►
From YouTube: IETF105-GROW-20190724-1550
Description
GROW meeting session at IETF105
2019/07/24 1550
https://datatracker.ietf.org/meeting/105/proceedings/
A
A
All
right
as
I
was
saying,
this
is
the
Gro
meeting
in
ITF
105.
If
you're
not
supposed
to
be
in
here,
you
could
stay,
it
might
be
fun
or
you
could
go
to
your
meeting
you're
supposed
to
be
at
if
you're
presenting.
You
should
definitely
either
stand
at
the
mic
and
speak
into
it
like
this
or
hold
the
mic.
A
A
Jarrod's
gonna
do
some
notes
for
us
good.
Now
we
move
on
okay,
you
see
you
found
your
paring
all
right.
There's
blue
sheets
they're
going
around,
please
be
sure
to
sign
them,
so
we
can
get
a
room
this
size
or
larger.
Next
time
we
have
a
minute
sticker,
jabber
scribe
we're
gonna.
Do
some
agenda
bashing
really
just
gonna
show
you
the
agenda
stuff
about
existing
drafts.
We
have
some
things.
This
is
I,
think
all
up
to
date.
Actually
even
we
hope.
Okay,
a
couple
things
in
ini,
ESG,
reviewer,
headed
that
direction.
A
There's
a
question
about
whether
we
should
adopt
something
or
not
I
think
that
might
not
actually
have
been
sent
out.
Yet,
if
you
have
a
draft
and
you're
wondering
gosh,
is
somebody
gonna
do
something
about
this
and
my
draft?
It's
so
cool!
You
should
let
the
chairs
know
or
ask
on
list
for
what
you
think
is
the
next
step
either,
if
it's
not
adopted,
ask
for
it
to
be
adopted.
If
it's
currently
got
a
bunch
of
comments
and
you've
dealt
with
all
those,
you
know
you're
ready
to
push
it
forward.
B
Our
brethren
ATS,
the
IDR
working
group,
are
looking
for
inputs
in
context
of
a
new
charter,
so
if
we
as
operational
community,
have
feedback
as
to
what
about
the
bgp
protocol
should
look
like
on
the
wire
and
in
its
behaviors,
now
it's
a
great
time
to
review
those
email
threats
and
provides
your
inputs
to
IDR.
So
please
take
that
under
advisement
I,
don't
know
when
the
chartering
process
will
stop.
But
I
expect
that
if
you
give
feedback
in
the
next
one
or
two
weeks,
it
can
be
taken
into
consideration.
C
C
Jones
cutter
again,
that
is
exactly
what
the
IDR
chairs
were
talking
with.
The
ad
is
about
this
morning
and
I
think
it
will
be
relevant
to
the
IDR
charter
discussion.
So
yes,
there.
We
hope
to
have
more
guidance
of
that
nature.
That
appears
in
the
IDR
Charter
and
possibly
in
the
charters
of
related
working
groups.
B
B
C
B
Will
not
be
an
update
tonight,
but
tomorrow
morning,
at
half
past
eight
in
this
very
room
do
liffe.
There
is
a
gathering
to
discuss
the
concept
of
documents
that
are
not
RFC's
and
I.
Think
that
would
be
interesting
to
the
operational
community
to
have
such
a
tool.
So,
if
you're
agreeing
with
that,
please
come
to
the
site
meeting
tomorrow
morning
and
provide
your
feedback
on
these
ideas.
B
F
F
So
previously
we
had
two
types
of
signals
for
roughly
prevention
and
another
signal
for
route
leak.
Detection.
The
first
rule
says:
if
route
is
received
from
provider,
route
server
or
peer,
it
must
not
be
sent
to
another
provider
appearing.
So
the
only
direction
that
is
applicable
is
customer,
so
you
may
send
your
prefixes
only
to
customers,
the
second
rule.
If
route
is
sent
to
customer
peer
or
route
server
client,
it
also
must
follow
all
the
down
rule.
F
F
Yeah
I
understand
so
yeah
yeah.
He
is
noise
from
outside.
Yeah
I
want
this
good.
Ok,
ok,
I'll!
Try
to
do
this
way,
so
we
decided
to
make
it's
a
joint
signal
and
modern.
We
will
have
two
signals.
There
was
a
argument
about
which
one
which,
which
is
one
is
better
attribute
or
community
attribute,
has
a
lot
of
advantages.
It's
more
reliable!
It's
memory
efficient!
It
has
a
reserve
to
use,
but
the
community
can
be
deployed
tomorrow
or
a
day
after
tomorrow.
So
we
will.
We
would
like
to
have
advantages
of
both
approaches.
