►
From YouTube: IETF106-HOTRFC-20191117-1800
Description
HOTRFC meeting session at IETF106
2019/11/17 1800
https://datatracker.ietf.org/meeting/106/proceedings/
B
B
C
B
A
D
A
A
A
A
A
A
F
A
A
Thank
you
all
a
so
my
name
is
Erin
Falk.
This
is
the
hot
RFC
lightning
talk
session.
If
this
is
the
first
one
of
these
that
you've
been
to.
Let
me
just
explain
very
briefly
what
it
is.
The
idea
here
is
to
give
a
forum
for
folks
who've
got
something
that
they
want
to
work
on
an
opportunity
to
find
collaborators
and
to
do
it
in
a
way
that
makes
that's
sort
of
an
efficient
use
of
time.
A
Lets
us
get
through
a
lot
of
ideas,
so
we
see
things
that
are
advertisements
for
boss,
which
are
new
work
coming
into
the
IETF
or
folks.
Who've
just
got
an
idea,
might
not
be
at
the
boss
stage
or
have
a
particular
paper
or
a
topic
or
a
research
area
that
they're
looking
for
people
who
share
that
interest,
so
I
think
you're
going
to
see
a
wide
range
of
stuff
tonight
we
try
to
do
this
at
every
ITF,
and
so,
if
you're
inspired
next
time
feel
free
to
shoot
an
email.
A
The
link
here
has
a
little
bit
of
context
for
the
session
and
contact
email
address
for
me
and
so
feel
free
to
reach
out
so
I
think
that
I've
probably
spoken
long
enough
and
so
Tommy's
gonna
be
our
first
speaker.
Let's
bring
on
Tommy.
Oh
sorry,
one
format
thing
so
the
talks
are
restricted
to
last.
No
more
than
four
minutes
when
the
speaker
starts.
A
I
start
a
little
timer
when
the
timer
quacks
were
using
the
Lars
Eggert
time
management
rules,
which
is
that
when
you
hear
the
timer
quack
I
would
like
everybody
in
the
room
to
clap.
So
I'm
gonna
count
on
you
for
help.
This
is
how
we
stay
on
schedule.
Great,
get
warmed
up
very
good.
Okay,
please
welcome
Tommy
Pali,
our
first
speaker
Tommy.
You
got
the
clicker
yep.
A
G
H
Right,
your
honor
all
right,
hello,
everyone,
I'm
Tommy
from
Apple
and
we're
going
to
talk
about
DNS
privacy.
It's
maybe
a
controversial
topic,
but
I
think
it's
one,
that's
very
interesting.
So
this
is
what
DNS
traditionally
looks
like.
You
have
a
resolution
that
you
often
do
to
your
local
resolver.
You
connect
to
the
name
that
you
get
from
that
some
people
are
concerned
that
the
local
resolver
may
be
seeing
things
and
be
able
to
profile
you
in
ways
that
you
don't
want.
If
you
don't
trust
it.
H
H
So
we
have
a
proposal
that
we're
calling
adaptive
DNS
privacy
to
try
to
have
more
different
resolvers
in
the
mix.
We
want
to
be
able
to
discover
many
decentralized,
OU
servers.
We
want
to
be
able
to
designate
doe
servers
for
given
domains
so
that
you
can
say
what
you
should
be
using
for
a
certain
set
of
names.
We
also
want
to
discover
what
the
local
policy
is,
so
that
we
can
do
the
right
things
on
the
local
network
and
then
have
a
good
algorithm
for
how
to
use
all
of
our
DNS
queries.
H
So
you
can
get
something,
looks
like
this
lots
of
possible
options
that
can
coexist
and
hopefully
make
sense
together
that
give
us
privacy
without
tying
us
to
one
option.
There
is,
of
course,
an
interesting
question
of
how
do
you
bootstrap
this
system,
especially
if
you
don't
want
to
just
use
the
local
resolver
without
just
trusting
someone
else.
Instead?
For
this,
we
have
a
second
document
which
you
may
be
interested
in.
It's
called
oblivious
doe.
It
takes
inspiration
from
the
work
on
oblivious
DNS.
H
It
allows
you
to
proxy
doe
queries
in
order
to
mask
your
client
IP
address,
so
that
no
one
DNS
server
knows
both
your
IP
address
and
your
query
content
picture
kind
of
looks
like
this.
You
have
one
more
hop
in
here,
but
it
gives
you
really
good
privacy
properties,
so
that
creates
a
whole
ecosystem
around
this,
that
we
can
hopefully
improve
DNS
privacy
for
everyone
and
have
it
be
very
scalable
and
something
that
even
the
local
networks
can
participate
in.
H
So
if
you
want
to
learn
more
about
this
or
get
involved,
we're
gonna
be
talking
at
the
ABCD
Boff.
We're
also
going
to
be
presenting
this
work
in
more
detail
about
how
the
protocol
works
at
deprive.
We
have
two
documents
here
and
we
also
have
a
github
where
we
really
welcome
any
issues
or
pull
requests
that
you
want
to
have
get
a
discussion
going.
We
want
to
solve
this
issue
for
operating
systems
in
general
and
anything
we
can
do
to
make
this
situation
better.
We
are
open
to
so
please
participate.
Thank
you.
Thank
You,
Tommy,.
A
I
The
afternoon
everyone,
my
name
is
Anil
I'm
from
Shanghai
University,
it's
America
I
to
be
here
to
share
work
with
you.
Our
work
is
deadline,
we're
transport
protocol.
As
we
all
know,
the
entire
Internet
is
becoming
real-time,
more
and
more
applications
as
deadline
requirements
for
their
data
transmission,
such
as
video
conferencing
and
cloud
vr
gaming.
The
deadline
of
this
application
can
be
divided
into
two
categories.
I
The
first
one
is:
if
the
application
is
push
based,
then
the
deadline
means
the
one-way
anytime
and
if
the
application
support
based,
then
the
deadline
means
the
rtt
and
those
applications
are
transferring
data
in
block
affection,
like
the
frame
in
the
video
conferencing.
