT meeting session at IETF115
2022/11/08 1500
https://datatracker.ietf.org/meeting/115/proceedings/
A
Good
afternoon
everybody
Welcome
to
Reef
session
ktf
115
and
we
are
about
to
start
so
sessions
being
recorded.
You
know
Dr
Panda,
please
wear
your
mask.
A
A
A
So
the
quick
update
we
are
going
to
have
a
presentation
from
Jordan
on
Facebook
update
and
give
a
registry,
then
other
flood
reflection
that
we
plan
to
adopt
after
this
meeting
and
then
Sandy
will
talk
about
early
Frank.
A
So,
what's
going
on
working
group
pending
Alvarez
reviews
base
specification
applicability
statement
model,
there's
been
some
progress
recently
we
need
to
finalize
Shepherd
right
up
for
KB
registry.
The
outer
EVP
and
flood
reduction
is
still
in
progress.
Ive
working
on
it,
we've
got
a
new
individual
draft
on
Leaf
rings
and
reef
desire.
Multicast
is
still
on
kind
of
Border.
We
do
plan
to
work
on
the
doctor.
We
update
the
chapter.
B
Okay,
hey
everyone.
My
name
is
Jordan
head.
Just
gonna
provide
an
update
on
how
the
rift
spec
is
progressing
as
well
as
several
others.
Following
this
next
slide.
Please.
B
So
versions,
you
know
the
draft
was
pushed
a
couple
of
weeks
after
it
f-114
following
the
most
recent
round
of
ad
review,
some
of
what
I
mentioned
here.
It
was
mentioned
there,
but
you
know
some
of
the
pairs
are
beating.
So
overall,
the
document
continuing
to
improve
I
won't
mention
every
detail,
but
here's
kind
of
a
rough
outline
of
what
we've
been
working
on.
So
you
know,
as
the
documents
of
all
the
couple
of
Concepts
need
to
be
clarified
in
text.
B
Just
you
know
as
an
example,
how
ties
are
serialized
on
wire?
You
know
a
couple
of
other
readability
components
like
adding
pointers
to
the
Reader's
Digest
for,
like
some
security
Concepts
beyond
that
svgs
and
ASCII
art
now
align
properly
and
what
I
mean
by
that
is
they
basically
tell
the
same
story
and
provide
the
same
value,
regardless
of
the
format
of
the
draft
that
you're,
reading
and
kind
of,
as
as
was
intended,
it's
kind
of
worth
noting
we
got
some
pretty
good
feedback
from
external
readers
as
well.
B
So
it's
it's
good
to
know
that
those
are
actually
helping
other
SVG,
optimizations
kind
of
helped
us
get
around
a
lot
of
the
ietf
tooling
challenges.
You
know,
there's
a
fair
bit
of
work
that
went
into
them.
You
know
I
think
they
provide
a
lot
of
value,
especially
for
the
more
complex
topologies,
but
it
was
it's
worth
noting
that
it
was
it's
pretty
difficult
to
actually
get
them
to
work
within
the
confines
of
the
existing
ITF
tools.
B
So
it's
something
I'm
following
up
with
some
of
the
tooling
folks,
and
you
know
a
couple
other
avenues:
I
digress,
updates
the
Ianna
consideration
section.
We
had
to
actually
push
the
new
version
of
the
Fritz
schema,
as
well
as
a
couple
of
little
little
phrasing
fixes
for
for
what
Ayanna
had
mentioned
to
us
next
slide.
Please
and
you
know,
and
then
the
usual
kind
of
you
know
editorial
changes,
grammar
spelling
flow
references,
Etc
and
then
you
know.
B
Last
but
not
least,
some
improved
kind
of
inclusion,
diversity,
sort
language
next
slide
and
as
for
what's
next
we're
continuing
to
work
with
Alvaro
and
AD,
and
you
know
we'll
keep
applying
as
soon
as
we
can
and
kind
of,
keep
keep
everything
moving
forward.
I
also
owe
avaro
and
and
the
chairs
and
the
shepherd
brief
thread
to
kind
of
rework
The
Shepherd
of
you.
So
we
can
get
myself
named
we're
officially
tied
in
as
a
co-author,
but
no
that's
it
next
slide
comments.
Questions
concerns
foreign.
B
I
think
Jeff
recovered
a
lot
of
this,
but
essentially
we've
got
the
you
know,
there's
nothing
terribly
exciting
in
new
versions.
It's
mostly
edits
due
to
Shepherd
review
things
like
you
know,
references
boilerplate,
Etc,
next
slide.
B
And
kind
of
Beyond,
you
know
just
waiting
for
the
write-up
Ayanna
actually
reached
out
a
couple
of
days
ago
and
pointed
out
that
you
know
we
need
to
specify
a
couple
of
ambiguous
things,
basically
because
specify
upper
bounds
for
some
of
the
values
and
things
like
that.
So
we'll
get
that
adjusted.
You
know,
along
with
the
shepherd
feedback
in
the
next
version,
next
slide.
B
And
that's
it
questions.
