►
From YouTube: IETF104-HACKATHON-20190324-1400
Description
HACKATHON meeting session at IETF104
2019/03/24 1400
https://datatracker.ietf.org/meeting/104/proceedings/
A
Patience
and
I
have
one
request
to
go
at
the
beginning:
I
have
one
request
to
go
at
the
end,
we'll
honor
that
and
then
other
than
that
I'm
just
going
to
go
through
them
pretty
much
in
the
order
in
which
they
show
up
in
the
github
org
and
work
my
way
down
through
them.
So
if
you
want
to
get
an
idea
of
what's
coming
up
next,
just
follow
the
list
there.
If
I
accidentally
skip
a
file
or
something
like
that,
we
can
go
back
and
do
it
just
kind
of.
A
A
If
you
use
more
time
than
unfortunately
I'm
going
to
have
to
cut
you
off,
and
we
didn't
actually
get
a
clock
going
here,
but
I'll
give
you
like
a
one-minute
warning
when
you
need
to
wrap
up,
and
so
please
be
respectful
of
that,
then
to
people
who
are
in
the
audience
listening
police.
Do
you
know,
keep
the
volume
down
if
you
need
to
have
a
conversation
which
I
know
you
know
hey
it's
a
good
opportunity
to
chat
with
people.
B
C
So
hello,
everyone
so
me
and
my
team
have
been
working
on
development,
quantum
networks
since
I.
Guess
it's
a
foreign
concept
for
so
many
people
here,
I
will
be
having
I
will
have
to
answer
what
at
what
our
quantum
networks?
Why
are
we
working
in
them
and
how
are
we
implementing
them
so
I'll
start
by
the
basics
in
classical
computing
I
think
the
most
basic
element
is
a
bit.
It
could
be
a
0
or
a
1,
but
in
quantum
computers
it's
a
bit
different.
We
don't
have
bits.
C
C
So
it's
very
problematic
when
it
comes
to
qubit
because
they
can't
be
measured
or
copied
in
classical
bits,
I
simply
copy
it
injected
into
the
network
and
transfer
it
over
the
internet,
but
for
a
qubit.
If
I
copy
it
and
injected
into
the
network,
it
will
be
downgraded
to
a
normal
bit,
so
it
uses
its
special
properties
and
we
can't
use
it
for
quantum
computing
anymore.
So
since
we
can't
use
the
normal
networks,
the
only
way
to
transfer
qubits
from
point
A
to
point
B
is
using
something
called
teleportation.
C
I
know
it
might
sound
like
a
complex
word,
but
we
use
something
called
quantum
entanglement.
So
we
integral
two
points
in
space:
a
and
B
together
using
this
concept,
and
then
you
can
teleport
the
qubit
between
them.
So
on
long
distances,
some
extra
devices
have
to
be
used
which
are
quantum
repeaters.
So
you
can
connect
to
quantum
computers
together
in
the
distance
of
100
kilometers.
But
more
than
that,
you
have
to
use
quantum
repeaters
and
here
more
complex,
entanglement
to
teleport
the
qubits.
C
So
the
first
real
implementation
of
these
concepts
is
going
to
happen
in
the
Netherlands
in
2020.
So
the
plan
is
to
connect
four
cities,
the
four
cities
which
you
can
see
on
the
screen.
So
hopefully
this
is
going
to
be
the
first
implementation
of
a
quantum
network.
What
she
really
can't
transmit
qubits
between
counter
quantum
computers.
C
So
it's
still
a
very
recent
topic
and
what
we
are
working
on
in
the
IETF
is,
first
of
all
bringing
some
of
the
concept
of
classical
networks
to
the
quantum
networks.
I
personally
and
the
others
have
been
working
on
like
adapting
the
concepts
of
routing
and
things
that
we
currently
use
in
that
internet
to
fit
for
the
quantum
networks.
C
Some
of
the
colleagues
over
there
have
been
working
and
developing
a
quantum
pin
how
to
measure
the
quality
of
the
quantum
connection
with
teleporting
the
qubits
over
the
specifications
and
requirements
to
establish
a
quantum
ping,
and
two
of
us
have
been
working
on
actually
a
permutation
of
applications
on
quantum
network
blockchain.
So
that
would
be
my
presentation.
I
think
this
is
less
than
three
minutes.
So,
if
anyone
has
any
questions,
I
hope
our
theoretical
physicist
over
there
can
answer
them.
C
A
D
E
Right,
Thank,
You,
Charles,
I'm,
Dave,
Wonka
I'm,
one
of
the
co-chairs
of
the
measurement
and
analysis
reporter
calls
Research
Group
or
about
three
years
old
will
be
meeting
later
this
week,
and
this
is
the
first
time
we
had
a
hackathon
table
to
focus
on
some
live
measurements.
That
will
give
you
some
results
in
the
meeting
later
this
week.
We
did
three
projects
at
our
table.
I'm
going
to
talk
to
you
about
the
first
one.
The
last
minute
L
is
going
to
talk
to
about
the
second
one,
and
then
we
have
a
slide
for
the
third.
E
So
what
did
we
do
well
we're
interested
in
doing
measurements
that
can
inform
the
engineering
efforts
in
the
IETF
and
also
inform
the
operations
of
those
protocols.
So
if
you're
at
Mapp,
Archie
you're
doing
one
of
those
things
bringing
real
measurement
results
that
have
something
to
do
with
the
engineering
or
the
operation
of
the
protocols
that
the
IETF
defines
our
goal
was
to
study,
in
this
case
the
privacy
and
security
issues
that
we
were
seeing
with
ipv6
in
deployments
today.
E
So
when
we're
doing
some
measurement
studies,
the
goal
was
to
the
plan
was
to
prepare
an
internet
draft
following
the
hackathon
using
the
anonymized
aggregate
results
that
we
prepared
at
the
hackathon.
So
we're
actually
hacking
measurement
results
instead
of
a
protocol
and
then
the
goal
is
to
inform,
as
I
said,
the
engineering
and
operation
and
practice
of
that
protocol.
What
we're
going
to
do
here
in
the
unique
opportunity
to
the
hackathon,
let
us
do
is
take
public
data
that
my
colleague,
Oliver
gusts,
are
from
team.
E
Munich
is
bringing
from
their
project
with
Akamai
internal
things
that
we
can't
share
to
protect
the
variety
to
protect
the
privacy
of
clients
that
touch
our
platform,
so
we're
gonna
get
together
here
and
find
some
way
to
compare
apples
to
apples
and
bring
a
combined
result.
