►
From YouTube: IETF106-MPTCP-20191119-1710
Description
MPTCP meeting session at IETF106
2019/11/19 1710
https://datatracker.ietf.org/meeting/106/proceedings/
A
A
A
Okay
and
no
takers,
thanks
for
crystal
footprint
area
note-taker
and
thanks
for
media,
take
care
of
each
other
and
just
reminder
when
you
stick
up
at
the
mic.
Please
say
your
name
before
you
start
to
making
comments.
So
that's
note-taker
can
write
down
your
name
in
the
minute
and
also,
if
I
knew,
submit
your
internet
draft.
Please
make
sure,
puts
MP
DCP
in
your
title
of
your
draft,
so
that
chairs
can
track
the
status
of
your
draft
and
also
in
this
meeting,
please
make
sure
write
down
your
name
on
the
brew
seat.
A
Okay,
so
this
is
used
or
not.
Well,
I
think
you
already
get
familiar
with
this
man,
so
this
is
a
reminder
of
ITF
policy
with
regard
to
your
behavior,
your
policy
I,
your
privacy,
your
intellectual
property,
your
copyright
and
so
on.
So
if
you
have
any
concerned
about
it,
please
take
care
of
it.
Please
no
take
a
look
at
carefree
and
you
can
find
same
contents
on
the
ITF
web
page
okay.
A
Now,
let's
move
on
to
the
working
group
status
and
it
tooks
for
so
long
time,
but
finally,
we
finished
66
to
a
24
bits
draft,
so
it's
in
in
RFC
editor
queue,
so
it
will
be
published
very
soon
and
I
really
appreciate
your
support
for
this
draft
and
thank
you
so
much
and
also
we
have
mentioned
this
several
times
already.
That's
know
we
don't
have
active
working
group
item
right
now
and
just
this
is
just
a
reminder.
A
So
this
list
of
the
milestones
of
our
working
group
and
then
we
have
done
many
stuff
and
then,
in
our
past
meeting
we
discussed
about
publishing
enhanced
API
draft,
but
our
consensus
was
that
we
don't
have
enough
energy
to
publish
this
draft.
So
we
skip
this
one
and
then
rest
of
the
item
was
MP.
Tcp
standard
track
protocol-
and
this
is
just
finished
right
now
and
as
a
result,
don't
have
a
Viking
to
fight
them
anymore
and
then,
as
you
know,
I
we
mentioned
several
times
already.
A
We
cannot
run
working
group
meeting
without
having
a
working
group
items
and
then
so
right
now
turned
80
continuing
disgusting
about
this
one.
If
we
cannot
hover
enough,
you
know
working
group
items
very
soon.
We
need
to
think
about
crossing
working
group.
I
know
we
still
have
a
summer
energy
in
this
community
because
you
know
every
time
we
have
meeting.
We
have
some
interesting
presentation
just
like
today.
Today
we
have
three
interesting
presentation,
but
in
order
to
keep
the
working
group,
we
have
to
have
a
some
working
item.
A
That's
create
standard,
that's
what
we
need
to.
If
we
want
to
maintain
the
working
group,
that's
what
we
need
and
then
we
do
not
make
any
decision
in
this
meeting.
But
we
still
you
know
discussing
and
then
I
think
you
know
we
might
make
some
decision
before
Don
Cuba
meeting.
That
means
we
might
cross
a
meeting
across
the
working
group
before
the
bank
Cuba
meeting.
If
we
don't
have
enough
energy.
So
if
you
have
some
items
to
work
on,
you
should
encourage
you
to.
A
You
know,
make
it
hurry,
submit
a
draft
and
then
announce
a
mailing
list
and
encourage
people,
read
the
draft
and
the
suppose,
a
draft
and
then
that
way,
I
think
we
can
think
about
know.
We
still
have
a
summer
energy
to
keep
the
working
group,
but
just
in
case,
even
if
we
close
the
working
group,
this
does
no
mean
IITs
do
not
work
when
PTC
P,
in
this
case,
most
likely
TC
p.m.
will
be
the
venue
for
the
NPDC
P
topics.
A
Christoph,
you
want
to
say
something:
okay,
I
see,
okay,
so
most
likely,
but
it
has
not
been
decided
yet,
but
most
likely
TC
p.m.
will
be
the
venue
for
the
MV
TCP
topic.
That's
a
case.
If
we
close
the
meeting
closed
the
working
group,
so
that's
the
status
of
this
working
group.
Any
comment,
question
suggestion
or
addition:
it's
okay
media.
You
want
to
say
something
just.
B
C
So
terms
of
implementation
updates,
we
released
iOS
13
in
September
and
if
you
have
seen
the
WWDC
presentation
in
June,
we
are
now
using
MPT
SP
for
Apple
music
to
stream
the
music
to
the
devices,
and
we
see
a
great
reduction
in
all
your
stalls.
Thanks
to
this,
we
are
also
now
using
it
for
Apple
Maps
when
people
are
searching
and
for
getting
directions
and
we
we
it
allows
us
to
handle
the
Wi-Fi
to
sell
transitions,
so
that
means
at
least
inside
Apple.