A
B
Okay,
last
but
not
least
for
me,
so
flood
reflection
got
some
updates
as
well.
This
time
next
slide,
please.
B
Okay,
basically,
we've
added
a
role
for
spine
nodes,
which
is
just
a
way
of
saying
a
way
to
specify
spines
as
a
specific
entity,
or
you
know,
non-flood
reflector,
client
or
non-flood
reflector
nodes
participating
in
Auto
FR.
So
this
lets
us,
you
know,
advertise
some
kind
of
generic
State
about
the
spine.
That
might
be
important
to
the
overall
picture
of
the
flood.
Reflection
topology
and
you
know
beyond
that.
It.
It
kind
of
lays
the
groundwork
or
you
know,
adds
the
capability
for
any
spine
specific
variable
derivation.
B
Should
it
be
needed
in
the
future.
We
added
a
bit
more
to
the
Telemetry
key
value
ties.
So
basically
we
can
now
carry
and
describe
State,
for
you
know,
kind
of
the
regular
or
internal
level.
One
Isis
adjacency
within
the
fabric
next
slide
slightly
more
detailed
view.
So,
basically,
the
items
in
bold
are
what's
new,
so
it's
kind
of
as
an
example.
B
If
we
were
advertising
KVs
from
a
flood,
reflector
client,
you
know
a
leaf
node
at
the
top
of
the
fabric,
we'd
be
able
to
see
the
state
of
you
know:
functional
versus
total
level.
Two
flood
reflector
adjacencies
to
the
actual
flood
reflectors
themselves,
functional
versus
total
level,
one
shortcuts
between
Leaf
nodes
for
any
deployments
that
that
leverage
them
and
then
you
know,
as
I
mentioned
the
new
one.
You
know
functional
versus
total
level.
B
One
you
know
air
quotes
regular
Isis
adjacencies
that
describe
more
of
the
underlying
Isis
topology,
so
you
know
the
operator
really
benefits
and
that
they
can
see
the
entire
fabric
from
the
leaf
and
spine's
point
of
view.
But
all
from
you
know
a
single
top
node
next
slide
some
more
SVG
stuff
for
this
one.
So
the
operational
section
had
an
SVG
that
sort
of
described
how
the
rift
underlay
you
know
the
base,
Isis
topology
and
then
the
flood
reflection
topology
all
fit
together
depending
on
location
in
the
fabric.
B
Previously
we
only
had
it
working
in
the
pdf
version,
but
kind
of
similar
to
the
base.
Spec
got
things
compliant
with
the
draft
upload
tool,
and
things
like
that.
B
So
now
it's
in
the
HTML
version,
so
I
think
it
keeps
things
kind
of
nice
and
crisp
that
slide
so
Jeffrey
mentioned
this
too
I
know
adoptions
waiting
on
the
retardering
discussion,
so
you
know
hoping
to
see
things
progress
on
that
front
and
as
for
future
updates
right
now,
we've
there's
more
KV
stuff
slated
for
the
next
version
and
then
we're
still
kind
of
noodling
around
with
a
multi-plane
flood
reflection
stuff
and
that's
it
for
the
for
now.
That's
it
for
me
any
questions
about
Auto,
fr.
B
C
C
Question
so
was
this
presented
in
the
LSR
at
all,
because
my
understanding
is
that
this
will
be
used
for
ISS
flattery
reduction
right.
So
what
do
the?
What
does
Sr
think
of
this.
B
The
questions
that
kind
of
came
up
there,
it
was
presented,
I
forget
which
meeting
it
was
a
couple
meetings
ago.
Basically,
their
only
question
or
concern
was:
does
it
overlap
with
any
of
the
other
working
groups
like
The
Auto
configuration
stuff
for
IDR
Etc,
and
it
doesn't
right
we're
not
actually
modifying.
This
doesn't
actually
modify
Isis.
It
we're
just
using
a
rift
to
kind
of
leverage.
B
C
So
if
Isis
implementation
wants
to
make
use
of
this,
do
they
need
any
specification
in
LSR
or
not,
or
do
they
need
to
interrupt
between
other
routers.
B
Things
like
that
so
for
there
is
an
LSR
spec
for
flood
reflection
itself.
This
is
just
the
auto
configuration
portion,
so
it
depends
how
you
want
to
do
it.
If
they
don't
need
the
auto
configuration
pieces
of
it,
then
they
don't
need
it
if
they
do
and
they
want
to
run
risks
then
yeah.
They,
you
know,
follow
this
spec
and
you
know,
along
with
the
lsr1
and
you
get
both
looks
like
Tony,
wants
to
add
something
foreign.
D
So
we
do
not
need
any
LSR
specs.
If
you
have
this
thing
implemented
with
Isis,
then
the
only
interaction
surface
is
really
pulling
out
the
analytics
right,
which
is
always
very
vendor
specific,
so
the
Isis
has
to
expose
with
already
does
in
some
form
of
fashion
right
worst
case,
you
do
CLI
scraping.
