►
From YouTube: IETF95-GROW-20160407-1620
Description
GROW meeting session at IETF95
2016/04/07 1620
A
A
Come
on
computer
there
we
go
all
right,
so
it's
the
grill
meeting,
I'm
Chris!
This
is
Peter.
Oh
there's
note
well
that
you
can
barely
read.
That's
intentional!
Oh,
maybe
that's
the
only
one!
Okay!
Now
we
have
a
den
of
bashing.
Anybody
want
to
add
anything
to
the
agenda.
That's
not
currently
on
it
or
something
you
desperately
like
to
talk
about
at
the
end.
A
Okay,
so
when
you
presently
have
to
stand
in
the
pink
box
for
Alvaro
and
whoever
is
going
to
present
the
other
draft,
sorry
I
didn't
catch
that
was,
must
beat.
We
have
the
filtering
threats.
Stuff
is
often
editor.
Q.
Roughly
problem
definition
is
often
editor
Q
as
well
or
no
sorry,
it's
a
nice
guy,
HD,
evaluation
and
I
think
ITF.
Last
call
now
what
I
understand.
A
A
A
A
A
C
A
C
C
C
Used
in
the
mts
network
and
which
optimized
over
a
network
traffic
and
the
selecting
so
is
optimized
service
for
our
user
service,
but
we
think
we
think
I
think
most
of
the
progress
which
you
in
the
IBM
just
network
cannot
be.
You
know
how
to
increment
a
similar
functionality
in
MPs
in
large
gear.
What's
wrong
the
last
year.
C
Our
native
IPL
background
is
to
your
challenge:
we're
isn't
because
lack
of
the
enabler
in
cab,
cab
handling
technology
and
its
levels,
Avicii
and
forwarding
past
control
mechanism
in
such
a
network-
and
another
reason
maybe
depends
on
the
background
of
the
operator
used
employed
network,
for
example,
most
of
mainstream
service
programs,
fixed
line
operator
in
China
most
set
up
their
IP
backbone
means
of
I
native
IP.
Instead,
MP
is
so,
for
example,
in
China
Telecom,
almost
a
90
percentage
of
internet
service,
internet
traffic
say
I
wrong.
C
C
We
have
helped
you
in
turn
separate
the
core
into
major
network
and
the
core,
and
the
major
network
covers
and
aggregate
public
software
for
high
precision
and
millions
of
our
IP
subscriber
millions
of
a
public
subscribers
I
position.
And
meanwhile,
the
core
network
is
a
full
mesh
switch
fabric
and
account
and
connect
can
as
thousands
of
virtual
network.
C
Is
the
girl
with
the
scale
of
network,
grows
up,
for
example,
when
we
are
use
the
unreasonable
control,
a
routing
policy
to
our
tutor
schedule?
Traffic
may
cause
network
vibration
in
the
run
time,
so
many
operators
realize
that
most
of
the
scheduling
checks
getting
work
are
temporary
high,
temporary
employment
in
the
case
of
traffic
and
congestion
control,
and
if
last
year,
network
traffic
re
adaption
happen,
it
means
you
have
to
reconfig
or
you
help.
You
increase
your
network
capacity
in
instead
of
traffic
theory.
So,
comparing
to
the
traditional
overall
network,
optimization.
C
Here's
the
big
feature:
here's
the
big
feature
of
the
overall
check,
steering
input
that
that
the
operator
try
to
implement
in
reality
is
that
just
a
model
you
can
down
the
capability
of
the
operator
have,
but
is
that
we
can
see
this
causes
of
three
parts.
The
other
12
is
the
quality
control
or
say
the
TE
server.
We
also
need
the
information
channel
for
the
link
estate,
information
and
the
routing
information
like
the
net
floor,
I
GPU,
officiating
state.
C
C
So
the
motivation
behind
this
informational
jobs
can
be
simplified
as
to
aspect.
One
is
improved:
the
utilization
of
network
network
locally
and
configured
the
new
config,
the
so
is
passed
and
the
related
policy
automatically.
They
don't
try
to
enumerate
all
case,
but
just
give
a
brief,
a
fast
go
fast.
So
the
use
case,
I
think
the
page
is
some
some
sentence.
C
Yes
is,
I
think,
the
temporary
and
I
sorry:
ok,
ok,
I
I,
continued
yeah.
C
Nevermind
yep,
we,
the
the
following
a
use
case
of
each
interviews,
can
be
category
two
dimension.