That's
hopefully
more
than
either
of
us
had
on
in
Rome.
E
So
the
idea
was
to
take
the
two
largest
known:
I
could
be
six
measurement
surveys
of
public
and
private
data
and
combine
them
together.
So
what
did
we
learn?
We
learned
that
as
of
this
year,
there
is
it's
easy
to
call
more
than
1.2
million
ipv6
router
addresses
that
have
eui-64,
meaning
we
know
who
manufactured
them
and
we
have
their
MAC
addresses.
This
was
discovered
accidentally
in
topology
and
reach
ability
studies
both
done
in
academia
in
industry.
E
The
interesting
part
is
the
public
and
private
results.
Don't
overlap
each
other
they're
complementary.
So
that's
good
and
justified.
The
use
of
of
the
time
here
at
the
hackathon
to
compare
these
results.
Another
interesting
result
we
had
was
older,
hit
lists
ones
for
last
year,
actually
produced
more
results
than
some
of
the
new
ones.
So
we're
gonna
report
the
results,
not
here,
but
we'll
show
them
after
we
think
about
them.
A
little
more
on
Thursday
and
I'm
gonna
turn
it
over
to
Al
for
the
last
minute
to
talk
about
his
project.
F
Thanks
Dave,
so
I'm,
mal,
Morton
and
I
worked
with
Alexander
on
this
IP
network
performance
and
capacity
measurements
comparison
setting
the
stage
for
this
is
that
Internet
access
speeds
are
going
up
and
up
and
into
the
gigabit
range
very
very
soon.
For
for
most
of
us,
some
of
us,
some
of
us,
have
these
speeds
now,
and
we've
been
looking
at
sort
of
ad-hoc
methods
of
measurement
for
IP
capacity
throughout
the
years,
bucola
speed
tests
and
so
forth.
F
So
we
we've
decided
to
look
at
some
new
ways
to
measure
this,
and
we've
been
looking
at
a
couple
of
different
platforms
and
opportunities.
Iperf,
of
course,
has
always
provided
TCP
measurements
and
UDP
measurements,
but
we've
got
a
new
platform,
basically
a
new
tool
that
has
a
search
algorithm
built-in
that
can
do
the
UDP
measurements
very
quickly,
and
so
what
we
did
was
essentially
created
an
additional
test
plan
for
looking
at
these
test
conditions
and
then
in
the
calibrated
lab
setup
that
we
have
courtesy
of
Intel
in
Hillsboro
Oregon.
F
Now
the
the
tcp
and
UDP
stuff
from
iperf
they
report
out
in
the
payload
bandwidth
and
the
UDP
st
that's
the
new
tool
that
reports
at
the
IP
layer,
including
the
header,
so
they're
they're,
actually
getting
close
to
the
normalized
values
from
100
to
1
gigabit
per
second,
and
when
you
put
in
the
correction
factors,
they
all
get
a
lot
closer.
But
the
key
story
is
right
here
at
the
bottom,
where
we're
basically
tcp
is
kind
of
running
out
of
gas
at
the
1
gigabit
range
in
the
200.
F
G
H
It
is
the
the
intention
to
have
a
very
low
latency
routing
protocol
coming
out
of
his
projects
in
the
microsecond
range
to
support
on
the
layer
to
the
TSN,
time-sensitive
networking,
and
so
what
we
did
is
to
have
everybody
cut
everything
to
get
it
very
fast,
so
no
lookup,
no,
no!
It's
open,
no
prefix
match
just
self
routing,
so
take
data
out
of
the
header
and
use
it
for
routing
and,
of
course,
also
cut
through
technologies.
So
the
packet
is
forwarded
before
being
completely
stored.
Just
everything
to
keep
it
very
fast.
H
We
used
a
crazy
feedback
from
this
meeting
crazy
addressing
we
mapped
the
telephone
number
into
the
prefix
of
the
ipv6
address.
It's
all
this.
It's
sounds
crazy,
but
already,
20
years
ago
proposed
by
Korea
telecom
authority,
we
summarized
everything
in
our
internet
draft.
This
is
what
we
did
here.
So
we
had
real
report
real
Hardware,
because
you
need
dedicated
switches
to
do
this.
This
research
devices
from
outcome
of
research
projects.
What
you
see
here
is
this
on.
H
The
display
shows
the
IP
address,
which
is
this
looks
like
a
telephone
number
and
it
is
delegated
to
the
next
lower
hierarchy
level
in
the
configuration
time,
and
then
it
is
already
or
routing
here
are
hierarchical
routing.
What
we
learned
here
is
very
the
only
technology.
Of
course
we
got
valuable
feedback
thanks
to
the
experts
from
Tony
from
David
and
Alistair
that
things
to
be
considered.
There
is
some,
let's
say
similarity
with
the
rift
project.
H
We
had
a
long
discussion
and
we
got
the
feedback
to
investigate
behavior
in
case
of
failure,
for
example,
what
we
also
noticed:
yeah,
of
course,
open
to
share
the
source
code,
but
it
doesn't
help
because
the
XD
hardware
is
so
expensive.
So
we
have
to
think
about
that.
That's
one
leg,
a
lesson
learned
and
to
wrap
up.
So
this
is
my
name:
Andrea's
forearm
and
Marion.
H
Uber
lyft
is
sitting
you
know
and
colleague,
and
we
have
one
remote
participant
and
what
we
will
try
next
is
to
make
to
forward
the
simulation,
because
this
is
cheaper
and
also,
if
possible,
in
relation
with
low
costs,
which
at
least
for
the
behavior
and
then
can
be
accelerated
in
the
in
the
real
hardware
and
for
more
information.
If
you
click
the
link,
you
will
find
papers,
articles
and
so
on.
Let's
concludes
my
presentation.
Thank
you.
I
I'm
here
from
the
the
six-man
group,
we
were
working
on
path,
MTU
discovery,
and
so
we
were
looking
at
a
bit
of
a
retake
on
how
path.
Mtu
discovery
can
work
in
the
internet.
That
could
be
six
and
we
actually
decided
to
do
a
reimplementation
of
a
thirty-year-old
RFC.
But
this
time
on
top
of
ipv6
this,
this
mechanism
works
on
my
normal
path.
Mtu
discovery
where
you
send
stuff
into
the
network,
can
you
get
ICMP
back?
We
add
an
option
to
keep
this
extension
header.
I
This
extension
header
gets
processed
by
Reuters
I
mean
works
like
a
record
route
option.
