►
From YouTube: IETF94-RTGWG-20151102-1520.webm
Description
RTGWG meeting session at IETF94
2015/11/02 1520
A
A
B
B
B
B
B
B
But
we
are
looking
for
November
is
to
get
LFA
manageability.
It's
been
in
editor
q.
For
quite
some
time
we
are
working
enlivened
with
some
other
drugs.
A
little
progress,
son
we've
got
remodel,
find
not
protection
in
routing
review.
They've
got
we've
got
really
good
comments,
but
they've
been
really
good
discussion
with
Stewart
and
outer,
so
they
are
getting
close
to
resolve
all
the
points.
B
B
C
B
E
Now,
in
some
cases,
you
do
not
really
want
that
to
happen,
because
the
shortest
path
towards
destination
might
not
be
the
best
part,
but
you
want
to
deliver
like
liquid
service
upon
the
packet,
so
you
actually
send
it
into
a
tunnel
or
on
a
particular
interface
for
Tito
scrubbing
whatever.
Now
you
can
do
that
right
now
and
in
the
main
way
of
doing
that.
Most
of
the
time
is
by
using
policy
based
routing.
E
So
the
draft,
what
I'm
going
to
be,
which
I'm
presenting
here
is,
like
you
know,
playing
in
the
idea
of
working
group,
but
some
of
the
implications
you
know
might
actually
hit
here
as
I
was
a
bigger.
You
know
routing
aspect
you
know,
and
some
of
the
context
might
actually
you
know
be
interesting
from
a
rocking
perspective.
E
So
the
first
thing
is
that
part
ID
created
an
incredible
amount
of
confusion
with
people
in
their
brains.
So
if
you
read
the
draft,
just
do
the
mental
swap
part.
Id
is
actually
in
direction
ID.
It
is
just
32
bits
or
128
bits
it
just
it's
a
big
data
and
it
will
identify
something
and
we
will
discuss
later
on
what
exactly
you
know
it
identifies
and
how
it
can
be
used.
E
So
again.
So
what
you
know
the
motivation
for
this
is
to
allow
route.
You
know
for
a
controller
to
tell
routers
what
to
do
with
traffic
and
where
to
send
the
traffic
at
all
which
tunnel
to
reduce
NT
traffic.
What
we
are
not
trying
to
do
is
to
signal
any
tunnel
encapsulations
or
set
up
tunnels,
or
you
know
these
kind
of
things.
E
So
when
you
look
into
PBR,
you
know
it
actually
has
like.
You
know
two
main
task:
if
you
configure
the
PBR
rule
the
first
thing:
what
you
do
is
you
configure
much
criteria,
so
you
actually
identify
the
traffic
that
interests
you.
The
second
thing:
what
you
do
is
you
identify
what
you're
going
to
do
with
the
traffic
and
those
could
be
two
things.
The
first
thing
is
you
condition
the
traffic,
so
you
pull
easy
to
shape
it.
Yuri
color
it.
E
You
know
on
all
those
things
and
the
other
thing
what
you
do
in
the
policy
based
route.
Is
you
send
traffic
to
an
interface
to
a
tunnel
and
so
on
words?
So
if
you
look
into
flow
spec
itself,
then
you
know
the
matching
is
very
well
covered.
Traffic
conditioning
is
very
well
covered,
but
the
traffic
steering
itself
is
only
covered
to
a
degree.
You
only
do
traffic
steering
into
the
vrf.
You
don't
really
go
beyond
that.
E
There
are
being
some
extensions
they're
having
some
extensions
to
it
and
so
not
towards
IP
address
so
in
essence
and
elsewhere
in
direction
ID
comes
into
play.
It
will
it's
actually
a
mechanism
which
will
allow
you
know
state
controller,
those
signal
towards
receivers
on
where
to
send
the
traffic
which
matches
you
know,
which
you
know,
which
thing
should
be
redirected.
So
this
is
would
be
a
wonderful
controller
and
we
have
our
matching
criteria.
What
you
see,
if
so,
in
this
picture,
we
actually
have
like
four
flows
back
rolls:
whoo,
1,
2,
3
and
4.
E
Each
of
them
actually
has
their
own
match
bacteria,
and
it
already
has
based
upon
RFC
5575,
some
conditioning
information
like
policing,
recoloring
and
so
on
words.
So
what
we're
going
to
do
now
is
we're
going
to
add
a
new
community
which
would
be
the
indirection
ID
again
32
bits,
128
bits,
opaque
data
and
it's
actually
an
abstract
reference
to
an
indirection.
So
the
controller
sends
this
thing
to
all
office,
which
appears
so
now.
E
These
peers
actually
have
this
information
so
now,
and
they
all
receive
the
same
for
the
particular
flow
they
will
receive
the
same
indirection
ID
so
flow
one
rather
bong
indirection
id1
powdered
for
another
14
rule
number
two.
It
gets
like
in
direction
attic
two
or
after
24.
You
know.
First
flow
set,
it
will
get
in
Direction,
id1,
Falcon
3
will
get
for
the
first
flow
set
also
in
the
region,
ID
1.
E
So
how
do
these
routers
actually
use
it?
So,
and
that
is
where
you
know
the
main
reason
we're.
You
know
why
I
wanted
to
introduce
this
work.
Also
in
this
particular
working
group
is
because
we
create
like
a
new
table
here,
which
is
the
indirection
table,
so
on
the
router
one,
for
example,
we
actually
create
a
table
which
is
indexed
based
upon
this
in
direction
like
this,
and
next
to
that
we
have
information
like
indirection.