F
We
will
have
community
tomorrow
and
in
the
attribute
the
day
after
tomorrow.
Maybe
later
so
here
is
the
specification
for
community.
It's
nearly
models,
OTC
behavior,
that
I
present
it.
If
the
idea
on
working
group,
so
we
suggest
to
reserve
this
class
for
well-known
transit
communities,
so
these
communities
should
not
be
stripped
on
ingress,
so
the
first
one
is
just
sub
class
for
do
community
and
the
video
is
carrot.
Is
the
autumn
system
number
how
this
weather
is
set
like
in
how
to
see
on
egress?
F
If
you
are
sending
it
from
provider
peer
or
internet
exchange
or
router?
You
are
setting
it
with
value
of
your
other
known
system
number
on
receiver.
If
you
are,
if
you
are
customer
or
beer,
you
are
also
setting
it
with
this
number.
So
it's
doesn't
matter.
If
it's
a
set
organ,
egress
or
ingress,
the
Whaley
will
be
just
the
same.
It
works
nearly
the
same
like
OTC.
So
as
we
discussed
for
in
this
example,
we
have
provided
on
left
side
is,
is
taking
transit
from
its
provider.
F
It's
also
has
a
pier
on
the
right
hand,
side.
The
signal
now
in
this
particular
example
community
is
set,
may
be
set
on.
Ingress
may
be
set
on
egress
it
doesn't
matter.
The
value
will
be
the
same.
So
when
it
goes
through
the
network
to
prevent
network
from
leaking,
it
does
not
need
to
match
the
class
of
the
community.
There
only
difference
from
yo
TC
having
not
either
or
ot,
see
happening
on
egress
at
the
level
of
ice
3.
Imagine
that
ice
2
is
not
filtering,
and
so
it's
a
liquor.
F
Unfortunately,
we
are
not
able,
with
community
image
in
a
single
community
means
image,
create
a
rule
that
says
image
for
class
and
doesn't
match
for
video,
so
this
rule
must
be
transformed
into
2.
Majors
first
check
that
is,
is
not
setting
the
community
second
check
the
class.
So
it
is
based
on
the
assumption
that
if
ice
2
is
adding
marking
its
also
makes
filtering
it's
very
important
and
we'll
get
back
to
this
few,
so
the
most
important
question
what
we
should
do
if
we
detect
a
root
leak.
F
F
The
previous
versions
of
the
draft
suggest
these
kind
of
behavior,
and
the
idea
was
that
maybe
there
is
a
customer
that
has
only
single
path
from
all
originator
to
a
t1
and
if
ralick
happens
in
this
part
and
immediate
upper
provider
accepts
it
and
then
next
level
provider
the
texas
and
drop
it.
We
may
end
up
that
roughtly
results
in
connectivity
loss
for
the
victim
and
for
this
case
with
the
thought
that
local
preference
zero
can
solve
the
problem
for
the
in
these
unreliable
topology.
But
it's
his
very
important
side
effects.
F
Actions
this
must
happen
at
the
same
time
to
make
network
unreachable.
So
it's
not
very
likely
situation.
So,
let's
go
to
the
side
effects.
First,
one
happens
because
of
when
root
Lee
is
propagating
through
the
upstream
providers.
They
are
still
able
to
detect
it,
but
when
such
prefix
goes
down,
the
network
lose
lose
this
opportunity
to
say
if
it
was
erratic
or
not
so
it
may
become
more
preferable.
We
are
getting
in
situation
with
priority
loops,
BGP
wages
and
in
some
specific
to
a
topologist,
even
stable,
BP
oscillations
with
Buju
pillows.
So
it's
not
tasting.
F
This
situation
can
be
fixed
with
additional
signal,
but
even
with
additional
signal
that
says,
for
example,
we
detected
a
root
leak.
We
cannot
guarantee
for
these
customer
with
unreliable
topology.
That
prefix
will
become
reachable
because
it
will
depend
on
the
on
the
policy
of
other
parties.
You
are
saying
there
I'm
propagating
a
root
leak.
Many
providers
will
just
drop
it.
Ok,
but
it's
not
the
full
problem.
The
biggest
problem
is
prefixes
that
has
a
specified
policy
that
restricts
its
propagation.
It
might
be
done
in
different
ways.
F
It
may
be
because
I'm
using
some
control
community
to
limit
the
ways
prefixes
is
propagated
by
my
upstream
problem
by
mega
upstream
provider.
Thank
you.
Then
it
may
be
most
specific
prefix
that
is
sent
only
through
internet
exchange
point.
So
there
are
many
kinds
of
such
policy
when
prefix
becomes
partially
visible
and
from
receiver
side.
