►
From YouTube: IETF102-PANRG-20180720-1150
Description
PANRG meeting session at IETF102
2018/07/20 1150
https://datatracker.ietf.org/meeting/102/proceedings/
A
C
C
A
C
C
C
C
D
D
We
had
a
few
sort
of
short
comments
about
the
Charlie.
Let's
go:
pay
money
has
changed,
work
to
be
done
in
the
research
group
from
the
mentally.
In
anyway,
we
wanted
to
basically
you
know,
I
get
the
Charter
in
a
shape
for
no
longer
being
proposed.
We
actually
were.
There
was
a
discussion,
the
IR
s,
G
right
after
London
about
whether
we
are
going
to
be
or
not
and
basic.
Is
that
garden?
Congratulations.
You
are
now
a
research
group
just
update
the
Charter
and.
D
D
D
D
D
We
also
bridge
the
gap
between
Internet's,
the
Energon
element
community
and
then
Muniz
and
actually,
what
should
say,
measurement
and
other
research
communities
right.
So
if
last
year
we
put
in
a
proposal
to
do
a
workshop
in
sitcom,
which
was
turned
down
because
they
like
okay-
well,
that's
great
you're
late.
Yes,
but
you
know
what
are
you
really
doing
and.
D
D
D
So
hiya
planning
travel
now
speaking,
not
as
co-chair
of
the
research
group
interested
in
this
space.
You
know
going
a
little
bit.
You
know
on
the
Charter
I
decided
to
kind
of
write
down
what
questions
have
come
up
at
least
sort
of.
In
my
thinking
about
the
space
we've
discussed
this
document
a
few
times,
you've
had
pretty
good
feedback
here.
It's
adopted
now
at
the
end,
I'll
ask
kind
of
what
what
we
should
do
next,
but
first
I'll
go
through
the
questions
very
quickly.
D
D
We
look
at
this
in
a
lot
of
opposition
to
the
idea
of
having
the
endpoints
talk
to
the
paths
turned
on
the
danger
of
the
type
of
the
things
the
endpoints
might
say
to
the
path
about
the
path.
So
you
don't
necessarily
have
properties
where
that
the
property
is.
This
is
a
path
that
is
acceptable
for
subversive
political
content
via
video
right,
like
that's,
that's
useless,
for
networking
to
specific
and
has
like.
D
Implications,
so
we
don't
to
do
that
right,
so
what
kind
of
like?
On
the
other
hand,
there
are
things
that
are
pretty
uncontroversial
so
as
have
latency
associated
with
them,
the
last
good
lots
of
things
that
we
know
how
to
measure
for
a
very
long
time
and
in
their
stuff
the
community
right.
So
how
do
we
define
these?
How
do
we
represent
these?
How
do
we
get
them
into
the
system?
Sometimes
that's
those
things
about
itself.
D
Sometimes,
that's
the
end
points
do
not
select
the
path
so
questions
one
and
two
I
think
are
kind
of
our
next
assignment
beyond,
like
those
once
you
have
information
about
the
path
you
need
to
be
able
to
use
the
path
to
figure
out
how
you're
going
to
networking
getting
packets
from
one
place
to
another
efficiently
or
better
or
cheaper,
or
you
know.
Why
are
you
doing
this
optimization?
D
D
It
wasn't
really
clear
that
questions
for
largely
going
to
be
addressed
by
the
taps,
working
group
in
the
IETF,
so
we're
kind
of
going
to
say:
that's
not
really
in
scope
for
us
to
work
on
here,
because
we
have
a
place
to
work
on
that
interfaces.
Transport
application,
support
the
use
of
path
awareness,
so
the
taps
api
has
an
explicit
notion
of
path
selection.
D
At
this
point,
it's
primarily
based
on
address
memory,
racing
or
interface
racing,
but
if
you
took
that
infrastructure
and
stuck
it
on
top
of
the
network
where
you
could
actually
be
passed,
selection,
it's
sort
of
trivial
to
show
like
we
have
this,
this
path
concept
in
the
taps,
architecture,
and
then
you
have
for
redesigning
transport
higher-level
protocols.