E
Id
means
you
know,
could
be
like
a
segment
routing
tunnel
now
recurse
that
one
in
direction
ID
to
could
be
like
an
EVP
n
es.
I
redirect
traffic.
The
third
one
could
be
manually
configured
the
fourth
one,
it's
just
an
IP
routing
table
lookup.
So
these
are
things
you
know
you
actually
place
in
this
indirection
table.
Now.
How
does
the
table
get
populated
is
true,
whole
set
of
different
kind
of
mean,
so
the
table
itself
is
decoupled
from
any
flow.
E
Specs
signaling
at
all
flow
spec
just
signals
an
ID
and
the
table
itself
is
like
independently
populated.
So
this
means
you
could
use
you
know
by
existing
technology
could
be
a
traditional
routing,
could
be
netconf
yang
could
be
xml,
/
XMPP,
you
know
manual
config,
all
these
kind
of
things,
but
it
gets
populated.
So
this
table
can
be
different,
/
router,
so
that
means
order
out
of
one.
For
example,
my
in
direction.
Id
1
means
it's
a
segment
rabbit
tunnel
going
from
router
one
two
or
three
out
of
five
to
our
charge
rate
or
router.
E
To,
however,
you
know
the
in
direction
ID
one
here
it
says
like
okay,
it's
also
segment
routing
tunnel.
There
is
going
from
rapid
to
towards
r
after
72
hours,
heart
rate
on
the
third
rail
trail,
Bracken
number,
three,
the
bottom
one
in
direction,
ID
one
actually
just
like
okay
I
need
to
redirect
traffic
here
towards
the
knowledge
ppt
eternal.
You
know
going
to
router
nine
to
one
strand
rate,
so
in
direction
at
least
the
same,
the
tunnel
information
itself
is
different,
but
device.
E
The
first
thing
is
on
the
router:
we
have
this
indirection
table
so
router
to
as
a
table.
What
you
see
there
are
two
tree
as
a
table.
What
you
see
that
these
tables
are
different,
the
indexes
are
the
same
in
direction
we
the
same,
but
the
way
it
is
an
old
increment
for
which
the
directive
stands
for
is
actually
different
earth
box.
So
imagine
or
router
to
we
have
a
packet
entering
so
we
identify
the
packet.
E
We
look
into
the
header
and
we
say
like
okay,
this
matches,
you
know
the
stuff
from
flow
spec
rule
number
one
we
do
a
match,
but
then
for
a
year
you
know
the
packet
needs
to
be
redirected
into
this
segment
dragging
tunnel
going
from
to
router
seven
and
then
to
router
eight
and
we
ship
of
the
packet
or
router
tree.
However,
assume
we
also
get
a
packet
and
it
also
matches
flow
spectral
number
one.
It's
just
like.
Oh
yeah,
you
know
it's
in
direction
here,
but
the
router
tree
it
actually
stands
for
the
K.
E
This
tunnel,
you
know,
is
actually
now
pointing
towards
an
rsvp-te
tunnel
from
Rapid
nine
towards
flat
rate
and
think
about
these
tunnels.
You
know
already
set
up
in
your
environment,
you
know
they
are
up
and
running.
So
that
is
what
we
have
here.
So
with
one
in
direction
identify
you
know
we
identify
like
different
tunnels
going
through
the
network
itself.
This
allows
you
that
on
each
of
these
you
know
routers
consuming
the
flow
spectral
that
the
next
hop
could
be
either
IP
address
could
be.
E
E
So
the
way
it
is
actually
constructed.
So
it's
a
it's.
A
community
32-bit
128
bits
value,
one
of
the
two.
We
have
a
cbt
copy
bit
in
there
if
you
only
want
to
redirect
traffic
or
if
you
want
to
send
a
copy
of
the
traffic
to
the
redirection
part,
and
you
have
a
thing
what
it's
called
like
tid
tunnel
ID,
which
means
you
know
in
this
case
only
reserved
like
two
bits,
meaning
you
have
like
four
different.
Unlike
these
in
case,
you
want
to
use
like
a
nest
of
tunnels,
not
between
each
other.
E
You
know
for
the
use
case,
actually
you
you
might
have
in
mind,
but
coming
back
to
this,
you
know
just
going
back,
because
there
is
more
to
focus
of
why
I
presenting
this
here.
That's
one
minute,
one
minute!
That's
my
last
slide
so
creating
this
tunnel
ID
table.
You
know
we
have
like
different
means
of
the
ring
data,
so
I
think
initially
it's
probably
is
going
to
be
like
manual
configuration
to
do
this,
because
you
only
have
like
a
limited
amount
of
tunnels
on
your
boxes.
You
pre-configure
them
in
the
future.
E
The
whole
thing
will
probably
be
orchestrated
now,
some
of
the
things
what
I
put
on
here
is
like
on
the
other
screen.
It's
like
how
it
could
be
populated
could
be
based
upon
extensions
to
routing
protocols,
and
that
is
you
know
where
I'm
gonna
leave
the
room
and
to
the
Microsoft.
Yes,
it
would
be
a.
F
Real
mean
from
me
I
are
concerned
related,
oh.
F
E
G
So
k,
patel,
cisco
speaking
as
a
quarter,
you
want
to
think
of
heterogeneous
forward.
You
want
to
think
of
heterogeneous
networks
with
multiple
different
forwarding,
cleans
and
or
you
want
to
think
of
a
network
with
a
multiple
tunnel
encapsulation
and
just
as
you
have
an
underlay
and
in
overlay.