One
is
temporary
case.
Another
is
a
persistent
temporary
case
means
that
traffic
is
scheduling,
happened
in
the
check
with
gem
and
ran
out
of
the
network's
resource,
and
it
happened.
Does
the
controls,
the
council,
the
SDN
controller
system
can
come,
he
assigned
traffic?
You
can
remember
all
the
schedule,
traffic
and
rope
under
owning
them
back
once
the
your
system
or
the
network
operation.
C
C
C
C
He's
the
similar
case
happen
in
there
a
similar
case
like
the
previous
just
different
in
the
is
happening,
then,
between
that
the
matching
Network,
for
example,
the
the
next
line.
The
next
pass,
became
that
you
met
your
network
congest,
you
can
choose
to
another
from
another
passed
from
the
original
national
accreditation.
C
Is
another
case
is
different.
Lies
in
the
is
happening
that
cornea
Kono's
example.
The
call
that
the
traffic
between
the
sea
to
the
corn
oats,
a
wonderful
when
the
link
below
congest
congest
link
be
happening
here
and
alternate
choice,
the
past
from
cqp
and
a
is
happening
in
the
core
network.
The
previous
line
is
that
became.
The
Metro
is
a
some
images
inside
that
the
a
s
and
the
press,
one.
Maybe
the
into
a
s-
is
that
another
typical
case
in
India
for
the
q
SP
or
operator.
C
The
middle
is
the
interface
between
that
US
v,
a
provider
interchange
exchange
point
in
one
in
the
different
city
once
locate
the
traffic
congestion
in
one
link,
we
can
choose
another
long-distance
into
a
speeding
exists
between
that
you
see.
T
is
another
case
for
the
transit
transit,
transit
AAS,
for
example.
Once
that
curious
of
JCP
here
occur.
Yes,
the
chest
day
will
be
preferred
and
the
same
shape
of
your
go
through
the
transit
is
a
instead.
C
Yes,
the
qos
yeah,
here's
the
OT
key
requirement
last
one
in
the
OT
key
requirements
is
derived
from
the
transit
networks
and
you
know
Jensen
company's
Network.
They
is
a
typical
counting
counting
provider
in
China.
Is
a
history
exist
with
ice
P
in
the
city,
a
B
and
C?
They
hope
to
Proteus
service.
You
can
choose
different
website
to
ice
P,
just
like
we
did
in
the,
for
example,
the
local
preference
setting
in
the
PGP,
so
they
can
choose.
The
virtue
are
pop
their
service
to
the
party,
so
soft
Weber.
C
On
the
contrary,
they
also
have
the
requirement
on
inbound
traffic
theory.
You
can
choose
when,
where
the
traffic
came
back,
that
mail,
which
you
have
choice
on
insurance
for
inbound
traffic
but
I
CP
may
choose
to
operator,
may
choose
ignore
even
a
probably
be
such
attention
to
affect
this
pass
so
another
that
negotiated
mechanism
is
needed
here.
C
You
can
summarize
that
for
the
native
live
in
your
work,
we
have
your
face
schedule
flow
automatically
with
fine
granularity
is
a
challenge.
We
also
need
a
new
policy
control
plane
and
two
to
provide
end-to-end
the
service
attitude
and
you
ended
the
traffic
scheduling
mechanism.
So
we
hope
the
further
work
and
will
continue
to
collect
more
requirements
in
teacher
Thank,
You.
F
F
In
many
cases,
I
think
there
are
components
that
you
have
as
requirements
that
are
actually
kind
of
impossible,
and
so
I
I
have
some
concerns
that
translating
those
requirements
into
into
functions
is
a
bit
problematic
in,
particularly
with
requests
to
a
scheduling
that,
with
fine
granularity
like
a
lot
of
scheduling
that
has
these
long
feedback
loops
results
in
a
tendency
to
polarize
your
traffic
ie,
the
one
that
looks
best
right
now.
A
Hi
Chris,
so
a
lot
of
your
presentation
seems
to
me,
like
you,
wanted
to
do.
G
So
I
just
want
to
control.
I
just
want
to
comment
on
the
control
system
that
you
need
to
do.
Oh
sorry,
I'm,
Peter,
Lawford,
the
control
system.
F.
Is
you
have
a
PID
regulator
and
in
this
case
you
have
infinite
gain
and
no
dampening,
and
it's
proven
to
be
unstable,
and
so
it
doesn't
matter
what
you
do.