We
have
Siri
music
and
maps
well,
first
party
users
of
MP
TCP.
C
Besides
that
in
the
linux
kernel,
there
is
now
a
community
working
on
up
streaming,
amputees
p
to
the
official
linux
kernel.
This
community
is
being
driven
by
Red,
Hat,
Intel
and
Tessa
res,
and
we
are
making
great
progress
and
we
hope
that
in
the
coming
months
we
have
the
first
part
of
MPGs,
be
in
the
upstream
Linux
kernel.
C
A
C
That's
right,
the
the
multipath
TCP
upstreaming
is
aiming
for
68
24
bits.
Our
implementation
currently
is
the
old
standard.
The
actually
one
point
here
is
as
Red
Hat
into
an
ant.
Srs
are
implementing
MPSP
from
scratch
here
in
the
Linux
kernel
now
and
because
they
are
not
usually
used
to
coming
to
the
ITF.
They
are
actually
reading
the
IETF
RC
as
it
is,
and
sometimes
are
finding
some
issues
that
are
a
little
bit
confusing,
and
so
they
have
actually
some
feedback
to
provide
to
the
base.
C
B
B
D
C
About
it,
I
don't
have
a
timeline
for
that
we
will
have
to
see.
One
thing
is
that
the
3gpp
ATSs
standardization
of
the
MB
TCP
proxy
uses
the
best
version
of
the
draft.
So,
ultimately,
if
we
will
want
to
support
that,
if
you
will
want
and
will
have
to
implement
the
best
version,
okay
I
think
more
magicians
will
have
to
respond.
Okay.
Thank
you.
E
Marcus
I'm
n
from
Deutsche
Telekom,
so
I
want
to
give
an
update
to
our
draft
about
robot
session
establishment
for
multiple
CCP.
It's
not
the
first
time
that
we
have
an
additional
contributor,
it's
Jia
from
who
or
why
and
she's
also
here
today.
So
in
case,
if
there
are
later
any
questions,
she
can
answer
as
well.
E
E
It
was
mainly
written
by
GL
and
now
be
merged
into
the
prophecy
establishment
because
it
had
had
more
or
less
the
same
goals
and
we
renamed
the
original
ro
be
extended
solution
from
the
first
draft
version
now
to
Roby
Eason
enhanced
sin,
what
it
means
we
will
see
in
the
next
slides
we
adding
two
new
solutions:
Roby
it's
for
in
this
report,
selection
and
Roby
timer.
The
initial
path
selection
solution
is
coming
from
the
merger.
E
With
the
initial
from
from
the
Metro,
be
the
initial
plan
selection
traffic,
we
have
now
an
approved
document
structure,
which
means
the
different
solutions
for
robust
establishment.
We
distinguish
between
solutions
which
require
multiple
TCP
protocol
adaption
or
does
not
require,
and
also
we
started
to
define
for
the
concept.
E
I
will
explain
this
further
in
slides,
just
as
a
short
recap:
what
a
service
session
establishment
about
we
stand
at
multi
pass,
TCP
and
RC
6
8,
2,
4
and
even
the
Biss
are
pretty
much
relying
on
a
functional
path
or
in
for
the
initial
session
establishment
of
the.
If
the
path
selected
for
the
initial
multi
pass
TCP
session
establishment
does
not
work,
no
communication
can
happen
even
if
there
are
other
working
paths
or
links
available
and
with
robust
establishment
with
the
robust
establishment
traffic.
E
E
Then
we
have
the
Aerobie
sim
that
was
already
part
of
the
first
version
of
the
traffic
that
makes
use
of
similar
changes.
Establishments
are
using
any
path
available
for
the
session
set
up
so
that
we
can
see
in
the
in
the
figure
here
as
well.
So
we
start
a
soon
request
on
on
a1
and
a2
simultaneously
and
the
the
request,
which
was
acknowledged
first,
will
be
then
the
elected
at
the
initial
path
and
the
other
one
or
once
our
resetted
afterward,
with
the
tcp
rst.
E
E
If
this
is
similar
but
yeah
as
I
said
it,
it
tries
to
to
learn
overt
over
time,
which
path
has
the
highest
probability
to
succeed,
and
we
have
to
two
examples:
how
it
could
work
if
I
start
on
the
right
figure,
the
figure
six
it
will.
Whenever
an
application
requests
multipass
TCP
session,
it
will
try
to
make
a
decision
which
path
could
be
the
initial
the
route
for
the
initial
connection
set
up.
If
it
is
the
first
connection,
don't
know
heuristic
available,
then
it
could
use
the
default
route
of
the
operation
system.
E
Otherwise
it
will
use
the
information
it
has
available
and
select
the
road
for
the
initial
path,
establishment
on
the
left
side.
It's
a
little
bit
simpler,
though
there's
no
decision.
If
if
a
heuristic
is
available
or
not
so
it
will
always
use
the
initial
part
selection
logic
to
select
the
proper
path.
