►
From YouTube: IETF102-INTAREA-20180716-1550
Description
INTAREA meeting session at IETF102
2018/07/16 1550
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
A
A
B
D
B
B
E
E
Right
so,
as
you
remember
last
time
you
know
down,
we
got
Ayanna
located
and
debris
code
until
the
reclaim
of
21.
That
is
good
thing.
We
got
an
extensive
review
from
teen
charm
or
is
not
here
today
and
from
the
Carol
I.
Don't
know,
then,
if
you
are
here,
why
didn't
I
pick
you
talk
to
you?
We
also
get
a
design
team
lunch
and
not
one
with
some
customers
between
Interstate
into
PB
and
we
are
seeing
when
we
are
talking
about
PVD
here
it
was
explained
because
of
foundation
work.
E
E
E
This
do
we
were
using
you
remember
when
we
copy,
we
insert
another
routing
header
running
advertisement,
header
after
the
PVD
option,
because
we
want
to
change
the
value
out
of
the
rotting
header.
We
were
using
a
a
flag
flag,
but
it's
confusing
because
in
a
Oh
we
also
ever
a
flag,
so
we've
been
emitting
to
our
flag.
E
Also,
when
we
started
with
dazzling,
the
PVD
was
more
a
PVD
in
the
option.
It
was
only
different
qualified
domain
name,
identifying,
basically
the
name
of
the
ID
of
the
provision
domain
and
as
now,
it's
actually
more
like
a
container.
It
makes
no
sense
to
call
it
a
PVD
ID
option
and
we
simply
call
it
PVD
option
right
and
before
Darwin
we
were
having
a
section
about
PVD
aware
us
behavior.
Obviously
we
need
to
define
the
behavior
with
a
new.
A
E
E
E
Okay,
thank
you.
So
the
issue
here
is
that
you
remember
when
some
people
need
a
hose
design
to
fetch
the
new
JSON
file
to
get
the
new
additional
information.
It's
triggered
by
two
thing
is
trigger
when
you'd
receive
a
route
advertisement
containing
the
PVD
option,
with
a
sequence
number,
which
is
different
than
the
previous
version.
Don't
thing
about
it.
E
You
are
the
big
conference
with
20,000
Wi-Fi
client,
and
suddenly
you
change
the
sequence
number,
because
you
want
to
change
in
a
server
or
whatever
do
using
a
TLS
server
we'll
be
happy
to
receive
20,000
a
society
s
connection.
At
the
same
time,
you
should
be
not
right,
so
we
need
basically
to
put
some
jitter
and
what
we
have
decided
is
to
use
four
bits
which
is
going
to
delay
field.
We
take.
E
So
maximum
15
times
to
maximum
30
and
we
take
to
the
power
of
this
number
and
which
is
up
to
50
days.
Okay,
if
you
use
a
maximum
value,
is
50
lines
and
each
and
every
hose
needs
to
wait
around
on
time.
So
if
it's
something
operating
you
put
50
days
and
resolve,
if
you
really
want
to
solve
quickly,
you
know
ready
to
take
the
heat
until
a
server
you
put
0
1
and
everything
is
updated
in
a
matter
of
seconds.
E
B
E
E
I
am
presenting
this
on
v6
ops
on
Thursday.
Just
to
give
I
was
about
to
present
it
on
six-month
today
and
the
authors
filled
in
the
draft
doesn't
change
a
lot
on
email.
It's
basically
fine-tuning
some
wording,
change
and
comments
during
lunchtime
from
irritable,
some
wording
there
and
reading
some
explanation,
but.
F
B
F
So
most
of
the
changes
in
the
latest
revision
are
based
on
our
experience
implementing
it.
So
what
that
basically
means
is
that
whenever
I
run
into
something
that
I
didn't
want
to
code,
I
just
change
the
draft.
Now
all
jokes
aside,
Isis
latest
revision
gives
implementers
more
freedom
with
regard
to
which
functionality
to
implement
and
which
to
leave
out,
so
salt
options
were
removed
because
it
turns
out
that
we
didn't
have
a
use
case
for
them.
F
After
all,
we've
also
remained
the
socket
options
to
stack
options,
but
they
still
do
roughly
the
same
thing
as
before.
Next
slide,
please
so
the
approach
we
took
with
options
was
that
we
try
to
make
all
of
them
optional.
So,
if
all
you
want
is
a
basic
proxy,
you
can
strip
stocks
version,
six,
all
the
way
down
to
version
four
functionality.
If
you
want
to
do
authentication
as
in
v5,
you
need
to
support
authentication
method
options,
authentication
data
options
are
orthogonal
to
these
and
enable
you
to
do
zero,
ITT
authentication.
F
F
F
Obviously,
if
the
protocol
running
on
top
of
stocks
tolerate
CFO,
then
it's
okay
to
use
the
fo
also
if
the
seller
only
contains
the
request,
none
of
the
initial
data
and
there's
nothing
for
the
socks
proxy
to
replace
so
tier
4
is
also
safe.
Now,
TLS,
with
early
data,
is
already
safe
to
use
on
top
of
tier
4.
So
there's
no
question
about
about
that
and
if
you
haven't
got
any
other
kind
of
mitigating
circumstances,
other
important
options
can
make
replays
almost
impossible.
F
The
item
potus
mechanism
has
also
received
some
attention
now
I'm
going
to
give
you
a
quick
refresher
about
the
mechanism.
So
after
everything
we
set
up,
the
client
has
the
window
of
token
suspend
this
case.
The
window
contains
tokens
1
to
100.
The
client
spends
the
first
token
by
placing
an
option
in
the
request
and
it's
token
1.
Likewise,
the
proxy
places
an
option
in
the
reply
informing
the
client
that
the
token
was
been
successfully
long
with
the
with
an
option
informing
that
the
shifter
was
that
the
window
was
shifted,
best
spent
token
C.
F
So
it's
now
all
tokens
to
through
101.
So
next
slide,
please
so.
Obviously
the
main
point
of
the
mechanism
is
that
a
token
can
only
be
spent
1.
So
in
this
case,
if
we
try
to
spend
talk
on
1
twice,
the
second
request
in
quest
1
go
through
the
mechanism,
also
tolerates
tokens
but
received
out
of
order.
So
in
the
figure
to
the
left
as
spending
token,
2
does
not
cause
the
window
to
shift
because
toker
and
1
is
still
pending
and
as
soon
as
token
1
is
received,
the
window
is
shifted
white.
F
A
F
F
F
A
F
Is
honored
again,
this
is
not
something
that
we
want
to
happen
so
next
slide
so
to
mitigate
this
issue.
The
administrator
could
take
some
extra
steps
to
prevent
replace.
So
if
the
idempotence
options
were
used
to
enable
the
blanket
use
of
TFO
simply
disabling
here
for
one
ms
l
before
the
proxy
research
should
be
enough,
the
issues
with
LS
early
data
will
seem
to
go
away.