G
The
reason,
traffic,
if
you
try
to
find
the
math
who
you
look
at
all
these
people,
that
do
traffic
engineering
with
their
favorite
traffic
engineering
to
you,
try
to
figure
out
what
they're
actually
doing,
and
you
end
up
finding
that.
Well,
what
they're
doing
is
they're
inserting
a
delay
in
the
control
loop
and
a
random
generator.
That's
the
human,
so
the
math
for
most
traffic
engineering
cases.
Is
it
sad.
The
delay
and
a
random
generator.
F
So
particularly
if
it's,
if
it's
about
working
around
resource
constraints
that
are
imposed
by
other
parties,
that
has
a
tendency
to
be
it
another
sort
of
deep
game
theory
exercise,
because
the
other
party
is
also
working
around
some
constraints
with
respect
to
their
posture
to
you
and
so
you're
actually
engaged
in
traffic
engineering,
arms
race,
and
so
those
are
things
that
you
know
get
fed
into
models
and
I.
G
H
Maria
for
a
while,
okay,
evil,
Peter
Peter's
assumption
is
that
you
throw
sufficient
over-provisioning
into
the
game
so
that,
while
okay,
actually,
you
have
high
quality
and
no
problems
from
the
start,
and
then
Peters
game
works
not
not
ever
no
well,
okay,
not
not.
Everybody
has
sufficient
resources
in
all
the
places.
All
the
time,
but
having
sufficient
resources
is
something
that
can
not
be
emulated
by
any
magic.
If
you
don't
have
sufficient
resources,
so
kind
of
my
question
would
be
how
much
spare
capacity
or
how
much
overloading
is
in
your
is
in
your
problem
set.
H
If
you
are
running
at
five
five
percent
spare
capacity
in
the
complete
network
for
half
of
a
day,
you
cannot
do
good
quality
service
or
even
even
even
a
tired
service
by
any
means
of
control
and
well.
Okay,
if
you
can
afford
a
serious,
serious
spare
capacity,
which
of
course
is
also
needed
for
securing
against
failures.
H
Well,
okay,
I
think,
I,
think
I,
think
Peters
notion
of
well.
Okay,
delaying
delaying
computations
and
tuning
of
unit
tuning
of
parameters
actually
works
in
an
extreme
example.
It
might
work
like
well,
okay,
you
measure
of
you
measure
for
a
week
and
figure
out
what
the
main
problems
in
the
load
versus
the
capacity
are,
and
you
compute
and
you
compute
the
best
distribution
on
fo
and
for
the
next
week.
You
are
at
least
approximating
a
good
distribution.
I
C
I
I
understand:
what
do
you
worry
about?
One
is
performance.
Another
is
maybe
the
granularity.
What
kind
of
gradual
on
to
achieve
and
the
time
at
a
time
a
requirement.
It
seemed
that
you
are
scheduling.
I
can
say
it
is
a
real
acquire
material
at
you.
In
the
cooperation
between
us.
These
Henson
Company,
a
you
know,
is
a
big
contests
are
provided
in
China
biggest
la
big
one
in
China,
a.
C
C
C
C
B
C
B
D
J
Hi
I'm
Ilana
I
did
this
with
russ
and
Sue.
We
actually
been
doing
this
for
quite
a
while.
We
presented
the
sittings
person
only
for
the
last
time
we
work
to
Berlin,
so
it's
been
a
while
last
time
we
talked
about
this.
It
actually
created
a
whole
bunch
of
discussion
on
the
middle
list,
I
think,
because
we
didn't
really
explain
the
intent
of
what
we
were
doing
about
the
mechanism.
So
the
same
thing.
The
changes
between
then
and
now
are
mostly
clearing
up
exactly
what
this
is
intended
to
be.
J
It's
there
in
green.
It
should
be
in
red
and
make
whatever
an
optional
mechanism
to
remove
a
prefix
without
overlaps.
We
know
that
there
are
people
that
do
this
in
their
network,
for
whatever
reason
and
all
we're
specifying
here
as
a
mechanism
you
can
use
to
maybe
automate
that
in
your
network
again,
this
is
not
intended
for
internet
white
deployment.
We're
not
saying
everyone
should
go
and
turn
this
on
everywhere.