So
this
is
work
that
was
adopted
in
the
working
group
at
the
last
I.
Think
so
I
think
that
we
should
be
sorted.
D
D
A
good
question
actually
came
from
from
some
of
the
other
people.
How
does
any
of
this
change
like
you
know
that
the
original
questions
were
posed
in
explicitly
in
the
Charter,
with
respect
to
end-to-end
path
awareness?
You
know,
the
server
in
point
knows
how
it's
gonna
get
the
clients.
Client
in
point
knows
how
it's
gonna
get
the
servers.
The
pier
is
know
how
they're
going
to
connect
with
each
other
I
love.
This
information
goes
all
the
way
down
to
the
end
a
lot
of
than
the
work
in
this
way.
D
So
we
had
presentations
two
or
three
meetings
ago.
People
who
work
on
Scion,
which
is
a
path
for
networking
architecture
or
the
way
that
that
usually
gets
to
play
this
tunnel
endpoint
tunnel,
endpoint
you're,
not
rewriting
the
applications
to
this
new
as
an
infrastructure,
you,
basically
the
what
the
architecture
gives.
You
is
basically
a
really
really
nice
EP
end.
How
does
all
of
this
work
change
would
not
has
sort
of
the
explicit
goal
of
the
work.
F
D
Category
change:
how
the
operations
change,
how
the
interface
has
changed
when
it's
like
in
week,
two
tunnel,
end
or
tunnel
ends
to
tunnel
and
write
the
questions,
seven
and
eight
or
layers
eight
and
nine.
How-
and
these
are
like
sort
of
these
were
also
raised
on
the
list
in
the
earlier
discussion.
How
can
you
effectively
operate
a
network
so
right
now,
network
operators,
they
kind
of
have
predictable
in
air
quotes.
D
Behavior
that
comes
out
of
you
know,
announcing
BGP
routes,
you're,
giving
Internet
traffic
engineering
through
bgp
on
a
pretext
level
you're,
making
an
assumption
that
all
of
the
traffic
and
the
flow
is
going
to
flow.
The
way
that
your
BGP
and
that
your
inter
domain
ecmp
is
distributing
that
traffic,
when
your
endpoints
now
have
the
ability
to
control
exactly
where
their
packets
go
and
all
of
that
traffic,
engineering,
knowledge
and
control
kind
of
goes
out.
The
window
and
you
set
up
in
the
worst
case,
an
adversarial
situation
right.
How
can
we
avoid
that?
D
That
is
a
sort
of
a
research
question,
and
our
operations
also
tends
to
be
a
little
bit
of
a
research
question
into
incentives
and
that
the
h1
right
like
how
can
we
align
the
incentives
available?
Various
people
who
could
benefit
from
this
architecture
in
such
a
role
with
actually
gets
deployed
right,
these
I
think
are
listed
in
order
of
increasing
difficulty.
D
At
least
that
was
the
intention
when
I
put
the
thing
together
is
we're
gonna
start
with
run
into
five,
when
I
have
to
do
some
work
on
three
at
some
point,
once
we
can
respect
that'll,
be
useful,
having
people
hunt
on
airplanes
and
then
go
to
seven
and
eight,
once
we've
been
successful
on
showing
that
the
technology
can
actually
work
yeah.
So
what's
next,
we
can
answer
in
for
questions
from
the
topic.
I
think
that's
the
right
thing
to
do
as
I
said,
I'm
sort
of
first
shot
at
this.
D
We
have
like
I
think
we
have
and
we'll
see
a
little
bit
in
Spencer's
presentation.
Next,
we
have
some
information
about
weapons,
might
find
paths
right.
So
there's
we're
not
starting
sort
of
often
the
ropes
there's
been
a
bunch
of
work
into
space
before
and
we
can
show,
if
collected
together
and
put
it
in
a
big
pile,
and
we
can
look
at
the
ways
that
path
properties
can
be
disseminated.