You
want
to
think
of
this
as
sitting
between
an
underlay
in
and
overlay
doing,
a
tunnel
abstraction
layer
and
so
as
to
form
a
tunnel
table
right.
So
it
hides
then
I'll
encapsulation
of
different
types
completely
through
the
indirection
ID.
F
Yeah
other
my
question:
you
that,
as
a
ton
of
the
encapsulation
can
be
hide
either,
but
the
problem
uses
a
unity,
traitor,
I'd,
indirection
ID
must
be
seen
by
the
server
or
the
controller,
so
I
mean
so
how
you
implement
at
this
one
configuration
and
also
I
propose
the
user
case
of
the
dominie's,
if
they
say
the
controller
or
the
tradition
of
the
flow
sebago
server,
so
how
the
server
or
the
four
as
a
further
server
other
controller,
how
to
get
to
know
the
in
direction
ID.
That's
my
microphone.
E
So
I'm,
yes,
also,
the
setup
of
the
ton
of
the
self
is
actually
completely.
You
know
beyond
the
scope
of
this
piece
of
work.
So
the
way
you
know
the
tunnel
in
order
in
the
in
direction
ID
is
the
in
direction
ID.
You
know.
If
you
set
up
a
tunnel
you'll
sign
in
that
in
direction.
I
did
you
can
do
it
manually?
You
can
do
it
orchestrated,
so
it
doesn't.
It's
completely
open,
don't
see
where
you
see
the
problem.
Okay,.
F
So
I
think
that's
that's
a
user
important
unusual
for
the
user
kisser
and
also
how
to
deploy.
This
is
the
feature
in
the
controller
or
the
other
configure
other
philosophy,
a
server,
because
you
will
I
think
this
really
important
away.
So
the
mail
last
the
comments
you
that
I
proposed
the
ceremony
routine
past
programming
in
the
spring
working
group
I
think
the
second
idea
has
also
had
also
the
minion
to
represent
her.
The
next
hopper
represent
I'll,
be
no,
please.
H
Yeah
the
snow
hats
so,
as
I
mentioned
an
email,
this,
the
in
direction,
ID,
I
think,
is
exactly
the
next
top
ID,
which
is
client
assigned
in
the
eye
to
RS
rib
yang
model.
I.
Think
it's
exactly
what
you
want.
It
does
absolutely
this
and
you
should
really
take
a
look
at
it
than
that
gang.
Well.
You
can
obviously
create
the
next
hops
using
rest,
conf
next
conf,
whatever
modification
of
I
to
RS,
that
we
have.
H
C
A
G
Kale
patel
cisco
one
more
time
just
to
clarify
it
may
not
be
ephemeral.
Also,
it
depends
on
the
policy.
B
So
now
heads
not
surprisingly,
we
are
presenting
in
pc
working
group
dropped,
which
is
exactly
the
same
terminology.
I
would
call
indirection.
Ideally
it's
a
policy
ID,
you
measure
the
policy
a
do
you
do
something
with
it.
So
we
also
talk
to
you
from
my
tourists
will
start
working
across
working
groups
on
creating
abstract
that
policy
model,
so
it
could
be
distributed
through
net
content
or
any
other
way.
D
My
name
is
a
mad
person
d
I'll
be
presenting
the
bgp
pic
next
slide.
Please,
where
I
can
do
to
me:
okay,
good,
so
I'll
talk
about
the
problem,
the
solution
and
one
thing
that
one
thing
that
I
got
comment
about.
An
edited
in
the
next
version
was
how
to
handle
hardware,
limited
capability
and
I'll
the
use
of
this
solution
and
its
applicability
in
recovering
from
various
failure
scenarios
problem.
D
So
I'm
not
sure
if
I
should
say
this
or
not,
but
bgp
is
kind
of
slow
because
a
serial
nature
things
have
to
be
understood
and
then
sent
to
the
other
one.
The
objective
of
this
architecture
is
to
be
able
to
recover
from
various
failure
scenarios,
as
well
as
from
venice
up
scenarios
in
time
complexity.
That
does
not
depend
on
the
number
of
bgp
prefixes
and
to
extend
this.
This
is
not
specific
to
bgp
prefixes.
D
It's
also
lickable
to
sue
the
wires
and
to
anything
that
basically
has
a
level
of
indirection
I
have
a
few
terms.
Basically,
a
leaf
is
a
prefix
configure
the
very
takes
the
path.
Everybody
knows
what
the
path
is
either
a
next
top
or
an
extra
plus
an
interface
a
path.
This
is
an
array
where
one
of
them
is
picked
based
on
some
sort
of
load.
Balancing
the
out
labeled
array
is
the
label
that
you
push
so
labeled
number
one
won't
be
pushed.
D
If
you
pick
Path
number
one
labeled
number
two
would
be
pushed
if
I
pick
Path
number
two
and
so
on
the
last
one
is
the
adjacency.
It's
the
basically,
they
are
two
encapsulation
solution.
So
bgp
pick
is,
to
be
honest,
is
not
a
bgp
feature.
It
is
actually
a
fifth
feature
and,
as
I
just
mentioned,
it
works
for
sudo
wires.
It
did
not
have
bgp
at
all.
D
The
idea
is
to
construct
the
forwarding
chain
in
a
hierarchical
and
share
and
do
the
hierarchy
and
sharing
as
much
as
we
can
so
that
we
avoid
having
to
change
too
many
things
for
one
small
thing
once
one
small
thing
changes
the
nice
thing
about,
it
is
that
it
is
internet
behavior,
it's
completely
transparent,
it
is
incrementally
deployable.
You
don't
have
to
have
it
on
all
your
router's.