All
the
TLS
sessions
are
killed
right
after
the
restart.
So
next
slide.
Please
so
now
I'm
going
to
give
you
a
quick
taste
of
our
message
library.
F
So
it
has
all
the
features
in
the
drafter
and
we've
written
in
c++,
but
it
also
has
C
bindings.
What
you
see
on
those
on
the
screen
is
the
C
API.
So,
in
order
to
create
a
request,
you
simply
fill
out
the
structure
describing
what
a
request
should
do
in
this
case
we're
trying
to
connect
to
some
cyborg
on
port
80
and
we're
also
trying
to
use
the
fo
the
structure.
The
structure
has
many
fields,
many
other
fields,
but
you
simply
leave
them
blank.
F
So
if
you
don't
want
to
use,
then
you
simply
call
Essex
M
request
back
and
if
the
structure
you
filled
out
is
OK
and
the
buffer
is
large
enough,
you've
got
yourself
a
serialized
serialized
request.
Then
you
simply
use
whatever
network
API
to
send
it
over
to
the
proxy.
So
on
the
proxy
side,
you
simply
receive
a
bunch
of
bytes
and
then
you
attempt
to
and
then
you
call
a
6m
request,
parse
and
now
this
will
give
you
an
error
if
you
need
to
receive
more
of
the.
F
F
The
function
is
dealing
with
the
other
message.
Types
are
very
similar,
so
next
slide,
please
so
we've
so
the
feature
complete
message.
Library
is
available
on
github.
We
also
have
a
utility
library
that
implement
implements
or
kind
that,
basically
a
motley
crew
of
all
kinds
of
stuff
that
we
deem
necessary
for
the
development
of
Sox
six
apps
we've
used
these
two
libraries
to
implement
the
basic
prototype
based
on
shadows,
socks,
but
a
full-blown
implementation
with
most
of
the
features
in
the
draft
agenda
works.
You
can
also
find
on
my
page.
C
E
F
Alright,
there's
one
other
use
case
that
we've
thought
about,
and
that
is
a
MPT
CP
deployment
on
mobile
phones
and
that
we
don't
that's.
That's
a
use
case
where
you
don't
necessarily
want
security,
but
the
latency
between
the
mobile
phones
and
the
base
stations
can
be
quite
high
and
that's
why
we
try
to
shave
off
as
many
ITT
as
possible.
So
so
it
will
make
the
case
that
the
latency
optimizations
also
makes
sense,
even
if
you're,
if
on,
if
you're
running
over
a
plain
TCP.
So.
E
E
F
D
D
Then
overall
I
also
wanted
to
see
how
this
relates
to
some
work
that
we're
doing
like
the
taps.
Transfer
services
of
a
group
looking
at
the
generic
API
is
I'm
kind
of
looking
at
being
here
that
allow
you
to
do
and
other
things
so
I'm,
just
kind
of
pitching
your
offering
of
beings
and
working
on
this.
Yes,.
A
G
A
G
A
G
G
G
Considering
that
the
beta
is
a
succeeded
to
the
location
of
enemy,
we
need
a
way
to
associate.
The
data
could
useful
to
the
beta
their
consumers
is
receiving
so
the
consumer.
The
producer,
our
twin
points
in
the
communication
in
the
edges
in
architecture,
the
consumer,
is
that
the
employment
inspection
data
in
the
producer
is
the
infamous
participated,
so
the
consumer
must
be
able
to
authenticate
in
the
location
if,
anyway,
the
origin,
so
which
means
the
reason
for
signature
can
be
usable
producer,
will
find
useful.
The
sign
the
data.
G
G
H
G
G
G
In
the
reply
is,
is
the
piece
of
data
a
is
actually
carrying
application
later,
that
is
requested
by
the
so
we're
still?
Probably,
the
units
in
interest
is
actually
the
PDUs
that
is
used
to
index
the
data
packet,
so
it
can
be
used
to
fetch
the
data
can
use
to
identify
the
data
packet
cache
and
also
in
the
transport
entrance
to
prove
identity
in
three
layers.
I
G
G
A
E
G
For
today
so
forget
the
UDP
encapsulation,
which
is
mostly
for
other
things.
But
what
you
see
here
is
a
TCP
header,
just
like
the
new
wave,
but
there's
no
TCP
protocol
in
the
endpoints.
So
we
are
using
the
TCP
header
to
carry
I
seen
semantics.
So
there's
no
TCP
protocol
like
20
points.
It's
just
a
habit.
It's
just
a
continual
fields.
G
The
phenomenon
could
be
important.
Here
is
the
name
sockets,
so
the
one
that
I
just
presented,
the
name
suffix,
is
in
the
mix.
That
is
pretty
much
a
secret
number
in
the
there's
generated
by
the
endpoints
by
the
producer.
Anything
they
said
named
in
the
beta
write
it
not
necessary
right
pipe,
so
this
number
can
be
generated
in
babies
and
raised
in
community
application,
but
into
it
semantically
Kari
index
information
about
the
segmentation
made
by
the
producer
in
the
then
there
are
other
invested
next.
C
G
If
you
seek
consideration
about
what
adjusted
before
about
the
data
same
philosophy,
interesting,
which
is
what
every
ever
get
apart
every
even
in
the
network
layer,
we
need
to
authenticate
them
in
the
provide
integrity
who
ever
seen
the
little
pocket,
because
as
soon
as
you
introduce
location-independent
communication
even
know
anymore
because
can
be
retrieved
from
multiple
physical
different
impacts.
So
you
need
a
way
to
authenticate
the
association
over
the
main
prefix
that
is
used
for
the
production
phase
by
being
very
useful
when
the
consumer
fetches
the
data,
and
so
this
is
a.
G
This-
is
visible
in
the
in
the
in
the
pocket
in
the
layer
in
the
network
layer
in
two
ways.
Essentially,
there
is
one
information
in
it
which
easier
we
basically
insert
a
signature
of
a
security
envelope
of
the
data
packet
inside
a
data
packet,
so
the
well
name,
the
consumer
fetches.
The
data
can
verify
the
signature
just
by
knowing
the
public
you
sort
in
check
and
they
can
perform
the
verification
tracks
today,
file
by
with
verifying
the
symmetry,
but
also
the
participe
of
the
of
the
association
of
the
name
in
the
data
that
eventually
available.
G
There's
an
optimization
that
we
call
layer
for
manifest,
which
is
very
useful,
which
basically
comprises
this
knowledge
instead
of
doing
this
basis
in
group
of
packets.
Well,
it's
an
optimization,
but
is
also
actually
did
before
thank
functioning.
They
were
using
our
software
because
it's
better
essentially
they
say
yeah,
it's
a
new
probable
units,
they're,
sparing
the
serials
of
recent
past.