J
We're
not
recommending
people
to
filter
anything
we're
just
saying
if
you
are
already
filtering,
you
may
want
to
use
this
mechanism
instead
of
manually,
going
to
put
filters
everywhere
so
with
the
objective
of
not
getting
any
spam
about
the
q6
examples,
I'm
using
a
TAC
example,
and
all
the
other
shows
is
somebody
else
on
here.
That
is
heaven,
as
in
two
runs
one
long
route
in
one
short
route,
they're
overlapping
runs,
and
so
this
guy's
advertises
this
long
on
the
short
up
this
way
and
only
the
long
run
the
other
way.
J
So
what
happens
is
of
course,
6403
gets
all
the
routes
from
everyone
else
and
assuming
that
6503
wants
to
go
this
way
because
of
the
policy.
Maybe
they
make
the
decision
that
they're
going
to
filter
the
more
specific
route
that
won't
change
their
dynamic
over
the
network
right.
They
will
keep
going
down
this
way
because
that's
what
they
decided
to
do,
and
in
fact,
in
this
example
and
I
know,
the
intern
is
obviously
more
comfier
than
this.
The
guys
behind
him
don't
see
a
change
either
right,
because
it
would
come
this
way.
J
Your
advertising,
your
block,
is
the
covering
ground.
So
when
you
go
read
the
draft,
that's
what
it
says
we
use
the
term.
The
term
bounded
and
all
this
means
is
we're
referring
to
a
community
that
we
use,
which
is
a
normal
community
that
you
can
put
and
define
yourself
in
your
network.
The
intent
is
that,
if
we're
going
to
filter
the
overlapping
around-
and
you
bound,
you
retain
the
route
in
your
network-
that's
the
intent
of
bounty
right
for
you,
two
to
bound
the
propagation
around
and
then
in
the
draft.
We
talk
about
fully
overlapping.
J
So
what
this
means
basically,
is
that
if
you
receive
the
route
from
both
overlapping
and
covering
routes
from
the
same
pier
and
they
have
the
same
way
as
path,
that
means
that
they're
functionally
the
same
and
so
like
in
this
example
over
here,
the
routes
coming
from
65,000
will
have
the
same
path
and
the
same
next
hop
for
65
thousand
and
three.
So
if
you
remove
one
the
longest
one,
there's
functionally
no
change
for
65,000
three.
J
The
draft
does
say
that
if
you
really
want
to
do
some
other
type
of
oil,
overlapping
filtering-
let's
say,
for
example,
routes
of
info
different
peers-
that's
up
to
you,
we're
not
telling
you
to
do
it
or
not.
So,
there's
four
obvious
steps
to
this
thing.
One
is
obviously
identifying
in
marking
your
black
routes
you
can
by
looking
at
them
or
by
having
the
router
and
a
lesson
you
know,
which
ones
are
overlapping
and
fully
overlapping,
for
example,
is
path
top
and
marking
them
with
this
bounded
community
again
this
is
just
community.
J
You
use
locally
significant
to
you,
then,
with
one
of
the
use,
of
course,
prefer
the
mark
routes
so
that,
if
we're
going
to
bound
that
more
specific,
we
don't
want
not
only
that
route
but
other
more
specific
routes
that
match
it
to
escape
Ras.
So
we
only
mark
the
overland
route
with
the
bound
of
community
and
we
make
that
be
preferred
inside
Ras
bonus
up.
How
do
you
do
that?
You
can
do
a
little
preference
if
you
want
to
or
our
preferred
mechanism,
of
course,
to
use
a
cost
immunity.
J
That's
going
to
prefer
that
in
the
draft
we
say,
use
the
absolute
value
of
the
community
inside
you
Ras,
you
can
optionally.
If
you
want
to
not
install
that
route
round
table
its
marked,
you
want
to
not
install
it.
You
probably
the
sale,
of
course,
so
you
can
filter
it
or
compare
it
using
the
community
and
then
what
we're
saying
in
the
draft
is
that
you
should
not
you're
going
to
go
through
all
this
stuff.
You
should
not
advertise
this
to
your
external
peers.
J
J
J
Same
longer,
prefix,
the
the
one
with
the
bounded
community,
because
we're
using
either
local
preference
with
Kazmir
would
just
be
preferred.
That
would,
of
course,
rather
see,
of
course,
we'll
get
that
as
well.
If
you're
implementing
that
should
not
advertise
external
e
than
65,000
for
would
only
get
the
shorter
prefix
again
for
what
we
talked
about
before.