F
We
are
not
capable
to
just
in
to
say
if
the
leaked
prefix
that
is
received
from
your
customer
is
poo
proper
cause
a
poo
in
direct
customer
with
unreliable
topology
of
if
it's
a
CDN
that
had
such
such
policy,
its
advertising
more
specific
and
you
are
going
to
ruin
its
connectivity
plus
ruin
your
own
connectivity
to
the
CDN.
So
this
is
just
a
highlight.
I
made
this
calculation
yesterday
and
it
shows
the
number
of
prefixes
and
its
visibility.
It's
based
on
right,
breeze
data
source,
all
small
producers.
This
has
length
more
than
24
I
ignored.
F
The
only
exception
may
happen
when,
at
the
time
of
early
adoption,
one
may
want
to
see
the
signal
before
making
a
decision
that
we
are
ready
to
implement,
implement
drop
policy.
It
can
be
done,
but
with
caution
as
we
discussed
previously,
we,
if
we
see
marking
from
a
neighbor,
I
oughta,
know
sister
number.
We
have
an
assumption
that
it's
have
drop
policy.
Otherwise
we
are
losing
the
signal.
So,
if
you
are
going
to
to
do
only
marking,
you
should
go
a
full
of
these
limitations.
F
They
will
be
described
described
in
the
next
version
of
the
drop.
So
here
is
the
result.
The
only
opportunity
to
mitigate
root
leaks
is
to
drop
them.
You
must
you
should
follow
this
rule.
There
are.
The
only
option
is
just
to
pass
it
by,
but
you're
accepting
root
leaks.
You
may
say
into
your
egress
traffic
to
a
black
hole.
F
What
are
the
next
steps?
We
really
need
your
feedback.
We
find
the
good
consensus
among
the
authors
in
the
version
that
is
in
mailing
list
with
the
miners.
Changes
will
be
uploaded
into
the
tracker.
I
hope
that
I
will
be
able
to
do
it,
and
during
this
week
we
also
will
require
ion
allocations
for
class.
F
This
will
say
that
we
have
well
known
trans
communities,
plus
a
sub
class
for
leak,
detection
prevention
and
so
on,
and
we
really
need
to
go
forward
to
have
a
fast
working
group
last
call,
because
if
we
will
eat
for
years
there
will
be
no
reason
for
this
solution,
because
I
will
have
a
more
reliable
OTC
or
a
spay
in
place.
So
thank
you
for
listening.
G
H
It
was
made.
One
of
my
two
points.
Second
point
is
that
large
communities
were
for
various
reasons
designed
to
not
have
resort
values.
So
one
of
the
things
you
need
to
be
aware
of
is
that,
regardless
of
what
working
group,
this
happens
and
you're
about
to
make
a
operational
change
and
having
I
Anna
carve
out
a
piece
and
know
sort
of
like
the
well-known
stuff,
for
you
know:
1997
ones
you
what
thunders
eventually
have
magic
behaviors
for
those
think
about
this
stuff.
F
B
In
a
different
draft,
which,
coincidentally,
is
on
the
to-do
list
of
the
chairs
of
this
working
group,
we
can
instantiate
a
registry
and
that
draft
will
allow
us
to
discuss
the
concept
of
a
registry
and
see
if
IDR
and
andro
agree
that
there
should
be
a
registry
and
once
that
registry
exists,
then
your
draft
can
just
request
a
value
from
that
registry.
So.
I
I
J
Then
medicine
work
online,
yeah
I
found
that
confusing,
but
I
was
just
about
able
to
pass
it.
My
understanding
reading.
That
is,
that
the
only
thing
that
is
mitigation
is
rejecting
and
you
may
or
may
not
choose
to
mitigate.
Yes,
that's
correct
I
prefer
that
construction
to
having
a
recommended
default
I,
don't
think
anything
other
than
dropping
leaks
is
mitigation
and
dressing.
Something
else
up,
because
mitigation
is
counterproductive
and
I
think
we
saw
that
in
origin
validation
for
years.
G
K
K
So
this
is
the
only
slide
that
I
have
on
these
two
drafts
right
we
presented
in
2017.
There
is
a
lot
of
backup
slides
in
this
presentation
on
the
motivation
why
we
are
doing
this
and
things
like
that
that
key
takeaways,
that
we
are
adding
important
advantage
points
over
the
existing
RFC
of
BMP.
So
in
the
existing
RFC
we
have
a
tree
being
pre
and
post
policies.
We
are
adding
block
creep
and
other
about
pre
and
post
policies.
K
We,
since
the
last
IDF
only
minor
changes
right
are
about
is
seeing
IE,
SG
review
and
feedback
was
processed
and
things
like
that
I
think
things
are
going
well
and
the
forelock
crib,
which
is
the
one
that
is
a
slightly
more
active.