I
mean
a
lot
of
this
stuff
is,
is
also
stuff
that
I'm
Spencer's
document
is
going
to
talk
about
ways
not
to
do
it.
D
Part
of
the
hope
here
right
like
this,
is
a
research
group,
because
if
we
knew
what
we
were
doing,
which
part
of
the
hope
here
is
that
we
can
sort
of
look
at
the
set
of
possible
ways
to
achieve
the
dissemination
and
all
of
the
way
is
that
Spencer
is
going
to
tell
us,
don't
work
because
we
try
big
countries
and
we
subtract
the
one
from
the
other.
We
don't
end
up
with
the
empty
set,
so
I
as
an
individual
would
ask.
If
there
is
anybody
interested
in
fees?
H
Just
it's
research
group
yeah,
just
because
you
just
won't
have
our
use
case.
You
should
see
if
it
let
anybody
right.
So
let's
say
you
have
an
any
user
to
have
multiple
different
interfaces,
and
you
want
this
an
host
to
make
a
choice
from
entering
path
which
ends
path,
J
choice.
What
is
the
bias
for
him?
It's
not
what
I'm
trying
to
do.
D
That's
like
sort
of
like
in
the
crazy
vision.
That's
phase
zero
right,
we're
kind
of
there
already
in
fact
like
so,
if
you're
on
many
mobile
devices,
if
your
monthly
connected
through
LTE
and
Wi-Fi
there's
a
policy,
that's
usually
pretty
crude.
That
says:
hey
I'd
like
to
minimize
my
cost,
so
I
don't
use
Wi-Fi.
First
in
the
note,
C
or.
D
D
May
communicate
wait,
so
you
can
do
this
right
now,
like
that.
The
cheap
way
to
do
this
is
you
take
your
home
router
and
you
give
it
like
seven
VPN
tunnels
that
I'll
exit
in
various
ixps
right,
like
you,
can
you
can
build
a
cheap
thing
with,
with
with
a
home
router
some
some
rules?
You
know
us
right
and
if
you
take
that
information
you
give
it
to
the
transport
protocol,
it
could
do
things
like
slow
spreading
rate.
D
That's
the
vision
for
the
virtual
interface.
That's
what
we'd
like
to
get
like,
actually
implementing
that
envy
in
the
naive
way
of
flooding
into
impact
information,
all
the
way
down
to
every
interest
where
you
have
path,
diversity
on
all
of
those
three
different
networks
or
any
perhap
basis,
is
clearly
on
scalable
right,
but
you
could
have,
on
the
other
hand,
right
like
so
that
the
thing
that
it
might
scale
is
again
this
this
hack
that
I
just
invented
with
with
the
various
with
the
internals.
D
You
can
do
something
else
when
you
put
tunnels
together,
I
mean
like
so
you
could
say
that
tor
is
example
as
a
circuit,
oriented
path,
aware,
network
networking
architecture
that
has
its
primary
goal:
immunity
right,
but
you
could
be
sort
of
like
sub-segments
together
in
ways
to
do
other
things
right.
So
the
one
thing
is
actually
missing.
Probably
from
this
from
this
I
wouldn't
call
it
a
paper
because
it
kind
of
is
but
from
this
draft
is
a
few
pointers
to
research
that
was
done
in
this
space,
especially
last
decades.
D
D
One
a
lot
of
this
routing
research,
the
source
routing
research,
was
stimulated
by
the
idea
that
the
routing
table
was
going
to
become
unscalable
and
when
we
just
like
and
then
Moore's
law
cut
up
and
after
mile
cut
out,
there
was
a
while
you're
gonna
blow
up
the
internet,
but
we
don't
need
to
blow
up
you
remember,
as
we
can
define
what
he
can,
but
a
lot
of
the
concepts
that
they
started
working
on.
There
are
I
think
really
interesting
a
little
bit
we're
trying
to
revive
it.
D
J
D
F
F
F
J
D
Right,
like
so
a
little
bit
of
inside
baseball,
I
guess
so
I'll
explain
some
of
what's
going
on
in
taps
is
basically
building
a
now,
as
we
found
out
on
this
time.