This
doesn't
change
the
dynamics
of
the
routing,
and
this
is
my
last
slide.
Yeah.
G
J
L
Gabriel,
cisco
are
excellent
proposal.
Arm
should
consider
it
now.
I'm
most
implementations
do
take
into
account
that
if
the
least
specific
route
and
the
most
specific
most
specific
route,
our
encounter
at
sea
and
the
next
stops
being
sane,
they
can
auto
aggregate.
I
think
you
can
extend
it
to
say
if
the
more
specific
route
comes
from
a
different
next
hop
and
you've
seen
the
instance
of
more
specific
route
coming
from
the
same
next,
stop
as
well,
you
can
order
aggregate
it
which
will
effectively
increase
the
scope.
J
J
M
Hi
hi
hello,
Nicholas
Antonella
from
your
way
I.
Just
if
I
know,
if
I
didn't
misunderstood
what
you
explained
it,
you
basically
mark
the
/c
64,
because
it's
more
specific
that
they
slashed
easy
one.
How
about?
If
you,
if
a
s
in
that
example,
if
a
s
65000
advertises
a
different
/
series,
64
block
of
the
bigger
one
of
the
CD
one,
for
instance,
I,
don't
know
and
a
different
one.
You.
J
M
Is
11,
but
how
about?
If
you
want
to
do
that
normal
way,
you
do
because
you
don't
want
to
export
the
/
city
one
and
the
other
asn.
That's
not
install
that
route.
He
needs
routers,
so
you
I
mean
you
normally
may
do
that
for
load
balance
proposes
you
publish
the
beer
blog
one
small
from
here
once
more
from
here.
But
if
you
discard
this,
you
are,
you
know
you
are
screwing
my
load
balancing.
Yes,.
J
M
J
G
So
good
love
her.
This
is
slightly
off
the
talk,
but
it
must
have
happened
on
your
mailing
list
that
I
didn't
read.
So
why
why
there
are
two
I
can
see
two
reasons
why
they're
more
specific
is
coming
out
of
the
65
000
a
they
don't
know
what
they're
doing
or
be
they
are
trying
to
do.
Traffic
engineering
is
there
any
other
reason
and
that
brick
and
my
next
question
is
going
to
be
how
what's
the
what's
the
split
between
don't
know
what
they're
doing
worse
is
they
do
traffic
engineering,
the.
J
First
question:
the
answer
to
both
questions
is
I,
don't
know,
I,
don't
know
if
there
are
other
reasons
and
I,
don't
know
what
the
split
is,
but
video
say
I
figure.
Those
are
the
true
reason
we
see
or
I
mean
everyone
sees
that
reporter
comes
from
jeff.
I
think
in
others
of
everyone
that,
though,
that
doesn't
aggregate
right
and
that's
or
the
one
that
will
you
leave
all
the
routes.
K
Geoff
Huston
and
the
first
one
wyd.
If
I
do
this,
when
I
actually
asked
one
of
the
poker
at
the
top
of
the
cider,
but
why
do
you
do
this?
Their
answer
at
the
time
was
so
no
one
else
can
hijackers
with
the
/,
24
and
I
kind
of
can't
argue
with
that
I
make.
It
seems
a
bit
crazy,
but
it's
true,
you
know
the
best
they
can
do.
The
attacker
can
do
another
/
24,
let's
play
routing.
B
K
Rather
than
more
specifically
so
bear
in
mind,
there's
a
third
thing:
what
kind
of
an
kinda
just
kind
of
gets
me
on
this?
Is
you
have
a
diagram
where
all
this
works
then
I
sort
of
think
I?
Yes,
I
can
do
a
diagram
when
it
doesn't
work
and
part
of
the
reason.
I
suppose
why
I'm
sort
of
getting
up
here
is.
K
That's
what
I
want
to
do
when
I
announce
this
stuff,
so
no
matter
what
you
do
I'm
going
to
counter
it
because
I'm
going
to
win
and
then
you
try
and
defend
so
I'm,
not
sure
this
sort
of
gets
us
anywhere
other
than
you
escalate.
One
step
up
and
I
go
hot
you're
escalating.
Let
me
escalate
so
while
well,
I
think
that's
fine
and
in
an
abstract
sense
you
could
knock
out
around
twenty
twenty-five
percent
of
the
advertised
routes,
600,000
down
to
400,000
yay.
