►
From YouTube: IETF102-ILNP-20180715-1345
Description
Identifier Locator Network Protocol
This talk provides an introduction & overview of the Identifier-Locator Networking Protocol (ILNP), focusing on ILNP as a superset of IPv6.
The core idea behind ILNP has its roots in the "8+8” naming architecture of
the early 1990s, and discussions within the IRTF Name Space Research Group (NSRG), circa 1999-2002. ILNP is defined as an IRTF Experimental protocol in RFC 6740 to RFC 6748, and supported by ongoing research from lab results based on running code. This talk will cover the key architectural concepts of ILNP and also the specific design & operation of ILNP as a superset of IPv6, with examples of how ILNP implements mobility. More information is available at the ILNP website.
Slides from presentation:
https://datatracker.ietf.org/meeting/102/materials/slides-102-edu-sessg-an-introduction-to-the-identifier-locator-network-protocol-ilnp-00
ILNP meeting session at IETF102
2018/07/15 1345
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
Just
do
this
hello.
Everyone
I
just
wanted
to
welcome
you
to
this
tutorial
Joel
Halpern's,
going
to
talk
about
identifier,
location,
network
protocol,
I'm,
Karin
Donahue,
with
the
education
team.
If
you
have
any
questions
or
comments
or
suggestions
for
future
tutorials,
we
would
welcome
and
I
believe,
did
you
get
the
link
at
the
end
of
the
slides,
there's
perfect.
We
have
a
survey
and
we
would
really
like
you
to
take
it
up
to
fill
that
out.
Thank
you
very
much
and
enjoy
your
day.
C
Thank
you
all
for
coming:
I'm
Joel
Halpern,
presenting
this
material
on
IL
NP,
basically
on
behalf
of
ran
and
saline
and
doing
tons
of
work
from
quite
a
number
of
years
on
it,
no
I,
don't
thank
you
we're
not
in
a
cave
and
unfortunately,
neither
of
them
could
make
it
here
this
time.
So
they
asked
me
to
present
it
I'm
happy
to
do
so
and
happy
to
take
questions
during
the
presentation,
because
questions
aren't
the
right
time
or
better.
C
If
the
question
is
too
detailed
or
takes
too
long,
I'll
ask
you
to
look
at
the
backing
material.
There's
lots
of
backing
material
that
I
will
not
be
presenting
today,
because
we
only
have
an
hour
and
if
that
doesn't
work,
find
me
during
the
week.
I
am
more
than
happy
to
talk
about
this
at
whatever
detail
during
the
week,
but
I
can
only
go
through
so
much
during
this
session.
C
I
want
to
really
try
to
give
you
a
view
of
what
this
is
why
this
is
and
what's
been
done
with
it
so
I'm
presenting
ran
labels
himself.
Co-Conspirator
he's
been
doing
this.
He
basically
came
up
with
the
idea
based
on
older
work
from
Mike,
Odell
and
other
people
and
then
came
and
gone
selim
at
the
university
st.
andrews
is
trying
to
work
using
a
whole
bunch
of
using
working
with
a
whole
bunch
of
his
students.
They've
done
a
lot
of
work
over
the
years.
This
is
based
on
the
writing.
C
Research
group
work
on
the
name
space
research
group,
all
sorts
of
work
that
was
done
in
other
parts
of
the
IETF
and
lots
of
men
discussions
so
background
IP
address
is
considered
harmful,
going
back
to
what
this
is
the
most
recent
one,
but
it's
a
paper
on
why
the
fact
that
IP
addresses
represent
both
location
and
identity
and,
in
some
cases,
semantics
and
other
things
combining
all
those
into
one
thing
creates
a
mess,
makes
our
life
much
more
difficult
than
it
otherwise
needs
to
be
now.
There's
a
long
tradition
of
combining.
C
C
That's
our
early
history
on
this
RFC
six
one
one
five
is
the
output
of
the
routing
research
group
produced
by
the
chairs
of
the
research
group.
This
is
not
a
consensus.
Rfc
one
of
the
nice
things
about
the
routing
about
the
research
groups
here
at
the
IETF
level.
Really
the
IRT
F
is
they
can
produce
results
that
aren't
necessarily
consensus,
and
this
is
what
the
chair
is
concluded
and
this
was
their
recommendation.
C
They
won't
read
it
out
carefully,
but
it's
worth
knowing
a
number
of
people
who
are
not
involved
day
to
day
with
the
il
NP
I
think
this
is
useful
stuff.
It
addresses
real
problems.
Note
that,
for
example,
Lisp,
which
I
actually
co-chair
was
another
effort
to
solve
the
same
problems
and
Lisp
is
really
useful
for
a
bunch
of
other
things.
It
turns
out
not
to
solve
internet
scaling.
C
It
is
a
set
of
architectural
concepts
and
constructs
and
approaches
and
ways
of
thinking
about
it
and
those
are
documented
in
RFC
and
a
set
of
specific
protocol
mechanisms
and
behaviors
that
realize
that
that
say,
okay,
if
you're
trying
to
do
this
degree
of
separation
while
not
having
to
encapsulate
well
not
changing
the
way
the
Corps
works.
All
the
other
things
you
need
to
have
an
operational
system.
How
do
you
do
it
and
it
goes
through
in
a
fair
bit
of
detail
and
there's
a
base
protocol
document.
C
There's
an
ICMP
document
for
some
extensions
that
are
needed.
There's
the
DNS
extensions
document.
It's
done
in
a
series
of
RFC.
Is
that
tell
you
how
to
do
the
protocol
after
you
have
the
architecture
so
engineer
the
leveraging,
the
current
environment?