It
records
the
the
lowest
empty
on
the
path
reducers
and
the
field
goes
to
the
end
and
then
I
guess
reflected
back
somehow
and
so
the
start
of
the
hackathon.
We
redefined
the
option
because
we
want
to
figure
out
a
way
to
get
and
the
packets
to
be
reflected.
So
the
option
for
mine
now
looks
like
this,
which
is
an
update
to
the
draft,
and
we
have
two
MTU
fields.
I
Mtu
one
is
that
the
MTU
of
the
host
sending
out
a
packet
to
try
and
measure
the
path
see.
What's
there
we
have
mq
2,
which
is
a
reflected
field.
We
we
realized
that
we
wanted
to
have
a
flag
here,
and
so
we
compress
the
MTU
to
space,
and
so
you
can
only
have
even
number
empty
use.
I
think
that's
okay,
so
we
get
a
respond
request.
Flagler
and
service
allows
us
to
have
hosts,
actually
actively
ask
for
a
response
and
just
mark
that
something
as
a
response
yeah,
and
so
what
did
we
do?
I
And
we
had
quite
a
busy
table
there.
Quite
a
few
of
us
and
and
so
we
didn't
update
to
the
drafts
just
as
a
starting
point,
and
then
we
worked
on
initial
implementations
for
host
for
Linux
and
FreeBSD.
We
go
to
functioning
Reuter
and
fermentations
in
VPP
and
p4,
and
a
bob
who's.
First
icon
is
working
the
sector
for
truck
and
we've
now
interrupted
and
tested
all
of
this,
and
it
seems
to
work
and
the
script.
I
have
says
that
we
have
fixed
the
Internet.
I
Well,
we
didn't
had
to
do
was
do
any
performance
testing
and
obviously,
this
being
a
hop
by
hop
option
and
being
loser
every
rooster
on
a
path.
Performance
implications
would
be
interesting,
and
hopefully
we'll
have
built
up
time
to
look
at
that
in
the
future,
and
we
didn't
try
and
crew
piss
off
further
into
any
operating
systems.
The
host
acts
we
did
were
just
looking
at
at
getting
the
packets
generated
and
processing
them
and
receiving
them
and
so
yeah.
I
So
this
is
good
and
we
think
this
has
been
really
really
positive
and
it's
managed
to
reinforce
the
the
work
and
the
working
group,
and
we
think
this
is
something
we
can
carry
forward.
It's
really
reassuring
to
see
that
this
could
be
implemented
four
times
in
like
the
space
of
the
day
and
then
interrupt
that
habit.
Work
and
we
don't
know
a
mr.
playable,
no
idea,
and
so
we're
gonna
continue.
Looking
this-
and
this
says,
we
I
see
a
six
man,
ICMP
limits
draft,
so
do
that.
A
J
Hello
and
welcome
to
our
presentation-
my
name
is
Martin
and
I'm
here
to
tell
you
about
the
light
color
networking
in
space.
So
what
is
delay
tolerant,
networking
well
in
current
setup,
usually
when
you
communicate
with
a
rare
computer,
the
other
party
has
to
be
online.
That's
not
the
case
with
delight
or
networking,
because
it's
very
difficult
to
achieve.
For
example,
if
you
have
a
ground
station
and
you
would
like
to
communicate
with
the
satellite,
there
are
all
kind
of
things
which
will
prevent
you
to
do
so
at
the
time.
J
You
want
to
do
that,
for
example,
the
earth
between
you
and
the
satellite.
So
what
we
are
doing
actually
is
the
preparation
for
DT
and
real
space
test,
which
is
to
be
loud
in
2020
in
2020,
there's
going
to
be
a
satellite,
hopefully
running
Micro
PC
and
implementation,
OTT
and
protocol,
and
what
we
are
trying
to
do
and
planning
to
do
is
that
ground
station
one
will
measure
some
some
sensory
data,
for
example
temperature.
J
It
will
encode
it
into
the
bundle
and
at
the
at
some
point
there
will
be
a
line
of
sight
with
the
satellite.
They
will
learn
about
each
other,
we
are
beacons
and
the
ground
station
will
transfer
via
a
bundle,
data
measured
satellite.
Then
eventually,
as
the
satellite
continues
on
the
orbit,
it
will
see
ground
station
2
and
again
they
will
learn
about
each
other
via
beacons
and
the
transfer
satellite
to
Micro
PC
and
build
a
satellite
Micronesian
implementation
will
transfer
the
data
they
are
bonded
to
ground
station
2.
J
J
J
J
Okay,
we
do
have
a
feedback
to
DT
embargo,
which
is
gonna
happen
or
the
session
is
gonna
happen
on
Tuesday
and
that
feedback
will
be
that
we
should
bring
IP
and
IP
and
D
draft
up-to-date,
because
it's
currently
expired
and
have
it
adopted
by
workgroup,
because
we
do
need
it
for
our
testing.
Then
we
found
out
that
actually
in
IP
and
D
draft
for
the
same
thing,
the
different
different
approaches
used
while
encoding
and
decoding
numeric
values,
so
I
peony,
is
using
Sdn.
We
and
the
rest
of
the
system
is
using
sea
bore.
J
J
A
K
Good
afternoon,
everyone,
my
name,
is
Alexei
and
I'm
here
to
tell
you
about
the
goals
and
results
of
the
EEP
new
hackathon
project.
First
of
all,
for
those
of
you
not
familiar
with
it
new,
it's
a
new
leap
method
for
bootstrapping
devices,
with
minimal
limited
user
interface
and
no
pre-configured
credentials.
K
So,
unlike
other
current,
it
methods,
we
don't
require
any
configuration,
so
you
can
think
of
any
sort
of
new
device
that
you
just
bought
from
the
store
and
now
want
to
bootstrap.
The
way
this
is
done
is
using
a
user
assisted
out
of
planned
channels
which
consists
of
a
step
such
as
scanning
a
QR
code
or
tapping
an
NFC
tag,
or
here
you
can
see
an
overview
of
the
EMU
architecture.
So,
on
the
right
hand,
side
we
have
our
new
device
on
the
Left.
K
We
have
a
server
that
we
want
to
bootstrap
it
to
and
in
the
middle
we
have
the
in
band
it
network
and
the
user
assisted
out-of-band.
You
know
there's
currently
an
open
source
implementation
available
on
github,
and
it
is
a
nice
mixture
of
C
code,
Python,
note,
JSON
server
side
and
C
n
Python
on
the
client
side.
One
of
our
main
goals
for
this
hackathon
was
to
fork
this
project
and
create
a
version
that
reduces
implementation
dependencies
and
have
it
running
as
easily
and
quickly
as
possible.
