►
From YouTube: IETF104-GROW-20190326-1610
Description
GROW meeting session at IETF104
2019/03/26 1610
https://datatracker.ietf.org/meeting/104/proceedings/
A
B
C
B
Have
the
special
room
with
John
again
I'm
Chris,
that's
Jobe,
that's
the
other
half
of
the
room
wave
to
them.
Hello,
okay,
told
you
be
fun
yeah.
They
are
remote
sort
of
remote,
semi,
remote.
Okay,
we
have
a
two
hours.
We
have
not
two
hours
with
the
presentations,
so
we
should
be
able
to
take
our
time
and
have
a
good
conversation
for
those
that
are
giving
presentations.
Please
that
might
be
close
to
your
face,
must
keep
the
mic
in
front
of
your
face:
here's
the
clicker.
It
has
a
left
and
a
right.
B
It
has
a
laser
all
right
as
operations
folks.
We
know
you
know
not
to
look
in
the
laser.
Okay,
all
right.
Here's
the
note,
well,
everyone's
read
that
good
okay
list
of
resources
also
get
out
your
Broncos
for
that
all
the
slides,
I
believe
are
uploaded.
If
anybody
is
presenting
and
has
sent
me
slides
in
the
last
20
minutes,
sorry
they're
not
up
there.
Yet
we
have
blue
sheets
which
are
going
around.
Please
sign
them,
so
we
get
a
better
room
next
time,
bigger
room.
B
Well,
the
column
is
nice,
I,
don't
see
those
people
or
they
can't
see
me.
We
have
a
jabber
scribe
already
Natalie.
Is
there
someone
ticking
who
would
like
to
take
minutes
awesome?
Thank
you
very
much.
All
you
have
to
do
is
email
them
to
Joe.
When
the
meeting
is
done
or
to
grow
desk
chairs,
do
we
have
anything
else
to
add
to.
B
We
have
seven
six
seven
six
drafts.
These
are
all
progressing
along.
If
the
authors
have
questions
about
them.
Oh
sorry,
five
of
the
six
are
progressing
along
those.
If
the
authors
have
questions
that
they
would
like
to
raise,
you
could
jump
to
the
mic
or
you
could
email
the
list.
If
you
want
to
move
on
from
tour
group
last
call
you
need
to
you
know
the
list
to.
Let
us
know
you
like
to
do
that.
The
last
one
will
list.
Mr.
Scudders
grow
BMP
Pierre
up.
Do
you
want
to
say
anything
about
that?
John.
C
C
B
F
F
Very
the
reason
is
to
have
the
boundary
if
and
plus
one
or
Adam
don't
say
doing
this
without
impacting
the
entire
network.
On
the
on
this
there's
a
scale
of
states
and,
of
course,
everybody
likes
fast
restoration.
This
just
for
reference
there
is
a
I
will
use
this
concept
during
the
presentation
so
basis
site
when
we
have
the
ASV
ours,
but
also
some
quarters
some
reflectors
of
optionally,
the
in
the
net
to
describe
how
what
is
the
proposal
we
used
it's
a
native
IP
network.
F
We
use
the
earth
math
multipath
whenever
necessary,
and
during
the
deployment
of
testing
of
that,
we
found
that
the
classic
approach
to
the
handling
and
beautiful
custom
search
short
errors.
So
one
is
that
next
hotel
doesn't
allow
us
to
requires
us
to
we
draw
a
path
by
path
in
case
of
the
external
team,
down
nothing
specific
except
it's
slow.
F
F
It's
a
strategy
to
address
you
just
pick
it
from
your
space
and
we
use
it
as
a
BGP
neck,
shot
with
a
internal
advertisement
for
the
path
which
are
received
from
one
of
given
subset
of
BGP
peers,
and
we
insert
this
up
to
all
these
address
into
the
I/o
GP,
because
we
need
to
resolve
the
BGP
next
code,
but
conditionally.
Only
if
they
at
least
one
session
from
the
set
is
in
established,
converged
state
state.
And
what
is
this?
This
subset
of
e
BGP
session
needs
a
little
bit
of
its
operator
decision.
F
F
F
F
What
is
happening
out
of
this
side?
I,
don't
want
to
propagate
all
these
paths,
because
I
want
to
hide
the
the
fact
that
I
have
so
many
asbr.
One
of
them
is
better.
One
of
one
of
the
path
would
be
selected
as
active,
but
I
need
to
compress
the
number
of
states.
So
there
is
another
abstract
next
and
this
abstract
next
have
sighs.
Okay,
this
is
new
address
and
I
use
it
for
all
sessions
with
a
VPS,
regardless
of
the
asbr
in
my
side.
F
So
it's
a
site-specific
and
route
of
like
to
advertise
just
one
puff
at
that
point
from
the
network
wide
perspective,
regardless
of
number
of
a
ASB
ours
here,
I
see
only
one
path
saying
there
is
exit
through
this
side.
That's
it
and
again,
because
this
is
next
hope.
I
have
to
insert
this
address
into
I
GP
I.
Did
it
shonali
saying
at
least
one
of
this
session
has
to
stay
up
to
advertise
this.
The
these
are
to
have
this
address
in
AGP
some
failure
cases.
F
F
F
So
we
need
one
more,
but
this
is
one
site-specific,
so
we
have
the
backup
from
the
other
side
it
doesn't
remove
the
need
of
withdrawing
path,
but
it
allows
to
react
on
these
failure
before
the
path,
our
withdrawal,
just
based
on
the
unavailability
of
the
BT
Phoenix,
and
this.
If
somebody
use
the
peak
edge
it
just
triggers
the
peak
edge
similary
how
this
oh
yeah.