K
Let's
go
to
the
pub
you
know:
everyone
wins
in
actual
fact,
you'll
have
650
thousand
routes
because
I'm
going
to
Blackmoor
in
and
through
various
ways
to
get
around
you
because
bgp
is
actually
a
negotiation
policy,
not
an
announcement
policy.
You
know
the
protocol
is
all
about
who
wins
and
you
know
who
gets
to
determine
this
so
I'm
trying
to
I
can
see
what
you're
trying
to
do-
and
it's
great,
the
pessimist
in
me
says
near.
J
Not
going
to
work,
ok,
so
to
personally
answer.
One
is
reason:
well
one
is
yes
again:
if
you
already
have
the
policy
we're
not
telling
you
to
go
full
term
or
if
you
already
have
a
policy,
you
can
use
this.
So
if
you
look
back
at
ITF,
whatever
number
it
was
in
product
last
time
there
was
a
guy
who
came
and
presented.
J
J
K
K
K
J
So
let
me
say
this
again:
if
you
already
are
doing
this,
you
can
use
the
spec.
You
listen,
I'm,
not
telling
everyone
to
go,
run
out
and
change
the
router
configurations,
I'm
not
advocating
the
Internet
to
go
filter,
local
IP,
realms
I'm,
saying
if
you're
already
doing
this,
then
you
can
use
this
mechanism.
That's
all
very
simple:
it's
bigger
WR.
H
Folk
well,
ok,
I
would
change
we
discussion
topic
slightly.
I
have
been
thinking
about
policy
a
long
time
and
once
in
a
while,
I
did
thought.
Experiments
were
a
feature
like
this
was
used,
I'm
not
going
to
dive
into
what
policies
can
be
used.
The
question
that
occurred
when
I
did
Hoover
thought
experiments
was
well
okay.
What
kind
of
feature
is
actually
implementable
in
the
bgp
implementations
saying
yes,
well,
okay,
we
introduced
the
generation
of
this
community.
H
Okay,
that
sounds
easy.
I
am
NOT
into
the
bgp
code,
and
I
guess
that
generating
and
with
drawing
attributes
on
a
route
depending
on
the
existence
on
the
existence
of
another
route
is
actually
something
that
is
not
straightforward,
and
so
I
really
would
like
to
ask
some
of
the
people
who
know
how
the
code
can
work
that
are
in
the
room,
whether
doing
something
like
bounded
and
getting
it
withdrawn,
right
and
so
on.
Actually
is
something
that
is
feasible
and.
D
N
Only
play
a
bgp
implementer
on
TV
anymore,
so
yeah.
I
guess
I'll
try
to
answer
that,
which
is
you
know
everything
can
be
done
in
software
right.
Sorry,
John,
Scudder,
it's
it's
not
completely
trivial,
I'm
pretty
sure
that
RFC
4271
still
says
you're
not
allowed
to.
But
you
know
you
know
you
can
ignore
things,
but
yeah
I
think
you
know
point
point
taken
that
it's
not
a
completely
straightforward
feature
to
implement
the
thing.
N
I
actually
came
up
to
the
mic
to
say
Jeff
mostly
already
said,
I
want
to
agree
with
his
comments.
I
will
add
on
I.
Don't
think
he
ever
quite
fully
expanded
out
the
the
example
which
is
you
know
that
the
rest
of
the
game
here
right
is
when
s6500
one
starts.
You
know
playing
his
filtering
game
is
a
a
65,000.
You
know
takes
that
/
61
and
explodes
it
into
whatever
the
you
know
hopefully
minimal,
but
maybe
not
it
could
be
all
the
64's
and
starts
advertising
them
same.
N
Now,
and
then
that
was
my
last
comment
was
to
reply
to
your
answer,
which
is
you
know,
spoken
like
a
true
arms
merchant
takes
one
to
know
one,
but
but
on
the
other
hand,
since
we're
here
in
the
grow
working
group,
if
you
don't
think
it's
a
good
idea
and
you're
not
advocating
for
it,
how
are
we
making
the
internet
a
better
place
by
by
publishing
this
RFC,
or
this
draft,
so
I
didn't,
say:
I.
J
N
N
The
follow-up
is,
of
course,
if,
if
all
you
were
doing
was
enabling
a
policy,
and
it
wouldn't
make
any
difference,
why
are
you
publishing
a
document
that
doesn't
make
any
difference?