K
In
addition
to
myself,
we
had
a
great
team
consisting
of
IETF
nuke,
Eduardo,
Anu
and
Alex,
who
joined
us
just
for
the
hackathon
and
every
now,
and
then
we
also
discussed
them
work
with
the
authors
of
the
draft
Thomas
and
moiety.
So
if
any
of
this
sounded
interesting,
please
read
the
draft
check
out
the
implementation
and
join
us
for
EMU
on
Monday
morning.
Thank
you.
B
L
How
this
is
a
protozoan
from
seung-hyun
University
in
Korea,
this
hackathon
for
RIT
as
a
project,
is
a
custom
project.
So
this
time
we
keep
working
for
the
validation
of
items,
have
major
three
interfaces:
a
consumer
facing
interface
and
another
facing,
and
also
the
ratio
registration
interface.
This
time
also,
we
implemented
translation,
stop
especially
policy
provisioning
using
a
machine
learning
scheme
such
as
decision
tree.
So
this
is
the
building
environment.
L
L
Also,
we
implemented
our
using
acoustic,
a
separate
C
and
a
talking
stop
and
that
we
implemented
our
substr
function.
Training
such
as
a
fiber
and
web
filter
using
Cheney.
This
time
we
proved
a
concept.
I
turn
as
a
framework
and
also
we
implemented
reservation
interface
specifically
at
the
company,
and
also
we
implemented
the
translation
based
on
NSF
database
management.
Also,
we
are
using
as
a
posse
provided
by
acoustic
that
talking,
especially
so
we
unloaded
the
our
coded
github.
You
can
access
tomorrow,
happy
hour
demo,
we
can
demonstrate.
Thank
you
for
your
attention.
L
L
Stop
this
time
we
proved
of
concept
or
IP
ye
OCB,
and
also
we
proposed
the
IP
version,
6
virtual
or
neighbor
discovery,
so
we
proposed
our
IP
version:
6,
never
discovery
for
IP
based
virtual
environment.
So
this
is
a
so
this
time
our
two
universities,
my
university
SKT,
you
and
also
some
here
university
to
professor's
and
three,
a
student
also
some
other
guests
from
other,
such
as
et
RI,
and
they
participate
in
this
project.
So
this
is
basically
the
particular
environment
case.
It
is
very
hard
to
demonstrate
the
Righteous
Kill.
L
That's
why
we
using
the
simulator,
for
this
is
called
the
sumo
sumo
easy
for
mobility
for
Vickers,
and
also
we
using
our
protocol
based
on
om,
the
plus
+
on
the
prosperous,
is
a
natural
simulator,
providing
five
Mac
IP
TCP
stuff.
So
this
time
you
can
see
to
our
se
you
and
also
to
Vickers,
audible
or
the
coverage
of
our
SEO.
We
can
apply
the
multi-hop
dat
using
our
protocol,
so
we
implemented
the
pker
structure
based
on
om
net
and
also
veins.
L
The
veins
is
provided
by
some
open
source
project,
so
it
is
implemented
or
she
be
so
from
this
the
project.
We
proved
that
the
basic
IP
wave
I
totally
doctor
11
OC,
be
based
the
IP
policy.
The
delivery
is
feasible
also
on
top
of
it,
we
can
implement
it.
The
neighbor
discovery
take
advantage
of
become
our
mobility,
so
we
design
in
implemented
the
IP
way.
We
are
vehicle
or
neighbor
discovery
based
on
am
net
and
sumo
paste.
L
Also,
next
time
we
try
to
propose
the
flexible
mobility
management
for
IP
wave,
so
this
is
appendix
so
this
is
open
source
environment,
or
so
we
unloaded
the
github
a
source
code
that
you
can
test.
Also,
you
can
see
the
code
tomorrow
also
or
happy
hour,
HEC
demo,
or
we
demonstrate
this
is
a
bit
ugly.
Thank
you
for
your
attention.
A
M
This
case
over
specifically
this
NTP
time
protocol
and
the
goal
at
this
heck
is
the
integration
of
NTS
in
different
implementations,
NGS
based
on
the
current
draft
NTS
version
17
it's
in
the
pre
final
state
and
we
guess
we
can
release
this
to
it,
soon
be
verifying
as
it
operability
between
different
groups,
NTS
implementations
and
currently
we
have
many
of
them.
So
we
have
about
five
working
client
and
server
implementations
of
mts.
We
have
also
as
implementations
that
has
to
work
in
progress.
M
We
have
implemented
implemented
in
different
languages,
so
we
have
C
C++
taller
in
person
and
also
we
use
different
NTP
limitations,
so
we
have
ntpd
NTP,
sag
Kony
and
as
our
entity
solutions
and
the
test
covered
two
parts
of
NTS,
so
we
have
here
a
space
key
exchanged.
Afterwards
we
have
entropy
authenticated
NTP
exchange.
M
The
green
path
in
this
variable
shall
successful
tests,
so
it
looks
pretty
good.
We
have
some
outstanding
tests,
but
we
have
no
critical
error,
and
yesterday
we
have
some
small
bug,
fixes
back-back
fixes
and
we
have
some
small
issues
with
optimizers,
as
the
last
of
my
version
have
back
in
the
TLX
part
of
a
function,
so
we
use
the
work
of
art,
but
in
general
probability
test
was
pretty
successful.
We
had
no
further
problems
and
the
specification
and
we
guess
we
can
release
it
very
soon.
M
D
N
You
so
this
year,
I'm
just
calling
all
tables
everything
DNS
we
did
so
much
stuff.
As
usual
DNS
we
didn't
work
on
one
major
project.
We
have
lots
of
things
working
in
parallel.
We
had
five
bigger
projects,
six
others
going
on
and
lots
of
people.
One
of
the
big
projects
that
we
worked
on
was
extended
error
codes.
Don't
you
just
hate
it
when
get
surveil,
and
you
have
no
idea
why?
The
idea
of
this
is
that
you
can
extend
the
error
codes
and
the
response
message
and
actually
find
out.
N
The
judge
has
been
around
for
a
couple
of
years,
but
the
work
that
got
done
here
wants
to
implement
it
in
to
open-source
resolvers
and
works
in
progress
on
a
third.
So
that
was
all
good.
One
of
the
other
big
projects
we
worked
on
was
Danish
cookies
cookies
themselves
have
been
around
for
a
while,
but
what's
been
recognized.
Lately
is
there's
interoperability
problems
between
the
different
implementations
of
this
at
the
open
source
code.