This
is
not
go
out
and
replace
all
of
the
routers
in
the
world.
I've
never
been
able
to
figure
out
how
to
do
that.
So,
let's
not.
C
C
But
that
means
that
if
you
have
multiple
interfaces,
you
have
separate
separate
IP
addresses
which-
and
she
was
full30
picking-
which
one
is
going
to
receive
things
or
send
things.
But
it's
not
too
useful
for
saying
wait.
I
was
talking
TCP
over
there,
I'd
really
like
to
be
using
it
over
here
doesn't
seem
to
work
right,
and
there
are
various
interesting
combination
to
try
to
solve
that
without
making
these
changes.
C
C
In
contrast
for
News
il
NP,
you
have
DNS
names,
you
have
no
to
identify
errs,
which
are
what
transport
can
work
in,
and
that
includes
things
like
IPSec,
which
yes,
IPSec
is
bolted
into
IP,
but
really
it's
a
transport
than
it
can
be
TLS.
The
IPSec
can
work
in
terms
of
the
node
identifier,
not
in
terms
of
the
locator.
The
network
works
in
terms
the
locator
and,
of
course,
your
interface
works
in
terms
of
its
MAC
address,
because
we
have
neighbor
discovery
and
that
can
do
all
the
things
it
needs
to
do
mean.
C
We've
in
fact,
gotten
further
and
further
away
from
embedding
media
addresses
into
the
lower
portions
of
IP
addresses.
We
still
do
it
sometimes,
but
we
see
lots
more
work
on
privacy
and
other
things.
That
say
you
know,
maybe
that's
not
how
you
want
to
do
it.
However,
you
want
to
do
it.
We
don't
care
your
locator
can
be
what
an
identifier
can
be,
what
they
need
to
do
so
we're
now
going
to
talk
about
engineering
and
on
this
slide
we're
going
to
use
the
abbreviation
ILM,
pv6
and
I.
C
Think
on
the
rest,
we
just
talk
I
L&P,
that
the
assumption
is
we're
talking,
Island
P
for
v6.
They
actually
did
do
as
part
of
the
early
work,
an
internet,
an
RFC
on.
How
would
you
do
this
engineered
for
ipv4?
It
uses
things
that
don't
work
very
well
on
the
internet
like
IP
options
that
need
to
be
manipulated
in
ipv4
it
just
you
can
make
it
work,
but
it's
not
a
good
choice.
Don't
there's
not
enough
bitching
or
dying
to
dress.
C
Don't
try
that
so
we're
gonna
focus
on
ipv6
we're
strolling
is
a
little
IP
option
for
one
thing,
but
it's
something
that's
only
into
end
and
it's
designed
so
on
the
wire
it
just
the
same
ipv6.
We
always
know
the
same
packet
format.
It
goes
through
the
routers
exactly
the
same
way,
with
this
same
prefix.
Look
up
that
routers.
Do
this?
No,
no,
no
specially
modified
routers
anywhere.
It
just
works
and
whether
that
frankly,
I
don't
care
if
they're
switches
that
are
looking
at
it
routers
it,
it
just
works
with
our
packet
handling
infrastructure.
C
Yes,
there's
a
one
of
the
RFC's
talked
about
interesting
proxies.
You
can
make.
You
can
add
things
if
you
need
to,
but
fundamentally
works
with
the
existing
infrastructure
all
the
way
through
it,
and
it
works
with
existing
binaries
and
you'll,
see
and
later
in
the
presentation.
How
that's
achieved
it
works
with
existing
applications
and
existing
binaries
I
was
a
little
surprised.
They
managed
to
pull
that
off.
I
thought
you'd
have
to
do
something
different,
but
they
actually
got
it
to
work.
C
So
it's
built
around
two
pieces,
the
locator
and
the
network
identifier.
They
are
each
64
bits
and
to
get
ahead
of
ourselves
for
bullets.
So
you
have
the
right
model
in
your
head.
You
put
locator
in
the
upper
part
of
the
IP
address,
put
the
identifier
in
the
bottom
part
of
the
IP
address
and
guess
what
routing
just
works
on
a
locator,
because
that's
what
it's
told
hey
it's
a
prefix.
It
doesn't
care,
so
locators
are
topologically
significant.
They
are
what
routing
will
pay
attention
to
just
by
its
normal
operations.
C
They
name
subnets
network
prefixes,
and
if
you
want
to
2/1
28
routing
on
a
part
of
your
network,
then
that
part
is
for
this
purpose.
The
thing
named
by
the
locator.
We
don't
really
care
how
you're
structuring
whether
it's
really
a
bridged
space
or
a
hybrid
bridge
router,
it's
the
64
bits
gets
you
to
the
edge
and
you
get
and
you
can
deliver
it
from
there.
The
node
identifier
is
stable
and
not
topologically
significant.
That's
the
whole
point
of
it.
You
get
both
of
them
from
DNS.
C
You
can
use
something
else
if
you
want
and
we'll
get
to
that
too,
but
you
get
them
both
and
then
you
use
them,
and
that
means
you
can
have
multiple
locators
associated
with
the
same
communication
session,
because
the
end
to
end
session
is
bound
to
the
network
identifier.
Now
the
question
please,
and
both
of
those
are
globally
unique.
C
C
If
you
want
to
see
how
the
point
is,
and
also
it
names
the
node,
it
does
not
name
the
interface,
you
pull
interfaces
out,
put
them
in
turn
them
on
turn
them
off.
The
identifier
stays
the
same.
It's
there's
a
lot
of
uses
here
at
the
IDF.
We
have
to
create
what
are
called
look-back
addresses
to
create
a
name.
This
is
stable
when
you
have
to
add
no.