But
that
also
means-
maybe
it
fails
when
not
enough
information
are
available,
the
that
what
a
former
Roby
xtour
for
for
extended
in
the
first
version
of
the
draft
and
that
enhance
the
sim
solution.
E
As
I
explained
for
sin
we
simultaneously
use,
or
we
make
simultaneously
use
of
all
paths
which
are
available
to
establish
a
multiple
TCP
session
that
will
guarantee
that
a
multipass
TCP
session
becomes
is
established
as
long
and
there
is
one
part
of
functional
in
the
system
about
the
destination
and
with
the
sim
solution.
So
without
the
e
this
means
we
have
a
lot
of
overhead
because
we
reset
all
paths,
but
one
linear,
obe
sim
we
make
use
of
all
the
is
all
all
the
simultaneous
establishment
and
me
the
MP
joint
cap
option.
E
E
E
E
E
E
It's
the
same
now
for
the
row:
B
IPS,
including
the
server
RTT
functionality.
So
we
could
make
use
of
the
RTT
information
available
by
congestion
control
and
to
make
a
more
sophisticated
decision
about
which
path
could
guarantee
an
initial
path
establishment,
and
that
is
exactly
what
what
we
are
doing
here
but
there's
the
case.
Imagine
that
the
IPS
is
working
in
a
mobile
phone
and
you
have
only
downstream
traffic
of
you,
so
your
transfer
traffic
from
the
Internet
to
about
the
mobile
phone.
E
Then
you
have
this
information
only
available
on
sender
side,
so
not
in
the
mobile
phone
itself.
It's
on
the
remote
host,
and
for
that
we
propose
the
ltte
an
option
to
be
defined
which
can
carry
the
RTT
information
from
the
remote
host
to
the
mobile
phone,
for
example,
though,
that
can
be
used
in
the
IPS
logic.
E
So
in
case
the
IPS
has
not
enough
information
for
selecting
a
proper
path
in
the
beginning,
so
there's
there's
not
enough
eristic
and
then
it
can
use
the
Roby
sim
in
the
beginning
and
afterwards,
then,
if
enough
information
is
available
in
the
system
or
in
the
IPS
logic,
then
it
can
fall
back
to
pure
IPS
and
there
is
a
combination
of
Roby,
timer
and
and
Roby
IOPS.
So
the
same,
if
not
enough,
a
heuristic
is
available.
E
Yes,
so
there
are
the
three
options
we
propose
currently
in
the
traffic,
so
that
is
the
in
the
middle
of
the
MP
joint
CAD.
What
we
need
for
the
ECM,
a
solution,
it's
on
the
top.
The
is
an
en
option
to
make
explicit
a
negotiation
for
the
ECM
solution.
We
are
not
sure
if
this
is
required,
but
if
explicit
negotiation
is
wished
and
then
we
then
we
needed
and
on
the
bottom
we
have
the
RTT
en
so
in
case
we
want
to
have
a
TT
information
from
the
remote
host
for
some
heuristic
functionality.
E
E
E
So
in
case
we
have
packet.
Loss
on
on
the
initial
part
or
part
of
the
hench
I
cannot
be
transmitted
then
for
multiple
TCP.
This
at
least
means
we
have
a
delayed
connection
for
the
Robie
timer.
This
depends,
then,
on
how
long
it
requires
to
to
trigger
a
next
try
on
a
different
path
for
the
Roby
sim.
It
has
no
impact
because
we
simultaneously
try
to
establish
a
connection
on
on
any
path.
E
So
at
least
if
there
is
one
part
available
without
packet
loss
in
the
system,
it
has
no
impact
for
Robbie
E,
sim,
it's
absolutely
the
same,
and
for
IPS
it
means
a
delayed
connection.
If
the
selected
path
by
IPS
is
affected
by
packet
loss,
if
the
initial
path
is
broken,
so
that
is
a
control
then
for
multiple
TCP.
That
means
there
is
no
connection
possible
for
timer.
E
It
is
then,
depending
on
how
long
it
it
needs
to
to
get
the
timer
expired
and
to
get
a
trigger
on
and
to
try
it
on
another
part
again
for
the
yoruba
sim.
It
has
no
impact
for
the
Eastern
as
well
and
for
IPS
it's
the
same.
There
is
no
connection
possible
now
in
the
third
row.
We
consider
what
it
means
in
terms
of
how
long
it
requires
to
establish
an
initial
path
so
for
multiple
TCP,
it's
depending
on
the
default
route
characteristics
for
the
Robie
timer.
E
It
depends
on
the
default
route
plus
possible,
tries
on
other
paths
for
OB
sim,
it's
depending
on
the
fast
part
and
the
in
the
system
for
robbery
same
the
same
and
for
IPS,
depending
on
the
selected
part,
which
could
be
the
default
route,
but
could
also
be
something
else
in
the
fourth
row.
Then
we
have
the
same
consideration
but
for
multipath
session
set
up
so
it
that
means
at
least
it
needs
a
second
path
in
vs.
system
to
provide
multi
connectivity
or
a
multi
path
usage.