F
We
have
some
artificial
links
there
so
same
with
the
internet
is
white?
Is
white
operation
one
session
no
changes
with
next
hope,
so
we
may
need
percent
the
update,
because
some
attributes
changes
like
originator
ID,
but
it's
as
if
the
next
sub
doesn't
change
before
were
defined
not
need
to
brett
programmed.
So
the
traffic
is
unimpacted
the
same
with
a
one
session
with
a
with
the
other,
a
s
I'll
say
all
session
with
one
of
ASB.
F
Ours
no
changes
only
if
we
lose
the
all
session
from
always
be
ours,
then
the
site
lose
the
connectivity
with
a
with
the
as2.
So
we
can
recall
from
IG
p,
decide
specific,
abstract
next
hops
and
then
traffic
will
move
to
the
other
side,
not
shown
the
picture
and
some
re
if
you'd
go
to
scale
out
for
the
peering.
F
That
allows
you
to
really
are
doing
that
for
you
for
the
boundary
for
redundancy,
maybe
for
road
trip
time
reduction
at
the
cost
of
scaling
up
the
control
plane
states.
So
we
need
some
constructs
and
practices
in
protocol
configuration
that
were
designed
to
contour
a
contour
that
abstract
next
hope
or
conditional
next
couple
or
whatever
you
want
to
call.
It
is
a
useful
and
we
can
use
it
to
control
these
bgp
scale.
As
in
this
example,
next
steps
I
need
more
feedback.
F
Thank
you
for
Robert
and
run
Robert,
actually
give
you
a
good
feedback
that,
indeed
draft
the
abstract
next
up
is,
is
tracking
the
state
of
the
of
the
BGP
session,
but
we
can
make
it
more
generic
and
track
other
objects,
which
is
true
and
I.
Think
this
this
should
heat
heat
to
work,
informational
document.
We
are
not
defining
any
new
standard,
just
sharing
the
practice.
Thank
you.
D
F
Entities
so
from
the
AA's
perspective,
we
can
use
similar
thing
between
a
ESS,
but
it's
rather
one
administrative
scope.
I
didn't
stop
through
the
doing
this
across
internet
and
the
total
independent
operation.
Operational
organization
should
work
because
it's
we
are
not
changing
BGP,
but
we
have
no
idea
P
right.
So.
F
G
It's
thinking
aloud
and
less
more
of
a
comment
and
less
than
a
question.
So
what
you
described
here
does
not
specify
the
protocol
either
changes
for
protocol
interworking.
This
is
more
of
applicability
in
a
sense.
How
do
I
build
and
run
certain
things,
so
the
question
that
they
have
is
what
is
to
standardize
here
that
and
whether
the
RFC
as
a
format
is,
is
the
right
format
for
such
type
of
information.
The
fact
that
it
describes
a
practical
and
applicable
use
case.
G
J
G
G
K
D
D
L
Yes,
so
just
a
couple
of
slides,
as
always
to
recap
that
the
it's
a
now
with
this
meeting,
it's
two
years
that
we
started
this
effort,
it
was
in
Chicago
and
in
March
2017,
and
the
draft
would
like
to
expand
the
functionality
of
current
BMP
from
a
tree
being
pre
and
post
policy
to
more
advantage
point
la
crib
other
about
pre
and
post
policy,
and
there
is
a
number
of
backup
slides
to
give
you
the
whole
context.
If
you
miss
the
the
start
of
the
whole
conversation,
there
was
a
lot
of
interest.
L
Pretty
much
I
would
say
no
opposition,
but
we
received
feedback
and
it
was
processed
and
as
far
as
we
know,
we
outers
know
there
is
no
outstanding
feedback.
You
know
not
standing
action
points
from
us
at
the
moment
and
since
the
last
ITF,
just
a
couple
of
things
happened:
I
process
the
feedback
from
John
Scudder
about
la
crib.
He
noticed
that
he
was
very
right
in
saying
that
the
period
down
doesn't
have
TLV
section,
so
we
the
same
use
that
we
had
for
the
tlvs.
You
know
what
we
wanted
to
kill
optionally
TLV.
L
L
Information
that
was
sent,
so
it
was
kind
of
you
know,
could
have
been
misread
like
the
router
that
sent
the
information.
Where
has
being
a
putter
about
it's.
You
know
where
the
pier,
where
we
are
going
to
send
this
information.
So
there
was
a
was
changing
into
an
es
and
that's
it
maybe
more
interestingly
I
would
say
it
is
really
the
most
juicy
slide.
So
I
start
from
the
bottom,
so
I
still
think
that
this
effort
should
be
wrapped
up
should
be,
let
it
go
and
we
should
move
forward.
L
L
There
was
a
lot
of
feedback,
also
in
him
in
Bangkok
right
so
in
which,
essentially
especially
about
the
lock
rig.
What
lock
really
means
right
so
shall
we
involve.
You
know
looking
at
the
feeb
in
any
way,
and
things
like
that,
yes
or
no
and
I
I,
think
I
was
already
providing
some
clear
feedback
back
in
Bangkok.
Also
I
will
just
stress
it
a
once
once
again
to
be
sure
we
are
all
online.
L
You
know
in
line
with
that,
all
in
sync
with
that
is
that
I
think
from
the
hallway
discussion
that
we
had
there
is
no
right
or
wrong
way
to
do
it
right.
So
there
are
use
cases
for
just
looking
at
the
LA
crib,
and
there
are
other
use
case-
is
to
involve
the
fib.
But
for
this
very
specific
draft
we
outers.