C
This
is
not
neither
face
made,
so
we
don't
have
to
make
up
an
interface
for
it
to
be
attached
to,
and
the
whole
key
here
is
the
upper
layer
protocols,
TCP
UDP,
quick,
whatever
you're,
running
TLS,
operates
in
terms
of
the
note
identifies
ignore
the
upper
64
bits
in
getting
that
to
be
backwards
compatible
so
that
you
can
talk
to
existing
modes.
Don't
do
this
does
take
a
little
work,
but
that's
in
there
too.
So
this
is
one
of
the
things
that
was
kept
in
mind
and
design
was
deployability.
C
You
don't
have
to
ensure
that
everybody
is
cut
over
the
first
day.
We
don't
do
flag
gates
so
nodes
that
speak
iymp
can
still
talk
to
existing
nodes
and
will
still
do
the
right
checksum
so
that
everybody
does
things
right.
You
won't
get
the
island
P
advantages
for
those
sessions.
That's
okay!
Let's
trade-off
because
not
everything
gets
updated
at
the
same
time,
so
the
locator
64
is
named
subnets.
You
can
change
your
location
during
the
session.
There
are
multiple
mechanisms
to
do
this.
C
This
actually
an
update
mechanism
that
has
exchanged
between
nodes
so
that,
if
you
are
moving,
if
you
know
you're
gonna
move,
you
can
tell
your
partner.
If
you
have
multiple
interfaces-
and
you
add
a
new
one,
you
can
tell
you're
communicating
partner
about
the
new
one,
so
classic
soft
handoffs
run
really
efficiently.
C
They
actually
took
the
entire
University,
that's
a
substantial
thing
and
cut
everything
so
bleeds
not
the
rest
of
the
infrastructure
had
one
second
lifetimes
everything
just
worked
and
the
lobe
didn't
even
go
out
much.
It
doesn't
even
cause
increase
significantly
I'm
sure
there
was
some
increase.
I,
don't
have
the
statistics
for
that
experiment
handy,
but
it
was
very
small,
so
you
really
can
use.
Dns
is
the
backup
it
does
work
as
a
backup
rendezvous
mechanism.
C
In
addition
to
an
initial
discovery,
so
we
have
communications
mechanisms
to
maintain
and
update
locations
and
we
have
DNS
as
the
fallback
and,
of
course
you
can
use
multiple
ones.
That's
in
fact,
part
of
the
point
here
is:
you
can
use
multiple
locators
simultaneously.
However,
many
you
want
with
many
partners
you
want
and
it
just
works.
Nothing
no
complications.
C
So
you
can
have
multiple
needs
now.
Why
do
you
want
them
either
for
privacy
or
if
you
have
different
sets
of
functionality
and
want
to
be
viewed
differently
or
you're,
doing
virtual
virtual
machine
things
and
want
them
to
be
viewed
as
separate
entities?
There
are
lots
of
different
cases
where
machines
really
want
to
be
seen,
as
just
a
machine
really
would
like
to
be
seen
as
having
distinct
identities.
Fine.
C
That
means
you
won't
get
session
handoff
between
them,
but
that's
you've
chosen
not
to,
and
it
works
just
fine
and
you
can
then
create
DNS
entries.
No,
you
don't.
If
you're,
if
you
the
initiating
end
of
a
communication,
you
don't
need
to
create
pms
entries.
If
you,
if
nobody's
ever
going
to
connect
to
you,
you
can
just
connect
to
other
people,
provide
them
things
and
when
you
change
your
connectivity,
send
them
the
updates
over
the
new
connectivity.
Even
everything
keeps
working,
so
we
can
use
tienes.
C
We
don't
require
any
additional
DNS
in
PMS
information
that
was
not
already
in
the
system
and
we
get
these
global
identifiers.
They
can
be
assigned
statically
and
so,
for
example,
if
you
have
a
server
that
you
want
well
known
in
your
environment
or
well
known
globally,
you
assign
a
static
identifier
and
everybody's
happy.
They
can
be
assigned
by
any
dynamic
means
you
like,
for
whatever
reasons
you
like
rule
is:
if
you
want
to
maintain
a
TCP
connection,
don't
throw
away
that
identifier
while
you're
in
the
middle
of
the
session.
Please
thank.
D
You
two
questions.
First
of
all,
are
these
n
IDs
was
the
problem
there.
Of
course
you
have
collisions.
How
is
that
going
to
be
resolved
if
the
names
are
going
to
be
allocated
region
IDs
are
potentially
being
generated
dynamically?
What
about
the
possibility
of
two
devices
clashing
by
rushing
to
the
scene,
an
idea
hose
out
resolved
so.
C
C
It
seems
to
be
good
to
be
the
case
that
we're
not.
We
don't
expect
you
now.
We
do
have
to
think
about
that
because
usually
it's
square
root,
four
billion.
We
are
now
talking
about
more
than
four
billion
nodes.
So
when
we
did
the
analysis,
it
was
definitely
the
case
now
it
it
may
not
be
the
case
and
in
particular,
when
we
get
to
50
billion.
So
that
may
not
be
the
other
thing
that
happens
and
the
real
answering
your
question.
C
When
you
establish
communication
with
someone,
you
include,
what's
called
a
nonce,
thank
you
and
if
you,
if
you
have
the
same
ID
but
a
different
nonce
than
someone
else,
you
will
be
considered
separate
session.
So
your
nonce
becomes
an
additional
set
of
random
bits.
They
qualify
and
distinguish
your
global
ID
from
anybody
else's,
and
you
maintain
your
announce
when
you
move
so
and
each
side
sends
announce
to
the
other
and
so
on.