G
G
G
So
in
this
client
you
can
have
a
consumer
end
point
in
the
producer
important
in
a
service,
you
can
have
a
consumer
end
point
in
the
producer
end
point
depending
if
you're
sending
data
or
receiving
the
year,
but
from
a
communication
perspective.
Consider
in
this
example,
one
consumer
endpoint
and
one
producer
endpoint.
So
this
is
say
a
so.
G
We
have
basically
two
different
forwarding
paths
in
HSE
and
one
is
the
forwarding
paths
followed
by
the
request
and
the
other
one
is
the
before
name
pass
followed
by
the
replies,
and
they
are
not
their
default
different
kind
of
semantics.
So
in
this
case
this
is
the
interest
path,
so
the
request
path,
the
forwarding
path.
G
So,
on
the
left
hand
side,
you
see
an
ingress
interface
and
a
request
packet
coming
in
and
so
the
floating
pipeline,
so
the
router
is
very
similar
to
a
v6
router.
The
only
difference
is
this
new
color
data
structure.
You
should
call
packet
cache
where
we
store
TEUs
for
a
certain
amount
of
time
which
are
instrumental
to
perform
the
the
packet
processing
operations.
G
So
the
first
operation
is
to
to
get
the
names,
so
it
was
a
the
prefix
in
the
full
name
out
of
the
data
packet
and
in
this
information
is
used.
The
full
name
is
used
to
perform
an
exit
exit
match
lookup
in
this
packet
cache
to
find
either
a
reply
or
request
matching
this
full
name
if
the
interest
pocket.
So
if
the
request
packet
finds
an
exact
match,
Luka
can
reply
with
the
data
packet.
G
The
data
packet
is
sent
back
to
the
through
the
same
input
interface
by
performing
address
translation
operation
into
the
reply
by
inverting
the
information
that
is
in
the
unit
in
this,
in
the
request
in
today
into
the
reply.
So
basically
we
we
copy
into
the
data
into
the
reply
destination
address
field,
the
the
information
contained
in
the
source
address
field
of
the
request,
so
in
because
remember
in
the
source
address
field
on
the
request,
we
have
a
locator
that
is
used
by
the
data
packet
flow
back
to
the
to
the
previous
requester.
G
So
this
is
a
first
optimization,
so
that
means
that
the
data
packet
can
be
can
be
sent
back
to
the
consumer
without
going
up
to
the
producer,
if
this
is
being
a
temporarily
cache
in
the
neighbors
of
the
other
consumer,
for
instance.
If
this
doesn't
happen,
what
happens
is
if
no
painting
request
is
available
in
the
rather
then
persistent
upstream,
just
like,
if
he's
in
a
regular
ipv6
packet.
G
G
Okay,
next
line,
so
the
data
path
is
a
little
simple.
So
you
can
see.
The
request
is
basically
is-
is
flowing
upstream
towards
the
producer
or
producers
by
using
the
v6
feed,
but
just
like
a
regular
v6
rather
class,
the
feminization
bucket
cache
to
reply
in
case
the
data
is
available.
Luckily,
the
data
pack
is
very
different
because
later
path,
we
don't
go
through
the
the
v6
feed
barrymore,
so
there's
no
operation
on
LPM
in
the
v6
FIBA
local
fit.
G
So,
in
this
case,
we
have,
on
the
right
hand,
side
an
ingress
interface
and
the
data
packet
is
flowing
from
right
from
right
to
left
to
the
ingress,
and
so
the
partial
in
the
pipeline
extract
the
the
name
prefix
of
the
full
name
and
we
copy
in
the
packet
cache.
You
essentially
need
to
find
one
or
multiple
carriers
that
are
making
this
the
name
in
their
pocket.
So
you
can
even
one
remaining
request
pending
for
the
same
full
name,
because
he
can
have
multiple
consumer
as
to
the
same
thing
as
a
isn't
multicast
for
instance.
G
So
what
happens
is
that
the
data
packet
is
flowing
back
is
is
clone
early
interfaces
that
are
identified
by
locator
in
a
different
pending
request,
so
that
when
the
data
packets
is
sent
back,
you
perform
a
another
translated
operation
on
the
destination
of
this
field
of
the
data
packet,
which
is
just
like
this
written.
So
for
every
single
data
packet
you
store
in
the
destination
that
was
filled
of
the
data
packet.
G
The
information
there
is
a
currently
available
in
the
source
address
field
of
each
in
the
request,
next
line,
of
course,
so
the
the
typical
deployment
of
this
kind
of
node
is
to
be
stored
within
an
ipv6
network
and
where
you
have
regular
v6
traffic.
So
you
in
in
order
to
to
get
HS
en
traffic
v6
traffic
out
of
the
the
network
in
process.
Is
this
spatial
fact
it's
in
a
different
way
into
the
final,
pointing
rule
which
is
basically
related
to
the
specific
network.
Configuration
of
your
use,
like
that,
you
must
be
able
to.
G
You,
must
be
able
to
recognize
that
one
IHSA
in
traffic
first
by
looking
at
me,
by
identifying
the
the
existence
of
an
a
location
independent
named
in
the
in
the
data
packet,
and
secondly,
you
need
to
understand
if
this
is
a
request
or
reply
and
depending
on
the
first
of
the
second
case,
you
found
the
traffic
you
can't
the
in
the
packet
towards
one
of
the
two
forwarding
paths
that
I
just
described.
So
there
are
many
ways
it
states.
This
is
still
an
open
problem,
so
we
in
our
implementation.
G
We
have
one
way
to
do
this,
based
on
classification
ratio,
but
can
also
be
done
in
other
ways,
so
next
line
so
a
little
bit
about
applications
before.
As
you
can
understand,
these
are
not
made
to
support
TCP
or
UDP
kind
of
traffic,
but
this
is.
This
is
actually
made
they're
not
described
in
detail
here
for
any
kind
of
a
transport
service
running
on
top
of
this
piece,
new
mechanism,
so
I
detailed,
but
essentially
we
have
a
new
transport
run
sitting.
G
On
top
of
this
request
reply
semantics
we
provide
already
now
implementation
to
mattingly
a
post,
socket
API
as
the
one
that
is
currently
being
worked
out
in
the
tax
work,
in
also
for
backward
compatibility,
a
socket
API,
it's
a
traditional
subject:
VI.
We
provide
two
application.
Two
different
kind
of
new
sockets
taxes
socket
the
socket
types.
One
is
the
producer
socket
when
one
is
a
consumer
socket
because
you,
the
socket,
of
course,
is
used
to
send
the
data.
The
consumer
socket
is
used
to
fetch
the
data
and
they
are
always
in
collection.