We
would
like
to
stop
at
the
LA
crib,
because
then
looking
at
the
fib
there
is,
you
know
a
lot
of
things
that
can
happen
which
are
vendor
specific.
L
There
is
a
lot
of
layer,
1
layer,
2
underlay,
maybe
also
the
vendor,
is
not
in
charge.
You
know
in
controlling
for
all
of
what
happens
at
layer
1
and
things
like
that.
So
in
other
words,
I
mean
there
is
a
kind
of
warm
that
we
prefer
not
to
open,
but
that
doesn't
mean
that
it
should
not
be
opened
so
maybe
to
serve
all
the
use
cases
where
it
would
be
great
to
know
about
balancing
hashing
algorithms
lag.
You
know
and
other
things
like
that.
Maybe
there
is
a
space
for
a
different
draft
right.
L
M
M
N
M
O
All
right,
you
go
schinsky,
Yahoo,
oath,
Verizon
media,
whatever
named
Harris
I,
think
this
is
way
overdue.
We
desperately
need
this.
Many
of
us
are
building
other
ways
of
collecting
this
data
via
streaming.
Telemetry,
no
get
really
nice
if
there
was
just
a
lock,
Reb
and
I
drove
out
of
BMP.
So
yes,
please,
how
do
we
help
and
also
for
the
FIB
thing?
O
That
is
also
becoming
very
important,
because
almost
all
magic
God
boxes
in
the
sky
need
to
know
the
FIB
in
order
to
even
calculate
what's
actually
happening,
because
very
often
the
rid
might
have
a
lot
of
stuff
in
it,
and
you
don't
know
what
sucks
your
program
to
the
hard
thing.
So
if
you
need
any
help,
what
text
or
whatnot
I'm
sure
there's
quite
a
few
people
in
this
room,
probably
including
me
that
would
love
to
help
with
that.
L
L
There
was
no
unified
way
to
get
all
the
vantage
point
with
a
single
protocol,
so
people
should
resort
to
two
or
three
different
ways
to
get
all
the
vantage
point
like
after
about
Audrey
being
and
la
créme
like
a
bgp,
BMP
and
yeah,
maybe
swimming
telemetry
for
others
about
or
maybe
see,
allies
screen,
scraping
or
other
things,
so
yeah,
Indian
yeah.
So.
O
Architectural
one
big
issue
will
have
any
multiple
ways
of
cutting
the
data.
Is
it
usually
they're
not
in
perfect
time,
sync
yeah?
So
if
you
have
a
single
protocol
that
can
actually
export
both
each
ribbon,
lock
Rev
edge
your
about
and
lock
fib
yeah,
you
could
actually
see
when
device
has
miss
programming
for
whatever
reasons
or
how
long
it
takes
for
device
to
program.
J
L
L
There
was
this
effort
in
which
it
was
called
BGP
part
marking.
Something
like
that.
It
was
one
attempt
at
trying
to
describe
you
know
the
problem
that
people
have
in
understanding
what
is
an
active
part
which
are
the
active
parts.
You
know,
because
you
could
have
ADD
part,
and
maybe
you
have
active
parts
and
you
could
have
multiple
active
paths,
but
not
all
the
ones
that
you
see
you
know
are
active,
so
there
are
others
that
are
not
active
and
essentially
that
draft
that
didn't
really
get
much
traction
and
support.
L
D
Commenting
on
the
process
reviewing
the
Jason
rip
out
draft
it
pass
working
group
last
call.
There
were
no
objections
to
move
it
along
and
related
to
lock
rip
Jeff
indicated
he
owes
you
some
comments,
but
I
also
think
that
John's
cover
has
an
outstanding
question
for
him,
because
she
made
some
changes
to
address
things
that
were
brought
up
and
we
never
saw
her
apply
whether
that's
addressed
the
concerns.
D
Some
of
you
may
know
what
the
purpose
of
maximum
prefix
limits
is.
They
are
a
safety
mechanism
where
you
can
configure
a
high-water
mark
and
if
that
trash
hole
is
exceeded,
the
system
can
shut
down
a
EVP
session
in
order
to
protect
the
innocence,
and
now
assumption
here
is
that
the
system
is
blind.
It
doesn't
know
anything
other
than
that
if
too
many
routes
are
announced,
something
somewhere
is
wrong
and
it's
safer
to
self-destruct,
and
this
applies
specifically
in
context
of
internet
routing,
usually
related
to
full
table
relics.
D
This
is
what
happens
when
maximum
prefix
limits
are
applied.
Free
policy
and
the
original
BGP
core
specification
talks
about
maximum
prefix
limits,
but
it
was
not
defined
in
a
very
clear
manner
in
my
opium
because
there's
different
types
of
prefix
limits.
If
we
look
at
this
image
from
left
to
right,
we
have
the
passing
of
time
from
bottom
to
top
there's
the
amounts
of
routes
being
announced
over
a
given
bgp
session,
the
steady
state
is
say,
a
set
of
routes
that
are
expected
to
be
announced.
D
The
operator
or
automation
system
made
a
error
and
a
leak
starts
to
occur.
So
more
and
more
routes
are
announced
to
the
neighbor.
At
some
point,
the
trash
hold
is
met.
The
session
is
going
now
and
from
that
moment,
on,
both
networks
are
protected
against
the
consequences
of
this
leak,
and
this
is
the
desired
situation.
D
However,
if
you
apply
maximum
prefix
limits,
post
policy
after
you've
gone
through
filtering
techniques
such
as
origin,
validation
or
peer,
lock
or
whatnot,
there's
a
fastly
different
image
again
from
left
to
right
time
from
top
from
bottom
to
top
the
set
of
prefixes
amounts
over
the
bgp
session.