If
you
want
to
provide
the
analytics,
you
need
the
information
out
of
Isis,
but
the
analytics
is
completely
optional.
So
no,
there
is
no
implications
in
terms
of
you
know,
scaling
or
implementation
or
specification.
D
D
20P
again
yeah
well
so
I
mean
this
idea
is
hanging
for
a
while
and
it's
a
very
short
road
to
fame
and
riches.
If
you're
right,
you
know
flood
reflection
for
SPF,
which
is
pretty
much
a
straight
knock
off
of
a
higher
size
stuff,
but
there's
just
simply
not
enough.
Customer
demand.
Those
SPF
networks
are
not
big
enough
and
people
aren't
used
that
way.
So
it
will
be
effort
wasted
yeah.
D
B
A
C
Okay,
so
Jordan
you're
going
to
provide
a
quick
update
on
other
uvpn
as
well
right,
no
slides
but
quick,
update,
I.
Think.
B
There
wasn't
any
actual
drop
that
they
just
go
around
and
there
is
some
stuff
slated
for
the
next
one.
It's
Auto
frapping
to
just
kind
of
take
the
take
priority.
There's
still
some
multiplane
and
data
center
interconnect
Gateway
stuff
that
we're
working
as
well.
As
you
know,
you
know,
as
as
the
thing
continuously
svgs
and
making
the
operational
section
A
little
a
little
bit
cleaner.
So
that's
for
next
time.
D
20P
again,
to
fill
in
a
little
bit
more
detail,
so
basically
the
as
far
as
I
remember
the
schema
version
and
coding
six
added
something.
What
where
we
know
that
we
needed
in
games
in
case
of
multiplying
fabric,
to
run
a
proper
Auto,
VPN
Auto
FR
election,
so
we
can
actually
distribute
the
FRS
or
the
RRS
across
multiple
planes.
Simply
for
what
you
call
it
resiliency
reasons,
and
but
the
election
in
itself
has
not
been
modified.
Yet
the
algorithm
to
take
that
into
account.
That's
pretty
much
the
the
thing
outstanding.
D
There
is
a
little
bit
of
work,
so
the
schema
everything
is
here
in
coding.
It's
just
you
know.
I
need
to
add
this
little
bit
of
to
the
algorithm.
We
will
talk
it
through.
We
know
how
it
would
work,
but
it's
just
right
now.
It
is
not
taking
the
plane
into
consideration
yet
because
the
plane
itself
adds
a
little
bit
to
what
Rift
has
to
advertise.
D
So
the
scheme
elements
are
here:
they're
optional,
but
if
you
run
a
multi-plane
implementation,
you
would
have
to
advertise
the
stuff
on
the
Note
tie,
so
other
people
can
figure
out
which
plane
they
are
in
and
you
know
who
what
is
the
plane
ID
and
there
are
multiple
planes
and
then
you
have
to
take
it
into
election.
D
You're
right,
so
the
algorithm
will
do
take
care
of
it
right,
so
it
will
basically
just
like
we
have
preference
and
we
have.
Then
we
have
system
ID
ordering.
We
will
basically
the
same
way.
We
would
take
the
highest
system.
Id,
lower
system,
ID,
second
highest
system,
ID
second
lowest-
will
have
the
plane
ID
in
front
of
the
plane.
Id
is
basically
nothing
else,
but
the
highest
system
ID
within
a
plane
and
that's
why
everybody
in
a
plane
starts
to
advertise.
D
Those
are
all
the
guys
in
the
planes,
and
you
know,
okay,
the
eye
with
this
higher
system.
Id
is
actually
the
plane
ID
and
then
the
other
guys,
who
don't
necessarily
see
all
the
guys
in
the
plane,
know
what
is
the
plane
ID
and
based
having
all
the
plane
IDs
that
can
actually
running
consistent
election,
where
the
FRS
or
RRS
will
be
basically
spread
across
the
planes
right.
So
first
you
take
the
plane
into
consideration.
They
continue
to
take
a
priority
within
the
plane
and
then
you
go
the
high
ID
low
ID.
D
E
Hello,
do
you
hear
me
yeah?
Okay,
thank
you
a
lot
of
time,
very
fortunately
hi
everyone
and
I'm,
sending
song
from
ZTE.
This
presentation
is
for
supporting
leaves
without
not
among
the
neighbors,
connecting
to
a
factory
network
using
Rift.
We
have
called
a
real
heart
number
Zone.
Next,
please.
E
Though
the
draft
has
been
presented
before
but
I,
let
me
explain
it
again
and
in
case
somebody
listens
at
the
first
time.
In
the
example,
Metro
Network,
the
spine,
Leaf
topology,
can
be
deployed
to
replace
the
traditional
topology
Rift
can
be
used
as
the
dynamic
routing
protocol
to
build
the
network.
E
E
This
figure
is
the
Leaf
part
of
the
network.
We
know
in
typical
Rift
protocol.
Every
leaf.