Any
of
the
updates
carry
the
nonce.
The
nonce
was
originally
invented
to
provide
security
from
off
path
attackers
because
we
have
location
updates.
C
If
you
didn't
have
something
in
the
on
past
communication
and
off
packet,
our
attacker
could
just
generate
messages
that
said
this
guy
loved
yours,
there's
no
located
so
the
nonce
protects
you
against
off
path
attackers.
It
also
qualifies
the
session
so
as
a
result
that
provides
extra
uniqueness
in
case
of
connection
and
the
point
that
was
just
made.
We
probably
need
to
emphasize
that
more
in
future
work
because,
with
the
current
state
scale
of
the
world
no
randomly
generated
things
are
not
going
to
be
globally
correct.
Okay,
so.
C
D
Well,
I've
got
two
questions
about
that
particular
issue.
First
of
all,
this:
how
do
you
solve
the
authentication
problem?
Actually
updating
the
DNS?
How
could
he
do
this
administrator
to
trust
you
to
update
information,
and
do
it
correctly?
Secondly-
and
perhaps
more
importantly,
instrument
device
leaves
the
network
somehow
be
types
itself
or
such
as
cell
phones?
How
do
we
make
sure
that
crap,
this
an
Indian
ass
gets
removed?
Okay,.
D
C
C
Whoever's
naming
straight
you're
in
may
care,
and
there
are
all
sorts
of
ways
if
you
do
them
with
short
lifetimes
and
refresh
them,
if,
if
you're
using
dynamic,
that
the
preferred
path
is
dynamic,
the
units
which
has
authentication
for
entries
if
you've
created
the
entry
with
dynamic
DNS
and
it
had
you
have
lifetimes
and
refreshes
and
things
will
go
away.
They
won't
hang
around
for
days
after
you're
done
and
if
there's
an
entry.
C
D
C
There's
also
provisioning
dynamic
DNS
to
the
entries
to
go
away
if
they're
not
maintained,
Venus
has
lifetimes
in
it
and
you
make
them
much
much
shorter
than
you
do
classic
server
entries.
You
don't
create
entries
with
lifetimes
on
the
order
of
30
days.
That
would
be
silly.
Oh,
the
point
is
this
is
the
this.
Is
we
have
the
tools
for
solving
this?
If
the
spec
doesn't
currently
say
that
it's
easy,
the
point
is
you're
correct.
C
We
should
make
sure
they
don't
live
forever
to
clutter
things
up,
but
presumably,
if
somebody's
got
dynamic,
DNS
remember
this
is
way
down
at
the
leaf.
It's
not
yeah,
like
com
has
too
many
doesn't
know
anything
about
these
Erickson
comm
probably
doesn't
even
know
anything
about
these,
it's
probably
further
down,
so
that
you
can
manage
them
in
appropriate
servers
and
distribute
the
load
of
maintaining
the
information.
C
F
Hey
dudes
remain
64-bits,
there
will
be
lots
of
collisions
and
I
think
you
explained
a
little
while
ago.
That's
gonna
be
okay,
yeah
and
because-
and
you
mentioned
the
knots
mechanism,
which
I'm
actually
not
familiar
with
the
details
but
I
believe
the.
If
there's
a
collision,
but
they
never
hurt
never
exposed
to
each
other
and
nobody
who's
in
touch
with
either
of
them
is
ever
exposed
to
them.
Then
who
cares?
There's
no
problem?
The
problem
is
if
two
people,
wind.
F
Same
if
I
wind
up
with
an
idea
that
I'm
using
to
contact
some
large
service
and
somebody
elseís,
unfortunately
generates
the
same
net
and
tries
to
contact
that
same
large
service
I.
Think
you
told
me,
there's
a
nonce
and
it's
going
to
be
different
mm-hmm.
Does
it
just
all
work?
We
can
both
communicate
with
that
side.
C
C
F
G
C
Catch-22:
it's
later
in
the
thing:
I
think
it's
not
it's
in
the
backing,
slides
that
the
mate
is
sent
if
the
nonce
is
sent
in
the
initial
communication.
In
fact,
the
way
we
detect
we're
talking
that
when,
when
you
talk
when
I
talk
about
backwards
compatibility,
you
send
the
nonce.
If
the
other
guy
is
not
speaking
I
LMP,
he
sends
you
back
an
error
saying
what
the
heck
are
you
doing
and
you
go
oh
you're
just
doing
classic
IP
I
won't
do
il
MP,
but
if
he
does
speak,
I
LMPD
understands
demands.
C
C
C
This
coupling
right
here,
but
very
few,
where
else
here,
that
the
application
is
trying
to
commit,
but
it's
not
tied
to
the
IP
address
to
the
locator.
So
if
you
move
or
if
you
have
multiple
locations
or
if
you
add
locations
things
up
here,
just
keep
working
just
natural.
Yes,
the
assumption
we're
making
is
there
is
an
enhancement
in
the
TCP
stack
in
the
il
NP
hosts.
There
are
one
five
steps
need
to
be
updated
in
the
world
to
support
this.
C
It's
now
tractable
back
when
I
was
doing
networking
in
the
first
place
in
supporting
16
OSS.
For
my
communication
software
putting
an
upgrade
in
a
you
know,
the
whole
stack
was
challenging,
but
there
aren't
60
different
operating
systems
yeah.
There
are
a
lot
of
variants,
but
they
new
is
the
same
kernel
code.
They
use
the
same
TCP
stacks.
C
So
there's
ongoing
research
and
implementation.
Rfc
is
67
40
through
67
48.
There
is
an
open
source
prototype,
that's
been
done
by
st.
Andrews.
They
support
on
commercial
DNS.