We
have
the
steady
States.
That
is
what
we
normally
expect.
D
So
it's
a
bit
asinine
that,
despite
our
attempts
to
construct
high
quality
filters,
it
negates
the
effects
of
maximum
prefix
limits
that
are
intended
to
counter
full
table
leaks.
There
are
different
applications
for
pre
policy
and
post
policy.
Maximum
prefix
limits
the
pre
policy.
Maximum
prefix
limits
predominantly
exist
to
protect
against
memory
exhaustion,
because
if
you
have
to
store
everything
in
adjacent
rate
in-
and
you
don't
place
a
limit
on
that,
then
somebody
could,
for
instance,
an
orange
millions.
D
You
and
pre
policy
is
very
effective
against
full
table
leaks.
Then
the
post
policy,
maximum
prefix
limits
can
help
protect
against,
for
instance,
fit
exhaustion.
If
you
have
a
device
that
has
a
limited
amount
of
fit
space,
also
rip
local
rip
exhaustion
or
to
enforce
contractual
agreements.
You
sometimes
see
this
in
context
of
layer,
3
VPN,
so
both
have
a
right
to
exist.
D
D
Think,
looking
speaking
for
entity,
it
would
be
very
useful
if
we
can
self-destruct
ebgp
sessions
when
it
crushes
goes
past
thresholds
which
we
know
should
never
be
met.
For
instance,
on
a
given
moment,
we
should
be
announcing
roughly
three
hundred
thousand
routes
to
level
free.
If
we're
announcing
six
hundred,
thousands
I
would
like
to
tear
down
the
sessions
because
whatever
is
happening,
it's
not
according
to
our
expectations,
out
of
the
vendors,
only
Birds
has
implemented
a
outbound
maximum
prefix
limits
and
it.
D
So
the
purpose
of
this
draft
is
that
we
want
to
describe
what
types
of
maximum
prefix
limits
exists
where
they
fit
in
the
conceptual
BGP
model.
We
describe
how
outbound
maximum
prefix
limits
would
work,
and
then,
if
the
shooters
become
an
RFC
at
some
points,
it
can
be
used
to
communicate
through
RFPs
with
the
vendors,
where
we
can
point
that
this
is
the
specific
feature
we
would
like.
D
In
addition
to
a
real
self
destruct,
where
you
tear
down
the
session
and
syntheses
notification,
essence
that
the
maximum
amounts
of
prefixes
was
hit,
there
are
other
options
to
deal
with
this
failure
scenario.
For
instance,
you
could
withdraw
all
your
routes
or
you
could
stop
announcing
you
and
a
la
rice
or
but
I.
Consider
those
out
of
scope
for
this
draft
offenders
are
free
to
implement
advanced
forms
of
high
watermarks,
but
I
wants
a
sort
of
baseline
that
is
specific
for
internet
routing.
D
J
H
C
John
Scudder,
so
partly
based
on
experience
with
first
of
all,
I
think
this
sounds
like
something
good
for
the
group
to
work
on
so
great,
so
sort
of
battle
scars
from
RFC
seventy
six.
Oh
six
I
would
like
to
suggest
that
you
reconsider
your
decision
to
ask
your
vendor
to
be
creative
about
what
to
do
when,
when
your
outbound
prefix
limit
is
exceeded,
I'm
relatively
certain
that
if
vendors
implement
that
as
self
destruct
as
you
put
it,
so
it
self-destructs.
C
C
I
think
that
it's
likely
that
now
putting
on
my
vendor
hat
that
my
annoying
customers
will
come
and
ask
me
to
do
something
more
sensible,
and
when
that
happens,
maybe
it
would
be
nice
to
actually
have
thought
through
this.
You
know,
as
an
industry
instead
of
having
every
vendor
do
a
different,
arguably
sensible
thing
we
can.
We
can
continue
the
conversation
you
know
later,
since.
D
J
D
O
O
Other
people
may
say
only
send
this
Goldin
list
the
saw
the
prefix
list,
maybe
that's
what
they
want
to
do.
I
think
this
definitely
is
a
discussion
that
needs
to
have
before
we
mandate
do
this,
and
then
we
have
a
best
document
that
updates
it.
But
I
do
absolutely
thing
that
we
need
this,
but
if
we
implement
this,
can
we
make
sure
that
this
limit
is
post
policy,
because
I
feel
that
in
series
highlights
those.
O
D
O
F
F
Kind
of
you
can,
you
should
well
okay,
I'm
I'm,
slightly
more
polite
being
a
little
bit
less
polite,
the
creativity
that
John
is
calling
for
may
open
ways
into
insanity,
be
careful
what
you
want
and
what
you
do
and
well
okay
kind
of
I.
Don't
expect
insanity
from
John,
but
I
could
imagine
other
people
really
doing
weird
stuff
and
also
customers
asking
force
the
creativity
that
is
insane.
P
Thomas
cows
viscom
I,
our
pant.
They
fix
limits.
Yes,
please,
but
please
go
one
step.
Further.
I
would
like
to
see
move
up
the
road
No
very
easier
in
the
capability
exchange.
I
would
like
to
understand.
If
there's
a
inbound,
graphics
limitation,
I
would
like
to
learn
it
and
not
sending
more
and
more
outs
than
they
are
accepting,
and
also
this
can
be
reused
on
BMP.
So
I
can
see
basically
how
much
outs
they
would
accept
and
I'm
not
sending
more
than
what
they
accept.
If.
M
Jeff,
as
we
already
talked
about
using
the
notification
shut
down
the
pass
long,
you
know
why
you
shut
it
down.
I,
threw
out
a
little
bomb
and
then
walk
away.