Node
has
not
found
the
connection
with
Final
in
case
I
leave
node
lost
all
the
North
Bank
the
connection
and
can
still
access
the
network
by
Southwestern
labor
lifted,
but
in
this
deployment
some
Leaf
nodes
never
have
Northbound
the
connection,
even
if
there
is
the
best
neighbors
haven't
not
bounded.
A
E
The
leaf
node
without
not
found
the
connection,
receive
the
prefix
sort
type
from
for
defaulted
World
from
the
East
West
neighbors,
understanding
the
prefix
not
tied
to
reach
to
the
div
node,
which
has
northbounder
connection.
This
is
a
change
for
the
flooding.
The
leaf
node
without
Northbound
neighbors
must
transfer
the
prefix
Source
type
received
from
an
Easter
with
the
neighbor
to
the
other
is
the
best
neighbor,
which
also
has
no
Northbound
connection
and
the
leaf
node
without
not
found.
E
Labors
must
transfer
prefix,
not
Pi,
received
from
a
Easter
Western
neighbor,
which
has
no
Northbound
connection
to
the
other
Easter
with
the
neighbor
other
flooding
function
and
the
SPF
algorithm
are
not
changed.
So
in
this
figure,
L4
and
L6
must
transfer
the
prefix
Source
type
received
from
L1
and
L3
to
L5.
E
E
And
this
chapter
has
been
presented
in
itf11
and
has
been
discussed
in
the
mailing
list
in
the
update
of
version
zero
one.
According
to
the
subjection,
a
suggestion
from
last
meeting,
we
added
the
description
of
external
prefix,
not
type
transference,
and
that
is
the
reef
node
without
not
found
Neighbors,
which
supports
the
leaf.
Transfer
capability
must
transfer
the
external
prefix,
not
tied,
received
received
from
and
used
with,
neighbor,
which
has
no
Nerf
boundary
connection
to
the
other
is
the
Western
neighborhood,
and
so
that's
all.
E
D
D
So
first
observation,
I,
don't
know
whether
I
made
it
already
last
time
you
kind
of
generate
a
corner
case
of
a
corner
case
and
solve
it,
because
if
you
don't
force
this
topology
the
L
notes
the
beliefs,
then
Rift
automatically
solves
the
problem
as
it
stands
today.
D
Imagine
that
a
low
405lo
6
are
not
forced
to
leave
right,
so
the
l05
by
ztp
will
automatically
end
up
a
level
below
so
it
will
start
to
look
like
a
sink
below
right.
So
it's
a
very
degenerate
Claw
lo4
is
only
single
home.
Lo6
is
only
single
homed
and
lo5
is
dual
homed,
but
based
on
normal
D4
route
advertisement,
the
whole
thing
will
work.
The
rift
will
support
the
stuff
today.
D
Now,
if
you
tell
me
that
you're
interested
in
the
corner
corner
case
where
you
force
all
these
notes
on
being
a
leaf,
then
yes,
what
you're
suggesting
I
have
to
think
about
this
new
addition
through
whether
that
can
possibly
Loop.
But
what
you're
suggesting
will
work
right
over
the
horizontal
link.
Support
will
basically
give
you
so
I
think
the
stuff
that
you're,
extending
as
the
flooding
is
already
in
the
spec.
D
The
only
thing
I
would
strongly
object
is:
can
you
flip
one
forward
with
the
procedures?
Please.
D
D
Otherwise,
things
won't
work
properly.
If
you
start
to
propagate
ties,
southbound
ties
multiple
hops.
You
will
end
up
in
trouble,
plus
it
breaks
the
whole
architecture
completely.
So
one
suggestion
how
you
solve
it.
Implementation
wise,
is
that
you
support
the
policy
or
you
can
take
a
prefix
that
you
got
from
southbound
and
you
redistribute
it
into
re
inside.
Basically,
you
say:
okay,
whatever
came
from
the
north
in
the
South
Thai
re-advertise
its
southbound,
so
it's
basically
a
prefix
redistribution
within
the
protocol,
which
will
work
fine
from
experience
so
to
Observation.
D
First
you're,
solving
a
corner
corner
case
that
you
kind
of
call
upon
yourself.
If
you,
if
you,
if
you
force
those
things
being
leaves
which
I
don't
see
a
reason
for,
but
may
you
may
have
one
and
the
second
one
is
that
you
cannot
reflaught
the
southbound
tie.
You
have
to
regenerate
it,
because
the
original
originator
has
to
be
one
level
away.
E
Okay,
let's
see
because
we
think
every
group
to
want
to
be
used
in
in
many
deployment.
Maybe
this
honor
case
will
occur,
so
we
think
that
it
may
be
a
use
case
for
Rift,
but
the
SEO
side.
E
We
use
the
dtp
latep
function
of
Rift,
we
can
generate
the
level
automatically
and
so
some
leaves
will
not
be
leaves
and
it
will
be
spine.
That's
a
way
to
solve
it,
but
we
know
that
in
some
Network
the
the
leaves
may
be
not
stable.
So
and
in
case
there
is
some
breaking
or
something
other.