We
did
process
the
feedback
of
Jeff
has
thanks
for
very
much
for
the
feedback
here,
the
plenty
of
feedback.
Now,
if
you
grab
through
the
document,
there
is
no
feeb
wording
anymore.
That
is
maybe
one
of
the
most
important
thing,
plus
many
other
things,
and
while
I
was
processing,
his
feedback
I
captured.
K
Something
else
that
we
have
TLV
about
vrf
and
table
name
and
table
name
is,
for
example,
if
you
have
a
slice,
a
partition,
specific
view
that
you
are
passing
of
the
lock
rib
and
vrf,
it's
vrf
right,
and
so
it's
possible
that
you
want
to
have
this
TLV
repeated
multiple
times
I
catched
it
while
I
was
processing
his
feedback,
so
some
tax
was
added
to
it,
and
that
is
pretty
much
it
right
now.
If
you
remember
last
time
at
ITF
104,
there
was
quite
some
queue
of
people.
K
You
know
asking
question
around,
you
know
what
goes
in
block
rape,
active
routes,
all
routes
installed
routes,
and
things
like
that
right
and
it
is
true,
this
draft
is
maybe
a
little
bit
under
it
under
determines
a
little
bit.
What
goes
in
lockrey,
but
any
question
that
you
would
still
have
around
that
I
ask
you
if
you
can
hold
your
tongue
for
the
next
next
presentation,
where
there
will
be
Camilo
here,
because
it
may
be
the
most
proper.
K
K
K
K
So
a
little
bit
of
background
or
history
behind
this
draft
is
that,
if
you
remember
like
one
year
ago
Hank
he
presented
you,
he
wrote
a
document
about
how
to
make
BMP
more
extensible.
I
myself
was
a
supporter
of
that
draft
and
also
we
met
in
person
with
Hank,
and
this
is
Interop
meeting
between
me
and
Hank
in
which
he
was
exporting
from
a
Nokia
device.
You
know,
BMP
content
with
the
following.
K
The
draft
and
I
was
collecting
it
with
pms
ECT
and
we
were
super
duper
happy
because
everything
worked,
but
then
what
we
did
is
a
little
bit
of
an
exercise
about
you
know
about.
You
know
rereading
the
draft
and
things
like
that,
because
it
was
quite
extensive
draft
right,
and
so
we
could
do
this
exercise
and
dissect
it
in
three
main
areas
that
the
draft
was
touching
upon,
and
so
one
was
very
easy.
K
It
was
getting
into
use
cases
and
we
should
you
know
it's
not
the
greatest
idea
to
compound
use
cases
with
other
stuff,
and
so
it
was
good
to
leave
the
use
cases
alone.
And
then
it
was
touching
a
on
to
other
areas,
very
important
one,
the
tlvs
introducing
tlvs
in
BMP,
not
introducing
but
making
sure
that
all
the
messages
have
tlvs
and
review
message
types
right,
and
so
we
are
here
right
so
in
the
tlvs.
K
K
K
You
know
complex
because
we
can
have
all
different
ideas
on
how
we
want
to
beautify
the
protocol
and
things
like
that,
but
we
can
kind
of
say
that
it's
a
you
know
an
engineering
effort
more
than
something
with
the
business
critical
sense
right,
and
so
that's
why
I
think
they
really
belong
to
two
different
drafts,
and
so
we
want
to
you
know
separate
in
the
two
things
that
said:
what
is
the
problem
statement?
The
problem
statement
is
simply:
it's
like.
We
have
a
table
with
four
legs,
but
not
the
all.
K
K
So
what
is
the
ideas
in
the
draft
is
to
add
TLV,
not
odd.
Allow
simply
allow
for
trailing
TLV
data
in
the
BMP
messages
and
in
order
to
do
that
and
I
leave
it
as
at
the
last
point
here,
like
a
bump,
the
version
of
BMP
to
for
essentially
I
want
to
leave
it
as
last,
because
you
know
I
want
really
to
stress
that
that
is
not
really
the
key
part
right.
So
we
are
not
in
a
critical
point.
K
You
know
we
are
not
twins,
inventing
a
whole
new
protocol
or
or
anything
like
that,
it's
really
just
backward
for
backward
compatibility.
So
that
said,
I
have.
There
is
one
open
question
in
the
first
release
of
the
draft,
it
was
on
purpose
left
under
specified
how
to
handle
the
p.r
down
right.
So
it's
just
written
that
also
the
pillar.
The
period
down
should
I
have
TLV
data,
but
the
truth
is
that
you
have
three
code
points
that
allow
for
trailing
data,
and
this
data
is
not
necessarily
in
a
TLV
format.