E
So
for
multiple
CCP
that
is,
depending
on
the
multi
path,
capital
handshake
for
the
initial
path,
establishment,
plus
the
MP
join
process
for
subsequent
path.
Here,
OB
timer,
it
depends
on
how
many
initial
pass
establishment
it
requires,
plus
the
following
am
patron
process
for
OB
sim,
it's
only
depending
on
the
handshake
of
the
fastest
path
in
the
system
plus
a
later
MP
join
your
OBE
sim.
It's
a
little
bit
different
there.
We
can
reach
the
highest
gain
because
we
can
there
make
use
of
all
the
pathway
or
all
all
the
handshakes
we
simultaneously
started.
E
So
we
don't
have
an
empty
joint
anymore
and
we
have
the
fastest
multi
pass
session,
set
up
over
all
the
solutions
in
the
Ruby
ECM
and
for
IPS.
It's
also
the
standard
anti
capable
process
plus
and
each
one
and
the
most
important
thing
now
in
the
last
row,
which
of
this
solution,
guarantees
at
least
session
setup
so
for
multi
pass
DCP
it's
depending
on
the
default
route,
Robey
timer,
sim
and
Isom
can
guarantee
session
setup
and
for
IPS
it's
depends
on
the
selection.
E
A
E
A
Then
I
think
this
is
useful
idea
because
know
you
can
establish
an
PTC
system
very
quickly
without
wasting
any
resources,
yeah
yeah,
and
so
my
question
is
know.
If
someone
wants
to
support
this
one
and
then
especially,
if
you
want
to
implement
this
one,
that
would
be
great.
So
anybody
have
some
interests
in
reading
draft
a
baby
in
draft.
Why
implement
draft?
Could
you
speak
up
a
little
bit.
A
E
A
E
So
if
you
check
my
presentation
from
the
IETF
104
there,
we
have
some
experiments
presented
and
there's
also
a
paper
out
where
we
did
several
tests
on
what
can
be
gained
and
there
there
are
huge
gains
you
can.
You
can
reach,
maybe
they're
the
only
question:
how
often
it
it
happens
or
you
need
it
in
the
in
reality.
B
E
A
A
F
Hello,
everyone
during
last
IGF
and
more
specifically
during
the
social
event,
I,
had
a
chat
with
Phil
on
the
problem
we
had
in
some
MP
TCP
deployments
and
performance.
So
if
thought
it
was
interesting,
so
here
are
our
issues.
What
we
tried
to
do
some
evaluation
of
performance-
and
we
don't
understand
everything
we
measure
just
to
warn
you
that
there
are
some
strange
thing
happening
and
Wow,
let's
go
through
it.
First
foundation
is
in
two
parts.
First,
one
is
MP
TCP
on
Geo
satellite
axis
and
the
other
one
is
on
low
Earth
orbit
satellites.
F
So
for
those
who
are
not
familiar
with
geo
satellites,
they
are
mostly
at
36,000
kilometers,
and
so
we
have
600
milliseconds
OTT
and
because
of
that,
we
usually
deploy
TCP
proxies.
So
what
TCP
proxies
do?
That
is
an
example
of
what
you
can
do.
They
basically
intercept
the
scene,
then
the
siak
to
the
asked.
So
we
have
a
quick
establishment
of
the
connection
between
the
US
and
the
first
proxy.
Then
we
have
a
specific
protocol
between
the
two
proxies
and
then
a
third
part
of
TCP
connection.
F
So
when
we
go
in
how
MP,
TCP
and
shakes
and
connection
establishments
work,
this
is
an
example.
We
follow
throughout
the
presentation.
We
have
a
stay
with
two
addresses
and
Osby,
so
we
have
the
connection
that
is
been
done
on
whatever
LTE
or
fixed
network,
and
we
have
the
exchange
of
MP
MP
capable
data
between
us
a
and
OS
b.
But
the
main
problem
is:
if
we
try
to
open
this
up
flow
on
the
satellite
link,
if
we
just
take
our
open
source
proxy
or
even
a
commercial
proxy,
if
you
don't
adapt
it.
F
Basically,
when
you
receive
a
scene,
please
press
n,
an
MP
join
it
just
forward
the
scene,
that's
what
we
measured.
So
the
problem
is
because
it
sends
a
syn
ACK
to
the
second
address.
The
MP
disappea
connection
cannot
happen,
so
we
have
done
some
modifications
to
our
open
source
proxy
to
be
able
to
establish
an
participe
connections.
F
So
what
we
do
is,
instead
of
finding
the
I
hope
it's
clear,
but
when
the
proxy
receive
the
thin
and
the
MP
join
instead
of
just
sending
the
thin
act
directly
through
the
address
we
just
hold
on
and
we
forward
the
scene
and
m-pin
join
with
the
token
through
OSB.
Once
we
receive
the
Finnick
with
MP
join
form
the
OSB,
we
just
wait
again,
then
we
all
lack
from
the
pepto
source
B.