We'll
come
back
to
that
and
we
talked
about
RFC
six,
one
one,
five,
their
estimate
is
that
they've
been
doing
that.
We've
been
doing
this
for
14
years.
It's
a
lot
of
examination,
so
here's
the
picture
of
how
we
encode
it.
C
We
put
the
locators
in
the
upper
part
of
the
ipv6
address
and
the
nodes
in
the
lower
part,
and
then
we
simply
tell
routing,
as
we
already
do,
how
big
its
prefixes
are.
That's
the
way
routing
works.
It
doesn't
try
to
we
once
upon
a
time
we
had
IP
addresses
where
the
routers
would
look
at
the
IP
address
and
guess
the
prefix
link
from
them.
We
called
classful
erasing
I
felt
forwarding
tables
using
it
there's
a
good
reason.
We
don't
do
that
anymore
turns
out
not
doing.
C
That
also
makes
this
kind
of
thing
much
much
simpler.
It
falls
out
for
free
from
the
other
work
name.
Resolution
is
DNS,
that's
supported
by
bind
not
DNS
and
by
NSD
already
defined.
One
at
least
I
know
has
been
in
for
many
years,
letting
talk
to
them
erasing
yeah.
We
could
do
that
an
advantage.
It's
really
not
hard.
You
can
do
etcetera
host
two
entries.
If
you
want
I
mean
you
can
also
use
some
other
mapping
protocol.
C
If
you
need
to
I
mean
we've
looked
at
what
you
would
do
in
mobile
networks
and
some
of
the
way
things
get
resolved
in
mobile
environments
uses
other
mapping
systems.
Okay,
that's
okay,
too.
You
teach
that
mapping
system
to
understand
about
this
combination.
You
get
all
the
benefits
and
well-behaved
applications
work
over
it
with
sockets
and
you
can
do
if
there's
an
application
specific
naming
protocol
for
whatever
reason
some
of
our
applications
have
them.
You
can
teach
that
I
can't
say
it's
free.
You
do
have
to
enhance
it,
but
you
can
deal
with
it.
C
C
So
the
resulting
thing
is
128
bits,
so
anything
that
handles
ipv6
just
sees
an
ipv6
address.
It
doesn't
need
to
know
anything
is
different.
There
aren't
four
routers
when
you're
using
a
LMP
between
the
ends.
There
is
not
even
any
recommended
modifications
of
behavior.
It
just
works.
It's
not
like.
Well,
maybe
it'll
work.
No,
it
works.
Neighbor
discovery
works.
So
again
you
can.
Do
you
asked
about
what
happens
if
you
collide
on
the
same
subnet?
Well,
we
already
have
mechanisms
to
discover
that
and
it
will
report
it
so
neighbor
discovery,
neighbor
solicitation,
all
just
works.
C
C
System,
os
and
kernel,
we
have
to
change
the
ipv6
stack
a
little
bit
yeah.
We
have
a
new
ICMP
message.
We
have
to
do
some
binding
information
so
that
you
can
tell
okay
which
a
locator
identifier
pairs
am
I
using
for
this
communication,
so
that
the
IP
stack
to
know
what
what
to
fill
into
the
packet,
because
the
transport
is
just
working
in
the
mid.
So
obviously
IP
has
to
fill
in
the
locators
extra
binding
information.
C
The
transport
level
code
has
to
be
changed
because
the
transport
checksum
is
now
calculated
unjust.
Indeed,
that's
how
we
get
the
ability
to
change
locators
the
tread
TCP
check
sum.
The
UDP
checksum
is
calculated
using
the
mid,
but
not
the
located.
So
from
the
point
of
view
of
TCP,
when
it
gets
a
bunch
of
different
pockets
with
different
locators
and
the
same
mid
and
confirmed
by
the
nonce
to
do
want
the
same
session,
then
you
have
to
then
you
can
just
use
the
checksum
on
that,
and
so
it
doesn't
have
to
go.
C
How
do
I
fix
it
up
to
for
these
different
ones?
It?
No
one
just
runs
the
Chuck
something
on
the
lid
and
everything
else
at
Jacksons.
So
that's
the
important
change
the
TCP
stack.
So
this
changes
the
IP
stack
and
changes
to
the
TCP
stack.
They
both
approve,
pretty
small,
the
binding
one
would
took
a
little
more
work
to
figure
out
how
to
do
but
they've
done
it.
The
TCP
is
a
for
obvious
reason.
C
It's
a
very
easy
change,
just
conditioned
on
being
told
by
the
IP
stack
that
this
is
my
LNP
session,
and
you
do
have
to
be
careful.
It
start
up
because
you
got
a
calculator
checksum
on
the
sin
and
the
sins
gonna
be
sent
assuming
it's
all
NP,
and
if
it
comes
back
that
you
need
to
use
regular,
you
need
to
go
back
to
the
default
check.
C
So
it's
those
kinds
of
things
that
make
it
a
little
interesting
but
they've
all
been
working
out
and
the
ipv6
binaries
just
work
and
you'll
see
why
and
how
the
basic
trick
is
you
tell
you
let
DMS
give
it
back
something
without
any
locators
and
the
app
doesn't
know
anything's
going
on
now.
You
can
and
we'd
like
in
the
future
to
expose
aisle
and
P
to
the
applications
for
those
things
that
want
to
take
advantage
of
it.
C
So
we
get
no
need
for
naps,
but
as
long
as
the
not
just
messes
with
the
upper
with
the
upper
64
bits,
we
also
don't
care.
If
you
did
not
now
this
checksum,
preserving
identifier,
changing
match,
that's
a
different
beast
II,
but
classically.