N
A
very
recent
draft
came
out
and
what
happened
at
hackathon
is
that
we
now
have
five
implementations
into
operating,
with
the
fix
to
make
them
all
into
operate
correctly.
Another
project
that
got
worked
on
here
was
doing
a
save
a
quick
because
we
all
need
another
transport
for
DNS
right.
There's
a
draft
that
came
out
a
couple
years
ago
proposing
doing
DNS
over
quick.
There
was
some
work
at
the
time,
but
it's
stalled.
What
happened
at
this
hackathon
is.
N
There
was
a
client
implementation
done
in
a
performance
tool
called
flamethrower
and
on
the
server
side
a
proxy
was
written
to
proxy
quick,
the
DNS
from
quick
to
UDP
and
hey.
It
works
oh
good
again.
Another.
The
last
of
the
big
projects
is
something
we're
calling
zot
and
it's
that's
because
it's
xfr
over
TLS
who
knew
Zots
a
thing.
That's
the
cute!
Little
guy
there
again,
a
very
recent
draft
came
out
and
what
we've
done
is
there
was
work
to
support.
This
was
to
improve
a
patch
which
enabled
dot
in
NSD
and
authoritative
nameserver.
N
So
we
put
OCSP
in
there
there's
pulling
new
work
on
the
chain
extension
so
that
should
be
merged
and
released
in
the
future
and
in
terms
of
the
transfer.
What
we
did
is
we
got
an
opportunistic
connection
between
NS
d,
acting
as
a
master
and
unbound
acting
as
a
secondary
and
actually
got
the
data
transfer
working.
N
So
that's
good,
but
what
we've
realized
is
that
the
authentication
model
behind
this
needs
a
bunch
of
work,
and
we
also
realized
that
this
is
a
really
nice
candidate
to
take
advantage
of
something
new
and
the
dns
called
stateful
operations.
So
we
we
have
the
outline
of
a
new
draft
about
how
to
do
this
over
state
corporations.
N
Some
the
other
projects,
I,
don't
have
time
to
go
into
detail
about.
There's
a
Yank
model
for
managing
zones.
In
the
authoritative
there
was
additional
work,
continuing
to
put
dough
support
into
stubbie
work
on
a
Danish,
tap
multiplexer
work
on
a
resolver
test,
bird
work
on
Dauphine,
DNS
Python
and
work
on
the
zone
MD
in
net
DNS,
so
all
sorts
of
good
stuff
going
on
this
is
our
team
list?
Apologies.
It
is
missing
a
few
names.
We
did
actually
have
five
first
timers,
so
I'll
fix
this
and
upload
the
new
slides
and
that's
everything
dinner.
O
Good
afternoon,
this
is
the
report
of
the
open
check
project,
so
our
plan
here
was
to
improve
the
open
source
implementation
of
shake
and
shake.
If
you
don't
know,
is
the
static
context.
Header
compression
protocol
that's
being
developed
at
the
LP
one
working
group
and
the
purpose
of
that
protocol
is
to
compress
headers
and
fragmentation
and
reassembly
for
extremely
constrain
networks,
which
payloads
typically
of
few
tens
of
bytes.
We
intend
to
carry
ipv6
UDP
coop
messages
over
those
kind
of
networks.
O
O
O
P
O
Code
base,
we
added
more
test
content
and
we
improved
the
functionality,
especially
on
the
compression
front,
where
the
code
was
not
fully
implementing
all
the
features
of
the
draft
so
far,
and
also
we
added
a
connector
that
the
decoder
has
a
simulator
of
a
link.
While
we
can
set
the
loss
rate
for
packets,
but
we
added
a
connector
so
that
you
can
use
actual
hardware
and
and
send
your
packets
over
a
real
Laurent
network,
oops
yeah.
O
What
we
learned
from
the
project
is
some
parts
of
the
spec
we're
not
clear
enough.
We
had
implementers
telling
us
they
didn't
know
how
to
implement
that
part,
because
they
hadn't
understood
it.
So
we'll
add
more
examples
to
explain
it
better
and
also
we
know
the
next
draft.
That's
coming
up
is
describing
the
rule
set
that
is
used
for
compression
fragmentation
and
the
context
of
parameters,
and
this
work
allowed
us
to
summarize
all
the
information.
Then
you
need
to
put
in
that
context,
and
this
is
useful
for
writing
the
next
draft
and
that's
our
team.
Q
Q
Q
So
what
we've
been
trying
to
do
this
hackathon
is
to
improve
the
spin
spin
dump
analyzer
tool
that
we've
implemented.
We
try
to
implement
some
aggregation
of
several
instances
of
these
measurement
tools
and
try
to
visualize
it
basically.
So
what
did
we
get
done?
This
time?
We
added
some
more
support
for
some
more
quick
versions
like
old
Google,
quick
versions
and
the
new
ITF
draft
18
version.
Q
So
we
have
been
looking
at
some
some
new
mechanisms
for
for
measuring
packet
loss
and
improving
latency
measurements
and
so
on
and
there's
some
really
interesting,
work,
they're
being
done
by
by
Mauro
and
others,
and
that
might
be
a
future
candidate
for
this
spin
up
tool.
That
we're
working
on
moving
on
from
here
is
that
we
are
preparing
a
demo
for
for
this
hack
demo
event
for
tomorrow,
where
we
will
be
visualizing
our
data.
Q
So
we
will
have
a
set
of
these
spin
dumping
instances
associated
with
a
bunch
of
routers,
and
then
we
can
display
end
to
end
and
segment
based
performance
monitoring.
So
please
come
check
it
out
and
yeah.
That's
basically
it
if
you're
interested
in
this
kind
of
stuff.
You
can
check
out
our
work
on
github.
It's
available
there
and
team
members
were
er,
er,
Co
Sylvester
me
Ronnie
and
Mauro
as
well
as
I
headed
down
there
I
forgot
you.
D
R
Right
so
I'm
gonna
talk
about
what
we
did
in
suit.
So
for
those
who
don't
know,
suit
is
working
on
an
architecture
and
metadata
format,
standardizing
this
type
of
stuff
for
IOT
firmware
updates,
and
do
that
in
such
a
way
that
it
also
applies
to
the
full
spectrum
of
IOT,
not
just
Raspberry
Pi,
that
was
stuff,
but
also
smaller
stuff,
like
micron
controllers,
that
are
much
more
constrained
in
terms
of
resources.
R
R
So
what
happened
at
the
hackathon?