The
block
is
the
minimal,
minimal
processing
units
of
application,
and
the
deadline
is
also
means
the
block
completion
time
this
these
applications
are
early
generating
and
the
transference
transferring
multiple
multiple
blocks,
concurrently,
so
different
blocks
of
different
importance
to
the
user.
I
Experience
like
in
some
scenario
of
video
conferencing,
the
audio
data
may
be
more
important
than
the
video
data.
However,
existing
transport
protocol
like
support
for
these
transmission
requirements,
even
though
such
some
researchers
has
proposed
the
deadline
concept
or
in
various
protocols
like
a
CDP
and
some
modified
a
TCP,
but
they
have
their
problems.
The
first
first
one
is
that
they
are
hard
to
deploy.
Second,
they
are,
they
have
no
optimization
for
the
deadline
transmission.
They
may
be
just
drop
some
stale
data
and
the
third
one
is
that
they
like
support
for,
pool
based
transmission.
I
As
a
result,
applications
are
forced
to
to
their
own
customized
the
solutions
like
specified
for
video
conferencing.
These
solutions
are
in
cross-layer
fashions
and
these
efforts
are
very
complex
and
redundant.
So
we
need
a
transport
protocol
to
provide
a
deliver
before
deadlines
device.
Our
solution
DTP
spewed
on
top
of
quick
and
is
very
easy
to
deploy.
It
can
rebuild
EDP
both
in
transport,
layer
and
HTTP
layer
and
way
TTP
provides
block
based
the
data
transmission
instead
of
quick
stream.
I
I
The
EDP
will
schedule
block
transmission
order
based
on
the
deadline,
priority
and
the
natural
convention,
and
it
will
also
drop
some
low
priority
or
still
blocks
if
necessary,
and
we
also
apply
redundancy
for
some
kill
packets,
of
a
block
to
about
the
transmission
delay.
The
congestion
control
is
responsible
for
controlling
the
RTT
below
the
deadlines
and
about
we
can
talk
about
the
implementation
of
DDP
later,
maybe
after
just
how
to
habit.
I
I
have
seen
if
you
are
interested-
and
then
you
can
also
refer
to
our
draft
here
is
some
part
of
our
evaluation
results
that
we
can
talk
about
it
later.
If
you
are
interested
and
we
are
looking
for
some
co-operators
to
develop
application
based
on
the
DDP
and
do
some
for
the
job.
If
you
are
interested,
please
catch
me
during
this
week
or
you
know
me
at
least
advice,
and
you
can
learn
more
about
our.
J
J
Now
the
particular
draft
that
I'm
working
on
is
trying
to
describe
machine,
readable
format
for
specifying
the
syntax
of
protocol
data
units
and
for
the
most
part,
that
looks
very
much
the
same
as
the
way
we're
specifying
the
format
for
binary
protocols
at
the
moment.
So
we've
got
a
packet
header
diagram,
followed
by
a
sort
of
description
list
of
each
field.
That's
in
that
packet,
header
diagram
and
essentially
what
the
drafters
trying
to
do
is
to
bring
some
consistency
to
this.
J
So
if
we
all
use
the
same
format,
then
we
can
develop
some
tooling
that
can
extract
these
diagrams
from
the
draft
and
then
we
can
start
to
do
some
interesting
things
with
that.
So,
first
of
all,
we
can
do
some
sort
of
simple
checks,
a
simple
but
effective
check.
So
we
can
check
that
the
diagram
matches
the
description
list
and
quite
often
it
doesn't
and
perhaps
we're
interested
in
whether
we
can
maybe
generate
some
partial
code
from
the
definition
of
the
protocol
itself.
J
No,
the
draft
that
mark
is
working
on
is
quite
different,
so
he's
proposing
that
you
write
your
internet
drafts
using
asciidoc,
which
is
a
markdown
variant
and
the
same
file
as
that
as
your
internet
draft.
You
then
add
a
formal
description
of
the
protocol
written
in
idris
so
with
dependent
and
linear
types.
As
I
say,
this
is
Mark's
draft.
So
if
you
have
questions
about
that,
you
should
ask
him,
but
what
that
then
gives
you,
then,
is
the
ability
to
generate
an
XML,
RFC
v3
file.
That
has
examples
that
are
correct
by
construction.
J
You
have
definitions
and
formal
languages
like
a
BNF
that
have
been
verified
and
you
get
a
formal
proof
of
that
application
of
pastels
law
and
much
more
now,
we're
aware
that
there's
a
lot
of
other
languages,
other
formal
languages
that
are
used
in
the
ITF
things
like
yang
or
ABN,
F
or
C,
D,
DL,
and
so
beyond.
These
sort
of
narrower
projects
and
drafts
that
we've
got.
We
want
to
try
and
bring
about
a
broader
discussion
about
the
use
of
formal
languages
in
the
ITF.
J
To
do
that,
we're
going
to
have
an
informal
side
meeting
we're
going
to
have
that
on
Thursday
night.
It
is
a
clock
at
the
winery
at
James,
that's
just
across
the
road
from
the
venue
you
could
downstairs
into
the
courtyard.
If
you're
planning
to
come
along,
then
send
us
an
email
and
we'll
make
sure
we
don't
miss
anyone
and
if
you're
interested
in
the
work,
but
you
can't
come
along
again,
send
us
an
email,
I'm
printers
in
the
hallway
and
we'd
love
to
discuss
it.
Thank
you.
A
K
So
there
are
three
core
problems
that
we
address
in
the
mesh
provision.
The
private
keys
of
the
device
provide
the
means
to
obtain
the
corresponding
public
key
and
to
secure
data
at
rest.
So,
oh
so
security.
Today.
What
we
have
to
do
is
we've
got
a
whole
bunch
of
different
applications
about
SSH
PGP
and
all
that
stuff
and
they're
all
silos.
K
They're
all
law
unto
themselves
and
security
tends
to
fall
down
between
the
cracks
between
applications.
We
need
to
be
able
to
join
all
those
applications
together
into
one
security
infrastructure
and
the
mesh
is
a
proposal
for
that
infrastructure.