This
is
potentially
a
good
time
to
bring
up
the
Twitter
/
BGP
thing
that
we
talked
about
the
past.
No
yeah
they
had
messages
is
a
perfect
way
to
carry
this
stuff.
What
that
feature
actually
meant
was
very
contentious.
I'm,
not
gonna,
talk
about
it
anymore.
Think
about
it.
Q
So
Warren
Kumari,
first
with
no
hats.
This
sounds
grand
now
with
the
ad
head
on.
You
should
probably
mention
this
on
the
IDR
list:
I,
don't
see
how
it
they
could
fuss,
but
it's
worth
just
making
sure
that
you
know
they're
informed
and
they
don't
come
along
at
the
end
and
say
this
changes
the
BGP
state
machine,
grumble,
grumble,
that's
good
feedback,
I.
D
R
N
Hi
Ron:
this
is
your
uncle
from
hallway,
so
the
first
draft
I'm
going
to
discuss
is
BGP
route
policy,
attribute
that
one
okay,
so
nowadays
already
BMP,
okay,
so
BMP
already
now
monitors
pre
policy
and
post
policy,
ribbing
and
local
rib,
and
also
Rapala
CM,
post
horsey
rebound
so
operators.
They
can
use
the
different
versions
of
the
ribs
to
find
out
if
there's
something
wrong
with
the
configuration
or
policies,
but
it's
hard
for
them
to
know
which
exact
policy
causes
issues
if
the
rig
doesn't
work
out
as
they
have
expected
it.
N
So
you
have
the
policy
name
or
policy
ID
which
we'll
discuss
a
little
later
and
then
you'll
have
the
timestamp
and
following
that
will
be
the
policies
okay,
so
so
by
the
by
looking
at
these
events,
in
order
of
time,
you'll
see
which
exact
policy
is
responsible
for
the
the
attribute
change.
Okay,
so
it
can
be
used
for
since,
like
policy,
validation
and
and
also
troubleshooting
yeah.
N
So
we're
post
proposing
a
new
message,
type
of
BMP
to
carry
this
route
and
policy
information
and
here's
the
message
body
looks
like
okay,
thank
you
and
we're
only
using
the
EMP
common
header,
where
the
message
type
will
indicate
a
new
message,
type
which
is
to
be
assigned
and
we're
not
using
the
prepared
hot
appear
header,
and
this
is
the
message
body.
So
it
consists
of
the
following
parts.
N
So
this
previous
hop
field
filled
in
here
is
used
in
combination
with
policy
distinguisher
here,
okay,
which
I'll
explain
a
little
bit
later
and
then
after
the
prefix
information,
you
will
have
a
counter
of
the
total
events
and
then
following
that
will
be
the
information
for
each
single
event.
First,
you
have
the
event
index
for
each
event
and
then
the
timestamp
and
the
following.
That
will
be
the
policy
information,
so
it
consists
of
the
policy
ID
and
policy
distinguisher.
N
Here
we
have
only
so
far
have
only
defined
two
types
of
policy
to
distinguish
once
the
impound,
the
other
is
the
outbound,
so
more
could
be
defined
later
and
fine.
That
will
be
the
pure
information
so
which,
which
exact
appear
the
policies
configure
around
configure
on
them.
So
suppose,
if
it's
a
it's,
the
policy
is
not
an
inbound
or
outbound
policy.
N
N
Okay,
so
yeah
one
more
thing:
the
previous
hopfield
here.
If
the
policy
distinguisher
refers
to
the
outbound
policy,
then
this
previous
hop
will
be
indicating
where
this
prefix
comes
from.
For
example,
two
peers
are
advertising
the
same
prefix
and
then
this
field
will
tell
you
which
exact
peer.
This
route
is
erected,
fro,
okay
and
then
here
say
a
single
simple
example
that
we
show
how
to
use
such
information
to
recover
the
route.
Propagation.
N
N
Q
N
So
the
same
prefix
is
advertised
by
c2
and
c1
and
after
received
at
p1,
it
selects
the
route
that
comes
from
cd2
so
for
p1.
Three
events
will
be
reported.
First,
one
is
inbound,
inbound
event
from
c2
and
also
the
invalid
bell
from
c1,
and
then
it
also
reports
a
album
event
which
indicates
that
the
route
I
advertised
to
the
roof
reflector
is
the
route
that
I
select
from
c2,
okay
and
and
then
stuff
like
that.
N
M
Jeff
has
we
started
a
little
bit
this
conversation
that
died
er
partially
for
context
for
other
people
that
should
go.
Look
at
the
narrow
draft.
Jacob
Heights
did
one
for
doing
tracing
for
events
using
time
stamps
and
trying
to
stick
that
the
BGP
this
case
we're
trying
to
stick
stuff
with
EMP
the
good
things,
no
we're
getting
three
of
our
route
types
stuff
in
one
message.
This
is
handy
for
correlation
the
policy
that's
being
referred
to
is
presumably
the
matching
element.
M
That's
the
rejector,
accepting
your
policy,
unfortunately,
in
a
lot
of
cases
and
rüdiger
will
speak
more
to
this
I
think,
since
this
conversation
did
come
up,
a
couple
of
ID
are
a
couple
gross
sessions
ago
about.
Could
we
actually
put
something
inside
the
BMP
the
same,
why
the
route
was
accepted,
rejected,
etc?
And
it's
a
good
thing.
This
is
a
good
example
of
the
of
the
thing
that
we
would
like
to
do.
M
The
problem
is
that,
if
you
think
about
the
types
of
things
you're
trying
to
trace
electic
Beach
P
communities
as
an
example,
no
they
could
be
changed
on
import
from
what
they
were
learning
as
they
can
be
changed
as
part
of
leaking
them
between
internal
routing
tables,
and
they
could
be
changed
again
when
you
actually
advertise
them.