We
actually
have
now
two
different,
two
independent
implementations
of
generator
for
this
metadata,
one
from
our
man
from
renesis.
We
also
have
new
parsers,
so
we
have
3
independent
implementations
of
parsers,
two
of
which
are
actually
running
on
actual
IG
devices,
so
small
microcontrollers
of
between
class
1
and
class
2
and
class
ya
class,
1
and
class
2
types
of
devices.
R
We
have
interoperability
between
the
manifest
generator
from
arm
and
two
of
the
parsers
that
we
have
from
INRIA
and
from
renesis,
and
we
also
demonstrated
a
full
workflow
that
we
want
to
enable
with
with
suit
architecture
so
where
you
actually
can
build
a
firmware
actually
generate
the
meta
data
according
to
the
format
actually
transport.
All
of
that,
through
a
repository
to
the
IOT
devices,
did
which
is
then
able
to
verify
and
then
update
the
former,
and
we
did
that
on
top
of
of
prior
training
on
these
devices,
yeah
yeah.
R
R
The
main
main
thing
we
learned
was
that
it's
really
important
to
just
specify
one
way
of
doing
things
in
the
stacks
else
is
just
like
kind
of
warps,
and
so
we're
pretty
happy
if
you
want
more
info
about
suit,
go
to
the
website
of
suit
and
also
just
check
out
the
this
distract,
and
we
also
have
another
pad
where
we
know
that
most
of
the
things
we
did
so
thanks.
That's
it.
S
Yep,
alright,
so
at
first
I,
wasn't
quite
sure
how
much
we
were
gonna
get
done
at
the
hackathon
until
I
saw
this
sign
here,
which
told
me
how
much
we're
going
to
get
done
a
ton.
So
our
group
was
trusted
execution,
environment,
provisioning,
and
so
this
is
the
t
p--
working
group,
which
tries
to
figure
out
how
to
provision
code
and
configuration
into
trusted
execution
environments
such
as
secure
elements,
Intel,
SGX,
try
so
on
and
so
forth.
S
This
was
our
third
hackathon,
and
so
our
plan
was,
since
the
drafts
are
still
in
progress
and
try
to
flesh
out
implementation
details
and
one
of
those
drafts
the
actual
protocol,
one
in
the
middle.
There
is
sort
of
known
to
be
under
specified,
and
so
we
were
trying
to
figure
out
what
are
the
areas
of
ambiguity
as
we
tried
to
implement
things,
and
so
we
tried
to
compare
our
implementations
and
des
validate
that
that
spec
is
actually
agnostic
as
to
which
type
of
T
EE
that
it's
intended
to
work
with.
S
S
S
We
updated
to
match
that
third
spec,
which
didn't
exist
last
IETF,
which
was
how
do
you
write
what
protocol
you
used
as
the
transport
for
ot
RP
and
it's
using
HTTP?
We
kind
of
design
it
on
the
fly
last
hackathon,
and
so
then
it
was
written
up
on
a
spec
last
hackathon
and
it
was
implemented
this
time
and
we
updated
it
to
match
recent
feedback
from
people
like
Mark
Nottingham,
and
we
implemented
another
mechanism
that
was
designed
in
the
middle
of
last
hackathon
that
hasn't
made
it
into
the
spec.
S
Yet
it's
just
documenting
github
so
design
in
progress.
We
filed
the
number
of
new
issues
there
tracked
in
the
github
on
the
issue,
tracker
for
the
draft
and
there's
a
summary
of
the
issues
which
we
will
talk
about
in
the
Oh
TRP
working
group
for
people
that
are
not
in
that
working
group.
If
you're
in
an
attestation
working
group,
one
of
the
issues
has
to
do
with
the
relationship
between
the
OPR
pre
protocol
and
SS
stations,
such
as
being
done
in
the
SE
eat
or
rats
proposals,
and
so
on.
S
So
the
relationship
between
those
and
that
is
on
the
agenda
for
this
working
group
meeting
coming
up
and
so
there's
our
team.
Last
hackathon
I
reported
that
our
team
size
had
doubled
from
one
to
two
and
this
time
it
doubled
again,
so
we're
making
great
progress.
Two
of
them
are
first-timers,
and
so
there
we
go.
That's
it
thanks.
T
All
right,
hello,
everybody,
I'm,
Kristy,
Lee
and
here
to
talk
about
a
CVP.
So
briefly,
a
CVP
is
a
JSON
transport
protocol
over
HTTP.
The
goal
is
to
provide
a
standard
protocol
and
data
model
for
verification
that
crypto
algorithm
implementations
are
correct
and
we
provide
an
extensible
data
model
to
cover
testing
of
new
algorithms
as
they
get
release.
So
the
goal
of
the
protocol
is
briefly
is
to
come
up
with
a
common
protocol
to
enable
technology
providers
to
interoperate
with
multiple
validation
authorities.
T
This
may
be
FIPS
140
or
Common,
Criteria
or
even
non-government
related
authorities.
So
the
plan
for
the
hackathon
for
us
was
to
implement
AES
GCM
SIV
as
an
extension
on
the
server
inclined,
based
on
a
draft
specification
that
we
had
written
for
a
s.
Gcm
SIV
on
github
here
are
some
of
our
various
drafts
right
now,
they're
just
internet
drafts.
T
The
current
data
model
as
well
was
extended
to
allow
for
GCM
SIV
testing
interoperability
between
the
demo
server
and
the
client
was
fully
completed
and
successful.
The
server
now
has
an
Aes
GCM
SIV
implementation
to
support
the
generation
and
validation
of
test
vectors
and
the
client
uses
boring
SSL's
GCM
SIV
implementation
integrated
with
Cisco's
open
source
client
to
actually
run
the
test
vectors.
T
So
what
we
learned,
we
now
have
a
protocol
that
can
remain
unchanged
while
accommodating
extensions
as
new
algorithms
and
we're
interested
in
forming
an
AC
VP
working
group
in
the
future.
So
quick
rack
up,
wrap
up.
Everybody
on
our
team
was
a
first-time
member.
Here
we
have
a
side
meeting
that
should
be
on
Tuesday.
So
if
you're
in
sit
in
this
space,
please
come
join
us.
We
want
to
put
together
a
Boff
in
Montreal
at
IOT,
105
and
as
well.
U
Okay,
so
the
topic
of
the
project
we
work
is
problem
in
testing
and
simulating
networks.
This
is
taken
from
one
of
the
oldest
and
most
common
drops.
Rfc's
from
the
bench
marking
me
told
you
workgroup
and
it
is
porous.