F
Then
we
then
the
thin
AK
and
the
MP
join
with
everything
from
auth
be
through
address
a
and
then
one
we
want
to
receive
the
enjoy
moment
with
a
we
can
for
that
to
us
me
and
we
have
MP
TCP
connection
established,
even
though
we
have
a
disappear
poxy.
That
was
the
first
step.
The
second
step
was
stranger.
F
If,
basically,
our
TCP
proxy
acknowledges
everything
at
MPT
CP
level,
though
the
problem
that
we
had
is
that
we
realized
that
if,
in
some
experiments,
we
had
the
loss
on
the
first
path,
and
so
what
happened
if,
in
this
case,
for
a
number,
we
lose
packet
2
and
with
our
open
source
experiment,
and
we
just
had
the
peps.
That
was
acknowledging
the
packet
5
asking
for
the
fifths
packet
and
because
us
a
receives
this
acknowledgement,
packet
2
is
removed
for
the
same
buffer.
F
Also.
There
is
something
in
that's
something
that
we
have
also
seen
if,
basically
on
the
area
v4
MP
TCP,
it
mentioned
that
we
need
to
have
at
MP
TCP
level,
one
one
unique
receive
buffer
and
the
problem
is
we
have
here
one
with
each
buffer
at
the
proxy
level
and
when
receive
buffer
at
the
Osby
level.
So
we
may.
We
have
sometimes
encounter
also
flow
control
issues,
so.
F
We
were
not
happy
with
the
results,
but
we
don't
like
to
have
proxies.
We
usually
need
them,
and
so
we
also
made
in
experiments
without
specific
accelerations.
So
we
just
have
this
pep
style,
which
is
an
open
source,
a
TCP
proxy.
We
have
just
used
an
MP
tcp
version
with
on
this
under
the
stack
and
we
have
huge
asymmetry,
because
we
have
66
hundred
milliseconds
on
saturate
thing
with
download
and
uploads
and
characteristics
that
are
close
to
what
we
have
in
public
accesses
for
internet
access
with
satellite
and
on
the
terrorist
er
link.
F
We
have
an
oddity
that
is
120
and
8
make
I
put
in
the
load,
so
we
have
done
some
downloads.
You
can
feel
the
bottom
left,
so
that
is
a
down
landing
time
at
the
function
of
the
5
size
and
4
MP
DCP,
using
both
links
and
participating.
Only
the
LT
link
or
the
emulated
somehow
tourist
holding
and
the
satellite
link.
F
Good
thing
is
even
if
we
can't
accelerate
these
MP
DCP.
If
we
just
that
it
goes
to
that
rate
networks,
without
acceleration,
we
have
gains,
despite
the
really
large
asymmetry
which
we
are
just
using.
The
basic
and
minimum
mataji
scheduler,
so
that
is
a
good
thing
for
Earth.
We
are
happy
to
have
this
kind
of
results,
because
we
see
lots
of
trends
in
using
MP
TCP
to
do
aggregation
between
it
with
different
networks.
That
was
the
first
part
of
my
presentation.
F
The
second
part
of
my
presentation
is
speaking
about
using
MP
TCP
to
link
aggregation
some
in
some
cases
we
have
the
huge
trend
in
low-earth
orbit
satellite
constellations.
You
may
have
heard
of
the
basics
one
or
one
web
or
Telesat
one.
What
are
many
around
we,
and
so
sometimes
there
is
interesting
doing
layer
to
irritation
or,
but
we
were
thinking
that
may
be
doing
that
at
MPT
CP
level,
which
would
make
things
life
easier
at
the
physical
and
link
layers.
So
we
have
done
some
measurements
to
see
how
MP
DCP
behaves
in
these
use
cases.
F
We
also
have
experiments
on
how
this
behaves
if
we
want
to
use
MP
TCP
for
end
overs,
but
I
will
present
the
case
where
we
have
basically
client
that
is
behind
the
box
with
MP
TCP.
We
have
two
different
satellites
available
and
we
have
them
participate
box
and
the
other
side's
and
server.
At
the
other
side,
we
have
a
variable
delay.
One
way,
delay
that
is
close
to
basically,
if
the
delay
is
moving
is
because
the
satellite
is
going
for
the
controller
and
going
further
away,
and
we
neglect
the
the
impact
of
end
overs.
F
Basically,
we
consider
that
we
always
have
one
satellite
that
is
going
closer
to
us
and
one
that
is
going
away,
and
so
that
is
a
simple
model
and
that,
but
it
still
somehow
close
to
one
of
the
constellation
models,
but
the
we,
even
though
we
either
jump
today,
you
have
only
couple
of
kilobits
per
second
successes.
With
these
huge
mega
constellations,
we
expect
to
have
more
decent
capacity.
F
What
we
have
done
is
comparing
MP
TCP
that
is
able
to
grab
to
have
8
megabits
on
both
links.
We
compare
that
with
a
single
pass
TCP
using
only
8
make
link,
but
also
to
have
a
fair
comparison.