A
standards
track
interface
for
doing
protocol
stack
agility,
which
it
turns
out,
is
actually
something
it's
you
can
think
of
it
as
a
as
a
generalization
of
happy
eyeballs
right,
but
in
in
a
more
challenging
environment
and
that
stuff
I
think
we
pretty
much
know
how
to
do.
We
actually
and
there's
this
shipping
code
in
this.
So
how
can
they
companies
layer
support?
D
D
Able
to
get
connectivity
and
they
have
properties
and
whether
those
properties
come
from
the
implementation
itself
or
something
that
the
implementation
measured
is
really
sort
of
irrelevant.
So
I
think
that's
four
is
pretty
solidly
in
tops
and
it's
pretty
silently
not
research
I
think
it's
wrong.
It
is
like
how
do
you
redesign
the
whole
stack
above
is
sort
of
a
new
interface
working
taps.
L
Thank
you
to
card
this
goes
to
questions
2,
&,
3,
I
think
we
should
not
be
too
quick
to
decide
what
are
and
are
not
appropriate
path
properties
to
be
able
to
report,
because
I'll
give
you
an
example
real
world.
For
about.
10
years
ago
we
were
IP
enabling
air
to
ground
communications
for
aircraft,
where
historically,
each
application
was
tied
directly
to
a
particular
communication
system.
So
if
you
were
doing
a
particular
application,
you
intrinsically
new
comm
system
was
going
to
go
over
and
we
were
inserting
Network
layers
in
the
middle.
L
There
was
a
particular
well.
This
was
for
a
military
user
and
there
was
a
particular
communication
system
involving
satellites
were
by
international
treaty.
Fire
control
traffic
could
not
be
routed
over
that
communication
system
and,
as
we
inserted
Network
layers
that
hid
it
from
the
user.
What
communication
system
he
was
going
to
be
using?
We
had
to
take
care
not
to
route
fire
control
traffic,
or
we
would
had
a
treaty
violation
that.
L
D
D
L
Tommy
Polly,
so
it's
kind
of
just
Spencer's
points,
four
questions,
four
and
five
or
four
first
well,
I
agree
that
four
definitely
matches
what
we're
doing
in
taps.
I
think
there
are
definitely
parts
that
we
achieve
that
we
don't
cover
there
yet
so
the
basic
way
of
how
do
we
interface
with
this?
We
have
concrete
ways
of
doing
this
today
for
like.
Oh,
this
is
the
access
type.
This
is
if,
if
it's
meter
or
not,
but
I,
think
there
will
be
areas
in.
D
L
Then,
for
number
five
I'm
almost
wondering
if
that's
less
taps
and
more
a
job
for
the
things
that
are
going
to
succeed,
MP,
TCP
and
MP
quick
these
right
now
they
have
some
ability
to
do
very
basic
protocol
adaptations
to
there,
but
we
haven't
probably
explored
in
those
groups.
What
are
the
ramifications
of
when
we
have
many
many
more
links
and
many
more
things
to
actually
do
the
scheduling
over
yeah.
D
So
so
I
know
that
there
was
a
Schwab
who
came
to
this
meeting
a
couple
of
times
ago.
He
did
a
little
bit
of
research
on
so
multipath
per
pack
and
scheduling.
You
know
in
a
in
a
path
aware,
transport
layer
any
basically
found
that
the
scalability
of
making
a
choice
was
such
that
if
you
had
one
path
of
his
trivial,
if
you
had
two
paths,
you
could
do
it.
If
you
had
three
paths,
you
were
screwed
right
so
so
like
reducing
the
that
actually
is.
D
D
H
H
D
D
Pamphlet,
like
specifically
said
we're
not
gonna,
bother
with
too
hard
the
same
thing.
The
way
that
it
does
route
dissemination
is
we're
just
gonna,
you
know
we're
gonna
pretended
asymmetric,
robbing,
doesn't
exist
and
we'll
add
it
later
so
yeah.
We
need
to
actually
that's
a
good
point,
but
I've
got
to
this
document
thanks
right.