So
what
we
have
here
is
a
picture
of
the
mesh
in
the
large
and,
as
you
can
see,
there
are
a
lot
of
moving
parts
there.
K
This
is
a
platform
and
it's
architected
as
such,
because
you
know
I'm
a
systems
guy,
so
the
mash
mesh
itself
is
that
green
box
in
the
middle,
and
that
rests
on
three
core
technologies-
UDF,
which
is
the
naming
technology
PGP
fingerprints
on
steroids
Deir,
which
is
a
cryptographic
message,
syntax
blockchain
in
Jason,
with
encryption
and
meta
cryptography,
which
is
a
way
of
going
beyond
the
crypto
in
PGP
and
Bruce
Schneier
Bluebook
yeah.
Most
of
us
have
been
using
the
same
crypto
for
25
years.
K
They
developed
a
whole
load
of
stuff
in
the
90s
that
we
don't
use
here.
I
want
to
start
using
some
of
that
and
I'll
be
talking
about
how
we
can
do
that
in
the
buff.
So
there's
a
lot
of
stuff
here
and
a
lot
of
applications
that
could
be
built
on
top
of
it.
I've
gotten
the
mesh
for
the
purposes
of
the
buff
and
starting
a
working
group.
We're
proposing
to
focus
on
just
warm
and
the
one
I'm
proposing
that
we
start
off
with
is
providing
a
end-to-end
secure
way
of
managing
passwords.
K
So
we
provide
an
end-to-end
secure,
password
vault
on
every
machine
that
the
user
has,
that
is
their
ubiquitously,
so
they
can
start
using
long
and
strong
passwords,
because
they're
going
to
be
available
on
everything
that
they're
going
to
use.
So
we
can
get
away
from
this
problem,
and
you
know
you've
all
seen
this.
K
The
problem
here
is
the
shortest
password
that
is
secure
is
far
longer
than
any
user
could
ever
be
expected
to
learn,
and
you
cannot
get
round
that
with
any
algorithm
hack
or
whatever
I
can
go
through
the
math
of.
Why
that's
not
possible?
We've
got
to
get
away
from
passwords
and
the
mesh
provides
a
way
to
do
that.
H
A
M
L
I'm
going
to
present
my
proposal
of
extinction
to
the
mother
of
C,
so
the
problem
is
that
the
more
FC
is
not
efficient
against
movin
are
given
our
abilities.
This
is
not
exactly
a
problem
because
of
mirth
because
they
are.
This
is
not
the
proposal,
but
a
new
vulnerability.
King
can
use
a
whitelist
a
traffic
by
the
mud,
and
this
needs
a
femur
or
a
mud
update
by
the
manufacturer
in
the
problem
of
the
point
of
a
point
of
your
security.
If
that
is
that,
we
have
only
one
point
of
failure,
that
is
the
manufacturer.
L
So
distant
can
take
a
long
time
to
be
done
by
the
manufacturer.
They
may
exist.
No
more
or
or
can
just
decide,
do
not
fix
the
bug
or
solve
the
problem.
So
the
good
part
and
the
good
part
of
the
history
is
that
the
security
of
the
operation
center,
every
time
discover
and
school
affirmation
about
botnets
and
all
the
vulnerabilities,
and
we
can
use
this
information.
L
The
behavior
of
the
botnet
or
or
vulnerability
to
protect
ourselves
and
another
important
to
stick
holder
in
this
part
is
the
ISPs
that
has
their
amended
if
being
used
by
DDoS
traffic
and
other
unwanted
traffic
and,
in
the
other
part
of
the
stakeholders.
The
end
user
won't
protect
a
security,
security
and
privacy,
so
the
proposed
extension
is
chose
to
use
they
about
vulnerabilities.
In
this
case,
we
are
talking
about
botnets,
and
the
data
about
botnet
is
being
collected
by
honey
pots.
L
The
administrator
of
the
network,
ask
for
a
block
or
an
auto,
only
alert.
So
in
the
in
the
end,
the
the
blocks
are
done
and
the
vulnerabilities
are
blocking.
So
thank
you
very
much
for
the
attention.
If
you
have
any
question
or
interest
on
the
discussion,
this
is
my
mouse
and
I'm
here
by
other
week.
N
Thank
you,
I'm
Richard,
Lee
and
thanks
to
Alan
vulcanizing.
This
event
is
quite
useful,
and
so
people
are
always
smart
about
the
past.
Sometimes
they
are
also
smart
about
now,
it's
very
dangerous
to
say
something
about
the
future.
It's
a
big,
especially
for
something
as
dynamic
as
unit
and
as
hugely
scale
as
internet,
but
fortunately,
or
unfortunately,
I'm
now,
on
that
Angeles
zone
now
today
is
released.
N
So
before
I
talk
about
the
future.
Let's
take
a
look,
a
brief
look
at
the
kernel
image,
so
Indonesia
has
been
really
really
successful,
but
it's
called
hard.
It's
based
on
a
few
principles,
concepts
and
technologies.
Fundamentally,
it's
based
on
something
called
statistically
multiplexing.
So
the
major
purpose
is
to
maximally
utilize.
N
The
link
on
the
protocol
stack
party,
it's
evolving
idea
today,
that's
more
or
less
capture
their
corners,
like
a
protocol
stack
except
in
POS,
its
message,
the
being
sit
somewhere
and
internet
is
huge,
but
it's
built
by
sweet
switches
and
routers
from
these
switches
and
routers.
You
can
only
expect
no
more
than
we
can
three
types
of
capabilities.
One
is
the
effort.
That's
the
most
popular
are
mostly
used
and
default.
Second
tip
serve
especially
for
the
some
like
voice,
video
applications.
You
can
expect
no
more
than
a
the
classes
of
services.
Another
one
is
chocolate.
N
Annealing
in
current,
it's
based
on
chapter
stealing
Express
departs
and
mean
hi.
We
can
also
provide
some
guarantee,
but
not
stupid,
and
sometimes
if
your
needlework
is
broken
can
provide
some
fast
reload.
But
if
we
look
at
history
of
all
the
internet
protocols,
you
Caesar
the
mostly
use
the
particles
are
pretty
old.