We
compare
that
with
them
with
a
single
TCP
that
actually
have
the
double
capacity,
because
otherwise
we
often
see
in
mp
TCP
comparisons
and
fair
comparisons,
because
with
a
better
performance,
but
that
also
just
because
we
have
no
capacity.
F
F
F
However,
when
we
moved
to
more
complexed
pages,
when
we
have
lots
of
redirections,
so
that's
the
part
where
I
don't
exactly
understand
what's
going
on,
because
we
see
that
fan
on
port
for
the
page
Lulla
per
page
blue
moon,
even
though
we
have
few
red
directions,
we
don't
see
much
differences
between
the
time
needed
to
download
the
page.
However,
for
the
pages
that
are
very
complicated,
we
thought
that
MP
TCP
was
not
doing
as
good
as
TCP
with
16
megabytes.
F
We
think
that
that
may
be
because
MP
TCP
opens
then
lots
of
sub
flows,
and
that
may
not
be
as
interesting
as
having
one
less
flow.
We
don't
know
if
it's
client
limits
or
if
it's
MP
TCP,
that
opens
too
many
things
that
still
something
that
we
are
investigating,
but
anyway
we
we
feel
that
as
an
easy
way
to
handle
end
overs
and
to
aggregate
capacity
in
the
Leo
constellation
news
case.
F
So
to
sum
up,
TCP
proxies
block
MPT
MPT
CB
traffic,
and
we
tried
that
quickly
and
adapting
TCP
proxy
is
not
that
simple
and
we
haven't
opened
all
the
Pandora
Box
we
may
encounter
when
we
want
to
do
that.
So
we
have
been
not
using
TCP
proxies
if
the
performance
are
good
and
they
are
actually
quite
good
and
also
we
realize
that
it
was
very
complicated
to
make
a
fair
comparison
with
the
diversity
of
the
web
pages
we've
seen
and
for
those
who
are
interested
in
SATCOM.
E
Mark
assignment
and
short
question
I
think
it
was
on
slide
10,
where
you
did
a
comparison
between
multiple
TCP
and
TCP.
Yes,
did
you
also
play
it
a
little
bit
softer?
Did
you
also
played
a
little
bit
with
the
buffers?
Did
you
did
you
play
it
with
the
tune?
Did
you
change
the
buffers
and
then
try
it
out
if
this
make
the
difference?
The.
E
E
A
F
A
G
A
F
A
That's
what
I
want
to
know
and
then
so
other
stuff
with
you
know,
can
you
share
the
TCP
dump
file
for
your
experiment
that
way
some
people
have
a
some
knowledge
about.
Mp
TCP
can
debug
that
traffic
and
then
from
you
might
be
able
to
identify
okay.
This
is
that
reason.
I
mean
it
could
be
a
bad
I
could
be
some
problem
of
being
big
TCP.
Okay,.
F
You
sure
we
can
I
can
share
or
I
can
share.
We
share
or
I
can
on
their
lists,
but
also
I
wanted
to.
We
thought
that
was
the
first
problems
we
were
in
country.
We
were
saying
more
and
more
problems
we
were
like.
Maybe
we
don't
need
any
acceleration
for
that
use
case
because
of
the
receive
buffer.
We
had
flow
control
issues
and
we're
like.
Oh,
maybe
we
just
don't
need
proxies
in
the
use
case,
but
I
can
show
you
the
problems.
If
that's
interesting
for
now.
H
Ok
thanks
for
cheering
thanks
for
your
coming
hello,
everyone.
My
name
is
Chi
from
Ching
hai
University
and
it's
my
first
time
to
attend
IETF.
So
please
excuse
me
for
any
anything
wrong.
Ok,
ok!
So
as
the
network
changes
and
develops,
we
see
more
and
more
information,
different
informations
that
we
could
adopt,
obtain
and
utilize
in
transportation
layer
to
improve
some
performance
like
stupid
or
say,
or
security
and
other
things.
So
today
we
are
bringing
two
simple
and
powerful
things
we
want
to
introduce
them
to
MP
TCP
and
to
improve
something.
H
So
now,
let's
start,
the
first
one
is,
is
to
introduce
prediction
and
we
have
submitted
the
draft,
the
predictable
multiparous,
tcp
extension.
So
what's
the
predictability
we
stay
here
as
the
in
the
emergent
networks,
we
see
more
and
more
predictable
nodes
are
involved,
so
by
predicable
we
mean
they
are
different
from
the
previous
nodes
as
they
have
inherent
and
thus
predictable
location
information,
some
of
which
are
at
the
fixed
locations.
Some
specific
locations
like
the
ground
stations
and
others
are
keep
moving.
H
They
are
continually
moving
at
the
expected
expected
speed
on
some
predefined
orbits
or
verifying
the
routings
on
earth
on
the
ocean
or
in
the
space
so
like
the
low
low
Earth
orbit,
satellites,
the
geostationary
satellites
and
the
air
aviation
crafts
and
the
ships
and
so
on,
they
have
predefined
routings
and
we
can
predict,
is
there
locations
and
the?