K
There
are
other
code
points
that
you
know
it's
not
specified.
You
know
that
they
can
add
a
trailing
data
and
things
like
that.
So
again,
even
there
you
have
not
an
even
surface
and
you
know
solutions
for
that
could
be
anything
right.
So
from
the
most
extreme,
like
we
have
a
new
message
type
to
we
duplicate
the
code
points
to
we
say
we
have
in
bmp4,
and
so
now
the
in
data
should
be
in
TLB
format
to
anything
right.
So
there
are
multiple
solution
to
this
problem.
K
L
So
the
goal
of
this
wrap
is
basically
to
to
mark
the
paths
in
B
and
P
with
our
state
on
the
on
the
local
rip.
We
will
have
this
information.
Maybe
we
can
allow
some
different
applications.
You
know.
So
that
means
basically
that
if
you
can
get
the
pad
mark
as
if
it's
installing
the
rip,
if
it's
a
backup
path,
if
it's
hopefully
unresolved,
maybe
and
stuff
like
that,
we
are
open
for
suggestions
on
what
else
to
mark
either
the
TLB
mechanism
that
Paulo
Lucinda
just
presented.
So
there's
some
faith
share
in
there.
L
This
is
actually
a
rehash
of
a
very
old
draft,
so
like
around
8
years
ago
or
7
whatever
or
six.
We
had
idea
of
of
using
that
path
and
communities
to
actually
mark.
So
you
see
that
path
to
get
all
the
path
from
from
a
router
and
then
using
communities
to
mark
what
type
of
path
they
were
if
they
were
passed
for
base
external
or
stuff
like
that.
L
However,
even
though
technically
there
was
not
much
of
an
issue
like
the
philosophically
is
not
such
a
good
idea
at
least
and
the
mind
of
many
to
monitor,
bgp
with
bgp
but
used
BMP,
which
back
then
was
not
very
much
sure
now
it
seems
to
be
so.
Why
not
to
put
it
there,
so
it
still
be
basically
right
now
we
have
so.
This
is
the
basic
idea.
The
rest
is
our
I
mean
we
have
the
first
iteration
over
these.
Maybe
we
can.
L
We
will
probably
improve
it
in
the
future
if
it's
interesting
or
not,
for
the
group.
So
right
now
we
have
the
pad
type
and
a
recent
string,
because
why
not
so
the
type
is
actually
a
bit
field,
so
you
can
have
more
than
one
type
for
a
specific
path.
I
mean
here
are
some
initial
ideas
of
this,
so
you
can,
you
can
mark
like
the
path.
Is
the
path
is
best
path
or
primary,
which
basically
means
that
if
you
have
multi
path,
then
you
can
have
multiple
paths.
L
You
can
mark
whether
a
pack
a
path
is
a
backup,
so,
for
instance,
if
you
don't
want
to
run
very
complex,
BGP
simulator-
and
you
have
this
information,
then
you
can
use
you
can
use
that
for
for
that
application,
then
you
can
also
have,
for
instance,
unreachable
next
hop.
So
you
can
say
there
are
path,
is
invalid
last
week,
actually
I'll
prove
retinas
and
on
the
ITR
that
we
really
were
not
sure
what
unreachable
next
segment,
but
whatever
so
there's
a
syllabus
external
path
there,
which,
after
six
years,
is
a
still
a
draft.
L
Even
though
people
use
it
I,
don't
really
have
an
application
for
that,
at
least
for
optimization.
But
maybe,
if
you
have
this
information,
you
can
you
can
verify
your
network
or
I
mean
the
more
information,
the
better
I
think
at
some
point,
there's
also
a
recent
string,
because
why
not
so
the
idea
here
is
that
if
you
can,
if
the
platform
actually
allows
you
to
tell
why
a
path
is
in
a
particular
state,
for
instance,
why
wasn't
the
best
or
wise
not
install,
or
why
is
unreachable?
Maybe
the
next
hope
is
a
multipath.
L
Is
a
multicast
address,
or
maybe
you
had
a
cap
in
the
number
of
backup
pad,
so
maybe
you
can
and
write
that
in
the
original
string
this
tends
to
be
controversial.
I
think
this
type
of
reasoning,
strings
but
parameterize
indeed,
but
I'm
addressing
this
information
might
be
very
complicated
and
it
could
be
useful
for
debugging
for
the
government
blogging
purposes.
So
again,
this
is
the
structure
we
have
now
we're
open
for
suggestions
or,
and
that's
it
for
that
side.