D
This
is
very
a
simple
example.
If
I
want
to
construct
the
forwarding
chain
here
are
the
two
destination.
This
is
destination
number
one,
and
this
is
destination.
Number
two
here
are
the
two
eat
usps.
Now,
if,
if
you
look
at
eager
speedy,
resolves
via
an
igt,
prefix
internally
learn
dryer
biosci
sauce
be
f,
and
if
I'm
running
segment
tracking
your
ldp,
it
will
also
have
labels,
on
top
of
it
path,
number
to
resolve
dry
and
I,
another
igp
prefix,
and
it
has
different
patterns.
D
Just
two
and
important
thing
is
that
the
label
here
is
different
from
the
label
here,
so
even
though
they
share
the
same
set
of
packets,
we
have
different
labels
because,
of
course,
the
local,
if
I'm,
using
just
VPN
or
even
/
next
up
the
the
labeled
advertised
by
this
guy
certainly
may
be
different
from
the
label
advertised
by
this
guy.
This
is
just
a
forwarding
chain.
If
you
see
I,
I
did
a
lot
of
sharing
and
a
hierarchical
and
we'll
see
how
we
can
benefit
from
this.
D
One
note
that
I
got
when
I
published
the
first
version
is
how
do
I
handle
the
case
where
I
have
limited
number
of
recursion
levels
number
of
hierarchy?
Let's
say
my
platform
supports
only
two
levels
and
I
have
three
levels.
So
I
came
up
with
an
example
in
the
version
that
I
did
it
in
a
very
quick
manner
and
it
wasn't
Lee
because
I
was
tight,
like
I.
I
did
a
couple
of
hours
before
the
deadline,
so
I
I'm
gonna
use
a
different
example.
D
Here
this
example
looks
scary,
but
it
is
not
actually
it's
a
simple
internet
option
see.
The
only
difference
is
that
in
case
of
option
C
to
avoid
having
3822
recursion
levels
here,
these
a
SVRs
would
advertise
the
core
routers
of
this
domain
into
the
igp.
Here.
What
I'm,
proposing
to
add
an
additional
level
of
indirection
is
that
this
a
SVR
would
advertise
the
core
routers
here
via
bgp
to
this
done.
So
that
would
just
add
another
level
of
indirection
is
fine
and
again
I
have
the
same
destinations.
D
The
same
English
PE
exception,
I
chose
three
years:
p
is
I
could
have
done
with
two,
but
basically
it's
just
something
that
I
did
a
bit
of
a
hurry.
If
you
look
at
the
structure
here
again,
I
have
three
levels
in
direction.
This
is
the
destination.
These
are
the
Pease,
and
these
are
the
SVRs
again
I'm
trying
to
share
as
much
as
I
can.
If
you
look
here,
I
am
sharing.
These
SVRs
are
shared
by
all
the
core
routers
here.
Each
one
of
them
has
a
separate
label.
D
These
bees
could
be
shared
each
each
of
these
pathways
could
be
shared
by
several
thousand,
even
several
hundred
thousand.
This
is
basically
a
forwarding
chain
if
I
have
three
levels
of
hierarchy:
one
level
here
at
second
level
year
and
a
third
level.
Dear
now
suppose
my
platform
supports
only
to
my
86
supports
only
two
levels
in
my
packet
budget
I
can
only
do
two
levels
of
lookups.
So
how
do
I
do
this?
We
need
to
do
some
sort
of
flattening,
so
I
took
this
guy
and
I
flattened
it
into
these
two
paths
lists.
D
So
I
ended
up
with
two
flattened
path
lists.
If
you
see
these
two
blue
or
collapse
into
one,
these
two
green
were
collapse
into
one.
How
do
you
flatten
it?
To
be
honest?
It's
a
lot
of
details
and
I.
Don't
think
that
I'll
be
able
to
cover
in
this
this
presentation.
But
the
point
here
is
that
I
I,
as
I
will
show
later
I
did
lose
a
little
bit
of
my
sharing.
I
did
lose
a
little
bit
of
my
sharing
okay.
Now,
how
do
I
use
all
of
this
now?
D
The
objective
of
all
of
this,
as
I
said,
is
to
recover
as
quickly
as
possible
in
a
complexity
that
does
not
depend
on
the
number
of
bgp
prefixes.
So
I'll
cover
a
certain
central
core
scenario.
I
had
a
path
and
something
broke
here,
so
I
need
to
change
that
internal
path.
I
did
not
lose
reach
ability
to
the
ESPE,
so
I
still
have
to
ego.
Sp
is
how
do
I
recover
what
happens
here?
So,
let's
look
at
the
forwarding
chain.
This
is
the
original
forwarding
chain.
D
I
was
going
via
interface
number
one,
the
link
broke,
so
what
I
did
is
that
I,
the
igp
recalculated
the
path
towards
the
same
PE
and
found
next
up
number
two.
What
happened
here
I
only
had
to
change
this
part.
I
did
not
have
to
change
any
anything
up
here.
This
could
be
again
a
hundred
thousand
leaves,
but
what
happened
is
that,
even
though
a
hundred
thousand
leaves
are
taking
now
a
different
path,
100,000
traffic
flows
are
taking
a
different
path.
Yet
I
only
changed
one
small
piece
which
is
this
fastest.
D
Let's
look
at
the
second
scenario.
Let's
say
I
have
a
primary
backup.
So
let's
say
this
is
my
primary
eager,
p
and
I
had
this
as
a
backup
PE.
Let's
say
something
happened
that
I
no
longer
reach
can
reach
this
guy,
so
I
need
to
use
the
backup
PE.