G
So
if
you
have
a
be
directional
communication,
you
basically
open
two
sockets
in
the
client
or
the
server,
so
the
producer
socket
performs
the
typical
layer,
four
operations
or
like
producer/consumer
sockets,
like
segmentation,
reassembly
plus
additional
operation
like
signature,
computation
in
integrity,
computation
and
the
other,
the
other
and
signature
verification,
integrity,
verification,
the
consumer,
the
consumer
in,
of
course
reassembly
as
usual.
So
you
can
have
different
types
of
circuit
just
like
today
that
data
gram
stream,
but
they
they
go
through
the
semantics
of
the
forwarding
path.
G
The
kind
the
HSE
in
approaches
to
to
go
directly
in
school,
currently
inside
ipv6,
install
over
running
on
top
of
IP
v6
or
next
to
ipv6,
replacing
completely
the
the
network
stack.
So
this
is
a
the
chosen
approach,
so
we
try,
which
turns
out
to
be
very
useful
for
many
many
innocence
deployment
source
implementations.
Instead,
we
own
prototype,
then
we
plan
to
open
sourcing
of
elements
in
the
for
essentially
support
to
application
like
HTTP
and
RTP.
In
the
first
place,.
B
Thank
you
very
much
value
from
this
mic.
I
have
a
question
for
cognizant:
Liga,
Archer
you're,
talking
about
the
packet
format,
Lea
flow,
and
you
talk
about
reducing
the
TCP
header
by
default.
Still,
you
say
this
is
not
TCP
traffic,
so
why
is
it
that
you
want
to
be
used
to
you
not
suggest
another
transported
there.
I
G
I
mean
to
to
have
a
high
hit
ratio
in
different
networks
like
and
also
the
TCP
header.
It
also
carries
semantically
some
of
the
some
of
the
things
that
you
actually
have
in
any
kind
of
transfer
like
segmentation
information
in
the
right
place
already,
so
it's
been
instrumental
to
implement
actually
all
the
features
we
need
it
for
ICM
semantics
in
the
transport.
So
then
there
is
a.
There
is
one
of
the
choices
who
made
them
for
convenience.
E
G
I
G
The
producer
socket
is
a
buffered
binds
to
a
prefix,
and
the
prefix
is
the
location
independent
number
so,
which
means
they're
the
producer
side.
The
producer
socket
alright
continue
to
use
the
term
socket,
but
the
previously
the
producer
endpoint,
basically
is
nothing
more
than
a
buffer
that
binds
a
short
memory
that
binds
to
a
v6
prefix
that
is
available
to
a
number
of
consumer
sockets
buying
into
the
same
prefix.
So
this
is
the
the
high-level
semantics
of
this
end.
Point.
E
G
We
don't
go
through
middle
box,
are
doing
connection
trackers
no
way
by
in
a
way
we
don't
care,
like
the
assumption,
is
not
to
go
through
any
middle
box
but
to
be
able
to
to
deploy
a
protocol.
That
is
not
there's
not
consider
it
as
an
attack
for
even
an
hour
even
see
swagger
like
Rory
Ave
is
basically
a
battleground.
A
communication.
G
Connectionless
is
a
connection
to
communication,
while
some
bento
box
daya
seems
by
looking
at
the
TCP
header
that
you
have
a
connection
TCP
connection,
so
we
can
modify
the
header
files
for
assuming
some
behavior
right.
So
we
don't
we
don't.
We
don't
support
the
new
scales
like
in
anytime.
You
have
like
a
corporate
firewall
at
the
Commission
tracker.
We
go
through
those
boxes
by
using
family
between.
E
G
I,
don't
have
any
any
special
choice
is
just
as
long
as
it's
worse.
The
thing
is
that
we
need
to
carry
semantics
in
our
protocols
satisfy
the
net,
the
architectural
we,
the
architectural
philosophy
by
carrying
names
inside
the
packet
header
in
for
essentially
the
reason
why
we're
using
this
is
header
we
use
this
errors
is
essentially
to
add
the
facility
to
deploy
this
kind
of
protocol
immediately.
G
You
may
think
about
between
a
new
header.
There
would
be
very
difficult
to
deploy
in
the
short
term
and
maybe
the
playable
in
the
long
term.
So
it's
a
it's
a
it's
a
convenience
for
the
plenty
of
reasons.
So
this
is
something
that
we
can
equate
today.
Instead
of
you
know,
redesigning
a
new
header,
it
would
be
hard
to
get
deployed
and
go
through
deployments
actual.
The
problem
today
is
usually
no
I
understand.
G
The
forwarding
engine
that
I
described
through
the
DNS
okay
I
just
go
through
v6
ATM
blue
cups
in
the
inverter,
so
which
means
that
when
the
packet
gives
it
goes
out
from
from
a
consumer
endpoint
with
it
already
as
the
location,
independent
names
in
the
destination
does
field
and
any
worse,
that's
an
address.
That's
right!
Where
does
that's
the
name?
The
name
is
a
Bob
there's,
no
other
name,
as
the
app
name
is
the
others
itself.
G
E
E
G
For
Royce
policies
are
basically
into
science
and
the
reply
is
empty
signs
like
you
go
through
the
usual
segmentation
into
your
size
segmentation,
so
it
was
more
especially
I
need
pocket,
so
the
maximum
size
of
something
that
I
can
retrieve
is
in.
This
is
1
K,
1
and
1/2
K
or
if
you
have
a
other
kind
of
segmentation
into
yes,
it
depends
on
the
NTU.
That
means
every
I
have
to
index
all
the
chunks
of
whatever
I'm.
E
E
G
You
have
one
request:
it's
just
like.
If
you
look
at
the
TCP
streamer,
you
have
basically
data
sent
and
the
acknowledgement
to
be
sent
to
say
to
the
endpoints
I
receive
the
data
here
is
exactly
the
opposite:
I
request
the
data
to
get
the
other
christa
beta,
and
then
I
get
the
beta
bag.
But
does
that
mean.
A
G
G
B
I
G
I
G
G
Your
congestion
control,
okay:
this
is
another
another
question.
There's
no
one!
There
are
many,
because
what
I
described
right
here
is.
It
is
just
the
network
piece
of
it
as
I
said
well,
several
strands
for
services.
You
can
reliable
transfer
services,
unreliable
transport
services.
Let's
take
the
example
of
a
reliable
transfer
services
similar
to
TCP
what
you
can
anticipate.
For
instance,
then
what
happens
is
gonna
have
a
producer
socket
that
is
a
producing
data
and
a
consumer
of
socket.
G
That
is
infection
data,
so
the
consumer
socket
would
be
responsible
to
basically
issues
as
many
interest
survey
requests
as
required
to
get
a
full
control
retrieval
of
the
data
from
the
producer
side,
okay,
so
the
congestion
control
and
full
control
were
you
today
perform
in
TCP
on
the
suicide
in
this
case
is
from
the
consumer
side
by
regulating
the
rates
of
the
request
sent
word,
but
the
consumers
got
it.