L
So
I
do
have
an
open
question,
which
is
what
paolo
mentioned
before,
which
is
what
relationship
does
we'll
discuss,
for
instance
with
with
the
local
rip
and
the
other
rip
drop,
so
I
believe
this
is
this
basically
come
from
the
local
rip,
but
to
actually
have
a
rich
received,
rich
information,
then
we
also
need
the
invalid
pads
and
I
don't
know
if
they
actually,
the
low-calorie
trap
allows
for
that
or
not.
You
know
if
it's
explicit
or
implicit.
M
Shared
much
Akamai,
can
you
go
back
a
couple
slides,
slides
to
the
reason.
Codes
are
back
another
one
I
think
that
yeah
this
this
one
so
I'm
just
curious
where
this
registry
would
live
because
it.
This
seems
like
it's
more
of
like
a
protocol
level
message
here
and
I'm
wondering
if
that,
if,
if
this
would
live
in
grow
or
an
idea
or
somewhere
else,.
H
H
Has
so
camilo
night
that
this
conversation
of
this
part
conversation
already
before,
so
this
is
partially
for
the
room.
I
like
the
idea
in
general.
This
is
overlapping.
You
know
some
other
questions.
You
know
we,
the
person
that
was
just
stuff
of
the
mike
head
about,
do
we
have
sway,
it
may
be
tracing.
Did
this
grout
actually
made
it
through
the
install
process?
No
intelligible
for
out
selection?
That
sort
of
thing
rudiger
has
had
similar
comments
before
as
well.
H
The
the
problem
with
a
lot
of
the
things
in
this
set
of
states
is
many
BGP
pipelines
and
implementations
are
not
fully
synchronous.
You
know,
use
I
can
get
her
out
and
it
may
go
through
several
stages
before
you
actually
get
some
of
these
things
potentially
set
like
reach.
Ability
of
next
hop
may
come
very
late
and
we
get
around
install
process.
Whether
things
eligible
for
multipath
is
another
example.
So,
and
whether
or
not
something's
elected
from
best
path,
so
people
have
drown.
H
Selection
happened
very
far
down
their
pipeline
and
the
problem
with
this
is
the
original
BMP
rip
in
at
least
it's
a
very
dumb
feature.
You
can
just
stream
this
stuff
out
almost
immediately,
if
you
want
to,
if
your
implementation
once
allow
it
or
come
very
late,
trying
to
actually
tie
these
things
in
is
try
and
actually
lock
the
implementation
down
to
don't
send
anything
out
that
tells
this
stuff
is
determined
and
I
think
you
might
find
that
actually
has
bad
impacts
and
some
of
the
use
cases
ribbons
used
for
today.
So.
H
H
Reach
ability
may
vary,
and
some
of
these
things
like
these
so
flags
if
you're,
actually
getting
these
as
part
of
your
local
rib
feed,
look
rib
is
only
giving
you
the
active
road
because
giving
all
the
state
as
it
changes
through
the
set
of
routes
they're
getting
altered
as
part
of
route
selection
changes
resolution
state
policy,
whatever
your
than,
if
it's
just
nothing,
but
there's
endless
churn
of
these
bits.
If
you
actually
exposed
them
that
way,
so
I
guess
my
two
pieces
of
advice
are:
some
of
these.
H
Things
may
have
negative
impact
on
the
implementation
if
you
really
care
about
them,
and
the
second
thing
is
in
a
low
crib
context,
you'll
end
up
with
a
agency.
If
churn
that
really
doesn't
give
you
potentially
allow
useful
information,
but
if
you
dance
back
to
the
string
slide.
The
second
piece
of
feedback
here
is
that
this
stuff
is
interesting.
H
It
overlaps
the
work
that
you
nuns
doing
for
10
cents
and
I
think
there's
some
value
out
of
having
that
conversation
continue
in
relation
to
that,
but
the
related
piece
of
feedback
that
sort
of
goes
into
should
be
put
the
state
into
EMP.
We
talked
about
this
night.
A
photo
for
BMP
is
a
difficult
place
to
put
this
stuff
in
because
it
can
be
you're
sort
of
jamming
up
the
pipeline
with
more
and
more
information,
EMPs
already
pushing
the
fire
hose
of
information
down
and
drinking
straw.
H
So
the
problem
we
have
here
is
that
strings
are
useful.
They
don't
compress
terribly
nicely
the
protocol
and
this
sort
of
just
fattens
up
the
p
views
with
information.
If
you
send
this
with
every
single
thing
might
be
interesting,
but
you
probably
want
as
a
telemetry
feed
for
interesting
things
rather
than
as
a
contiguous
tree.
J
F
F
F
L
That
is
all
parts
and
ribbon
information
of
the
log
grip.