It's
only
the
more
than
one
generation,
for
example,
like
ipv4,
is
still
a
like.
You
stay
mostly
used
states
already
20
38
years,
even
for
ipv6,
it's
only
24
years.
So
it's
all
generation.
N
If
you
have
a
baby,
and
now
it's
already
public
graduate
from
university
so
and
many
people,
especially
from
academia
they
are
thinking,
can
we
do
something
inside
the
network?
So
that's
about
the
canary
so
and,
as
I
said
earlier,
it's
really
predictable.
You
know
about
some
saying
the
future.
He'll
I
mean
ITT.
There
is
a
Folk
School
on
nano
work
2030.
They
have
worked
the
formal
for
one
year
or
so
they
have
identified
something
and
this
triad.
N
To
summarize
it,
the
first
one
is
that
very
large
volume
and
tiny
instant
communications,
for
example,
holographic
fiber
communications,
and
you
may
or
may
not
know
that
right
now
we
are
starting
to
deploy
5g
one
application
is
air
via
a
RV
is
next
to
radio
communication,
but
what
we
will
be
after
a
aria,
so
people
tend
to
agree.
The
holographic
type
of
communications
will
happen
sooner
or
later,
and,
for
example,
we
could
have
spotted
something
called
a
holographic
teleport
so
that
we
require
really
short
latency
and
after
meeting.
N
O
Good
evening,
everyone
I'm
leon
from
china
mobile
and
I'm
going
to
talk
about
computing
first
networking
today.
So
I
start
with
some
background,
so
edge
computing
very,
very
popular
nowadays
and
emerged
by
the
introduction
of
5g
cool
network
which
actually
intrinsically
supports
this
and
I'll
start
with
some
facts.
In
china
mobiles,
we
have
more
than
six
hundred
rows
of
CDN
and
those
known
can
be
upgraded
to
virtual
at
CDN,
which
have
a
common
interaction
layer
and
then
actually
can
naturally
upgrade
it
to
edge
computing
nodes.
O
So
we're
talking
about
from
the
city
level,
to
county
level
and
even
to
the
unset
level
of
the
infrastructure
layer.
We
are
going
to
see
more
than
maybe
tens
of
thousands
or
maybe
hundreds
of
thousands
knows
that
used
to
be
only
router
based
infrastructure,
but
nowadays
is
going
to
be
integrated,
router
and
then
back
to
back
with
some
IT
infrastructure
installed.
So
the
network
original
are
designed
to
optimize
connectivities
and
we
treat
cloud.
O
The
data
centers
are
users
of
the
network
users
of
our
connectivity,
but
nowadays,
if
we
have
the
district
district,
the
distributed
manner
of
edge
computing
deployed
we're
actually
talking
about
changing
the
role
of
edge
computing
or
cloud
computing
to
be
the
from
the
user
of
the
network
becoming
a
part
of
the
network.
So
seeing
the
characteristic
of
edge,
confusing
it's
actually
very
limited
in
terms
of
resource,
and
it
requires
heat
or
genius
require
resource
distributed,
because
not
all
the
edge
computing
nodes
can
have
GPU
or
other
Asics
capabilities.
O
But
some
of
application
really
require
that,
and
we
have
other
requirements
like
we
need
to
dynamically,
distribute
the
traffic
to
overcome
the
the
outcome
of
the
limited
resource
caused
by
the
edge
computing
node.
So
the
question
raised
here
is:
how
does
the
network
helped
application
to
find
the
optimized
edge
computer
note
and
how
does
Network
help
as
computing
knows
to
offload
the
traffic?
O
So
that's
what
siphon
is
designed
for?
So
it
actually
designed
to
be
able
to
distribute
computing
resource
status
across
the
network.
So
we
can
use
that
status
to
do
habit,
routing
habit,
algorithm,
considering
the
computing
status
and
the
network
status
together.
So
it
helps
at
computing
to
provide
local,
insensitive
equivalence
services
and
dynamic
traffic
computing
offloading
and
also
seamless
switchover
between
edge
nose
with
flow
affinity.
O
So
we're
going
to
talk
about
this
more
in
this
meeting
and
we
have
a
site
amazing
holes
in
the
morning
on
Thursday,
in
room
VIP
a
from
starting
from
8:30,
so
welcome
to
join
us
and
have
a
discussion,
and
there
are
three
draft
after
the
uploaded
on
the
website
and
if
you're
interested
have
a
look
and
if
you
have
any
further
information
that
you
need
further
information,
please
don't
not
be
hazarded
to
cut
me.
Thank
you
very
much.
Thank.
G
P
Hello,
Joe
I'm,
going
to
talk
about
single
slide
only
so
this
this
is
about
loops
local
optimizations
on
path
segments
loops
is
sorry.
The
the
goal
of
looms
is
to
provide
a
local
in
network
recovery
over
some
specific
segments
to
optimize
the
packet
delivery.
A
very
typical
user
scenario
is
over
the
when
there
might
be
multiple
overlay
segments
and
for
certain
segments
there
are
higher
loss
compared
to
the
others,
and
it
des
contributes
to
the
overall
loss
most.
In
that
case,
we
possibly
can
provide
a
so-called
local
in
network
recovery
by
either
retransmission
or
FEC.
P
So
we
had
above
in
last
IDF
meeting.
There
was
quite
a
strong
interest
showing
that
the
standardization
of
the
work
who
was
required
so
we
are
going
to
meet
in
in
this
meeting,
to
discuss
more
detailed
design
issues
well
include
the
encapsulations
and
the
detailed
retransmission
operations,
possibly
sketch
FEC
version
then
clearly
outlined
the
work
to
be
done
in
loops,
so
who
might
be
interested
of
course,
the
transport
protocol
designers
in
this
time,
because
we
are
going
to
attach
a
little
bit
more
on
the
encapsulation,
so
Tunnel
protocol
designers
and
also
the
FEC
experts.