Thanks.
D
Is
that
is
more
editorially
intentional
and
it's
that's
a
mistake.
Actually,
I
wasn't
thinking
hey,
you
know,
there's
this
vision
and
it's
very
very
clear
and
we
just
need
to
figure
out
how
to
do
it.
I
was
thinking
more
yeah.
It's
like
this
is
yeah
right,
like
a
property
definition,
question
mark
property,
access,
dissemination,
dot,
dot,
question
mark
right,
but
like
ok,
so
these
are
the
these
are.
D
An
idea
I
would
start
personally
from
documenting,
what's
been
done
before
you
know
start
with
Spencer's
document.
You
know
taking
the
mind,
especially
there's
a
section
that
he
has
of
like
you
know.
These
are
the
things
we've
learned
read
that
take
that
and
then
look
at
work.
That's
been
done
and
try
to
see
all
the
stuff
that's
been
done
doesn't
fall
into
the
traps
that
we've
fallen
into
and
then
go
from
there.
That
would
be.
F
D
N
D
N
F
N
J
D
D
D
O
Like
is
that
depends
on
what,
ultimately,
is
the
goal
here?
Do
you
wanna
also,
we
make
this
document
sort
of
a
reference
collection
of
references
of
previous
work,
as
I
can
see
elements
of
myth.
Even
there's,
v6
ops
is
disgusting
stuff
that
is
potentially
relative
relevant
to
this,
like
provisioning.
D
Domains
so
I
saw
this
with
my
you
know,
with
Marie
a
trainer
hat
on
and
I
saw.
This
is
primarily
just
a
set
of
questions
right
like
so
then
documents
that
tried
to
answer
these
questions
within
reach
out
to
other
work
right,
so
we
can
talk
about
whether
that's
whether
we
want
to
actually
stick
the
pointers
in
here
or
not,
but
but
I
was
always
my
radio
is
that
we
could
get
to.
F
F
F
F
So
I've
done
eighth,
the
first
major
update
honestly
to
this
to
this
draft
of
national
one
just
prior
to
ITF
102
I
added
pointers
to
IAB
documents
on
protocol
design
and
adoption
Thank
You
IAB
for
producing
those
documents.
These
are
not
specific
to
path,
aware
networking
but
they're
worth
reading
I
added
section
2,
which
is
a
summary
of
lessons,
learned
that
Bill's
on
new
contributions
and
I
now
have
contributions
on
it,
serve
QuickStart,
trig
Tran,
which
was
there
before
and
shim
six
and
based.
You
know
basically
provided
by
different
folks.
F
F
So
this
is
a
summary
of
the
summary
of
lessons
learned
section:
that's
in
201
I
have
a
question
about
what
the
right
level
of
detail
is.
I
have
a
section
on
lessons
learned
in
each
contribution
like
if
you
look
at
4.1.2
in
the
draft
and
I
have
a
summary
section
section-
and
this
is
a
summary
of
the
summary
section
so
I
like
to
understand
where
people
think
the
right
level
of
detail
to
include
the
draft
is,
but
basically
these
are
the
lessons
that
I've
got.
F
This
is
a
summary
of
the
summary
of
the
lessons
that
we've
learned
from
the
things
that
have
been
contributed
so
far.
The
benefits
of
adoption
must
be
big
enough
to
overcome
inertia.
If
you
can't
trust
middleboxes
middleboxes
can't
help
you
benefits
must
be
big
enough
for
operators
to
justify
deployment.
F
Current
operational
practices
can
prevent
deployment
of
good
ideas.
/
connection,
state
and
intermediate
devices
is
an
impediment.
Increasing
distance
from
sources
of
information
makes
path,
information
less
useful
and
many
applications,
don't
know
the
things
that
they
need
to
tell
path,
aware,
networking
transports
and
the
ideas
that
we've
had
or
how
to
do
this
previously
next
slide.
Please.
F
Be
ok
with
it
at
the
end.
If
people
have
ideas
about
that
and
just
to
be
clear
what
I'm?