So
after
the
decision
process
is
made
if
that
decision
was
made
after
the
decision
process,
I
don't
see
why
not,
obviously,
on
my
paper
me
doing
the
paper.
I
don't
do
the
code,
so
I
obviously
think
that
is
hard
for
two,
but
I
don't
see
why
not,
for
instance,
we
get
something
that's
very
important.
We
can
also
add
a
type
for
it,
but
obviously,
with
this
is
travel
of
implementation
versus
what
we
need.
G
G
N
Thanks
for
saving
me
several
minutes,
and
can
you
hear
me
clearly?
Yes,
okay,
so
I'll
do
a
very
quick
update
on
this
draft,
and
so
this
one
was
previously
presented
once
on
ITF
103,
that
we
use
PMP
to
carry
the
ruling
detection
information
and
where
well,
nothing.
Much
has
changed
in
sense
since
then,
but
I
would
like
to
bring
up
a
major
information
here.
N
N
The
difference
between
my
draft
and
other
works
is
that
the
detection
of
root
Lake
happens
a
actually
at
the
local
airs.
As
far
as
I
understand
and
after
I
have
also
discussed
with
Alex.
There
has
no
no
message
that
can
be
used
to
detect
the
existence
of
rulings
that
is
happening
at
the
local
s.
Okay,
so
are
typically
a
Lakes
can
be
detected
in
the
upstream
or
downstream
areas,
but
not
in
the
local
s.
Okay,
so
that's
where
this
draft
stand.
N
Okay
and
a
very
simple
example:
sorry
very
simple
idea
for
this
dropped
that
we
want
to
use
a
IOT
t
RV
to
carry
the
business
relationship
to
to
state.
What's
the
current
relations
and
then
and
then
at
the
EMP
server
side,
it
can
combine
the
our
DT
of
ease
from
the
ingress
and
egress
notes
and
then
compiling
to
analyze
these
two
relationships
to
detect
if
any
rolek
has
actually
happened
within
the
ass.
So
very
simple
idea:
any
questions
on
this
draft.
N
How
this
draft
should
go
because
after
I
have
been
talking
to
some
of
some
of
you
like
Jeff
and
I'm
booty
girl,
there
still
exists
some
gaps
where,
with
whispers
back
to
certain
perspectives,
so
yeah
I
would
like
to
see
how
how
help
the
working
group
suggests
the
draft
to
go
and
a
quick
recap
of
the
idea.
So
this
drafts
this
draft.
This
draft
wait
does.
Is
that
a
it
recalls,
the
processing
of
a
group
policy?
And
this
is
an
example
that
shows
you
what
is
exactly
recorded
using
this
proposal?
N
N
Here?
We
list
several
use
cases.
First,
it
can
be
used
to
do
route
policy
validation,
for
example,
you
configured
a
new
policy
to
do
maybe
maybe
rule
ich
prevention,
so
you
add
new
filtering
conditions
and
then
you
can
use
this
function
to
record
if
the
prefixes
that
you
care
about
are
actually
correct
or
correctly
a
future
or
not.
Ok,
and
also
it
can
be
used
to
do
root,
cause
analysis.
So
the
root
cause
analysis
is
not
used
right
after
the
policies
policy
is
configured.
N
It
is
actually
used
after
a
certain
period
of
time
when
the
policy
was
configured
because
some
of
the
issues
are
only
triggered
when
certain
environment
conditions
change
so
that
you
don't
know
when
the
what
what
policy
is
actually
responsible
for
the
current
issue.
So
you
need
to
have
such
record
within
the
device
and
then
to
locate
the
actual
real
cost.
Okay
and
third
example
is
the
route
propagation
trace,
and
there
is
one
example
in
the
draft
that
tells
you
how
you
can
do
that
with
the
data
recorded
okay.
N
So
here
I
listed
the
three
major
changes
from
the
first
version.
The
first
one
is
that
we
designed
a
new
message
structure,
which
consists
of
two
parts
which
first
one
is,
is
the
basic
information
part
which
is
fixed
lens,
and
it
tells
you
the
the
basic
background
information
about
the
event
and
then
the
second
part
is
flexible.
It
can
be
customized
according
to
user
requirements,
either
its
bandwidth
consideration
or
your
visualization
requirements.
N
Okay,
and
the
second
change
we
made
is
that
we
designed
a
new
structure
to
represent
the
route
policy,
basically
to
allow
the
policy,
training
and
policy
recursion,
and
the
surfacing
is
that
we
add
a
optional
TLV
to
to
allow
users
to
freely
express
whatever
I
think,
whatever
they
want.
Okay,
so
here's
the
message
that
the
message
body
looks
like
I
consists
of
the
prefix
information
and
then
the
event
information.