The
outside
world
policy
is
a
giant
program.
M
So
if
you
want
to
actually
try
the
tracer
program,
you
know
you
do
sort
of
the
commode
of
adding
print
statements
at
the
places
where
things
happen
to
some
extent,
I
think
so
the
thing
that's
trying
to
be
solved
here
is
good.
We're
trying
to
actually
trace
what
happens
to
the
route.
We
probably
when
something
we're
like
tracing
a
program,
so
basically
print
statements
on
whatever
you
make
an
alteration
to
route
properties.
You
want
to
keep
track
of
that.
You
probably
don't
want
to
take
that
giant
list
of
output.
M
Even
if
the
idea
of
how
do
you
record
one
spot
inside
your
policy
like
inside,
if
your
draft
is
no
the
way
we'd
want
to
do
that
into
BMP
itself,
just
because
you
couldn't
keep
enough
stay
to
the
route,
your
routing
table
the
same
least
or
this,
so
it
makes
the
same
comment
I
made
to
Jacob
when
he
was
talking
about
timestamp
and
sticking
this
in
BGP.
No,
that
messes
things
up
here.
M
If
we
wanted
to
keep
the
interesting
stuff
that
full
list
of
things
that
we
changed
not
in
BMP,
maybe
we
take
this
problem
space
and
move
it
out
of
EMP
and
move
it
to
telemetry.
So
you
instead
at
sorry
for
completely
changing
your
proposal.
He's
trying
to
talk
through
know
the
problems
that
we
have
with
just
a
single
item
if
it
becomes
something
more
like
telemetry
and,
like
your
draft
does
say:
oh
you're
not
looking
to
do
this,
for
every
route
in
your
table
doesn't
scale
nicely.
M
It's
good
for
tracing
specific
things
and
zama
tree
means
that
you
can
tell
your
routing
system
router
by
router.
Please
watch
for
this
prefix
that
go
by
and
send
me
this.
The
bug
information
when
it
passes
through
the
system
and
central
collector
can
still
pick
it
up,
not
saying
there's
not
a
huge
amount
of
text
that
we
need
to
do
to
define
this
sort
of
thing,
but
I
do
think
that
telemetry
might
actually
be
a
better
solution.
Space
than
BGP
well.
N
Thank
you
for
the
suggestion
of
considering
today
me
treat
your
PC
as
another
option
and
well
III.
Think
it's
an
optional
thing
approach
to
do
that,
but
I
can't
really
tell
which
one's
better
right
now,
but
I
think
it's
possible
and
it's
discussing
yeah
and
another
thing.
I
forgot
to
mention
sorry
here.
The
policy
ID
part
well,
I
I
want
to
clarify
that
the
policy
here
is
not
the
policy
itself.
It's
just
a
a
indicator
that
you
can
use
to
map
to
your
actual
policy.
M
M
M
The
chain
of
policies
will
often
end
at
a
common
set
of
conditions
that
just
say
yes
or
no
and
doesn't
give
you
a
hint
as
to
how
the
route
has
changed
over
that.
That
course
of
things
so
to
some
extent
I'm
not
arguing
this.
Is
it
useful,
I'm
saying
this?
Is
you
know
the
barest
level
useful?
Maybe
we
should
do
you
know
the
more
complex
thing
all.
F
F
The
thing
that
the
thing
that
I
have
been
asking
for
in
this
area
actually
is
trying
for,
for
my
use,
cases
to
scale
things
down
and
kind
of
what
I
am
asking
for
is
well.
Okay,
please
give
me
the
opportunity
to
tag
on
on
the
drop
root,
primitives
an
optional
exit
code,
so
that
I
can
mark,
for
which
of
he
25
reasons.
I
will
reject
customer
roots.
I
can
actually
mark
why?
T
So,
thank
you
for
the
information
and
ideas
you
have
here,
as
I
mentioned
to
you
privately
I
I
want
to
mention
the
same
comment.
So
pardon
me
if
I
repeat
my
advice
but
I
think
after
the
rest
of
discussion
may
be
helpful.
One
of
the
things
for
some
of
the
actions
that
ridiger
just
described
you
might
do
and
use
the
new
functionality
for
net
comes
push
of
information
based
on
an
action
I.
Just
simply
encourage
you
to
look
at
those
options
because
I
know
you're
serving
customers,
and
you
may
want
to
look
at
it.
T
Part
of
the
problem
is
we
haven't,
moved
the
BGP
yang
models
quickly,
but
we
need
more
eyes
in
review.
So
perhaps
this
is
a
double
tree
if
you
think
the
yang
models
are
better
for
telemetry
or
some
of
the
push
mechanism.
If
you
don't
know
what
the
push
mechanism
is
from
yang,
that's
a
an
ability
to
push
out
telemetry
or
information
based
on
information
that
may
fit
some
of
the
your
things
it
may
fit.
Some
of
the
customers.
T
N
U
O
Penske
Yahoo
Verizon
media,
whatever
it
seem
like
you're
trying
to
solve
two
different
problems
here.
One
is
a
problem
of
what
actually
happens
when
the
route
is
received
to
it,
whether
it's
acceptable
dropped.
The
other
one
is
why
the
what
you
could
almost
get
from
just
come:
nothing,
but
comparing
BGP
ribbon
and
bike
lock,
Reb
effectively
the.
Why
is
a
lot
more
complicated
than
a
single
policy,
at
least
on
my
network?
We
use
chains,
we
use
recursion,
we
use
chains
in
recursion
and
recursion
and
chains.