E
There
are
something
other
have
happened,
maybe
and
the
the
devil
will
be
broken
and
it
will
be
generated
at
the
end.
So
that's
why
we
don't
want
to
use
80
people
here
and
in
our
side
if
we
use
the
new
link
capability,
that
can
be
a
strictly
configured
in
limited
nodes,
so
it
may
avoid
the
wrong
flooding
of
the
the
tie
that
can
avoid
the
looping.
But
it's
just
another
size.
For
now,
and
maybe
in
some
cases
it
will
still
cause
the
loop
as
your
size.
No.
D
Yeah
nope,
because
the
flooding
on
the
in
multi-plane
case
on
the
Rings
read
the
section
on
the
the
flooding
procedures,
or
rather
the
flooding
Scope.
When
you
have
a
ring
at
the
top
of
the
fabric,
you
try
to
introduce
exactly
the
same
Flooding
at
the
bottom
of
the
fabric.
That's
what
you're
doing
here!
You're
ringing
the
bottom
of
the
fabric,
just
like
we
ringing
in
my
multi-plane
case,
the
top
of
the
fabric
and
the
same
flooding
Scopes
will
apply.
That's
why
I
told
you
stuff
will
work,
but
you
you
cannot
regenerate
the
South
tie.
D
You
are
not
solving
any
looping
problems.
The
only
valid
reason
to
do
what
you
do
here
is
to
say
that
I
do
never
want
those
leaves
to
be
transits,
but
actually
you
already
violating
it.
Ll6
is
a
Transit
here
and
think
about
the
failure
of
link
between
lo1
and
lo4.
Yeah
lo5
will
become
Transit
anyway,
otherwise
you're
black
hole
ink.
So
even
the
corner
corner
case,
you
think
you're
solving
you're,
not
solving.
D
If
you
just
leave
those
things
not
being
forced
to
believe
you
achieve
exactly
the
same
behavior
that
you
get
with
all
the
changes
that
you
try
to
introduce.
Additionally,
here
so
I
mean
brutally
speaking.
I
think
this
is
largely
Superfluous.
It
could
be
done
right
at
the
top
of
the
fabric.
We
had
no
choice,
we
had
to
rank
those
things
and
we
had
to
get
the
flooding
Scopes.
D
If
you
want
to
do
that,
yeah,
okay,
we
just
repeat
the
extensions
for
the
top
of
the
fabric
rings,
so
you
can
basically
read
that
and
compare
with
what
you
try
to
extend
and
you'll
find
that
you're
pretty
much
doing
the
same
thing,
but
I
really
don't
see
the
value.
D
Could
be
done,
but
I
don't
see
the
use
case
that
you
ultimately
ultimately
are
solving
unless
you're
willing
to,
for
example,
black
hole
l04
if
lo1
lo4
links
dies
and
then
lo5
really
because
it
behaves
like
a
leaf
and
never
forwards
through
it.
It's
never
Transit.
But
then
you
will
black
hole
that
low
forward,
and
basically
you
don't
have
ring
resiliency
anymore
and
you
have
ring
resiliency
anymore.
If
you
just
leave
Rift
as
it
is,
you
will
get
it
by
yourself
by
itself.
Okay,.
B
A
D
D
A
A
Home
yeah,
so
when
link
between
L1
and
L4
breaks,
you
change
the
relationship
right.
D
Well,
the
ztp
will,
basically,
the
lo5
will
look
at
the
offers
it
got,
it
will
got,
l06
and
l04
lo4
will
withdraw
it,
offer
lo5
will
say
I'm
still
good
I
got
from
lo6,
whatever
level
I'm
one
below
the
lo4
or
you
realize
it
has
no
value
offers,
because
the
only
valid
offer
to
for
for
for
level
was
from
l01.
The
offer
is
gone.
D
D
D
Right,
so
you
need
all
these
special
flooding
Scopes,
just
like
the
top
of
fabric,
cannot
become
lower
on
top
of
fabric.
That's
why
we
need
the
ring
flooding.
So
now
the
zero
becomes
like
top
of
fabric.
You
have
to
do
all
the
ring
flooding
which,
if
we
take
the
top
flooding
rules,
Scopes,
which
is
kind
of
what
they
suggest.
Here's
extension.
It
will
work,
but
it's
just
overly
complicated
because
Normal
riffs
by
changing
the
levels
and
ordering
those
guys
will
provide
you
all
the
correct
forwarding
behaviors.
No,
the
link
breakages
here
yeah.
D
So
that's
so
that's
kind
of
my
point.
If
you
really
want
to
force
those
guys
all
on
the
leaf
and
zero
okay,
then
you
need
this,
but
why
why
you
do
that?
The
only
reason
would
be
you.
Don't
you
don't
want
it
to
ever
be
a
transit
for
traffic,
but
then
you
cannot
protect
loanlo4
failure,
because
lo4
has
only
yellow
five
as
Transit.
If
lo5
refuses
to
be
Transit,
you
have
no
ring
protection
anymore
right,
so
I'm
like
yeah.