N
So
one
thing
I
would
like
to
mention
specifically
here
is
that
how
we
define
and
event
because
previously
there
have
been
discussions
on
how
how
we
can
use
this
functionality?
Some
people
has
mentioned
that
it
can
be
used
to
do
programming,
execution,
tracing
or
say
debugging,
but
I
thought
about
it
and
and
and
and
I
would
say,
this
draft
is
not
positioned
should
not
be
positioned
in
that
place.
I
would
say
this
draft.
Does
the
tracing
at
the
policy
group
policy
level?
N
Okay,
so
that's
two
things
from
the
actual
programming
execution,
so
here
the
event
will
be
defined
as
one
group
policy
execution,
whether
it's
mapped
or
a
matched
or
or
not
matched
okay.
So
whenever
a
policy
is
executed,
there's
one
event
generated,
and
this
is
what
each
single
event
looks
like
a
consists
of
major
a
four
parts.
The
first
part
is
environment.
It
tells
you
the
time
stamp
in
event
index
and
other
basic
informations,
and
then
we
have
the
second
part,
which
is
the
event
results.
N
It
tells
you
the
of
the
BGP
attributes
right
before
and
after
the
policy
execution.
So
this
party
is
optional.
You
can
have
both
preamp
and
post
policy
attributes
uploaded
or
you
can
have
just
one
of
them
or
you
can
have
none
of
them.
Okay
and
then
the
third
part
is
the
event
course,
which
is
a
policy
part,
a
policy
part
which
tells
you
what
is
responsible
for
the
results.
Okay
and
according
to
Rudy
Gers
suggestion
and
now.
N
I
also
believe
that
we
should
exchange
the
course
and
the
results
the
order
of
these
two
parts
and
I'll
do
it
in
the
next
and
the
last
part
is
the
event
notes.
This
is
the
optional
TOB
I
mentioned
just
now,
and
I'll
show
you
in
the
later
slides
how
we
can
use
that
okay.
So
these
three,
the
the
the
the
event
results
course
and
those
parts
are
or
optional
so
the
attribute
she
always
looked
like
that.
N
So
if
you
have
both
pre
and
post
policy
attributes
carried
in
one
message
and
then
you
can
have
a
direct
one
by
one
attribute
comparison
for
before
and
after
policy.
The
reason
that
I
add
the
true
policy
attribute
here
is
that,
because
sometimes
we
miss
the
recording
of
certain
events,
so
there
are
not
necessarily
direct
relationship
between
the
two
events
that
are
reported.
So
with
the
pre
policy
attributes,
you
can
have
a
a
very
direct
and
straightforward
with
a
visualization
of
the
results
before
and
after
the
palace
execution.
N
Okay,
so-
and
this
is
a
new
policy-
a
struct
structure
that
we
have
defined
previously-
we
only
allow
one
policy
expressed,
but
now
we
add
the
recursion
and
chaining
consideration
here.
So
basically,
what
we
do
is
that
we
use
a
C
bit
and
Arbit
to
indicate
the
relation.
The
relations
of
the
next
policy
to
the
previous
one,
so
you
can
stack
like
I
would
say
unlimited,
but
it
it
doesn't
necessarily
have
to
be
like
that.
N
You
can
stack
our
policies
one
after
another,
with
relationship
indicated
with
the
C
and
our
bit,
and
there's
also
one
a
policy
match
flag,
which
is
called
the
M
bit
to
tell
you,
if
part,
if
a
policy
is
actually
matched
or
okay.
And
finally,
we
have
the
policy
optional
TLV
here,
I'm
thinking
is
that
it
can
be
a
string
type
trv
that
can
be
used
to
freely
express
whatever
the
user
wants.
N
For
example,
the
user
may
dumped
him.
They
don't
want
the
attribute
TLV
or
don't
want
the
policy
ID
TLV
that
we
defined.
He
only
wants,
maybe
a
conclusion
from
the
device.
Of
course
this
is.
This
conclusion
is
also
generated
properly
from
the
execution
results,
but
he
only
wants
such
conclusion,
so
he
can
use
the
optional
govt
to
use
to
express
that
so
yeah,
that's
one
possible
usage
of
the
option.
Hiv
and
well.
N
B
Thank
you
very
much
in
the
interest
of
time,
I
would
suggest
that
feedback
is
provided
through
the
mailing
list,
because
we've
reached
the
end
of
our
session,
we're
actually
over
running
by
two
minutes,
we're
missing
cookies.
Thank
you
very
much
for
your
two
presentations.
I
will
see
you
all
in
Singapore
at
ITF.
106
see
you
next
time.
Thank
you.