Now
all
traffic
destined
to
this
next
to
this
destination
needs
to
go
via
a
new
PE.
Originally
it
was
going
on
by
a
PE
one.
Now
it
is
going
out
by
a
PE
2
again,
I
could
have
100,000
a
place.
D
How
do
I
fix
this
in
a
time?
Complexity,
that
does
not
depend
on
the
number
of
these
destination.
Let's
see
what's
happening
here.
This
is
my
forwarding
chain.
This
is
my
backup
PE.
This
is
my
primary
PE.
Now
remember,
each
of
these
VPN
leaves
and
again
I'm
using
VPN
leaves
because
they
are
labeled
has
a
different
outgoing
label,
which
means
this
guy
here
advertise
an
outgoing
late,
but
that
is
different
from
this
guy
and
I
could
have
here
a
hundred
thousand
of
this.
So
how
do
I
do
this?
D
Originally
we
used
to
follow
this
path,
which
means
we
matched
an
IP
packet.
I
push
this
VPN
label.
Then
I
push
this
ldp
level
and
goes
out
here
now
suppose
this
guy
dies.
What
do
I
do?
All
we
have
to
do
is
to
correct
the
staff
list.
The
reason
is
that
all
leaves
that
were
looking
at
this
path
list
are
sharing
the
same
set
of
toughness,
so
all
I
need
to
do
is
to
fix
one
practice
only
or
maybe
fume
10
15,
20
30,
but
definitely
not
100,000,
definitely
not
two
hundred
thousand.
D
The
third
part,
which
is
pece,
link
protection
so
on
p2,
I
pre-programmed,
a
backup
path,
saying
that
if
this
link
failed,
please
leave
out
the
packet
this
way
so
on
this
link
failed.
Now
my
primary
now
remember
on
visiting
days
it
takes
a
little
bit
of
a
while
until
PE.
One
knows
that
the
link
has
failed,
especially
if
I'm
using
next
hop
self,
so
when
they
take
a
little
bit
more
than
a
small
time.
I
need
this
guy
when
it
receives
the
traffic
and
the
text.
D
This
link
has
failed
that
it
would
rear
out
the
traffic
via
the
second
PE
and
PE
2
would
definitely
send
the
packet
out
this
failure
scenarios.
It
relies
a
reliable
local
failure,
detection
we
can
do
it
in
50,
millisecond
or
sub
50
min
a
second
time
frame.
Let's
see
how
we
construct
the
forwarding
chain
again
on
this
node.
D
So
again,
I
have
a
very
similar
routing
chain.
Again,
the
same
sharing.
The
only
difference
is
that
the
backup
label
is
unlabeled
on
this.
When
this
node
sends
the
packet
out
at
sends,
a
packet
has
unlabeled,
but
if
it
pushes
it
back
to
PE
2
is
to
push
the
VPN
label
that
p
III
advertised
for
this
prefix,
so
forwarding
chin
is
very
similar,
except
that
I
have
a
different
label
structure
again.
This
node
fails
this
field.
This
is
petey
no
longer
reachable.
D
All
we
have
to
do
is
to
ear
out
the
traffic
via
the
second
p.
Again,
what
I
did
is
I
changed
a
single
pathways.
I
did
not
have
to
change
a
possibly
very
large
number
of
bgp
prefixes
this.
This
work
again
as
I'm.
Try
as
I
mentioned,
is
applicable
to
both
bgp
and
anything
else
could
be
sued
words
that
BTW.
I
D
Could
be
anything
else,
although
in
this
draft
we
focus
only
on
bgp.
One
last
thing
is
the
benefit,
or
how
do
we
gracefully
handle
the
the
loss
of
a
number
of
levels
of
recursion?
So
remember
this,
this
scenario,
I'm
not
sure
if
it
is
realistic
or
not,
but
we
don't
like
this
is
not
the
point
here
that
I
had
here
three
levels
of
recursion
suppose
this
a
SVR
dies.
D
So
this
is
my
forwarding
chain.
This
is
my
original
forwarding
chain
before
they
SVR
dies
when
it
died,
I
only
need
to
change
one
purpose.
Only
in
the
entire
hierarchy,
I
did
not
need
to
change
anything
else.
Just
because
something
failed,
I
need
to
change.
One
thing
only
could
be
few
again.
As
I
said:
it's
not
going
to
be
one,
but
it
is
not
going
to
be
a
hundred
thousand,
maybe
20
30,
maybe
100
220.
D
Pick
point
here
is
that
which
you
can
gracefully
recover
from
the
loss
of
your
hierarchy
of
levels
of
basic.
However,
eventually,
if
you
don't
have
enough
hierarchy,
you
will
end
up
losing
completely
bgp
pick.
This
is
my
last
slide.
Basically,
so
what
did
we
do
here?
We
just
have
a
simple
design
that
actually
runs
very
efficiently
could
run
on
a
simple
line
card.
It
works
for
any
recursion.
What
does
v
GP
or
anything
else?
D
I
D
I
Okay,
the
world
is
more
complicated
than
that
right,
frr
place
in
inter
chelsea
redundancy
people
have
all
kinds
of
interesting
things
that
are
on
top
of
the
whole
thing.
So
it's
not
like
you
can
take
that
and
build.
You
know
world-class
router,
okay,
yeah.
B
From
our
perspective
will
be
informational
in
every
event.
They're
here
built
very
similar
solution.
I
think
would
be
very
useful
for
service
provider
for
people
who
try
to
understand
how
it
works
and
I
would
really
like
I've
met
to
kind
of
enhance
this
for
explain.