You
can
see
very
simple
to
understand
what
the
project
was
about
now
in
today's
world.
There
is
a
lot
of
progress
in
creating
models
and
right
young
models
and
implementations
which
represent
the
device
on
their
test
in
this
situation.
But
what
is
missing
is
a
model
of
a
host
or,
as
in
this
case,
a
tester.
U
So
the
problem
with
that,
obviously,
is
that
you
don't
have
a
move.
The
define
Network,
if
you
have
part
of
it,
which
is
very
important,
which
you
cannot
model
but
in
in
the
case
today,
you
have
to
run
a
pink,
which
is
application
which
will
send
data
and
will
simulate
some
traffic
or
you
can
run
a
iperf,
but
there
is
something
that
can
be
done
and
has
been
done
in
simulation
world
like
despise
there
you
have
models
for
sources
and
sinks.
U
There
is
also
simulation
primitives
in
VHDL,
which
simulates
source
and
sinks
so
that
people
can
write
tests,
and
this
has
been
existing
for
a
while
in
the
network
simulation
environment
like
like
on
net
PP,
they
had
their
software-defined
networking
concepts
from
10
years
ago,
so
they
were
able
to
reproduce
topology
with
all
the
packets
being
generated
received
and
do
that
in
the
model.
So
today
we
have
in
target
in
RealNetworks
the
device
under
test.
U
This
is
the
real
NATO
switch
actually
have
young
models
so
that
what
we
are
working
on
is
creating
a
model
for
controlling
a
traffic
generator.
So
the
problem
is
today:
you
have
to
buy
a
traffic
generator,
which
is
proprietary.
It
is
not
interoperable
and
if
you
create
a
test
case
for
it,
you
can't
really
simulate
it.
U
So
you
lease
you
have
to
here
the
expensive
hardware
at
the
end,
so
you
can
test
it
and
if
you
have
a
simulation,
you
cannot
just
take
that
simulation
and
reuse
it
in
the
top
team
that
will
be
generating
the
test.
So
we
want
to
go
from
Software,
Defined
Networking,
to
model
Defined
Networking,
and
by
using
that
concept
we
mean
that
also
the
data
plane
can
be
modeled.
U
These
are
some
details
except
from
the
draft,
so
this
is
a
to
take
the
traffic
generator,
because
there
are
three
components:
these
are
some
of
the
Leafs
you
can
configure
and
when
you
can
feel
that
when
you
commit
your
configuration,
then
the
generator
will
start
generating
waveform
yeah.
So
if
you
see
this
commit
at
the
end,
it's
very
important.
U
It
has
to
be
simple,
because
hardware
people
also
have
to
use
this
API,
and
there
is
a
test
case
which
we
use
to
demonstrate
how
it's
using
the
draft,
and
then
one
thing
that
is
very
easy
with
the
model
of
a
traffic
generator
is
that
you
can
actually
write
the
test
script
that
is
universally
validating
devices
that
implemented
the
advantages
that
you
can
create
a
loopback.
You
can
create
a
traffic
and
you
can
check
the
status
in
the
analyzer
a
bit
and
then
you
can
easily
validate
the
traffic
generators
that
you
kept.
U
Your
network
actually
generate
what
they
are
supposed
to
generate
without
what
writing
a
single
line
of
code.
So
this
test
case
we
committed
as
part
of
the
work
on
the
project
and
now
I
want
to
give
word
to
specialists
from
the
simulation
environment,
which
sees
also
advantages
in
using
such
model
in
their
development.
V
Everyone
I
am
Attila,
it
seems
like
I,
have
five
seconds
per
slide,
so
I'll
just
slip
through
all
of
them.
The
sub
project
I've
been
working
on
is
a
software
simulation
based
implementation
of
these
models.
Some
of
them
are
standard.
The
ITF
photos,
some
of
my
drafts
and
the
major
part
of
it
is
a
tiny
Python
script
that
loads
an
XML
representation
of
the
whole
network
in
the
configuration
of
the
hosts
and
that
conforms
to
these
models
and
generates
different
energies
interests
file.
V
Basically,
it's
the
same
information
just
with
the
different
format
and
when
it's
executed.
It
looks
like
this.
This
is
one
of
the
example
scenarios
you
can
see
the
insides
of
the
one
of
the
hosts
and
the
switches
and
the
queuing
inside
one
of
the
interfaces
and
I
think
simulating.
This
kind
of
thing
is
great
because
it
makes
experimenting
with
new
concepts
easier
and
faster,
and
we
also
made
it
in
a
new
young
model,
but
that's
not
important
right
now.
These
are
all
names
by
the
way.
V
L
I'm
Matt
suno.
So
let
me
talk
about
this
support
project
bravely,
so
this
activity
is
related
to
the
IGN
research
group,
because
support
is
open-source
software
enabling
content,
eccentric,
networking,
shishun
communication,
which
is
available
at
this
website,
and
the
objective
of
this
project
is
to
enhance
support
functions
such
as
extensive
wording,
engine
which
is
a
compatible
with
XI,
XI
and
X
message,
which
defined
in
this
draft
and
in
networking
management
and
so
on.
L
So
we
had
a
three
goals
in
this
hackathon
1
Y
is
2
implementing
XI
XI,
an
info
which
can
be
used,
discover
information
about
network
topology
and
the
in
network
cache.
The
specification
is
described
in
this
draft
and
the
second
ace
developing
transport
program
for
network
coding
in
coordination
ways
in
network
cashing
in
order
to
achieve
high
performance.
The
requirement
is
described
in
the
struct
and
third
solder.
It
is
to
implement
the
safety
code,
which
is
a
Python
package
to
enable
faster
and
easier,
safer
application
tools.
L
So,
as
our
achievement,
we
built
the
basic
function
and
for
a
mark,
and
we
enhance
the
safety
code
to
specify
the
optional
optional
out.
Here.
We've
had
a
flexp
and
testing
will
be
a
huge
org.
This
is
a
our
team
members
and
we
have
several
rings.
So
if
you
have
any
question,
please
ask
us.
Thank
you.
W
Hi
I'm
Karina
she's
come
from
NTT
communications.
My
presentation
is
about
adults,
interrupt
from
it
thoughts.
What
good
I
had
made
a
presentation
three
times
in
a
hackathon,
so
you,
however,
you
may
not.
You
may
not
know
about
what
is
Otto
trees
in
a
dot
either
the
doors
open
streets
Schulich,
so
it
standardized
the
signaling
or
dealer's
prediction
between
two
or
more
organizations.
W
Dot
protocols
is
consists
of
two
major
to
draft.