Furthermore,
the
because
the
relative
position
between
Zim
can
be
predicted.
The
state
of
the
let
work
Nick's
can
also
be
predicted
so
in
to
be
specific.
H
The
predictable
information
are
include,
but
not
limited
to
disease,
for,
like
a
some
link
handover
like
if
you
have
an
air
aircraft,
that's
connection
to
the
internet
with
the
air,
with
the
air
to
ground
state,
a
port
back,
that's
the
connecting
to
the
stations,
the
base,
the
base
stations
on
the
ground
and
the
aircraft
is
kept
moving.
It
will
hand
the
link
will
hand
over.
So
we
can
predict
as
a
time
the
handover
happens.
The
link,
bandwidth
and
delay
before
and
after
the
handover
happens
and
for
some
on
the
off
switch,
and
that
is
some.
H
H
Here
is
an
example
to
show
a
scheduler
using
prediction
information,
as
we
can
say
in
this
scenario
and
aviation
crafts
connection
to
the
internet,
with
MPT,
CP
and
Xia
simultaneously,
using
air-to-ground
broadband
and
satellite
communications,
and,
what's
more
because
of
the
complicated
topography
like
some
mountains
or
rivers,
the
base
stations
on
the
ground,
maybe
they
don't
appear
at
anywhere
on
the
ground.
So
it's
leads
choose
a
this
community
of
ATG
broadband.
That
means
the
ATG
broadband
link
may
be
intermittent
intermittent
and
therefore
defaces
equality
of
the
hoe
MP
TCP
performance.
H
So
we
want
to
introduce
the
protection
informations
and
we
have
a
modified
vision
for
a
version
of
the
MPCA.
We
call
em,
it
is
a
PP.
The
default
scheduling
principle
of
MPT
CP
is
minimum
our
TT
principle
and
we
introduce
these
these
two
prediction:
information.
That
is
the
time
the
next
to
disconnection
happens
and
the
time
the
next
reconnection
happens,
and
they
are
denoted
as
time
down
and
time
up.
So
it's
very
simple
body
modification,
the
original
RTD
was
calculated
by
the
default
method,
is
called
the
RTC
original.
H
Of
course
we
reserved
it
and
when
the
ATG
link
is
connected,
the
rttp
equals
to
RTD
original.
However,
when
the
link
is
disconnected,
it's
calculated
as
our
TTP
equals
RT
t
original,
plus
RT
T
time
up
and
time
now,
and
actually
the
rttp
will
notice
the
extra
time
that
you
use,
if
you
send
a
package
now
the
packages,
the
RTT
of
this
of
packet,
so
the
MBTA
PP
schedules,
the
packets
in
the
minimum,
Artie
T
P
principle-
so
please
pay
attention
here
is
how
we
the
impulse
mint
we
can
get
through
the
modification.
H
What's
more,
some
extension
to
TCP,
you
may
be
also
be
be
needed,
and
we
are
thinking
about
this
working
on
this.
If
you
have
any
suggestions,
please
tell
us,
and
so
what's
more,
we
are
here
to
encourage
the
community,
the
whole
community
to
find
more
ways
to
use
these
prediction
informations
for
the
emergent
predictable
network
nodes
like
if
you
find
some
new
modifications
that
you
can
make
with
prediction
information,
or
you
think
we
should
build
a
framework
to
a
to
a
battalion
and
to
use
these
prediction
information.
I
D
I
I
The
ITF
has
had
efforts
in
the
past
to
try
to
say
if
I
know
something
about
changes
to
link
state
for
some
reason
and
a
number
of
years
ago
this
was
I,
know
I
noticed
that
I've
lost
my
cell
cellular
signal
on
my
cell
phone.
You
know,
I
know
that
so
I
could
tell
the
TCP
on
my
cell
phone
and
it
could
use
that
information
in
some
way
that
hasn't
worked
in
the
past,
but
you're
not
talking
about
the
same
kind
of
TCP
that
we
were
talking
about
in
the
past.
I
You
know
you're
not
talking
about
one
path,
which
is
what
a
lot
of
the
work
that
I
was
involved
with
was
doing
so
this
might
be
different
enough
to
where
this
could
succeed.
Where
previous
attempts
in
this
space
have
not
succeeded.
I
am
in
path
aware.
Networking
research
group
I
am
draft
author
for
a
draft.
The
the
file
name
for
the
draft
is
what
not
to
do,
and
it
describes
one
of
my
own
failures
in
this
draft.
So
I
would
love
to
talk
to
you
about
that
and
about
how
this
might
be
different.
I
H
H
The
accountability
is
a
very
important
part
of
trust
for,
for
example,
the
source
of
just
validation,
that
is,
SAV
mechanisms
are
developed
to
prevent
nodes
to
spoof
other
nodes,
IP
addresses
and
to
do
something
malicious,
and
this
us
it
can.
All
the
IP
address
in
the
net
will
be
accountable
and
such
improve.