O
I
have
no
idea
how
you
would
express
that
in
this
type
of
a
message.
What's
more,
as
we
add
communities
perform
action,
remove
communities
in
the
single
policy
in
a
policy
change,
obviously
not
an
or
curtain
so
to
Jeff's
point.
This
level
of
debugging
is
incredibly
useful,
but
I
don't
think
you
can
encoded
enough
of
it
into
a
BMP
message
like
this.
Some
other
form
of
like
basically
trace
option
type
equivalent
or
like
gdb
like
this
is
actual
debugging
info,
where
that's
great
that
you
exited
on
policy
X
but
I
have
seven
terms
in
there.
O
Three
of
them
call
other
policies
that
have
17
other
terms
like
honestly.
I
just
need
a
dump
of
every
value
that
you
came
in
and
the
action
and
then
will
debug
how
the
hell
those
values
came
in
there.
I
just
don't
see
how
you
can
make
it
useful
enough
to
put
all
that
in
BMP.
But
what
happens
like
accept
it
or
not?
That's
useful,
but
then
I'll
just
compare
the
two
tables.
O
The
one
thing
that
that
wouldn't
do
is
the
reason
for
rejection,
like
in
rudaba's
case
but
I'm
willing
to
bet
that
you
have
the
same
issue
where
you
have
change.
You
have
subroutines
and,
like
you
need
to
know
at
what
point
the
exit
could
happen
and
when
you're
doing
recursion,
do
you
pop
out
the
bottom
exit
code?
The
top
all
those
occurs
in
between
things
get
funky,
so
I'm,
not
sure
that
you
can
actually
get
this
via
this.
We
are
not.
O
O
I
mean
so
giving
a
clear
example
policy.
We
have
a
chain
that
says
strip
that
communities
block
bad
routes,
might've
set
local
press
do
something
else
right,
that's
pretty
typical
Cheney
and
then
block
bad
routes
has
a
bunch
of
terms
in
it.
That
say,
are
you
invalids?
Are
you
to
small
prefixes?
Are
you
I
in
a
reserved
like
bunch
of
what
stuff
steps
in
it
the
block
bad
may
exit
because
of
term
three,
which
is
you
try
to
semi
default
and
that's
bought
that?
O
But
do
you
pop
out
that
the
policy
that
did
it
is
block
bad?
Do
you
say
that
its
ingress?
Do
you
say
that
its
term
three
of
walk
bad?
Is
it
blog
bad
because
you've
sent
me
an
adult
community
on
it
that
we
specifically
told
you
not
to
all
of
this
is
actually
meaningful
information.
If
you
want
to
send
the
customer,
why
and
I
think
for
that
you
need
a
stacked
up,
I
mean
so
there's
people
that
you
don't
need
that
for,
but
then
there's
people
that
use
recursion
that
would
need
that
and
I.
O
Don't
think
you
want
to
build
a
protocol
that
basically
says
anybody.
That's
do
anything
even
remotely
interesting
with
your
policies.
You
can't
use
us
so
I'm
kind
of
I'm
kind
of
torn
whether
this
actually
belongs
and
BMP
or
not.
We
need
this
absolutely
I.
Just
don't
know
this
is
the
best
vehicle
for
it.
Okay,.
N
So
I'll
explain
a
little
bit
about
how
I,
like
the
current
policy
format
here
so
for
our
advices.
Typically,
we
don't
use
recursions
under
the
group
policies.
So
that's
typically
just
a
word
policy
for
each
peer
or
each
peer
group,
and
then
there
will
be
multiple
item.
Sorry,
it's
no!
It's
quarter
notes
under
the
row
policy
and
then
under
each
note
there
will
be
if
match
and
apply
sentences.
So
it's
it's
not
like
recursively
policies.
N
D
D
D
We
sends
a
BMP
message
to
the
collector
that
has
the
policy
ID
and
the
notes,
attributes
or
note
ID.
It
may
be
hard
to
correlate.
What's
the
actual
policy
at
that
moment
on
that
box
was
because,
if
I'm
reef
sending
a
new
configuration
to
the
Box
every
few
seconds,
then
I
don't
know
which
version
of
the
policy
was
in
use
when
the
BMP
message
was
created.
D
We
just
comment
on
the
outcome
of
the
policy
and
it
will
be
up
to
the
operator
to
say
value.
One
means
it
was
rejected
because
it
wasn't
bogan
value.
640
means
it
was
accepted
because
it
was
made
it
through
the
entire
policy
or
but
there
will
be
up
to
the
operators
so
yeah
anyway.
I'm
very
interested
in
this
work,
so
I
would
really
like
to
encourage
you
to
continue
working
on
this.
Thank.
N
You
and,
and
and
one
more
thing,
I
forgot
to
mention
that
I
think
yes,
adding
the
outcome
of
the
policy
processing
to
Thea
to
the
message
is
important,
because
sometimes
a
policy
is
is
processed
incorrectly,
but
it
doesn't
change
the
attributes
then
currently
in
this.
In
this
version
it's
not
recorded
but
yeah,
it
should
be
so
yeah.
The
outcome
should
be
added
into
the
message.
P
Thomas
Carr,
Swisscom
I
think
as
well.
It's
very
important
that
we
get
visibility
into
our
policies.
I
think
with
yeah,
sorry,
I
think,
with
adjacency
deep
in
with
a
locally
purchasing
say
rip
out,
we
know
exactly
which
prefix,
which
attribute
we
have,
in
which
rate,
and
now
it
is
information
we
get.