D
E
This
situation,
so
I
think
that
we
should
not
use
dtp
here
but,
as
Tony
said
configured
prefix
redistribution
may
be
considered
rich,
so
I'm
not
I,
think
much
more
about
it,
but
I
will
so
I
think
that
if,
if
we
think
that
the
new
link
capability
will
broke
the
algorithm
of
the
drift
fundament
fundamention,
so
we
we
may
not
use
it,
but
we
have
found
another
way
to
solve
it,
though
it's
a
corner
case,
but
but
in
my
opinion
or
in
our
customer,
they
think
it
may
be
a
deployment
for
a
it.
D
D
Yes,
it
will,
because
you
just
have
to
disable
the
next
top
if
you
really
need
yes,
so
I
give
you
that
right,
I
mean
frankly,
the
ztp
runs
within
well-implemented
finite,
State
machines.
My
experience
runs
within
milliseconds,
so
at
most
tens
of
milliseconds
right
to
get
all
the
accelerated
hellos
and
Z,
because
we
put
the
ztp
on
the
hellos
and
the
hellos
get
accelerated
on
failures.
My
experience
is
it's.
You
know.
D
D
So
let's
assume
it'll
default
right,
so
you
cannot
install
in
lo4
D4
entry
that
has
next
stop
lo1
and
lo5,
because
that
may
lead
to
looping
behaviors
okay,
because
you
are
not
on
the
claw,
the
the
whole
you
you're,
not
in
valley,
free
routing
anymore,
the
whole
Rift
trick
of
being
able
to
take
any
path
is
because
it's
completely
value
free.
This
is
not
value
free.
This
is
flat.
You
can
go
up
and
down.
D
So
you
won't
be
able
to
accelerate
convergence
in
the
generic
case
on
this
little
little
thingy
that
you
show
here,
depending
how
you
treat
the
020503
links.
If
you
follow
Rift
today,
you
will
not
loop
with
your
extensions.
D
I
cannot
tell,
but
in
the
generic
case,
if
you
start
to
introduce
this
flag,
you
can
only
install
shortest
path,
because
you're
not
invalid,
free
routing,
anymore
you're,
not
guaranteeing
that
okay,
so
I,
don't
think
you
can
accelerate
convergence
either
practically
speaking,
depending
on
a
million
things
right,
you
can
pre-compute
stuff
and
accelerate,
but
practically
speaking,
I,
don't
think
you'll
be
gaining
much
okay.
So
that's
that's
my
input
to
the
whole
thing.
A
The
Sunday
so
I'm
completely
with
Tony
here,
look
in
the
implementation.
You
cannot
use
faster
hash
because
no
endpoint
is
the
network
is
a
key
distance.
They
cannot
have
array
of
next
hubs.
You
would
have
to
compute
something
like
look
free
alternative
which
Rift
not
suitable,
for
where
L1
would
be
primary
and
L5
backup
to
switching
Hardware.
Otherwise,
you
still
need
to
invalidate
next
stop
in
rips
and
re-download
the
routes
which
kind
of
defeats
the
point.
E
Yes,
I
think
from
our
experience
on
the
IDP
protocol.
We
think
that
the
protocol
converges
time
will
larger
than
the
backup
table
or
convergence.
So
we
think
about
the
solution,
but
it's
great
if
the
dtp
is
changing,
will
still
consume
so
little
time.
So
it's
great,
but
we
have
not
tested
on
it.
We
think
we
can.
We
can
do
more
workouts
about
it.
D
Cindy
so
I
liked
it,
but
why
don't
you
you
know
implement
this
I
mean
that's,
not
the
massive
change,
I
I
advise
you
to
look
at
the
ring
flooding
on
top.
So
you
verify
what
you're
doing
here,
because
you're
changing
flooding
Scopes,
which
is
cool,
but
it's
basically
the
same
problem,
so
you
can
Implement
a
variant
of
it
and
then
a
run,
a
rift
just
doing
the
election
in
your
case
and
run
this
and
see
what
is
the
difference?
D
I
mean
you
don't
have
to
bring
absolute
numbers
but
show
that
in
our
implementation
with
3
we
see
3x
Improvement
and
then
we
chat
but
I
think
you'll
be
a
little
bit
disappointed
and,
like
I
said,
don't
forget
unless
you
compute
Loop
free
Alternatives,
like
Jeff
said
you
really
go
into
igp
Loop
free
Alternatives,
that's
literally
what
you're
going
into
you
cannot
just
naively
install
the
other
one
lo5
next
stop.
So
you
have
somehow
either
extend
the
whole
Solution
by
LFA.
D
That's
that's
a
heavy
thing
or
and
don't
forget,
because
you're
flooding
southbound,
you
will
not
see
the
topology,
so
I,
don't
think
you
can
even
compute
LFA.
Okay,
you
cannot
understand,
you
don't
have
the
full
topology
LFA
is
very,
very
reliable.