Why
do
you
need
nesib
redundancy
in
BGP?
Otherwise
it
doesn't
work.
Why
do
you
need
to
know
Cindy
see
protection
and
it
would
be
extremely
beautiful?
Dancer
service
providers
have
deployed
this
technology
and
different
vendors
are
from
informational
perspective.
It
is
very
needed
work,
so.
I
B
D
J
Hello,
this
is
Pushpa
sheesh
from
juniper,
I'm
here
to
present
the
first
version
of
this
draft
called
micro
loop
prevention
using
spring.
So
here's
the
agenda.
We
call
will
go
through
the
problem
statement
and
a
little
bit
of
solution
and
then
cover
some
facts
about
multiple
events
and
ring
partial
deployment
and
the
protocol
extensions
the
need
for
it
that
we
have
proposed
in
this
chart.
J
So
to
start
with
a
problem
statement,
we
have
taken
an
example
network
to
show
whoa.
You
know
where
how
microbes
can
form,
but
essentially
micro
loops,
are
nothing
but
transient
loops
causing
cause
because
of
you
know,
unsynchronized
fav,
updates
on
different
nodes,
and
so
and
and-
and
there
are
facts
that
certain
properties
are
more
prone
to
micro
loops,
especially
the
ring
type
of
topologies.
J
So
in
this
example,
if
the
the
link
between
node
3
and
4
goes
down,
we
can
see
there
will
be
some
microbes
occurring.
So
here
we
see
the
new
S
paved
paths
which
is
supposed
to
happen
after
the
nigp
convergence,
and
in
this,
if
you
really
see
there
is
not
only
a
local
max
look
attached
to
the
North
3,
there's
also
some
remote
macro
loops
between
the
note
29,
1
and
6
right.
J
So
essentially
this
the
purpose
of
his
adoptive
SOG
might
look
and
and
using
spring,
but
but
this
solution
is
based
on
RFC
five-cent
15
in
oscillation,
which
is
called
the
near
end.
Tunneling
solution
and
the
basics
of
that
is,
is
found
other.
This
proposal
is
founder
of
the
fact
that
the
tunnel
to
from
any
you
know,
andressa
soares,
router,
all
the
way
to
the
nearest
peeler
is
always
unaffected
by
the
link
event
link
or
not
even
work
right.
J
So
based
on
this
fact,
we
come
up
with
the
solution,
and
this
is
in
the
poll
in
the
next
few.
Slides
I'm
would
actually
go
through
a
series
of
events
which
this
draft
car
and
his
proposals.
So
the
first
point
first
step
is
for
the
solution
to
work.
The
all
the
notes
should
be
configured
with
the
single
inner
max
convergence
delay
and
Bay,
and
using
that
value
there
are
a
couple
of
steps
that
the
solution
proposes
again.
J
So
when
the
link
goes
down
you
basically
again,
the
assumption
is
that
the
the
plr,
the
nearest
Paylor,
will
have
some
form
of
you
know,
backup
using
LFA.
So
in
this
case,
say
I'm,
just
looking
at
node
3,
which
is
which
is
supposed
to
be
the
their
nearest
paler
for
all
the
discs,
all
the
sources
on
the
left-hand
side.
J
If
it
is
changing
instead
of
changing
to
the
new
next
hops,
they
will
tunnel
the
traffic
to
the
to
the
old.
The
nearest
scale
are
basically
nearest
paler
on
the
old
SPF
path.
So
that's
what
this
orange
no
troubles
look
like,
and-
and
this
will
keep
this
will
keep
like
this
for
the
next.
You
know
the
DAR
integral
of
time,
which
is
equivalent
to
d.
So
after
time
d,
each
of
these
non
plr
nodes
will
start
converging
to
the
new
new
escape
path,
and
this
is
something
hautelook
will
look
like.
J
K
J
I
J
J
I
J
J
Yep
I
think
this
is
the
slight
propping
will
answer.
So
if
you
really
see
before
the
event,
which
is
before
t0,
they
were
all
using
a
single
you
know.
Kind
of
a
tunnel
to
the
destination
destination
is
again
0,
0
5
and
in
sahaja,
Yogi's
thousand
to
two
thousand
five
and
index
is
like
this
similar
to
the
what
the
node
numbers
are.
So
there
are
two
kind
of
things
I'm
trying
to
show
you
here.
J
J
There
should
not
be
003
a
002
so,
but
you
actually
are
using
when
it
instead
of
one
we're
going
to
use
two
labels,
so
first
label
is
to
take
you
to
the
nearest
peeler,
which
is
cow
thousand,
for
you
know
thousand
three
here
and
then
the
second
one
is
basically
the
the
label
from
from
the
thousandth
or
basically
003
to
the
final
destination.
005.
J
Ok,
so
the
the
other
Taiwan
opponents
want
I
wanted
to
show
here
is
that
the
node
0
0
8.
It
doesn't
find
a
change
in
next
top
because
of
SPF
after
running
SPF,
so
it
doesn't
do
anything
yeah.
It
just
keeps
on.
You
know
sending
the
same
thing
and
the
third
row
you
see
in
this
table
is
basically
you
know
how
the
nearest
plr
reacts
to
the
yield
three.
C
C
Like
a
little
its
I
don't
office,
quite
1
a-c
lindam,
a
cisco
I
seal
this
first,
but
what
happens
when
d-max
convergence
timer
times
out?
What's
to
keep
you
from
having
micro
loops
at
a
second
order
between
the
between
those
all
the
nodes
that
are
tunneling
to
know
its
rate?