P
Q
Q
He
wanted
to
provide
the
security
that,
if
it
was
compromised
and
messages
when
exchanged
in
the
past
were
actually
not
compromised
if
one
of
the
keys
was
compromised,
and
it
also
wanted
to
provide
the
property
of
deniability
in
the
sense
that
there
was
no
way
to
create
proof
of
authorship.
If
someone
managed
to
compromise
the
conversation,
but
of
course
what
he
as
I
said,
was
created
in
2004
and
since
there's
a
lot
of
cryptographic,
things
have
happened.
Q
So
we
wanted
to
obtain
of
that.
So
basically,
what
out
here
how
it
looks
like
is.
Basically
you
have
a
two
participant
communication,
as
I
said
in
which
Alice
and
Bob
want
to
talk
to
each
other,
they
would
request
to
have
an
OT
up
conversation.
That
is
further
a
process
in
which
they
authenticate
each
other
in
a
Dinello
way,
by
exchanging
the
long
term
keys
and
also
by
generating
a
shared
secret
and
by
then
the
eye
shadow
actually
to
exchange
the
message
in
we
teach
each
other
by
using
the
lover
ratchet
and
using
encryption
keys.
Q
That
will
be
used
per
message.
Only
so
no
keys
will
actually
be
compromised.
So
basically
right
now,
what's
the
state
of
oti
in
it
special
voice
that
we
finish
the
specification
in
the
sense
that
we
finish
the
cryptographic
specification.
We
are
still
doing
an
implementation
on
C
and
in
:,
and
some
people
are
actually
also
doing
this
implementation
with
Java
and
:,
and
some
people
are
interested
in
doing
of
actually
doing
an
implementation
once
this
shop.
Q
What
we
want
to
do
on
what
I
am
presenting
right
now
is
that
it
will
be
really
interesting
to
have
an
RFC.
There
has
been
some
interesting
in
the
past
to
do
an
RFC
for
previous
version
of
all
ta
v3,
but
we
think
that
is
very
important
to
actually
do
a
passion
for
of
the
RFC
about
here,
because
this
is
something
that's
Korean.
Q
Secure
messaging
applications
I
actually
based
upon
like
the
signal
protocol,
so
actually
having
an
specification
of
the
authorization
for
in
this
current
form
would
be
actually
something
very
interesting
for
the
secure
message
in
world.
If
you
want
to
learn
more,
please
check
out
of
repos
that
you
can
see
here
and
thank
you
very
much.
You
can
reach
me
on
my
Twitter
account
or
the
Twitter
account
or
vote
here.
Thank
you.
R
Hi
I'm
Stu
he's
Bob,
so
I'm
gonna
motivate
he's
going
to
talk
about
the
work
in
the
host
identity
protocol
to
try
to
get
the
job
done.
So
fundamental
problem
is
gap
between
physical
space
and
logical
space.
I
see
an
unmanned
aircraft
there
I,
don't
know
who
it
is.
I
don't
have
any
way
to
contact
the
operator.
If
some
emergency
situation
arises
and
I
need
to
contact
the
operator.
How
can
I
do
that?
So
in
the
United
States,
the
Federal
Aviation
Administration
is
expected
to
propose
a
rule
next
month.
R
There
was
some
work
done
in
ASTM
International.
Just
last
week
we
were
there.
They
released
their
first
version
of
the
standard,
it's
good
as
far
as
it
goes,
but
it
does
not
make
the
information
immediately
actionable.
It
doesn't
make
it
trustworthy.
It
doesn't
allow
me
to
do
a
one
button
press
and
get
in
contact
with
the
operator
and
so
we're
looking
at
taking
various
IETF
standardized
protocols,
leveraging
them
to
address
this
application
and
I'll
leave
the
details
to
Bob.
E
What's
needed
here
in
extremely
constrained
environment
Bluetooth
4
is
a
basic
minimum.
Bluetooth
for
broadcast
messages
is
the
mineral
basis
for
the
communication,
extremely
constrained
message,
format
extremely
constrained
content.
What
can
we
do?
We
need
to
provide
trust
with
the
identity
to
pair
with
physical
location
data.
E
It
turns
out
that
the
host
identity
tags,
which
are
valid
IP
v6
dresses,
can
be
used
over
this
broadcast
medium
boot
Bluetooth
to
promote
to
provide
provable
ownership
by
using
the
host
identity
for
the
signatures,
give
us
full
mobility
and
multihoming
when
we
need
to
communicate
directly
to
the
operator.
However,
we
find
where
that
operator
is
just
tell
them.
Please
abort
your
mission.
Now.
E
Timing
on
this
can
be
very,
very
critical
as
drones.
These
things
can
move
quickly
in
the
airspace
or
they
may
be
hovering
so
secure
registration
protocol
is
also
needed
to
be
able
to
register
the
devices
on
a
first-come
first
owned
the
ID
so
that
you
say
I
had
this
ID
for
this
device.
It
may
be
long-lived.
The
hobbyist
has
an
ID
long-lived.
A
delivery
company
may
have
the
ID
for
every
single
delivery
mission
so
that
the
people
cannot
observe
to
say.
Oh,
this
is
a
UPS
craft.
This
is
a
of
Amazon
craft
so
forth.
E
What
we're
doing
right
now
we
have
the
US
use
case
draft,
which
is
out
that's
two
authored
and
for
myself,
I
bought
through
the
hierarchal
hits
draft
as
well
as
extension
for
the
the
raishin
protocol,
as
well
as
getting
together
the
new
crypto,
we're
going
to
need
for
this
lightweight
environment
new
in
terms
of
what
we've
used,
but
it's
NIST
approved,
crypto
stuff.
It's
been
out
for
a
little
while
I'm
using
a
EDSA,
kmac
c-sharp
see
shake
k
AK,
while
that's
waiting
for
the
light
crypto
competition
to
finish
we're
looking
at
actually
signing
messages.
E
These
broadcast
messages
because
there's
no
state
these
are
messages
coming
out
on
a
broadcast.
We
had
a
hackathon
on
yesterday
and
today
we
made
some
progress,
Adams
doing
some
great
work
of
coding
and
expect
to
see
him
in
the
cold
lounge
to
continue
working
during
the
week.