You
know
reliant
on
the
fact
that
you
see
full
topology
so
see
what
happens
if
you
switch
a
rift
and
just
let
it
do
the
ztp
and
rebuild
the
adjacency
reload
and
what
you
can
get
out
of
your
extension.
D
If
you
lose
the
stuff
and
you
just
redistribute
the
prefixes
and
you
recompute,
the
prefix
to
the
neighbor
without
losing
adjacency
I
will
see
how
much
you
can
get
out
of
it
because
right
now
it's
you
know
bit
handwave.
You
can
get
your
point,
but
it
will
I
I
would
be
far
more
interested
in
the
solution.
If
you
show
me
very
significant,
you
know
half
order
of
magnitude
gain
in
terms
of
you
know:
failover
protection,
okay,.
E
So
I
think,
let's
be
don't
consider
the
solution.
We
just
consider
the
situation
is
this
situation
or
the
deployment
need
to
be
considered
or
or
this
situation
makes
sense?
How
do
you
think
about
it?
And,
despite
of
the
solution.
D
Oh
I
think
that
what
you're
showing
here
is
deployment
is
stuff
that
I've
already
seen
and
people
were
talking
to
me.
The
ring
deployment
at
the
bottom
is
absolutely
valid,
no
discussion,
so
the
topology
you're,
showing
is
reality
and
I
saw
the
stuff
like
this
I
saw
it
double.
Ringed
I
saw
already
all
this
stuff,
yeah
yeah,
the
the
topology,
is
absolutely
of
value.
No
discussion.
E
Okay,
okay,
we
can't
consider
about
it,
and
maybe
this
will
not
be
an
extension
for
rifter
or
just
list
some
solution
and
compare
it,
and
we
can
do
some
concluding
about
the
drift
fast
convergency
or
other
somethings.
E
D
So,
if
we're
really
looking
for
work,
okay
and
relevant
work,
these
rings
are
mostly
lambdas
from
what
I've
seen.
Okay
and
this
lambdas
are
building
either
at
the
bottom
or
the
top
of
the
fabric,
and
they
are
content
rotating
and
actually,
when
you
cut
the
link,
you
only
lose
half
the
lump
does.
So
it's
very
very
weird
thing
and
we
have
an
extension
for
Rift
that
deals
with
the
stuff.
D
If
we
want
to
go
into
this
direction
on
the
charter,
we
can
have
a
chat
with
problems
that
poses
and
what
kind
of
solution
that
needs.
So
we
could
start
to
work
on
the
stuff.
But,
let's,
let
me
be
honest-
the
customers
that
we
had
that
got
extremely
interested
wanted.
The
stuff
built
basically
pulled
out,
so
we
never
built
the
stuff
at
the
end.
D
But
the
problem
is
valid
when
you
as
when
you
start
to
deal
with
rodents
and
counter
rotating
rodents,
doing
Special
reuse
on
lamptas,
probably
to
cryptic
you
know,
but
yes,
the
ring
topologies.
Is
they
present
unique
problems?
Yes,
that
you
can
do
funky
stuff,
not
in
this
case,
but
at
the
top,
especially.
D
D
Yeah
but
it's
a
different
case
like
I,
say
it's
mostly
top
of
the
fabric,
when
a
lot
of
fabrics
are
being
hooked
up
on
a
on
a
big
counter,
rotating
rod
and
ring.
That
is
the
real
stuff,
because
there
you
really
need
the
lamb
dust
right
and
the
rodent
Behavior
starts
to
matter.
D
Do
you
just
more
and
more
levels?
That's
why
I
have
so
many
levels
in
Rift
and
the
ztp
is
so
fast.
It
really
doesn't
matter
see
the
whole
GTP
has
been
optimized.
So
when
you,
the
whole
GTP
is
built
asymmetric
for
for
the
failure
case
right.
So
it
comes
up,
it's
slower
when
it
goes
down.
It's
really
really
fast.
Imagine.
D
D
Right
so
you've
got
here
clear
structure,
yeah
yeah
I
mean
it's
at
least
I
mean
even
if
we
don't
do
so
those
extensions
it
already
merits
a
draft
just
to
talk
people
through
okay.
How
do
you
deploy
these
rings
and
what
are
the
operational
considerations
right?
So
you
want
an
implementation,
for
example,
with
horizontal
links.
Yes,
for
that
reason,
and
then
it's
absolutely
valid
operational
consideration.
What
will
be
the
convergence
times
right?
D
So
if
you
go
all
the
way
to
the
extreme,
you
could
actually,
in
the
leaf
case,
change
the
FSM
right,
like
Do,
not
drop,
keep
on
forwarding
and
so
on,
but
your
relief
life
will
be
good.
You,
you
readjust,
so
those
are
efficiency.
Improvements
for
ring
deployments,
but
yeah
those
are
valid,
consideration
or
discussion.
You.
A
Know
can
we
start
some
exchange
of
emails
on
the
list,
I
think
it's
a
valid
problem
and
actually
interesting
solution.
We
know
that
convergence
and
Rings
is
horrible
unless
you
use
tuna
link
or
some
label
stuff,
it's
very
painful.