What
I'm
actually
asking
is
is
the
level
of
detail
in
section
2
in
the
draft,
the
appropriate
level,
because
I've
got
more
detail
in
the
individual
contributions,
and
you
know
we
can
certainly
summarize
the
we
could
certainly
summarize
the.
D
F
I
mean
you
know,
I
was
also
pleasantly
surprised
and
I
think
this.
You
know
this
is
blindingly
obvious
to
everyone.
Who's
ever
made
these
mistakes,
but
that's
not
everybody
in
the
world
or
even
everybody
at
the
IETF,
so
I
think
it's
more
valuable
than
maybe
we
thought
it
was
at
the
first
meeting.
So.
F
Yes,
actually
actually
I
think
maybe
backing
up
one
from
there:
okay,
yeah,
okay,
yeah
you're
you're
in
the
right
place,
I
just
got
lost
or
II,
so
I
wanted
to
think
about
what
the
goalposts
for
this
document
are
I'd
like
for
it
to
be
useful
advice
from
Panaji
to
the
IETF,
because
of
the
things
that
Brian
and
I
were
just
talking
about.
Where
things
you
know,
we
keep
doing
things
like
this
to
ourselves.
F
F
So
I'd
ask
for
people
to
be
thinking
about
what
other
guidance
can
we
give
to
the
IETF
I
would
also
like
this
document
to
be
useful
advice
to
Panaji
the
way
I
see
this
working
and
I'll
be
interested
in
Brian's
opinions
is
that
the
open
questions
are
there
to
guide
Panaji
research,
and
this
document
can
help
identify
other
open
questions
that
would
guide
pen
RG.
So
my
question
is
what
other
guidance
can
we
give
ourselves
yeah
next
slide.
F
F
Yeah
and
the
slides
are
painting
slowly
here
for
me,
so
thank
you
for
being
patient
I'm
wondering
if
there
are
more
paths,
aware
networking
lessons
that
we
that
we
should
be
documenting
and
part
of
that,
but
not
the
only
part,
is.
Are
there
more
places
to
look
for
lessons?
I
know
that
I'm
looking
at
incest
with
Martin
providing
a
contribution
there,
he
just
hasn't
been
able
to
produce
that
before
the
meeting.
But
there's
always
after
the
meeting.
F
F
G
C
F
I
think
this
was
probably
my
point,
my
opportunity
to
point
out
that
this
document
is
a
lot
better
from
zero,
zero,
two
zero
one
and
that
at
least
a
major
portion-
if
not
all
of
the
improvement
is
due
to
text
that
was
provided
by
other
people.
So
you
know
so
as
I
say,
my
inbox
is
always
open.
Okay,.
D
F
N
N
User
MP
TCP
in
a
current
scenario,
which
have
not
physical
interface,
such
as
a
Wi-Fi
plus
30
or
DSL,
plus
RT,
which
are
very
limited
and
also
even
for
this
to
case
we
think,
is
not
really
can
be
completely
support
possible
networking,
for
example
they
between
the
even
physical
interface.
If
the
true
path
is
joined
somewhere,
then
the
MGP
CPU
will
not
hire
anyways
so
the
to
have
some
solution
to.
N
N
In
this
scenario,
so
if,
for
example,
one
Operator
owns
the
network
from
end
to
end
definitely
a
product,
him
can
try
to
provide
support
even
for
one
access
interface
and
also,
this
is
very
popular
capturing.
The
five-year
joint,
because
in
the
five
young
parent
we
are
service
provider
normally
have
age,
computing
or
crawl
net
connect
power
to
access
network.
So
for
that
scenario,
operator
can
try
to
change
their
topology
or
over
a
pollution
or
network
to
provide
enormous
support
and
another
popular
two
parodies
that
we
have
from
multiple
operators.
N
N
N
So
why
we
won't
do
this
is
because
the
penalties
for
using
metal
flows
are
concerned
by
the
MP
TCP.