While
he
can,
we
need
to
progress
these
drafts.
A
lot
more
testing,
both
here
and
coming
up,
stewing
and
and
Adam
are
located
at
the
the
Griffiths
Air
Force
Base,
or
the
old
Griffis
base,
the
UAS
test
site
for
the
FAA
or
they'll
be
doing
testing
there.
E
S
A
S
Paul
Condon
I've
been
talking
about
data
center
congestion
control,
the
last
couple
of
us
and
I'm
trying
to
find
where's
the
best
place
where
we
can
get
more
interest,
more
people,
more
review
of
our
dress
and
development
of
them.
I
showed
this
picture
last
time,
but
I
liked
it.
So
much
I
wanted
to
show
it
again:
data
center,
congest
ins,
different,
unique,
then
the
congestion
we
experienced
in
the
big
I
internet
so
there's
much
difference,
delay
bandwidth,
different
switches,
architectures,
the
networks
are
much
more
homogeneous.
S
There's
a
lot
of
high-speed
links
with
compute
and
storage,
that's
very
in
close
proximity
and
the
traffic
profiles
are
a
little
bit
more
predictable,
a
little
bit
more
understood
and
there's
typically
fewer
people
managing
these
things,
maybe
even
one
set
of
people,
so
the
congesting
management
can
be
different
in
a
data
center.
So
you
know
where?
Should
we
consider
these
things?
If
you
look
at
the
ICC
RG,
we
could
say
yep
they're,
they're
Charter
does
mention
data
centers
as
a
possible
starter.
There
they're
often
quite
full,
with
their
agenda
all
the
congestion
work.
S
S
That's
coming
up,
so
a
couple
of
the
key
questions
that
we're
trying
to
answer
this
time
around
what
are
needed
from
NYX
themselves
to
do
better
congestion
control,
there's
a
draft
on
an
open,
open
congestion,
control
framework
where
we
can
negotiate
capabilities
in
a
more
open
fashion
and
and
try
to
determine
how
the
network
and
the
NIC
can
work
together.
How
can
the
network
itself
actually
participate
in
congestion
management
in
a
different
way
or
a
more
enhanced
way?
One
thought
is
about
configuring.
S
That
parameters
are
very
complicated,
so
can
we
use
AI
in
a
way
to
help
model
that
there's
a
draft
on
that
and,
of
course,
shortcutting
the
whole
control
loop
could
could
really
reduce
some
of
the
reduce
some
of
the
delay
in
getting
congestion
and
pressure
on
some
of
our
buffers.
So
there's
a
fast
feedback
draft
as
well.
There's
some
other
interesting
topics
like
maybe
should
we
be
measuring
things
differently
in
the
data
center
as
well?
So
we
have
a
couple
of
drafts
to
review.
Discuss
join
us.
T
T
T
I'm
going
to
talk
about
the
later
people
approach
me
I,
just
gonna
give
you
some
ancient
history,
the
academic
view
of
my
topic,
which
is
something
that
God
called
RCP
route
control
processing.
What
I
mentioned
doing
automating
BGP
programmable
internet
routing
is
what
I'm
calling
it
in
the
academic
world.
We
see
that
nothing
really
has
changed
for
maybe
a
decade
I
feel
confident.
T
That's
not
the
case,
and
the
reason
I'm
here
tonight
is
to
invite
people
to
come
up
and
tell
me
what
it
is
that
we
on
the
academic
side
don't
get
so
this
is
what
we
see
so
in
2004,
someone
came
up
with
a
bright
idea
that
you
could
invent
a
a
God
box
that
would
take
control
of
an
entire
transit
network.
You'd
flip
all
of
the
bgp
connections,
all
the
AI,
all
the
internal
ibgp
mesh
onto
this
box.
And,
of
course,
if
the
box
went
down
your
network
went
down.
T
This
was
the
kind
of
academic
idea
of
how
to
make
networks
better.
Needless
to
say,
this
didn't
fly
this
next
slides
from
the
same
paper
and
I'm,
putting
that
up,
because
I'm
gonna
come
back
later
on
and
add
my
overlay
to
it
to
make
it
show
how
I
can
hopefully
make
a
network
change
to
where
we
want
it
to
be
without
it
having
to
take
it
apart.
First
from
the
same
paper
that
their
idea
was
to
stage.
T
T
I
RS
CP
same
idea
a
bit
better
they've
got
route
reflectors
in
there
and
this
one's
interesting,
because
it's
non
intrusive,
but
in
fact
that
doesn't
do
much
more
than
just
optimize
IDP
paths,
which
is
pretty
much
what
you
can
do
without
that
art
solution
when
they
want
to
do
anything
brilliant,
then
they
take
over
and
you'll
see
the
big
black
lines
are
just
physical
links.
There
data
paths,
not
control
inks,
so
that's
the
world
that
the
academics
still
live
in,
there's
been
no
real
change
in
that
there's
another
slide
here.
All
these
people
doing.
T
There's
people
like
Jennifer
Rexford
from
Princeton
and
a
whole
bunch
of
people
from
AT&T
wrote
this
stuff
twelve
fourteen
years
ago.
So
so
my
take
on
this
is
that
cannot
be
the
end
of
it.
Why
did
this
not
go
anywhere
and
so
I
kind
of
tried
to
encapsulate?
Why
that
didn't
work
and
I
think
the
problem
was
going
back
in
in
those
days
you
couldn't
get
enough
external
routing
State
without
without
breaking
into
the
ebgp
feeds.
So
the
first
question
is:
how
can
you
do
that
and
I
think
we
can
solve
that
today?
T
There's
it's
no
secret.
How
are
we
going
to
do
that
and
then
the
second
question
which
they
didn't
really
address
is
if
they
knew
that
they
could
do
something
better.
How
would
they
actually
push
that
into
the
network
without
again
breaking
the
mesh?
So
those
are
the
two
two
sides
of
it.
The
third
piece
is:
how
do
you
know
what
it
is
you
want
to
do?
But
for
me
that's
a
different
problem.