So
if
we
can
solve
it
and
given
how
fast
drift
is,
you
know
it
could
be
a
good
solution.
D
Okay,
so
I
mean
I,
don't
have
customers
pushing
me
right
now
with
huge
Rings
right.
So
yes,
in
the
interesting
work
topic
will
be
like
kind
of
forward-looking
and
you've.
Send
the
you
know
that
really
pesters
them
okay
I,
would
still
encourage
first
in
operational
draft
and
then
seeing
what
numbers
we
have
on
the
convergence.
Then
we
start
to
consider
like
hacking
more
into
the
protocol,
which
works
fine
without
these
extensions
right
now.
Okay,.
A
D
Yeah
yeah,
okay,
so
up
to
the
chairs
to
drive
discussion
more
than
happy
to
participate.
Okay,
okay,.
E
It's
great,
if
you
can
provide
some
tests
test
the
number
or
analysis
about
the
converging
time
about
the
dtp,
because
we
can
test
of
audience
of
device
in
the
ring,
so
we
can
see
the
convergence
time.
What's
the
number
of
it?
Could
you
please
do
some
tests
on
it?
We
can
do.
D
It
also
yeah
I,
don't
have
cycles
for
that.
You
may
get
someone
like
Jordan
who
may
have
to
burn
some
Cycles.
You
want
to
play
I
simply,
don't
have
Cycles
to
give
you
those
numbers.
I,
don't
have
customers
pushing
me
on
the
stuff.
I
have
I,
have
converged
number
internal
numbers.
Obviously,
but
I
cannot
disclose
those,
and
for
that
case
we
could
do
something
and
show
it,
but
I
don't
have
the
cycles
for
it.
D
E
Okay,
okay
I
understand,
basically
that
it's.
E
D
D
Somewhere
between
20
to
100
millisec,
but
that
does
not
mean
anything
because
you
have
to
flood
out
first
sorry
for
my
French
right.
So
until
you
flooded
stuff
out,
you
don't
have
anything
anyway,
so
the
adjacency
bring
up
time
is
actually
not
even
that
interesting.
What
is
far
more
important,
do
you
have
the
database
all
right
and
then
we
start
to
talk
about
Healthcare
fast.
Can
you
flood
and
the
discussion
becomes
incredibly
complicated,
very
fast
because
it
depends
on
how
big
is
your
fan
out?
How
much
CPU
do
you
have
blah
blah
blah
right?
D
D
You
know
the
Qatar
guys
may
get
very
interested
in
the
stuff
because
they
have
the
full
framework
for
them
to
bring
the
stuff
and
play
with
the
stuff.
They
have
our
implementation.
You
know
they
have
access
to
it,
so
they
can
actually
convert.
No
compare
python
is,
of
course,
slow-ish
all
right,
but
you
they
can
play
with
the
python
they
can
play
with
our
stuff.
They
can
play
it
with
Isis
bgp,
just
basically
doing
just
like
this
right,
so
Katara
guys
poke
them.
They
love
this
kind.
A
D
Software
is
really
good,
I
think
tommaso
and
those
guys
will
be
more
than
happy.
You
know,
there's
a
bunch
of
phds
going
on
and
they
have
a
superb
primary
that
can
very
quickly
bring
the
stuff
up.
Okay
and
frankly,
I'm
doing
most
of
my,
like
you
know,
hammering
the
stuff.
Now
you
know
around
containers
in
their
framework,
not
even
using
you
know,
like
large
amount
of
machines
or
anything
like
that,
so
that
that
that
will
be
a
great
idea.
D
Let's
involve
Tomaso,
let's
start
to
talk
through
this
issue,
we're
in
quantify
the
ring
Behavior
versus
bgp
or
whatever
normal
Rift,
with
these
extensions
I
think
that
will
be
a
very
productive
thing
to
do
and
quantify
not
in
terms
of
like
five
milliseconds
versus
seven
milliseconds.
It's
like
okay,
what
is
the
total
load
message?
Wise
and
processing
wise,
and
you
know
a
failure,
scope
that
you
put
on
the
network
on
a
failure
in
each
of
those
scenarios,
because
this
is
then
what
tends
to
behave
better
right.
We
will
start
to
talk
about
these.
D
Implementation
is
a
little
bit
faster,
CPU
and
so
on
will
never
finish,
because
it's
just
basically
watching
a
lava
lamp,
but
the
way
they
pull
the
metrics
the
metrics
are
very
objective.
You
know
they
basically
discount
a
lot
of
these
variables,
which
are
very
hard
to
quantify.
Okay,
long
discussion,
very
productive
thanks.
A
A
The
by
Yokohama
potentially
so,
let's
start
this
discussion
after
this
ATF
and
see
how
we
converge
we
just
because
Jeffrey
will
be
more
than
willing
to
include
rings
in
a
new
Charter.
D
A
A
So
I
was
hoping
to
come
to
California
last
week,
but
and
they
even
had
an
agent
schedule
with
our
CTO
internal
meeting,
no
trouble
thanks.