Fairness
means
that,
no
matter
how
many
foes
you
have,
if
they
are
sharing
in
the
one
part
nak,
where
pennyways
cannot
be
more
than
one
TCP
session
so
first
to
overcome
this
difficult,
the
network
operator
need
to
provide
for
that
special
user
who
won't
use
about
pastors
about
MP
PCP.
N
Today
also
mention
that
use
tunnel
is
the
wrong
way
and
also
we
can
use
the
current
across
a
nation
s.
We
are
technology
too,
to
provide
the
disjoint
path
in
access
network,
all
human.
We
can
use
the
Sdn
corporation
different
boundaries
to
a
specific
user.
I
also
I
have
another
draft
which
values
are
in
balancing
why
the
flow
now
with
us
PCP
that
can
also
be
used
so
for
Matins.
Is
that
because.
N
I
need
to
matrons
to
make
sure
that
those
not
pass
existed
and
useful.
So
when
are
the
different,
a
Platinum
Lee
have
a
BGP
motorhome
connection
and
they
also
have
another
pass
technology
use
the
inside
domain,
so
the
apartment
when
they
try
to
provide
the
password
networking
they
need
to
maintain
such
a
topology,
and
also
you
have
to
make
sure
that
narpath's
not
disjoint
and
now
available.
This
is
not
just
a
teeny
tiny
work
have
a
lot
of
research,
also
social
disease.
N
So
a
baby
step
for
this
approach.
We
think
we
need
between
host
better
then.
What
is
a
purpose
for
this
first
purpose
is
that,
of
course,
the
distribute
the
past
product,
such
as
second
version
information,
another
pass
information
to
end
the
user
before
the
in
Internet
attacker
can
be.
You
can
do
such
things,
so
second,
desserts
and
the
user
can
tell
apart
what
his
service
expectation,
such
as
how
many
paths
I
want
use
and
families
for
each
also,
the
we
want
to.
N
N
Maybe
we
want
to
reduce
her
staff,
so
in
general,
what
message
will
be
exchanged
is
that
we
want
to
use
the
user
expectation
and
also
want
to
have
the
distribution
for
the
network
information
such
as
a
matter
pass
and
the
past
property.
This
is
foreign
to
the
demands
of,
and
also
we
won't
have,
the
features
such
as
how
to
use
a
mod
pass
that
and
nitrous
he
is
a
second
which
can
be
more
efficient
for
the
violence.
P
Yeah
I
mean
those
who
are
would
enough
use
you
and
I
last
when
they
were
like
trying
to
connect
to
ice-t
n.
So
the
concept
of
you
and
I
at
that
time
was
that
you,
you
know
you
could
have
you
have
to
our
network
and
that
gets
terminated
and
then
you've
got
a
very
limited
interface
that
you
know
allows
you
to
request
a
chance.
Like
you
know,
access
control.
P
F
P
P
F
D
D
D
D
D
I
D
The
thing
so
I
would
I
would
think
that
what
we
useful
here
is
something
isn't
an
explicit
you
and
I,
but
is
basically
a
way
for
you
to
be
able
to
tell
an
endpoint
that
you
so
you're,
okay,
that
a
user,
an
integer
price
or
a
server
could
connect
to
some
sort
of
thing.
That
is
basically
a
tunnel
broker.
D
That's
doing
this,
this
data
for
him
stuff
in
tack
to
that
about
the
properties
that,
at
once,
in
terms
of
these
Patrick
and
now
only
it
gets
up
yes
achieve
an
impact,
and
so
like
basically
saying
okay
well,
we're
gonna
put
the
Uni
in
the
in
and
icons
that
back-end
I.
Don't
think
that
is
a
great
idea.
I
do
think
that
the
defining
the
vocabulary
for
this
interface,
where
you
could
talk
to
a
tunnel
endpoint
and
tell
us
what
kind
of
of
properties
you
want
is
probably
not
a
bad
idea.
D
I
K
Q
It
is
going
to
be
work
to
be
done
there,
but
you
could
see
sticking.
A
well-known
kind
of
this
is
like
here's
a
this.