Okay,.
I
A
I
G
Is
I
mean
this
is
true
anytime?
You
have
a
new
transport
protocol
in
the
internet.
Any
time
you
have
red
bag
deployed,
but
be
tiring
and
people
scare
about
that
because
they
say
this
is
new,
and
so
you
go
through
the
ITF,
which
is
what
I
am
here
right
now
to
say,
don't
be
afraid
there
was
something
to
describing
in
this
framework
so
that
you
can
extend
what
is
about
okay.
E
D
I
think
this
is
an
interesting
experiment
for
that
category
of
things.
You
definitely
are
trying
to
do
API
that
specifies
whether
or
not
you
have
reliability
or
unreliability.
You
are
dealing
with
messages,
not
just
stream
to
Datagram
feel
like
messages
with
interesting
data
in
them
and
then
going
forward.
D
One
of
the
things
that
we
currently
don't
have
in
those
api's
on
that
we
need
to
add,
is
better
support
for
things
like
multitasking,
that
you
do
have
truly
consumer
and
producer
models,
so
I
think
that's
gonna
be
and
I
think
actually
that
will
fit
much
better
to
what
you
have
then
either
sockets
today.
Thank.
G
Everything
is
represented
here,
such
as
many
aspects,
including
which
transfer
layer
in
the
in
a
socket
API
I
still
continue
using
the
term
second
API.
Just
in
there
are
many,
many
working
groups
were
listening,
so
I
considered.
So
this
is
a
first
step
to
disclosure.
Some
of
the
work
you've
done
at
Cisco
in
many
other
things,
making
school
do
other
work
so
that
the
full
pictures
would
be
more
clear,
but
it
can
be
done
in
one
single
shot
of
many
other
things
too
in
the
question
is
raised
in
today.
Absolutely.
H
Okay
and
keep
you
awake,
okay,
the
title
of
the
draft
is
IP
fragmentation
considered
fragile
and
that's
probably
not
thrilling
to
you.
What
will
be
thrilling
is
the
last
slide
in
the
last
slide,
we're
going
to
make
some
recommendations
a
that
people
not
develop
new
applications
that
rely
on
an
IP
fragmentation,
because
it's
fragile
also
it
points
out
one
application
that
is
absolutely
critical
to
the
internet
that
relies
heavily
on
IP
fragmentation.
H
There's
a
recommendation
to
those
folks
to
maybe
reconsider
their
reliance
on
IP
fragmentation
so
that
the
world
can
migrate
to
ipv6
without
too
many
more
problems.
Anyhow,
what
are
we
going
to
talk
about
we're
going
to
describe
fragmentation
really
quickly,
I'm
going
to
assume
most
of
you
understand
how
it
works
already
we're
going
to
describe
health
fragmentation,
reduces
reliability
and
we're
going
to
talk
about
those
recommendations.
H
Okay,
how
does
fragmentation
work
well,
I
PV
for
fragmentation
defined
in
the
ipv4
spec.
Fragmentation
is
allowed
at
the
source
and,
if
you
do
not
set
the
DF
bit,
fragmentation
is
allowed
downstream.
By
contrast,
an
ipv6
fragmentation
is
allowed
at
the
first
fragment
only
there
is
no
DF
that
it's
always
assumed
to
be
don't
fragment,
dentistry,
there's
something
that
ipv6
and
ipv4
have
in
common.
The
upper
layer
header
appears
only
in
the
first
fragment.
So
when
you
fragment
a
packet,
you'll
see
the
upper
layer
header
on
fragment
one.
H
H
Okay,
thank
you.
The,
when
you're
fragmenting
at
the
source
only
either
if
you're,
ipv6
or
ipv4
with
the
df'
set
the
source
has
to
do
one
of
two
things.
It
either
has
to
refrain
from
sending
packets
that
may
be
fragmented
in
the
case
of
ipv6.
It
has
to
refrain
from
sending
packets
that
are
less
than
1280
bytes
long
or
it
has
to
maintain
a
running
estimate
of
the
pam
to
you.
H
Now
there
are
two
ways
to
do
this:
one
is
with
PMT
you,
the
PMT
ID,
the
other,
with
PLP
MTU
D,
and
with
some
extensions
by
a
query
therapist
at
or
Virginia
it's
not
defined
for
UDP.
Now,
let's
talk
about
how
PMT
UD
works,
the
source
does
an
initial
estimate
of
the
PMT
you
that
is
equal
to
the
MTU
of
the
first
half
but
then
start
sending
packets.
H
If
one
of
the
packets
is
larger
than
the
actual
PMT
you
a
downstream
node
can't
forward
the
packet
it
drops,
the
packet
sends
an
ICMP
ptb
back
to
the
source,
the
source
updates,
its
PMT.
You
don't
estimate
this
whole
strategy
relies
on
the
intermediate
node
being
able
to
get
an
ICMP
ptb
back
to
the
source.
H
Pl
PMT
works
a
little
bit
differently.
Again,
it
produces
an
initial
PMT.
You
estimate
it
sends
probe
packets
in
various
sizes
and
relies
on
Reyes
for
acknowledgments
the
source
updates
the
PMT.
Accordingly,
it
doesn't
rely
on
the
ICMP
PDB.
It
may
accept
them
if
it
gets
them,
but
if
there
were
lost
PL
PMT
still
rocks
now.
Why
does
fragmentation
break
things?
Well,
it's
because
that
upper
layer
that
way
a
floor
header
only
appears
in
the
first
fragment.
This
will
mess
up
those
balancers.
H
It
will
mess
up
prior
us
and
it
would
mess
up
other
Miller
boxes.
Think
for
a
minute
about
a
fire
were
that
is
blocking
our
packets
that
except
those
that
arrived
on
TCP
port
80.
Well,
it's
fine
on
the
first
fragment
when
it
knows
of
the
packet.
By
the
way,
this
is
a
stateless
fire.
Well,
it's
fine!
If
the
first
packet
is
on
TCP
port
80
second
packet,
it
has
no
idea
that
it
was
a
TCP
packet
packet,
hello,
the
TCP
header,
isn't
in
the
second
fragment.
H
Okay,
others
are
problems
with
fragment
presentation.
There
are
some
real
known
security,
vulnerabilities,
overlapping,
fragment
attack
attacks
are
resource,
exhaustion
attacks.
Then
there
is
the
possibility
that
your
ICMP
PTP
is
going
to
be
lost,
and
there
are
many
reasons
why
that
happened.
We
won't
jump
into
them.
Probably
the
most
common
is
common
bugs
and
consumer
CPA
equipment.
A
H
H
H
Is
the
solution
that
most
people
rely
on
today?
Next
is
transport
layer
solutions
like
PLP
MTD
from
TCP
or
UDP
in
process
right
now
now?