I
think
this
is
what
so.
J
C
J
J
J
This
is
a
slide
on
handling
multiple
events.
We
have
added
text
in
the
draft
for
how
to
handle
multiple
events
as
well,
and
there
are.
There
are
cases
which
you
know
we
have
called
out
that
it
won't
be
not
be
handled,
for
example,
this
multiple
network
events
which
are
not
part
of
the
same
acer
LG,
you
know,
cannot
be
handled,
and
then
we
have
some
detailed.
J
B
M
Another
can
arrange
yes,
we
need
the
solution
for
microbes,
otherwise
it
kills.
The
benefits
of
phosphor
would
because
you
have
protected
your
traffic
in
50
milliseconds,
but
then
you
have
a
GP
command
runs
and
you
may
have
microbes
for
five
rolls
or
250
500
milliseconds
of
two
seconds.
So
if
you
can
focus
reboot,
you
should
care
for
France
conversions
and,
although
you
should
care
for
microbes
or
avoidance.
B
M
K
I
K
It
makes
sense
and
like
a
forwarding
mechanism
and
spring
like
how
it
is
like
when
a
marty
was
first
done
like
it
was
then
for
ldp,
the
mine.
So
those
things
like
there
are
two
options
provided
like
how
it
fits
into
spring
and
weather
weather
like
there
is
a
tlf,
a
hand
like
of
our
better.
It
can
complement
the
LFA
and
what
I
gb
support
is
needed
and
conclusion
so
so.
K
Because,
like
when
you
see
a
marty
was
done
for
ldp
scenario
like
a
jeep
extension
is
divided
defined
and
the
same
extensions
with
the
added
few
extra
thing
we
can
support
it
for
spring
and
then
like.
Since
there
is
a
a
topology
label,
you
can
define
for
our
tree
topologies,
so
that
is
red,
blue
and
rainbow.
We
can
easily
support
a
amarte
in
spring
and
your
marty
is
not
completing
the
LFA.
It
complements
so
having
too
good
for
our
algorithm
for
a
spring
is
not
a
bad
thing.
K
I
came
and
chains
are
very
minimal
and
for
spring,
like
there
are
two
options
provided
in
ldp
extension.
That
is
like
one
label
we
are
indicates
both
prefix
and
empty
Eddie.
Other
one
is
like
a
separate
label
for
our
empty,
aiding
or
a
prefix
label,
so
these
two
can
be
easily
supported
in
spring,
so
this
would
be.
The
label
stack
like
first
first
option
is
like
a
wreck:
traffic's
label
for
a
destination
where
you
want
to
redirect
the
packet
and
then
the
followed
Bank
normal
payload.
K
K
When
case
of
like
default
profile,
we
have
a
fixed
mt
IDs
that
is
like
red,
blue
and
rainbow
that
is
like
allotted
by
IA
and
may
in
case
of
non
default.
We
need
to
have
a
consensus
like
okay.
What
empty
will
be
using
like
what
free
MTD
will
be
used
across
the
domain
for
all
three
and
the
same
thing
has
to
be
advertised
and
I
feel
like
it's
useful
I
like
there
are
cases
where,
like
both
tlf
and
my
tias
advantage
and
disadvantages,
so
in
case
of
mr
t,
like
computational,
is
very
efficient.
K
C
L
L
Okay,
so
when
we
were
defining
the
ospf
and
access
data
model,
we
realized
the
requirement
of
all
local
government
kitchen
model,
so
the
the
routing
protocol
can
use
it
for
authentication.
L
The
kitchen
model
is
defined
as
a
list
of
keys.
Each
Key
has
a
key
string,
descent
except
lifetime
and
also
the
crypto
algorithm
related
with
the
key
they
say
the
model
has
been
presented
before
so
the
overall
structure
stays
the
same
with
Ian
change
it.
So
what's
new
this
time
last
IETF.
The
major
comments
we
got
is
the
kitchen
model
doesn't
support
any
encryption
for
the
key
string.
So
we
I
did
a
key
encryption
using
the
advanced
encryption
standard
key
right.
L
L
L
Authentication
and
encryption
policy,
it
can
be
used
by
it's
already
being
used
by
ospf
and
I
access,
and
can
we
also
used
by
other
route
other
protocols
and
routing
protocols
when
we
design
them
model?
We
made
it
the
way
in
a
way,
so
it
can
be
easily
used
and
augmented
and
so
far
we
think
the
model
is
much
foreign
complete
and
we'd
like
to
request
a
working
group.
Adoption.
B
On
the
third
day,
same
drug
is
going
to
be
presented
in
security
area
and
grace,
encirclement
and
hopefully
they'll
agree.
It's
important
and
I
think
in
this
area
we
all
agree
it's
used
everywhere
and
important,
and
when
security
area
comments
have
been
unrest,
I
believe
we
could
proceed
with
working
group.
Adoption.
N
L
So
any
questions
for
this
model.
C
So
we
added
those
two
more
things
based
on
discussions
with
brian
weiss
and
the
reason
we
added
that
one
algorithm
that
nobody
supports
said
is
it's
the
one
that's
recommended
for
TCPA?
Oh,
so
that's
why
you
might
wonder
why
we
added
something
nobody
supports,
but
it
that's
why
you
made
it
so
we
made
it
a
feature.
O
Damn
bogdanovich,
I
would
recommend
looking
into
removing
the
time
you
know
the
time
leaves
out
of
the
model,
because
it
might
be
useful
to
define
a
separate
bowl
that
would
be
deal
with
in
a
lifetime,
because
we
are
looking
you
don't
want
to
have
different
lifetime.