The
accountability
of
the
internet
savvy
and
the
sava
here
they
have
own
RFC's,
so
Marquez
here
is
being
faced
with
some
challenges:
cos
2
pi
D
forged
the
control
pack,
it's
sent
by
some
malicious
hosts
with
a
forge
an
IP
address.
H
H
So
here
is
what
we
have
extended.
So
here
we
define
an
IP
address
is
trusted
if
it's
protected
by
Savio
Sabha
and
if
one
on
one
sub
for
the
source
IP
address
and
the
destination
IP
address
are
both
trusted.
We
call
sub
flow,
a
trusted
one
and
then
for
the
control,
packets
and
of
MVT
they
be,
they
will
be
prone
will
they
will
be
preferentially
sent
Susie
trusted
sub
flows?
That
means
MPT
to
be
my
time
simultaneously
utilize
different
paths,
some
of
which
may
be
protected.
H
That
is
trusted,
and
some
may
be
not
so
for
the
control
packets
that
may
be
used
to
to
carry
out
some
like
security
issues.
It
can
be
protected
if
we
send
all
the
control,
packets,
the
venerable
control,
packets,
so
the
the
trusted
pass
and
if
there
is
no
trusted
pass
mbta
we
will
just
performs
as
usual.
So
this
is
what
and
for
to
be
specific,
the
details
we
have
the
first.
We
have
extended
the
original
eight
at
address
option
to
carry
a
more
flag.
H
That
is
a
trusted
address
flag
passively,
when
you
add
one
address
so,
and
secondly,
to
make
sure
that
both
parties
for
n
communication
paths
know
that
the
sap
flow
is
trusted
or
not.
We
have
proposed
a
new
option.
That
is
a
just
trust
to
notify
they
trusted
just
proactively
and
third,
we
will
record
the
this
in
trusted.
Informations
in
one
table
called
trusted
past
spending
tables
to
maintain
the
adjusted
sub
flow
state
so
that
when
we
want
to
send
a
control
package,
we
can
send
it
from
the
trusted
one.
H
So
here
is
the
trusted
adjust
notification.
We
have
extended
the
address
packet,
so
we
have
a
tea
flag.
If
you
notice
the
flag,
tea
is
to
indicate
the
whether
the
adjust
is
trusted
and
the
flag
yi
as
original
is
to
guarantee
the
reliability
and,
what's
more,
the
truncated
H
Mac
is
Chung,
is
a
Mac
H
Mac
of
address
and
the
trustor
flag,
which
can
protect
that
as
a
Chester
flag
is,
is
true,
and
this
is
how
it
was
interacted
and
the
second.
This
is
a
chested
connection
notification.
H
A
H
Actually,
for
the
for
the
trusted
one,
we
do
have
some
a
priori
intention
and
we
have
a
paper
that
is
not
paying
and
being
published
and
for
backups.
This
is
a
framework
that
we
have
implemented
implemented
Church.
We
call
it
a
trusted,
MPT
CP
and
we
have
three
new
objects,
like
trust:
either
just
notify
to
mark
and
the
chassis
the
pass
choose
and
add
into
the
original
MP
TCP
architecture,
and
we
have
also
evaluated
it
and
it
works,
and
it
it
will.
H
You
know
the
CIM
civil
uses
of
it
with
slower
than
the
regular,
a
regular
MP
TCP.
So
do
you
have
any
suggestions
for
next
step
or.
A
J
J
Sorry
wonder
if
you
believe
there
is
any
potential
benefit
to
exploring
stronger
notions
of
trust
than
merely
a
verified
address
such
as,
for
instance,
an
ipsec
encapsulating
security
payload
tunnel
that
might
have
been
built
at
layer
3
between
a
pair
of
those
source
and
destination.
Ip
addresses,
obviously
there's
additional
overhead
associated
with
that,
but
there's
also
a
greater
degree
of
trust.
J
H
Actually,
with
the
development
of
security
or
network,
the
internet
serves
for
the
applications.
So
as
last
two,
if
the
application
doesn't
know
whether
they
the
Trust's
as
is
a
pass
or
the
connection,
is
trusted,
it
cannot
do
some
selections
or
according
to
it.
So
if
we
can
do
more
know
implement
more
stronger
security
on
the
transportation
layer,
we
can
tell
the
application
layer
whether
it's
secure
or
not.
So
the
application
can
also
utilize
this
one,
and
is
that
clear
and
maybe
another.
K
H
K
G
Holy
from
to
that.
G
G
I'll,
speak.
Ok!
Ok,
ok,
thank
you
that
that's
a
crude
here
from
application
layers.
We
need
some
trans
information
from
IP
address.
If
we
have
a
true
IP,
there
is
not
forc
IP
like
some
DDoS
services
can
be
avoid
from
CDN
services.
So
this
is
why
application
need
much
informations
not
only
by
the
IPSec
and
by
other
support,
hidden
from
network
layer.
So
this
is
why
we
have
savvy
infrastructures
in
the
fact
for
China
there
we
have
sir
net
and
sir
net,
who
we
have
established
to
establish
some
savvy
structures
to
support
it.