Finally,
the
recommendations,
the
first
recommendation
is
for
application
developers.
Please
do
not
develop
more
applications
that
rely
on
ICMP
fragmentation
to
do
so
well
in
the
world
of
ipv6.
Only
second
for
network
operators,
please
don't
filter
those
ICMP
pdbs
and
you
should
not
deploy
equipment
that
discards
all
packets
that
contain
extension
headers.
We
suspect
that
a
good
deal
of
that
is
deployed.
H
But
there's
a
metal
recommendation:
there
is
one
application
DNS,
particularly
with
DNS
SEC,
that
receives
very
small
requests
and
answers
with
very
large
responses,
responses
that
may
be
larger
than
1280
those
responses
they
can
go
either
over
UDP
or
TCP.
Typically
they're
over
UDP
performance
gets
pretty
lousy
when
it's
over
TCP,
because
you
have
to
do
all
the
connections.
H
E
So
I
would
be
effective
if
this
stuff
doesn't
work.
However,
I
feel
that
if
we
say
don't
do
this,
then
developers
who
build
networking
applications
they're
just
gonna,
say:
oh
I,
know
I'm
just
gonna
do
my
own
fragmentation
and
they're.
You
know
they
may
or
may
not
end
up
making
the
same
mistakes
that
we've
made.
So
it
seems
useful
to
say,
look
if
you
really
want
to
do
fragmentation.
E
Do
it
like
this
or
like
you
know,
don't
do
it
unless
you
have
to,
but
if
you
do
have
to
do
it,
you
know
I,
don't
know
what
the
best
practices
are.
Presumably
like
include
a
sequence
number
somewhere,
because
otherwise
your
parents
are
doing
support
overlapping
fragments
right,
there's
been
like
a
history
of
security
vulnerabilities
right,
so
you
know
I,
don't
think
the
recommendation
can
just
be
used
TCP.
It
might
be
used
TCP
or
Creek,
but
you
know
I
think
we
have
to
say
something
I
think
you're
right.
It.
B
E
J
E
H
I
A
I
Support
this
draft
also
the
something
else
to
think
about
having
we
used
translations
thinking
about
them.
If
I
have
a
network
in
which
the
m2
use
are
all
the
same
length
return
on
the
middle
of
the
problem,
what
usually
happens
is
like
ringing
up
the
fight
making
inside
the
torrents
attack.
Another
header
on
so
I
think
something
very,
very
usually
doing
here
would
be
recipe
for
those
who
want
tunnels.
This
is
just
how
how
to
go
about
doing
it
without
without
having
to
resort
to
fragmentation.
This,
of
course,
interaction.
The
tunnel
strap
is
working
yeah.
B
E
E
E
So
one
fact
many
types
of
limited
are
in
fact
appearing
for
many
years
home
office
moment
for
networks.
Well,
if
it's
networks,
vehicle
networks,
wedding
services,
networks
that
was
bad
and
networks,
its
control
networks
of
one
kind
or
another,
since
I'm
it
works,
MIT
agent
works
in
a
rumored.
Whether
they
exist
is
another
question:
enterprise
networks
and
caps
and
writes.
We
fought
for
many
many
years.
Data
centers
hosting
centers
and
some
of
those
are
distributed
located
in
one
place.
Work
slicing
is
appropriate
topic
in
some
places,
see
deals.
E
E
E
E
E
E
E
I
think
in
reality
is
we
have
to
do
that.
Such
protocols
are
not
required
to
operate
across
scenes
at
all
and
for
me
ensuring
the
most
adduction
from
that
is.
We
need
to
define
a
limited
domain
boundary
in
some
way:
half
methods
of
security
funding
to
membership
within
that
boundary,
and
that
is
intentionally
where
the
document
stops.
The
worst
is
TBD,
and
our
main
focus
this
week
is
to
garner
interest
in
this
topic.
You
can
write
to
the
author's
addresses
are
in
the
list
in.
E
Ago,
talking
about
a
piece
of
the
option
for
being
original
addresses
in
a
TCP
heck
in
case
it
was
like
a
proxy
or
something
involved
and
I
became
of
interest
to
me,
because
I
was
looking
at
figuring
out
where
I
came
from
in
a
provider
network
where
there's
many
ways
of
my
little
balance
etc.
Those
finding
this
information
rate
for
me
and
ultimately
for
reading
this
document,
it
was
decided
it
was
done
because
I
wouldn't.
E
A
I
I
Reasons
for
saying
that
we
actually
do
have
one
domain
that
was
only
diffserv.
It's
been
used
over
time
since
and
X
works
quite
well,
which
is
the
boundary
of
an
operator's
network
or
administrative
domain.
I
suggest
that
we
actually
do
have
the
protocols
that
do
a
pretty
good
job
of
defining
the
main
membership
of
that
particular
sort
of
domain
in
practice
likely
the
usual
sorts
of
domains.
E
I
E
D
E
E
And
what
does
it
mean?
Then
I
mean
generally
I
think
this
is
a
way
to
be
I
think
it's
it's
a
way
to
be
lazy
or
it
could
say.
Well,
you
know
we
don't
want
to
think
about
all
these
other
use
cases
really.
This
is
only
gonna
be.
This
is
one
of
the
things
that
we
hear.
A
lot
is
like
this
is
never
going
to
talk
to
the
internet.
That's.
A
C
E
Suggest
that
II,
that
one
way
of
doing
that
is
to
make
the
failure
mode
absolutely
right.
So
if
you
say
look,
if
you
do
this,
it
would
just
like
totally
fall
over.
That
is
actually
a
safe
way
of
doing
this.
If
it
almost
works,
then
that
is
a
recipe
guarantee
actually
a
future
hacker
at
the
borders
of
these
things,
like
not,
for
example,.
C
C
So
so
we
have
the
ideas
perfect
couple
years
back,
there
was
an
eye,
a
tick
bath
proposal
this
year
and
they're
all
sort
of
discussions
about.
Oh,
we
are
privacy.
We
should
probably
do
something
about
that,
but
didn't
get
into
a
whole
lot
of
detail,
so
I
started
thinking
in
it
about
histone,
said:
okay,
let's
write
down
some
more
stuff
and
see
and
throw
it
out,
though,
and
see
what
fruit
comes
back
and
he
goes.
What
did
I
do.
C
Some
identifier
is
out
there
that
maybe
they
are
associated
with
you,
video
streaming
or
you
watching
a
movie
while
you're
wandering
around
and
and
you
don't
you
want
to
be
able
to
keep
that
stuff.
So
the
analysis
here
it
says:
okay,
let's
assume
that
we
have
an
identifier,
that's
long-lived
and
see
what
are
the
privacy
issues
with
that,
and
then
we
can
always
improve
it
by
saying
okay,
some
traffic
can
have
short-lived
ephemeral,
identifier,
strong
and
so
that's
sort
of
horoscope
things.