Is
a
user
one,
here's
a
well-known
segment,
ID
I'm,
but
in
the
case
of
type
of
power
and
then
at
different
points,
maybe
be
able
to
look
up
and
do
the
translation
there's
a
lot
of
different
ideas
of
taking
it.
Q
It's
not
just
one
service
provider,
you
can't
just
go
to
one
service
provider
and
see
here.
This
is
what
I
want
and
the
questions
do
you
have
a
description
on
the
package
saying
what
your
desire
do
you
assume
that
you
don't
have
more
local
service
road?
This
is
what
I
want
and
if
then,
whenever
it
goes
to
the
next
service
provider,
a
tile
gives
you
the
best
ever
because.
Q
Q
That
has
control
the
difference
with
MPLS
and
with
sublet
routing
is
it
that
goes
away
mm-hmm.
It
doesn't
take
time,
throw
32-bit
about
20
bits,
plus,
you
know
throw
your
label
stock
on
and
off
you
go.
You
don't
have
to
have
set
up
time.
It
doesn't
have
to
look
like
an
interface
it
can
just
be.
This
is
how
I
want
to
go
directive
and
I,
though
it.
Q
L
Tommy,
just
based
on
some
of
our
things
that
we
actually
just
saw
here
about
my
my
deployability
and
the
fact
that
we
have
one
interface
in
multiple
interfaces.
I
was
thinking,
of
course,
about
so
the
work
that
we're
doing
and
pvd's
in
which
it
would
allow
us
to
through
one
physical
interface
of
course,
have
things
that
can
be
represented
as
virtual
interfaces
or
essentially
different
local
addresses
that
have
truly
different
meanings
in
different
provisioning
domains.
L
Would
it
be
useful
to
have
a
distinction
in
our
terminology
as
we're
talking
about
the
different
paths,
also
of
things
that
are
different
paths,
because
they
have
truly
different
provision
grains
and
they
cannot
like?
We
are
kind
of
guaranteed
that
these
are
really
different
paths
all
the
way
down,
even
if
they
are
purely
if
they
don't
relate
to
it.
L
The
physical
interface,
as
opposed
to
things
that
we
think
are
pathless
that
are
more
just
like
a
routing
equivalence
that
we
may
want
to
toggle
somewhere
down
the
network,
but
are
equivalent
from
the
perspective
of
these
services
that
they
offer
so
I.
Don't
know
if
it
I'm,
not
necessarily
is
just
saying
we
should
incorporate
provisioning
remains
as
a
top-level
thing
here,
but
we
should
think
about
our
powers.
D
D
D
Can
know
or
things
that
you
need
to
know
in
order
to
schedule
a
packet,
do
you
know
basically
just
just
inputs
or
packets,
and
they
also
have
topological
properties,
and
they
might
also
have
a
third
class
of
property,
which
is
fuzzy.
It's
a
fuzzy
sort
of
constraint.
Based
on
you
know
what
actually
I
don't
understand
this
property.
L
Will
you
two
is
like
what
I
was
saying
for
the
provisioning
domain,
yeah
right
so
saying
that
this
there's
a
fundamental
difference
about
what
I'm
able
to
do
over
this
network
potentially,
and
maybe
the
unlink
up
at
the
end
of
the
day.
Yep
and
I
can
actually
do
multi
path
of
one,
but
I
have
knowledge
that,
oh
this,
because
the
classic
example
of
a
different
provision
being
that
we
have
today
is
a
like
a
cellular
APN.
L
D
L
O
J
D
F
B
D
Your
own
network
pretty
easily
right
like
it's
spring,
we've
done
it
with
chipped.
It's
it's
it's
as
many
times
as
you
need
forever
for
free
and
yeah,
there's,
not
research,
but
the
real
thing
that
changes.
This
is
an
inter
domain
scenario,
where
the
inter
domain
scenario
really
is
an
intra
means
we're
missing
this
as
as
Dirk's
library
or
missing
this
network
network
interface.
How
do
you
do
this
in
the
world
where
there's
no
network
network
interface?
No
common
right
like
so?
Is
there
stuff
that
flies
out
of
this?