If
you
me,
if
you
want
to
change
the
locators
from
the
ones
that
you
use
internally
to
ones
you
as
externally,
go
ahead.
It'll
just
work,
no
special
tunnels,
so
it
like
the
ILA
guys
complain
about
Lisp
that
it
adds
and
end
caps.
C
Well,
not
sure,
that's
a
big
deal,
but
since
people
think
it's
a
big
deal,
I
want
to
point
out
there
ain't
no
in
caps
here
your
decision
with
that
matters
for
your
use
case
and
no
proxies
who
need
for
any
proxies
for
anything,
no
enhancements,
no
changes
to
the
routing
behaviors,
no
special
routers
just
does
its
thing
and
you
get
mobility.
You
get
a
whole
lot
of
features.
Go
ahead.
Question
please
yeah.
H
C
Is
in
the
clear
because
if
somebody
want
corrupted,
no
I
mean
if
somebody
can
drop
packets
anyway,
if
they're
on
path
they
can
corrupt,
they
can
do
all
sorts
of
things.
Fundamental
premise
on
security
is:
if
you
want
to
protect
against
on
path,
attackers
use
IPSec
protection
with
the
mid,
not
a
violent
to
the
mid,
not
to
the
locator
but
use
IPSec,
so
we're
worried
about
it.
So
for
real
security
use,
real
security,
we're
not
gonna
pretend
yet
ILM
P
by
itself
gives
you
real
security.
C
However,
it
does
give
you
protection
against
off
path
attackers,
since
they
can't
guess
your
nonce,
even
if
they
know
everything
else,
they
can't
do
that.
So
that's
the
fundamental
security
distinction
we're
making
here
is
between
on
passing
off
path,
attackers
and
where
is
it
put
in
a
destination
option
because
it's
only
needed
by
the
destination,
so
it's
an
ipv6
destination
option
that
is
actually
marked
as
generate
mirror.
If
you
don't
understand
this
so
that
we
can
get
interoperability.
C
That's
why
it
is
specifically
a
destination
option
so
that
it
doesn't
get
in
it's
not
completely
ignored
by
the
routers
yeah.
Yes,
okay,
and
there
are
those
who
argue
that
even
that's
not
doable,
but
if
that's
not
doable,
we've
got
a
whole
lot
of
other
problems,
because
we've
got
an
awful
lot
of
work
here.
That
assumes
we
can
do
some
a
level
of
extension
on
the
IP
header.
But
that's
not
my
problem
for
today's
discussion.
We
can
discuss
that
in
another
working.
C
C
C
C
So
the
point
is
this:
was
their
test
setup,
one
correspondent,
node
and
a
mobile
node
moving
between
two
locations
with
the
option
for
putting
emulated
when
delay
in
either
the
source
or
the
destination
side
like
make
sure
does
this
work
the
same
better
or
worse
than
existing
mobility
mechanisms?
In
these
two
cases,.
C
So
that
what
they
compared
since
night,
your
vertical
thing
was
mobile.
Ipv6,
because
if
you
wanted
to
compare
cellular
mobility,
you
need
a
lot
more
infrastructure
costs,
HANA,
which
is
unfortunate.
But
it's
the
way
it
is
so
they
compared
mobile
ipv6
with
the
with
il
MP,
and
they
did
both
Island
P
with
both
hard
and
soft
and
off.
And
you
can
see
here
that
a
the
first
thing
you
notice
is
Island.
P
is
far
more
predictable,
very
stable,
very
tight
error
bars
on
their
observations.
C
It
just
doesn't
happen
because
you're
just
able
to
use
both
things
during
the
soft
handoff
and
it
works,
but
even
with
soft
handoff
snip
v6,
even
when
the
land
to
land
case
experiences,
measurable
loss
and
you
can
decide
whether
the
amount
shown
here
is
significant
or
not
I'm
not
going
to
argue
about
that.
But
it
was
certainly
measurably
different
and
in
the
hard
handoff
case,
il
MP
was
no
worse.
C
C
C
C
Iom
pain
was
better
than
the
same
or
better,
and
the
point
is
this
is
running
code.
This
is
not
how
this
is
Linux
code
open
source
yeah.
We
had
to
do
the
modifications,
but
we
did
it.
It
works.
It
runs
and
there's
been
a
lot
of
other
tests.
They've
been
running
with
various
things
in
the
testbed
for
quite
some
time.
C
So
they
like
to
describe
it
as
a
dressing
without
a
addresses,
the
point
being.
We
use
the
term
address
as
the
overloaded
term
for
a
thing
which
is
both
a
specific
identified,
communicating
party
and
at
a
specific
place
in
the
topology.
This
separates
identifiers.
We
have
locators.
This
is
not
the
only
protocol
to
separate
by
any
means
a
number
of
other
alternatives.
In
fact,
it
can
be
argued.
Even
the
mobile
IP
separates
them,
because
it
creates
the
base
the
address
and
the
co-located
and
the
care
of
address,
which
is
the
topological
location
and
with
mid-six.
C
You
have
triangle,
elimination,
that's
what
they
were
talking
about.
What
did
you
do
route
optimisation
to
try
to
make
things
work
better?
It
turns
out
it
didn't
help
much
for
this
problem.
It
makes
other
problems
go
away,
so
mobile
mobile
ipv6,
but
it
doesn't
solve
the
actual
handoff
problem
for
obvious
reasons,
and
so
it
gives
us
much
cleaner
way
to
bind
to
higher
levels
of
the
architecture
without
without
coupling
to
topology.
C
I
C
The
there
is
actually
work
on
how
to
use
il
NP
with
multicast
ran
just
actually
worked
very
hard
on
multicast
stuff.