T
So
here's
the
strategy-
and
this
is
so
simple,
I
kind
of
feel
like
people-
must
be
doing
this,
but
on
the
academic
side
we
don't
see
it
so
to
get
that
external
routing
state,
you
can
either
look
at
BMP.
The
problem
BMP
is
some
implementations
seem
not
to
be
reliable,
but
some
ad
path
seems
pretty
much
up
to
the
task
and
most
of
them
the
route
is
out
there.
If
the
border
routers
can
do
that
today
and
the
second
solution
and
I've
already
heard
tonight,
people
say
yeah,
we
do
that.
T
But
but
we
don't
know
this
on
academic
side
is
just
pick
the
second
best
route
and
push
it
out
there
with
higher
preference,
not
difficult
I
thought
about
trying
to
extend
BGP,
but
that's
not
a
bright
idea,
not
unless
you
remember
the
ITF,
so
here's
my
diagram,
so
that
was
the
one
I
put
up
at
the
beginning
and
there's
the
mesh.
My
blue
path
is
me
pulling
into
my
god
box
all
of
the
additional
path
information
and
the
second
piece
is
pushing
it
back
out
again
by
just
sending
routes
with
high
local
preference.
T
T
T
A
T
A
D
Explicit
in
band
measurement
is
the
topic
of
this
short
speech.
The
first
case
of
a
special
measurement
is
the
spin
beat
spin
beat,
is
a
RTT
measurement.
It
is
implemented
on
quic
protocol.
The
spin
beat
idea
is
to
create
a
square
wave
signal
on
the
data
flow
using
a
beat
whose
length
is
equal
to
L
tt,
an
observer
in
the
middle.
Whatever
is
located
that
can
measure
the
end-to-end
ITT
only
measuring
the
line
on
the
square
wave.
D
The
new
idea
that
we
propose
is
the
round-trip
packet
loss,
measured
on
production
traffic
between
client
and
server,
how
it
works.
The
client
makes
a
train
of
production
packets
using
the
packet
loss
bit
and
these
market
packets,
bounces
between
client
and
server,
to
complete
the
two
rounds
between
client
and
server,
client
and
server
reflects
market
packets
by
marking
production,
packets
flowing
in
the
opposite
direction.
A
new
server
comes
to
the
market
packets
during
the
two
rounds
and
comparing
the
numbers
can
measure
the
feet
de
los
ISM.
D
D
Okay,
how
it
works
the
client
generate
the
Train
of
market
packets
using
the
packet
was
beat.
The
server
reflects
these
packets.
The
client
reflects
the
market
packs
again,
and
the
server
again
reflects
so.
We
completed
two
two
rounds
between
client
s
error.
Not
all
the
packets
are
market.
Maybe
you
can
see.
D
This
is
the
opposite
case.
When
the
speed
is
different
is
more
in
one
direction
towards
the
other
direction.
Okay,
the
server
the
server
is
put
in
the
middle.
He
can
only
count
the
pockets.
The
pockets
of
the
first
train
is
compared
to
the
parts
of
the
second
train,
so
the
packet
loss
is
statistically
measured.
Statistica's
because
is
the
round-trip
is
between
the
server
and
not
between
the
client
and
the
server.
But
status
is
the
main
things
the
same
things.
D
How
to
know
more,
we
have
an
academy,
quick
measurement
and
spin
damper
in
cooperation
with
Erickson.
The
draft
was
presented
will
be
presented
in
a
TS
w
meeting
Thursday
and
tomorrow
we
will
be
present
in
the
academia
P
our
with
the
spin
dump
demo.
The
mailing
list
is
possible
to
use
the
t,
SW
g
many
lists
or
dial
it
may
as
a
maraca
cheerio
fabula
europe.
If
your
color
and
ricardo
system
tomorrow.
U
To
Tommy's
first
talk
in
the
beginning,
so
in
the
IETF
IRT
F,
we
have
gotten
used
to
using
the
term
considerations
for
something
well,
that
is
actually
what
we
haven't
really
thought
enough
about
in
the
beginning,
but
actually
turned
out
to
be
very
important,
for
example
like
security,
human
rights,
perhaps-
and
so
this
is
another
say
topic
that
I
think
it's
very
important
so
well,
the
tussle
between
internally
control
points
and
the
issue
of
conciliation
and
centralization.
So
after
the
well
in
the
post,
Snowden
area,
our
response
to
the
emulation
was
okay.
U
We
have
to
rethink
our
stance
on
encryption
and
and
privacy
and
basically
made
a
decision.
Okay,
that
we
want
to
ramp
up
our
so
encryption
work
we've.
So
we
knew
before
it's
terribly
important,
but
so
we
improve
TLS.
We
are
doing
things
quick,
for
example,
so
in
also
defending
this
against
a
lot
of
criticism,
so
very
important
work.
U
On
the
other
hand,
we
know
that
in
there
are
other
problems
that
are
be
concerned,
so
so
surveillance
based
economies,
platforms
attract
users
all
across
the
web,
and
so
at
some
point
it's
even
the
case
that
the
technology
that
we
are
producing
to,
for
example,
protect
the
users
against
on
past
attacks,
where
I
could
actually
be
used
against
the
users
in
some
circumstances.
So,
for
example,
well
we
can
protect
the
say
connection
the
channel
from
on
pass
surveillance,
but
it
doesn't
keep
platforms.
U
Of
course,
from
you
know,
collecting
profile
data
trading
this
and
even
even
trading
this
was
say,
instituting
that
I
actually
on
the
pass.
So
we
all
are
aware
of
that,
and
so
we
have
started
to
react
to
this,
so
we
are
rethinking
or
in
enlarging
the
the
threat
model.
We
are
kind
of
trying
to
establish
some
principles
so
what
it
means
to
develop
technology
for
the
user.
We
are
rethinking.
Is
there
something
wrong
with
our
technology?
U
So
some
of
these
problems
they're
not
caused
by
internet
technologies
per
se,
so
that
you
have
to
understand
and
say
how
the
web
works,
how
the
economy
works
in
that
space.
So
we
talk.