C
The
graph
talks
a
bit
about.
Why
do
we
have
ID
locators,
but
and
why
do
we
want
to
expose
the
locators
and
fundamental
needs
about,
in
my
mind,
about
latency
bandwidth
and
sort
of
isolation?
I
can
keep
local
traffic
logo.
I
can
minimize
stretch
from
marauding
perspective
and
hence
I
want
the
cases
when
I
want
to
be
able
to
share
those
locators
to
make
my
traffic
be
as
direct
as
possible,
but-
and
this
is
sort
of
my
initial
shortlist
range
like
so
so.
C
What
are
the
considerations
about
when
you
want
to
and
well
you
might
want
to
to
share
it
a
friend
at
the
party
or
the
party
next
door,
a
streaming
video.
We
already
know
that
you're
physically
at
the
party
right.
So
do
you
have
an
issue
with
sharing
your
IP
locator?
Well,
probably
not
because
you're
willing
to
share
DPS
location.
C
C
I
mean
well,
that's
not
that
different
from
a
privacy
perspective
than
saying
I'm
gonna
send
around
horse
trots
in
OSPF
right
somebody,
somebody,
it's
part
of
that
trust
domain
is
going
to
be
able
to
see
the
whole
link
state
database
and
know
where
everything
is
so
being
everybody
locator
that
doesn't
change
anything
from
perspective.
It's
really
about
defining
a
domain,
and
then
the
other
case
is
like.
Well,
somebody
owns
a
bunch
of
trucks
right.
C
They
want
to
be
able
to
know
the
GPS
locations
of
those
trucks
being
able
having
them
be
able
to
know
the
IP
locators,
but
doesn't
have
any
privacy
issues,
but
these
are
other
cases.
One
okay,
I'll
leave
this
party
right.
I,
don't
want
my
friend
to
know
what
I'm
doing
the
rest
of
my
life.
All
right,
I
wanna
be
able
to
choose
when
making
Sita
and
I
don't
want.
You
know
the
trucking
company's
competitors
shouldn't
be
able
to
tell
the
IP
locators
of
the
trucks
right.
A
C
Again,
a
strike
I
think
there's
this
one
thing
which
is
just
about
the
location
privacy.
Can
you
tell
the
IP
locators
of
some
device?
Some
to
my
name
product
is
another
one,
which
is
the
mere
fact
of
something
having
moved
at
a
given
point
in
time,
if
I
can
get
a
notification
that
this
identifier
moved
from
this
room
to
the
hallway
and
I'm
standing
here,
watching
I
can
figure
out
who
walked
out
the
board
correlate
just
those
new
many
events
with
other
information,
and
so
they
can
actually
be
privacy.
C
C
Split
and
you're
willing
to
use
that
to
deploy
some
fixed
points.
Basically,
some
anchor
points
in
the
network.
Thinking
you're,
never
like
you
have
agents
or
whatever
you're
saying
that
my
locator
is
always
gonna,
be
Antarctica,
know
what
matter
in
the
world.
I
am
right.
I
can
use
these
already
located
what
to
say.
C
Well,
you
can't
watch
me
move
even
less
than
today,
because
even
when
I'm
going
to
go
access,
Google,
whatever
I'm
still
I'd
be
located,
don't
Arctic,
no
matter
where
that,
in
course,
more
stretch,
your
lower
performance
whatever,
but
there's
trade-offs
to
do
in
this
place.
That's
it
who
wants
to
contribute
here
or
elsewhere,
and
if
you
have
friends
that
are
working
on
this
dog,
you
know
this
is
not
an
internet
area
problem.
It's
an
internet
problem.
B
E
E
This
document
adds
additional
functionality
into
those
IP
re
clients
to
hopefully
mitigate
some
operational
issues
or
an
operational
issue
that
we
are
experiencing
right
now.
The
catalyst
to
write
this
draft
was
yet
another
discussion
with
appear
in
the
industry
around
the
benefits
of
IPRA
versus
people.
E
Argue
against
was
its
lack
of
sort
of
session,
state
tracking,
something
that
you
get
intrinsically
with
PvP
and
as
keep
arrives
so
in
a
PPO
eland,
both
sides
of
a
PB
session
will
seem
LCP
echoes
to
the
remote
peer
which
gives
them
disability
connectivity.
So
if
that
path,
for
whatever
reason
temporarily
phone,
both
sides
had
visibility
and
they
can
act
accordingly.
So
the
BMG
can
can
clear
the
D
release
the
lease
they
can.
E
Tear
down
a
subscriber
station
and
the
city
can
restart
this
pppoe
client
and
all
the
to
trigger
that
authentication
with
radius
and
around
also
the
benji
can
allow
a
subscriber
to
really
educate
and
connect
as
quick
as
possible
in
our
community,
and
you
don't
really
have
that
sort
of
session
tracking
at
the
moment.
So
what
several
ng
winters
have
further
have
done
will
take
it
upon
themselves.
To
do
is
implement
their
own
kind
of
physician
tracking
using
our
neighbor
discovery
again.
E
E
Issues
well,
CPE
can
be
left
with
a
stone
deity
in
lease
for
up
to
world
lifetime.
Of
that
least,
it's
something
Avella
prefix,
delegation
or
I
can
be
for
address
and
it
would
still
be
for
trying
to
for
packets,
viola,
B
and
G
to
the
Internet
and
the
B
engine
will
just
be
dropping
them
on
the
floor.
So
what
happens
today
is
that
after
half
of
the
life
time,
so
in
this
example,
it's
set
to
an
hour.
So
after
three
minutes
the
t1,
timer
would
expire
and
well
our
period
client
would
sing
a
dhp.
E
E
B
E
Well,
I
just
wanted
to
know
if
that
behavior
for
the
bng
not
responding
to
requests
as
a
bug
according
to
the
standard
and
therefore
could
be
argued
that
it's
easily
fixed,
you
might
have
checked
yeah.
That
is
that
that's
a
bug
right,
so
I
guess
dependency
axis
nodes
implementing
LDR
a
would
probably
in
sort
the
option,
a
detour
and
watching
three
seven
information
into
the
renews
as
well.
So
technically,
the
Benjy
could
probably
authenticates
based
on
a
renew
so
I.
E
E
E
How
frequently
will
seem
that
after
we've
had
our
films
check
the
limit
being
the
number
of
consecutive
failed
attempts
before
we
consider
the
health
check
failed
and
then
the
behavior
to
trigger
once
that's
failed
the
person
flag
that
allows
our
about
forces
actually
or
above
this
are
ideally
health
care
clients
to
just
passively
check,
as
opposed
to
sitting
probes
and
in
a
lie
at
all.