He
did
one
of
the
first
ipv6
multicast
implementations
back
back
when,
but
mostly
from
at
red.
As
a
transmitter,
you
could
you
could
use
il
NP
so
that
if
you
change
your
location,
you
wouldn't
break
the
sessions,
but
since
the
multicast
trees
are
tied
to
locations,
so
they
get
good
a
good
distribution
effects.
C
There
isn't
we
aren't
claiming
much
benefit
in
the
multicast
space?
There
are
ways
to
use
it,
but
what
it's
really
useful
for
is,
if
you
want
to
have
unicast
communications
tied
to
the
multicast
you
can,
you
can
have
the
unicast
still
do
things.
So
if
receivers
are
moving,
they
get
the
multicast
from
there,
but
they're
there,
a
unicast
feedback
which
you
want
to
try
to
your
older
state
about
them.
You
still
have
so
there
are
ways
to
leverage
the
combination,
but
mostly
we're
not
trying
to
focus
on
multicast.
J
C
C
Part
of
it
is
that
some
folks
consider
that
maybe
changing
the
host
stock
is
actually
a
big
deal
and
the
different
host
stock
vendors
there
for
each
vendors
is
generously
the
whole
stock
builders
are
sort
of
all
waiting
for
somebody
else
to
decide
to
do
it
and
they've
all
said.
You
know
we
could
do
this,
but
why
should
we
put
it
in
until
somebody
else
is
putting.
H
C
G
G
C
When
I
have
looked
at
but
I
it's
it's
certainly
more
complicated.
I
would
like
to
say
yeah
you
get
them
all,
but
no,
it's
not
that
simple
and
are
some
more
interesting
that
one
of
the
RFC's
is
more
interesting
deployments
and
this
fall's
it
to
the
even
more
interesting
deployments,
but
there's
actually
analysis
on
the
server
end
on
how
this
can
be
used
for
data
centers
in
ways
that
are
highly
beneficial.
H
H
C
Want
to
judge
there's
a
survey,
but
now
yes,
so
the
key
is
the
TCP
initially
computes
the
checksum
over
just
the
name
does
at
il
NP
thing
sends
the
destination
option.
The
other
guy
doesn't
speak.
The
speak,
IPP
il
MP,
get
back
an
error.
Recompute
the
checksum
before
you
resend
anything
I
see
you
have
a
back.
C
It's
a
dirt,
but
it's
very
very
fast.
It's
one
round-trip.
It
does
introduce
one
round-trip
of
latency
relative
to
exist
for
a
communication
with
existing
hosts,
and
that
may
be
one
of
the
reasons
some
people
didn't
like
it.
It
is.
It
doesn't
reduce
that
latency
and
quick
is
trying
to
move
to
zero
round
trips,
not
even
one
that
is
I.
Don't
know
any
way
to
change
these
behaviors.
F
C
If
I
do
not
find
I,
64
and
L
64
records,
then
I
just
won't
bother
trying
to
do
il
MP
and
so
I
can
have
a
heuristic
which,
for
particularly
for
early
deployment,
will
let
you
avoid
the
one
extra
round
trip,
so
I
didn't
think
about
that
when
I
was
first
responding,
but
one
of
the
common
heuristics
and
it's
not
mandated.
This
is
why
this
robust
interoperability?
If,
if
you
do
a
DNS
lookup
and
ask
for
a
964
record,
if
it
isn't
there,
you
will
be
told
by
the
DMS
server,
because
DNS
supports
this
stuff.
C
It
ain't
there
and
at
that
point
you
can
say
I
can
reasonably
guess
it's
not
there,
and
since
it's
not
critical
to
me,
I'll
just
use
normal.
If
it's
really
important,
if
I
really
want
to
get
it
anyway.
I
might
try
anyway,
to
see
if,
for
some
reason
he
supports
il
MP
without
a
DNS
entry,
so
there's
multiple
choices
in
behavior,
but
we
can
therefore
cut
out
the
extra
round
trip
in
many
cases,
I.
G
G
C
I
think
I
think
the
right
answer
is
to
use
the
dns-based
heuristic
and
for
most
cases
now.
The
interesting
thing
is,
if
you're
using
sip
and
it's
giving
you
I
P
addresses
rather
than
DNS
names,
then
you
might
want
to
use
the
DI
LNP
for
the
first
connection
to
see
if
you
can
use
il
MP
because
being
able
to
move
while
talking
is
a
really
important
feature.
I
hope.
F
C
C
Two
issues:
if
you're
on
a
high-speed
train:
first
off
you're
generally
the
connector,
in
which
case
you
use
the
update
mechanism,
not
the
DNS
mechanism,
and
so
with
your
correspondence.
You
can
update
it
easily
fast
enough
more
than
one
round-trip.
It's
not
a
problem
now
and
there
is
provision
for
both
moving
in
the
design
as
long
as
it's
soft,
handoffs
and
usually
even
look.
One
second
is
a
fair
distance
or
a
high-speed
train,
but
it's
not
actually
that
it's
actually
a
long
time
in
terms
of
network
communication
time.
C
So
in
terms
of
communicating
parties,
keeping
the
updates
going
is
quite
easy
in
terms
of
dns
updates.
Obviously,
if
you're
getting
a
new
locator
every
10
seconds,
don't
be
stupid,
don't
send
a
DNS
update
every
10
seconds,
mostly
you're,
connecting
out
you
don't
need
to
keep
new
parties
connecting
in.
If
you
think
you
need
it,
send
it
every
once
in
awhile,
send
it
you
don't
don't
mess
with
your
DNS
server.