Time
ranges
in
those
supported
actually
defined
in
different
models.
I
mean.
C
O
C
C
L
L
So
we
know
our
rib
is
an
entity
that
contains
all
the
routes
calculated
by
routing
protocols,
so
for
all
other
implementations,
smaller
or
larger
ones.
They
also
part
a
rib.
The
routing
config
model
defines
the
basic
building
block,
building
blocks
for
rape,
and
it's
expect
is
expected
to
be
augmented
with
small
rape
attributes
so
on.
This
is
what's
currently
defined
in
the
routing
config
model,
so
it
defines
a
list
of
rot,
but
for
each
route
you
can
see
here
on
the
next
hub.
Options
are
very
limited.
L
Besides
these
specific
next
hop,
so
y'all
next
hub
can
only
be
either
an
one
outgoing
interface
all
an
IP
address,
so
it
doesn't
support
things
like
ecmp,
which
is
a
very
common
use
case
for
route
attributes
in
it
only
defines
the
job
preference
here.
So
obviously
we
need
to
augment
this
model
with
small
stuff.
L
As
I
mentioned,
the
drugging
config
model
tense
about
the
ecmp,
so
I
did
multipath
option
to
the
next
hub.
Choice
list
is
defined
as
a
list
of
passengers
and
for
each
part,
you
need
to
define
a
unique
combination
of
interface
and
IP
address.
So
by
this
we
can
support
ecmp
and
since
the
outing
config
model
and
if
and
only
defined,
a
real
preference
for
each
dot,
and
so
we
are
adding
more
relevant
passage,
reveals
magic
tag
and
repair
path.
L
L
So
put
everything
together:
this
is
the
extensions.
This
is
the
extensions
tree
that
we
are
eating
and
we
are
augmenting
to
every
path,
so
it
consists
of
a
magic
attack
and
the
repair
path.
So
the
thus
the
tree
structure
showed
in
this
slide
is
B,
is
augmenting
to
every
next
help
options
and
for
the
multi
pass
next
help
options.
L
L
We
also
augmented
the
static
route
configuration
with
multiple
next
help
support,
so
it's
also
defined
at
the
list
with
the
list
of
next
Avengers
the
next.
How
is
a
combination
of
interface
and
address,
and
also
our
preference
and
tagged
related,
which,
with
each
path
and
the
examples
showing
here,
we
are
just
using
IP
with
before
I
saying
example,
we
the
same
argumentation
for
ipv6.
L
So,
in
summary,
we
have
define
the
Maori
attributes
and,
together
with
the
resulting
config
model,
we
defined
a
very
generic
model
is
common
enough
and
we
believe,
hopefully
all
when
their
implementations
support
all
these
attributes,
and
it
also
include
pretty
much
all
the
basic
attributes
required
for
a
rib
model
in
question.
So
comments.
N
Jeff
I
was
speaking,
is
nitrous
chair.
The
work
you're
doing
here
is
very
interesting
because
two
things
first
one
we
know
the
net
matta
KU
thick
module
was
very
simple.
It
needed
this.
The
second
thing
is
the
I
terrestrial
data
model,
one
of
the
big
things
that
it
differs
from
from
the
net
mod
config
is
more
complicated.
Next
top
types
we
are
far
enough
along
in
the
ITER
s
model
that
we
have
been
requested
to
look
at
potentially
refactoring
to
refactoring
against
the
routing
config
module
to
remove
redundancy.
This
makes
design
sense.
N
My
request
to
you
is:
could
you
talk
to
the
rib
data
model
authors
with
the
nitrous
about
this,
simply
because
I
think
that
it
actually
makes
sense,
for
you
know
our
two
groups
to
come
to
a
common
understanding.
What
the
next
top
should
be,
and
potentially
what
may
happen
for
my
dress
is
that
our
model
largely
disappears,
and
that
would
actually
a
good
thing.
It
means
that
we
succeeded.
O
O
Modernizing
there
is
a
de
Sam
device
model
design
team
within
the
routing
group,
and
we
publish
the
draft
that
you
know
how
how
young
models
should
be
organized
to
represent
the
device,
and
within
that
model
you
know
the
how
the
routing
configuration
is
done.
Doesn't
you
know
correlate
nicely
what
we
have
proposed
so
because
it
their
impact
the
idea
rest
with
model
and
also
partially
you're.
So
that
is
something
that
AC
will
present.
I
believe
on
thursday
yeah.
L
Okay,
yeah,
the
major
thing
is
this
model
is
anti
reich
augmentation
of
the
routing
config
model.
So
if
you're
on
some
implementation,
that's
how
ready
is
a
powder
loading
config
model.
This
one
is
very
easy
to
do.
Just
add
more
of
you
more
calm,
arab
attributes.
Yeah
do.
O
Yeah,
that's
a
that's
a
question
because,
as
trail
and
PBB
are
using
inner
layer
to
routing-
and
you
are
you
put
in
ipv4
address-
you
know,
and
it
seems
like
you
made
it
very
IP
specific.
I
think
this
you
know
the
same.
Augmentation
could
be
just
reused.
Extra
model
could
be
used
for
other
layers,
so
making
it
you
know
less
IP
layer.
Three
specific
would
be
useful.
C
Yeah
there's
one
other
model:
I
don't
know
is
robbing
the
room
he's
not
there's.
Also
a
local
route
model
from
open
can
fake.
That
also
has
overlap.
This
Center
has
static
routes
and
aggregates
in
that
so
I'd
like
to
maybe
get
them
all
together.