Basically
when
enter
were
at
which
rail
policy
and
at
which
time
what's
very
important,
is
that
it's
well
aligned
with
bgp
yang
model
so
that
we
can
take
a
reference
basically
to
the
configuration
of
the
young
policy.
P
P
N
N
So
both
cases
are
the
the
for
the
forged
a
s
passes
and
it
could
be
utilized
by
some
malicious
attacks
to
to
to
to
lead
to
the
results
that
block
or
to
block
the
road
from
S
4
1
3
4,
because
when
a
receives
the
s
pass
and
detects
a
loop,
it
will
drop
the
route.
So
in
this
draft
we
propose
we
propose
a
enhancement
to
the
inbound
and
also
outbound
policy
to
analyze
such
airs
passes
when
there's
dope
detected.
Ok,
let's
look
at
the
two
examples
so
for
the
inbound
policy.
N
N
We
all
check
the
s
pass
to
see
if
itself,
for
example,
is
if
it's
placed
in
the
original
position
of
the
s
pass
and
then
you
can
do
the
LV
to
see
if
that
prefix
belongs
to
itself
or
if
it's
placed
in
the
20
place,
then
it
can
check
the
business
relationship
with
the
s
with
s
that
is
next
to
it,
to
the
left
and
to
the
right.
So,
for
example,
here
2200
well
check
its
relationship
with
300
and
600
to
see
if
they
are
in
the
right
place.
N
So
this
kind
of
checking
does
not
need
any
business
relationship
referring,
but
but
but
it's
only
limited
to
certain
hijack
checks.
So
it's
not
covering
all
the
cases
yeah
but
yeah
it
can
serve
us.
It
can
find
out
some
hi
Jackie
cases.
So
that's
the
inbound
policy
enhancement
and
for
outbound
see
when
a
a
s
rope
is
detected.
It
will
check
again
the
s
relationships
locally
and
and
see
if
there's
some
euro
Saints
happening
and
yeah
for
the
next
steps.
F
F
Well,
okay
kind
of
for
doing
for
doing
a
s
puff
filtering
I
certainly
would
suggest
look
into
that.
That
may
be
a
very
good
source
for
feeding
into
the
policies
that
you
want
to
run.
I
cannot
tell
whether
you
are
attacking
more
than
can
be
done
by
ASPA,
but
ASPA,
quite
certainly
is
a
very
written
is
a
very
solid
concept
and
well
okay.
The
information
base
that
you
will
get
that
that
way,
we'll
be
very
trustworthy,
and
so
that's
something
that
should
certainly
not
be
ignored.
D
What
I
was
wondering
if
we
know
that
it's
not
a
regular,
yes
path,
loved
you
to
a
route
League,
but
it's
a
forged
s
path
because,
for
instance,
it
fills
the
ROV
check
or
it's
an
impossible.
Yes
adjacency
in
the
SPF,
then
then,
what
because
I'm?
Not
not
the
cases
we
want
to
accept
such
a
routes
anyway.
I
would
think
so.
The
only
follow-up
action
that
I
can
efficient
would
relate
to
your
previous
presentation,
where
in
BMP,
for
instance,
it's
marked
differently
as
malicious,
a
SPF
low,
proud
of
them
very
good.
N
D
N
C
B
C
Of
what
I'm
trying
to
understand
is
what
we're
fixing
but,
and
then
the
sort
of
semi-related
follow-up
I
guess
is
that
it.
It
seems
like
a
lot
of
this
could
be
addressed
probably
already
today,
by
just
looking
in
your
routing
table
to
say.
Show
me
all
of
the
you
know,
looped
paths
that
have
been
you
know
not
selected,
because
they're,
looped
and
sort
of
in
in
an
ideal
world.
C
N
M
Has
so
John
the
path
poisoning
was
a
use
case
I
suggested,
but
it
was
not
originally
part
of
their
core
use
cases
to
restate,
but
I
see
out
of
the
draft
really
what's.
There
is
a
taxonomy
of
the
different
types
of
looping
conditions,
based
on
what
you
see
in
the
path
for
your
own,
a
s
and
those
side
from
the
fact
that
they're
currently
numbered
use
cases,
no
there's
still
a
useful
taxonomy
for
analyzing.
M
What
type
of
look
you're
seeing
being
able
to
have
this
in
your
route
output
also
know,
for
example,
our
implementation
shows
you
by
a
route
failed
to
be
installable.
You
know
ship,
for
example,
a
s
loop
is
one
of
the
conditions.
This
is
effectively
a
sub
condition
of
an
a
s
loop
being
able
to
see
in
a
consistent
multi-vendor
fashion.
What
kind
of
loop
this
is
is
a
useful
thing
where
would
expect
this
to
eventually
show
up
after
the
work
has
become.
M
B
Sorry
Chris
tomorrow,
Google
I,
think
a
list
of
requirements
would
be
great.
What
are
you
actually
trying
to
do
here
and,
if,
like
so
as
Jo,
said
like
so
you
drop
you
get
this
route?
What
are
you
gonna
do
drop
it
accept
it
do
something
else
with
it
that
should
be
part
of
the
requirements.
I
want
to
do
this
thing,
so
I
can
agree
to
accept
the
route
or
I
want
to
do
this
thing
so
I
know
I
absolutely
should
not
accept
it
and
then
also
jobs.
B
N
V
But
such
a
detection
heavily
relies
on
a
precise
lease
of
agencies,
and
there
is
a
problem
and
the
problem
is
I
access
which
are
not
present
in
the
path,
and
there
is
a
lot
of
indirect
I
agencies
there
and
they
are
not
easily
can
there
is
no
way
to
easily
retrieve
them
at
least
for
not
all
axes.
So
there
is
my
comment.