C
K
Hi,
just
a
couple
of
comments
of
thinking
about
this
in
in
relationship
to
multipath
TCP,
which
in
many
ways
is
kind
of
similar
to
this.
It's
related
absolutely
certainly
related
in
there.
So
multi
party
city
took
the
decision
that
you
would
you
kind
of
have
a
split,
identify
and
located
at
you
from
the
point
of
view
of
an
application.
You
continue
to
use
your
existing
TCP
address
as
the
application
identifier
and
you
have
their
network
locators
if
you
like,
as
as
IP
addresses
underneath
you've
got
so
it's
kind
of
very,
very
similar
to
this.
K
So
I'm
Carter
I'm
trying
to
think
firstly
from
motivation
for
why
people
would
deploy
il
NP.
What
what
are
you
assuming
you
if
you
want
mobility,
the
sort
of
mobility
that
multiple
TCP
gives
you?
What
will
be
the
motivation
for
somebody
to
then
use?
They
got
on
one
Island
P
to
get
further
benefits.
What
further
benefits
would
you
get,
but
that's
the
first
okay.
C
So
there's
a
there
there's
some
backing
slides.
You
can
look
at
later.
The
basic
idea
is
this
is
a
common
mechanism
which
can
be
used
by
multiple
upper
layers
and
then
the
upper
layers
can
add
extra
value
if
they
are
aware
of
issues
with
multiple
paths,
so
we
could
do
some
things
so
that,
for
example,
multipath
TCP
is
good
at
keeping
track
of
different
congestion
behaviour
across
different
connect,
different
communications
with
a
little
cleverness.
You
could
you.
C
We
believe
we've
tried
to
talk
with
some
of
you
folks
and
it
hasn't
worked
well,
but
I
think
it
would
be
worth
having
the
exploration
that
if
multipath
TCP
was
aware
of
Iowa
and
P
state,
you
could
do
some
really
nice
congestion,
congestion
separation
so
as
to
get
the
congestion
behaviors.
That
MP
TCP
very
nicely
adds
TC
that's
wrong.
If
it's
something
else,
it'll
be
some
different
congestion
over
the
multiple
mechanisms
that
IL
MP
makes
available,
communication
style
MP
makes
available
so
it'll
simplify
the
use
of
multipath
TCP.
C
L
K
So
in
terms
of
deployability,
so
it's
a
multiparty
speeds
in
all
kind
of
Apple
OS.
So
that's
been
very
successful
in
in
terms
of
the
Linux
kernel,
so
you
can,
you
can
download
download
it
then
getting
in
the
mainline
kernel
is
proved
to
be
surprisingly
difficult.
So
I
wondered
if
there
would
be
some
lessons
but
mutual
lessons
to
learn.
There
I
think.
C
C
F
Been
almost
a
fan
of
il
NP
from
its
very
earliest
days
and
I
think
what's
different.
Now,
though,
I
haven't
paid
attention
to
it
for
many
years.
I
think
the
story
now
is
much
better
because
you
don't
have
to
explain
how
it
works.
A
variety
before
and
I
realized
sounds
much
better
now
than
it
did
last
time.
I
was
trying
to
understand.
C
F
C
E
F
C
F
F
I'm
trying
to
imagine
the
future
if,
if
we
decide
that
this
is
good
and
it
catches
on
and
we
get
it,
the
changes
into
the
500s
kernels
that
are
needed
and
goodness
comes
from
Island
P
and
then
we
go
into
the
future.
Is
it?
Is
it
because
we've
done
that?
Are
we
going
that
at
some
point
have
to
do
something?
That's
wider
than
128
bits,
I.
C
C
F
E
E
C
E
C
I
C
Spite
of
working
with
the
guys
who
everybody
who
worked
with
Bob
Moskowitz
on
inventing
hip
I,
don't
know
of
any
significant
deployment
of
him
and
the
hip
infrastructure
because
of
the
way
the
big
problem
with
hip.
As
far
as
I
know-
and
the
only
thing
that's
ever
worried
me
about
it-
is
that
it
requires
a
new
infrastructure.
It
doesn't
just
live
with
VMs
and
that
creates
its
own
set
of
complications.
C
C
I,
probably
should
the
reason
is:
I
have
preferences
that
gets
into
controversies
about
how
ila
works
and
whether
the
trade-offs
and
using
routers
are
the
right
trade-offs
and
whether
doing
address
rewriting
on
the
fly
is
the
right
way
to
make
things
work
and
while
I
have
opinions
there.
These
are
issues
on
which
reasonable
people
can
and
do
differ,
and
they
aren't
here
to
rebut.
So
I,
don't
think
it's
fair
for
me
to
go
into
details
about
that.
I.
H
C
/
v4
does
a
separate
checksum
than
TCP
/
v6,
already
yeah,
so
happy
eyeballs
before
versus
v6
would
work
exactly
the
same
way.
Sure,
okay
and
if
you
get
an
ILP,
if
you
get
INL
records
and
your
DNS
lookup,
then
on
your
v6
side
you
use
il
NP
and
on
your
v4
side
you
use
before
because
that's
what
you
do
on
v4
and
whatever
works
works
so
happy
eyeballs
should
be
just
fine
yeah.
H
C
If
your
multihomed,
you
have
two
locators,
because
otherwise
traffic
can't
be
routed
to
both
of
them.
I
mean
you,
don't
assign
the
same
prefix.
There
are
people
who
try
to
work
it
by
assigning
the
same
prefix
to
two
places
and
really
strange
things
happen.
Sometimes
it
does
the
right
thing
and
frankly,
if
it
works
for
you,
you
can
do
that.
It'll
work
with
il
NP
an
LMP
won't
care.
It's
just
not
usually
a
good
idea.
So.