The
answer
is,
of
course
it
does
make
a
difference.
When
you
introduce
a
convenience
feature
and
even
though
you
you
know,
are
saying
a
disclaimer.
What
you're
actually
doing
by
introducing
a
convenience
feature,
is
you're
encouraging
people
to
turn
it
on.
M
And
so
my
comment:
is
you
keep
referring
alvaro
that
if
someone
has
that
policy,
I
understand
that
this
is
the
policy
to
filter
more
specifics.
However,
the
point
of
more
specific
specifics,
as
Peter
mentioned
this
actually
DTE,
and
now
we
have
the
white
communities
as
part
of
that
I
collected
lot
of
feedback
from
operators
and
they
wanted
to
have
one
value
which
would
enforce
te
and
hops
away.
So
if
that
community
is
attached
to
64
and
you
just
delete
it,
you're
making
those
lives
miserable
right.
J
So
again
that
guy
over
there
3
may
already
have
that
policy
because
he
doesn't
care
doing
what
6500
is
the
way.
It's
you
know,
maybe
he's
not
getting
paid
or
whatever
it
is.
If
the
white
community
somehow
says
must
never
delete
the
route
or
something
right
and
still
in
hops
away,
then
this
policy
should
be
applied,
because
obviously
that
is
a
must
right.
The.
M
J
A
Chris
tomorrow,
so
I
this
strikes
me
is
a
lot
like
VA
when
I
read
it
originally
and
it
kind
of
still
in
the
same
concept,
so
sort
of.
Let
me
reduce
my
routing
my
individual
routing
domain
review,
but
still
maintain
reach
ability
right,
which
you're
doing
you're
saying
there's
overlapping
routes.
I
want
to
get
rid
of
the
small
stuff.
I
want
to
save
some
space
time
memory
speed.
What
have
you
I
still
want
to
be
able
to
reach
those
places?
That
seems
fine
right.
A
I
think
all
the
conversation
about
you
know
she
did
this
or
not.
People
like
you
said
people
already
do
it.
If
you're
trying
to
document
a
process
or
procedure
or
configuration
I
think
that's
interesting,
I,
don't
know
that
I
would
do
it,
but
you
know
that's
that's
cool
right.
Everyone
does
their
own
thing.
I!
Think
that
this
it
falls
into
like
a
consenting
adult
problem
like
right.
A
You
know
you
decide
to
lose
route
to
your
network
so
that
you
can
save
money
time,
speed
whatever
cool,
but
at
the
end
of
the
day
you
still
have
to
reach
for
it
the
rest
of
the
internet.
So
there's
some
complexity
that
you're
implicitly
or
explicitly,
adding
that
may
they
may
not
save
you
money
like
you,
think
it
does
I.
Would
you
I
think
I
think
your
paper
document
sort
of
says
that,
but
at
the
end
of
the
day,
if
your
documented
configuration
or
set
of
configurations
that
that's
that
seems
fine
right.
L
O
Understand
what
you're
I
understand,
what
you're
proposing
up
here
I
the
benefit
it
could
provide.
However,
like
Jeff
I
have
my
doubts
about
its
efficacy,
in
the
sense
that,
if
you
think
about
these
lines
on
the
graph,
there
may
be
assumptions
made
about
the
the
interconnectedness
of
these
guesses.
That
is
not
necessarily
true,
or
at
least
not
necessarily
true,
in
certain
regions
of
the
world,
meaning
there
are
a
lot
of
internet
exchange
points
out
there.
O
J
O
Okay,
well,
then
same
comment
still
applies,
though,
because
he's
going
to
need
to
be
able
to
see
or
65,000
for
or
other
a
esas
as
they
form
these
adjacencies
are
going
to
need
to
see
that
more
specific
as
the
interconnection
mesh
increases
in
density,
so
I
just
again,
I
I
doubt
the
efficacy
of
this
proposal
in
courtroom.
Is
it
quick
Ruettiger
because
we're.
H
O
H
Yes
again,
I
know
we
are
not
an
eye.
Dr
IDR
should
shoot
and
needs
to
decide
whether
the
Baja,
the
feature
can
be
done.
The
any
feature
that
is
introduced
can
help
people
to
create
bad
policies.
I
know,
I
could
use
the
feature
for
things.
I
cannot
do
without
it
that
that
are
beneficial
for
my
network
and
do
not
have
any
external
consequences,
except
that
I'm
serving
my
customers
better.