A
lot
about
you
know,
consolidate
Conservation
Center
is
Asian
recently,
of
course
you
know
some
some
particular
topics
that
create
quite
some
sensation,
but
I
think
it's
really
important
that
we
try
to
take
a
step
back
and
try
to
see
the
bigger
picture.
So
what
role
are
we
playing?
U
What
role
is
our
technology
stack
playing
in
the
bigger
system
so
consider
technical
and
economic
factors?
So
one
topic
here,
for
example,
is
control
points
in
the
network.
So,
of
course
we
don't
like
the
ship.
On
the
other
hand,
we
need
some
control
points
so
right
now,
some
of
other
technologies,
I
used
in
a
ideal
way
so
DNS,
for
example,
is
used
a
lot
to
implement
control
so
there's
something
that
we
can
probably
improve
there.
Also
when
we
say
for
the
user.
So
what
does
it
actually
mean
so
from
an
application?
U
Q
A
A
V
At
the
very
end,
we've
been
working
in
a
in
an
area
called
pluggable
transports
where
we're
we're
trying
to
provide
solutions
for
people
who
are
subject
to
pervasive
monitoring
and
surveillance
and
and
censorship,
and
we've
been
working
on
this
for
quite
a
while
and
and
for
five
six
years
now
and
and
so
we're
interested
in
in
what's
coming
and
what's
going,
how
active
the
sensors
are
and
how
active
our
community
is
and
and
we're
looking
at
a
couple
new
issues
that
have
that
arise.
Now
that
much
work
has
been
done
and
things
are
constantly
changing.
V
So
here's
a
brief
update
on
the
status
of
some
of
the
pluggable
transports
that
have
been
developed
and
used
over
the
last
number
of
years.
Some
of
these
have
some
not
mentioned
here
sort
of
fallen
by
the
wayside.
These
are
the
ones
that
are
still
active
that
provide
different
sorts
of
technologies
to
avoid
evade
censorship.
V
One
of
the
biggest
problems
that
we
now
have
with
so
many
technologies
in
the
in
the
field
is
reaching
a
network
peer
that
understands
the
kind
of
communication
you're
trying
to
do,
and
so
how
users
understand
how
to
interact
with
bridges.
With
these
communications
bridges
is
a
big
problem
and
we're
working
on
user
experience
solutions
for
that.
How
you
select
and
find
the
right
bridge
at
rendezvous
with
it,
and
then,
lastly,
which
form
a
pluggable
transport.
You
need
for
the
kind
of
surveillance
regime
you're
in
this
machine
learning
stuff
is
really
brand
new.
V
It
becomes
stems
out
of
some
work,
that's
being
done
by
a
larger
group
who
monitor
how
these
technologies
are
used
in
in
the
field.
We're
sort
of
mobile
device
experts,
so
we're
interested
in
how
mobile
devices
can
can
help
us
in.
In
this
effort,
but
also
do
so
in
a
privacy-preserving
way,
we've
involved
in
a
little
hackathons
thing
today,
working
on
setting
up
an
environment
to
start
doing
more
research
in
in
in
the
user.
V
Experience
with
these
things
we'll
be
at
that
at
the
hackathon
happy
hour,
hack
demo,
happy
hour
here
tomorrow
night,
you
can
come
over
and
see
us.
Our
work
is
encapsulated
on
a
nice
website
called
pluggable
transports
info,
and
we
do
have
an
draft
out
there.
That
starts
to
talk
about
this
stuff
anyway.
Hit
me
up
if
you'd
like
to
get
involved
with
this
sort
of
thing
be
happy
to
have
people
help
us
out
I'll,
be
here
all
week,
thanks
very
much
I
guess,
that's
it.
F
F
Interesting
thing
was
happen,
decided
I
was
shown
in
the
last
IT
airside
meeting
I
have
allowed
to
have
along
side
of
you.
That
is
doing
all
the
I.
Any
such
group
and
interesting
is
that
on
the
g20,
it
is
trends
to
discuss
about
this
in
the
ITF.
But
deep
at
the
g20
meeting,
the
league
leaders
agreed
that
just
playing
early
elected
things
do
not
work.
They
realize
that
and
that
they
need
to
communicate
with
some
other
stakeholders
included
tech
community
to
how
to
implement
the
decentralized
technology.
That
is
a
little
interesting
sec.
I
think
so.
F
The
their
focus
is,
of
course,
the
consumer
protection
and
the
prevention
of
the
use
of
the
technology
for
the
criminal
activity
or
whatever.
So
we
have,
we
need
to
respond
to
these
activities.
So
last
four
months
since
the
last
I
hear
ya,
we
had
some
land
similar
workshops,
including
a
DFA
workshop
at
scaling,
Bitcoin
conference
that
in
Toledo
PB,
terrible
and
also
I
am
writing.
The
I
and
my
colleague
attended
several
panel
discussions
to
describe
the
the
what's
going
on
and
what
we
want
to
do
under
calling
for
the
colleagues
to
work
with.
F
So
we
are
planned
to
have
some
open
meeting
in
March
to
too
much
2020
on
a
much
nice
to
tense
in
Japan
Tokyo
Japan,
so
I'm
going
to
provide
an
update
to
the
people
who
want
to
involved
so
and
we
are
going
to
have
a
mailing
list,
I
promised
in
the
last
item
in
the
meeting,
and
we
are
going
to
have
it
and
I'm
going
to
let
you
know
and
I'm.
Please
find
me
and
drop
me
an
email
if
you
have
an
interest
on
this
activity.
Thank
you
very
much.
A
Thank
you,
sugar,
okay,
one
more
call
for
Jake
Holland
looks
like
he
didn't
make
the
meeting.
So
that
was
our
last
talk
thanks.
Everybody
I
hope
that
you'll
consider
doing
a
lightning
talk
at
the
next
ITF,
also
I'm,
doing
an
informal
social
event
on
Thursday
night,
Pecha,
Kucha
and
other
sort
of
humorous
lightning
talks
feel
free
to
join
the
informations
going
out
on
the
106
attendees
list
and
I
hope
everybody
can
come.
Thank
you
good
night
enjoy
your
dinner.