Would
force
allowed
to
check,
as
opposed
to
our
using
BFD
echo,
that
we
prefer
and
in
the
alternative
target
dress
forms
backs
our
horses
until
I
use
this
IP
insolence
Defoe?
E
E
E
E
If,
for
whatever
reason,
the
client
cannot
use
service,
PFT
echoes
soil,
health
check
and
or
it
is
forced
to
buy,
that's
our
right
to
flag,
then
it
can
use
part
and
maybe
discovery
for
that
health
check
pro
by
default.
The
default
gateway
affects
both
the
ultimate
target:
first
cousins
and
then
again,
if
the
arkarow
torrents
cannot
or
what
is
forced
to
to
not
user,
it
can't
passively
more
till
the
art
and
over
I
know
the
cache
told
us
we
were
post.
E
Now
so
a
health
check
has
felt
we
withdraw
for
this
document
specifies
battle
running
for
different
our
babies.
The
default
0
is
renew.
This
is
considered,
probably
the
most
the
surface
approach.
This
would
just
seem
to
renew
back
to
the
original
server.
So
I'll
pull
that
previous
slide.
It
may
not
necessarily
work
for
all
BG
vendors
because
they
may
not
necessarily
both
news,
so
there
it
is
in
line
with
broadband
forms
to
our
one
four
six
and
that's
what
they
currently
mentioned
should
do
so
then
the
next
one
is
rewind
behavior.
E
This
still
doesn't
solve
that
problem,
where
you
still
require
solicit
or
the
speller
fourth
education,
but
it
is
possibly
beneficial
for
apologies.
Well,
that
may
have
note
will
be
a
gene,
spar
and
another
one
could
and
almost
subscribe
to
command
right.
If
the
previous
BG
is
phone,
for
example,
the
third
option
is
solicit.
This
is
probably
the
one
that
I
see
be.
The
most
common
is
sitting
the
one
that
I
saw.
I
was
using
on
our
network,
and
this
behavior
skips
the
previous
two
phases,
and
it
goes
straight
to
sending
the
solicit
a
discovery.
E
E
The
thought
develop
approach
in
this
document
at
the
moment
is
we
experiment
lease.
This
is
more
of
a
scorched-earth
approach
in
the
fact
that
it
would
immediately
expire
the
lease
it
will
send
out
any
dhcp
release
to
the
network
to
clear
out
any
possible
sale
states
within
the
network,
how
many
proxies
and
relays,
and
then
it
will
move
against
the
solicit
and
discovery
face
to
allow
the
authentication
and
stealing
the
soils.
E
E
Johnny
speaking
at
a
general
coming
about
a
couple
things
over
so
first
we
have
four
different
recovery
mechanisms
and
you
have
like
tunable
parameters
in
DHCP,
and
how
do
you
expect
people
implement
up
correctly
when
you
have
so
many
combinations?
I
mean
it's
easy
to
say.
Okay,
we've
got
like
16
combinations
of
ways
to
do
this
and
we're
gonna
clamp
our
vendors
if
they
get
any
of
them
wrong,
but
it
seems
easier
to
say
well.
We
only
have
one
way
to
do
this
into
the
more
likely
to
result
in
success.
E
I
feel
so
I
would
think
everyone
looks
good
to
have
lots
of
options.
You
might
be
able
to
gain
consensus
on
your
own,
the
graph
or
easily
it's
like.
If
you
chop
them
down,
it's
actually
gonna
be
more
yeah.
Thank
you
for
threat
of
rust.
I
have
been
told
to
just
keep
it
simple,
so
yeah
I
couldn't
be
convinced
easily
to
drop
the
Forester,
because
I
don't
believe
that
they'll
be
too
useful
sitting
on
my
network
and
I'm,
not
sure
other
networks.
They
would
be
useful
so
again,
based
on
community
feedback,
I'll
be
I.
J
So
what
means
we
are
not
positioned
in
the
work
is
that
that
we
are
an
individual
submission
truck.
We
are
open
to
any
what
buckets
like
to
work,
but
what
not
the
access
management
services
about?
It
is
a
framework
for
configuring.
In
selecting
protocols
for
multi
axis
use
cases,
it
has
a
control
point
which
is
application
level
control
point,
and
then
it
it
has
names
of
aggregate
in
user
points.
J
The
control
plane
itself
is
is
agnostic
over
the
use,
tunneling
protocol
so
use
of
protocols
like
multi-party,
see
it
that
it
does.
It
discovers
and
selects
networks
of
proxies
configures
the
baths
independently
for
uplink
and
downlink,
and
is
able
to
adapt
the
scheduling
based
on
network
conditions
using
the
partners
amongst
communities
it
has.
It
is
not
rooms.
Concept
is
under
discussions
in
a
number
of
standard
standardization
and
Industry
forums.
The
first
first
went
in
to
a
TPP
where
there
is
a
study
on
access
traffic
steering.
J
Switching
and
splitting
is
there
part
of
solution,
one
in
well
as
protein
allowance.
It
is
part
of
discussions
on
unlicensed
integration
with
5g
networks
in
Mercer,
for
it
in
this
part
of
the
discussions
of
integrate
the
European
5
chief
with
Wi-Fi
in
it
see
it's
part
of
the
requirements
through
the
use
case
in
order
to
produce
cases
of
multi-axis,
and
it's
also
part
of
the
Wi-Fi
allows
discussions
of
Wi-Fi
cellular
integration.
J
Applications
that
work
using
a
special
plan
where
we
have
network
for
Nixon
many
shell,
which
is
sitting
in
the
network.
It
doesn't
need
to
be
a
purchased
network
with
them
in
the
enterprise
network,
an
element
that
that
is
the
topology
and
the
connections
and
then,
of
course,
the
counterpart,
the
client
connection
manager,
which
typically
seats
in
the
lancet
most
and
then
it
also
the
proxies
that
that
that
are
configured
according
to
the
messaging
to
exchange
over
the
control
and
in
the
host
side.
We,
of
course,
of
this
is.
J
Missing
in
Caesar
WebSocket
waste
in
JSON
code,
ETLs,
say
cute.
The
user
plane
has
means
of
aggregating
multiple
different
tunneling
protocols,
multiple
TCP
in
the
case
of
multiple
multiple
TCP
protocol
level,
to
the
PC,
also
so
spirit
regimes,
multiple
TCP,
the
use
cases
that
are
benefiting
from
homes.
Of
course,
the
integration
of
LT
+
R
alpha
with
enterprises
public
venues
because
of
when
2.10
can
be
attained.
Enterprise
side.
J
It
is
beneficial
for
introduction,
hurts
presence
noses
in
five
shape
deployments
and
then,
if
you
see
something
for
reducing
for
existing
fixed
assets
and
joint
traffic
steering
across
multiple
networks,
so
this
was
just
mean
frontal
community.
That
there
is
a
cost
of
work,
is
ongoing
and
if
you
are
interested,
please
comment.