One
is
a
silicon
Arad
on?
Second
one
is
the
data
charitra.
However,
on
top
of
that,
we
are
tested
a
new
individual
draft,
which
is
a
outcome
of
the
last
IDF
hackathon.
Then
I
redid,
the
interoperability
testing
between
two
individual
implementations,
one
is
auto
source
of
the
other
end
is
proprietary.
In
our
we
made
a
demo
of
ceará
use
cases.
W
You
know
here's
what
we
got
we
get
done
so
here
the
matrix
of
the
result
of
internet
interoperability
testing.
Then
we
have
the
over
90
percentage
of
this
coverage.
Then
also
the
you
may
notice
or
orange
says
what
we
did
in
this
hackathon.
Then
we
tested
a
new
features
of
filtering
control
or
version
each
other,
even
in
attack
time.
W
Then,
as
I
said
Adi,
those
dots
is
about
DDoS
protection
requests
from
one
organization
to
another
organization.
So
what?
If
a
neuron?
What
is
a
disease
get
attack?
Then
they
ask
for
help
to
add
other
organization.
However,
in
current
specification
over
the
dots
to
each
other,
there
is
no
chance
to
convey
control
of
a
she
else
in
other
people's
organizations,
mainly
because
the
China
is
over
TCP.
W
W
Here's
what
we
learned
so
at
the
third
protocol
enables
of
mitigation
request
on
ACL
contour.
He
do
another
hostile,
Network
situations,
then
issues
funding
I,
keep
on
all
three.
A
cotton
were
all
addressed
by
the
effort
of
Raj
group
members.
Then
we
are.
We
really
feedback,
the
invitation
conduction
constellations
to
the
but
grid.
So
here's
a
mentee
members
and
it
is
remarkable
that
are
we
have
included
four
new
members
and
they
contributed
to
the
Oasis
project
thanks
so
much.
D
D
A
X
Hello,
I'm,
Teresa
and
I
present
taps
taps
is
for
transport
services
and
the
idea
is
that
we're
providing
an
abstract,
API
or
a
common
abstraction
on
top
of
different
transport
protocols
and
then,
for
example,
an
application
specifies
some
services,
some
transport
service
as
it
needs.
For
example,
it
needs
reliable
in
order
data
transfer
and
transport
system,
underneath
we'll
figure
out,
for
example,
whether
to
use
TLS
over
TCP
or
whether
to
use
quake,
or
they
can
also
be
other
transport
services
that
are
supported
and
other
transport
protocols.
X
Our
drafts
are
still
work-in-progress,
so
we
wanted
to
get
some
more
implementation
experience
and
we
worked
on
a
new
implementation
of
tabs
called
PI
attempts,
which
we
had
written
from
scratch,
based
on
price
and
as
in
IO
and
at
the
beginning
of
this
hackathon
pi
tabs
only
supported
TCP
and
TLS
UDP,
quick
and
SCTP
were
planned
to
support.
So
we
wanted
to
see
how
far
we
can
get
within
the
hackathon.
Now
TLS
1.3
works
with
our
implementation,
but
this
is
thanks
to
Python,
I,
think
I/o
and
the
underlying
open
SSL
library
that
actually
yeah.
X
So
we
just
provide
security
context
with
some
parameters
and
we
can
sense
ability,
force
and
then
TLS
works.
So
we
can
now
use
TCP
and
TLS
with
a
now.
Api
and
UDP
is
work
in
progress.
There's
code,
it's
not
merged
yet,
but
we're
working
on
it
and
we
have
an
idea
how
it
might
work
to
integrate
quick
into
our
prototype,
which
is
quite
exciting
and
for
SCTP
we
haven't
had
the
chance
to
actually
integrate
it.
Yet.
X
So
what
we
learned
is
that
finding
the
right
abstraction
is
non-trivial,
API
is
a
hard
right
and
even
within
Python,
as
in
kayo,
we
started
out
with
a
more
high
level
abstraction
that
worked
well
for
TCP
and
for
TLS
over
TCP,
but
then
we
had
to
eventually
basically
go
down
a
layer
within
a
sink
I.
Oh
and
now
we're
implementing
the
protocols
on
top
based
protocol
abstraction
what
we
call
it.
So
if
the
protocol
is
implemented
using
async
IO,
we
think
that
now
it
will
be
possible
to
integrate
all
those
transport
protocols
into
our
prototype.
X
Let's
see
what
we
can
do
about
protocols
where
this
is
not
so
easy
or
it
is.
This
is
not
the
case.
We
will
take
our
implementation
experiences
back
to
the
tabs
working
group,
so
team
members
are
myself
and
Philip
and
Max.
We
are
we're
all
at
at
you
Berlin
and
our
table
is
right
at
the
front.
So
if
you
have
any
questions
about
tabs,
you
can
look
at
our
github.
You
can
write
us
an
email
by
the
way
we're
all
first
timers
at
the
idea
of
hackathon
had
a
great
time.
Y
Y
Okay,
when
I
have
hours
for
school
picture
of
the
day
now
we
should
already
win
for
that.
So
these
gentlemen's
of
the
participant
had
an
important
role,
though
what
were
we
doing?
We
were
doing
cams
monkey
testing
on
the
routing
protocol,
which
is
probably
like
the
first
that
I've
seen
with
this
rift.
That's
a
specialized
protocol
for
special
data
center
topologies
metro
building,
cabling,
which
is
coming
out
becoming
more
and
more
important.
Y
Y
We
have
an
open-source
implementation,
so
one
of
the
implementations
actually
open
source
and
github.
We
generate
a
configuration
of
large
data
center
topologies,
actually
not
that
large
I
ran
the
whole
thing
in
AWS
everybody
in
the
cloud.
We
did
a
lot
of
testing
of
correct
initial
convergence
and
then
we
monkey
cows
tested
this
stuff,
which
means
we
were
just
breaking
fingerstick.
Y
M-More
there
was
a
lot
of
preparation,
so
we
already
came
with
the
framework
to
bring
all
these
things
up
and
there's
much
time
here.
Material
is
out
there,
including
like
videos
and
so
on.
So
that's
all
generated
out.
We
have
a
lot
of
cool
pictures,
so
the
tool
generates
you
know
shows
the
topology
is
what's
coming
up,
what's
breaking
and
so
on
different
resolutions
lot
of
SVG
stuff.
So
we
run
the
whole
thing
in
network
namespaces
I
think
we
started
with
containers.
There
were
far
more
tedious
to
get
the
stuff
done.
Y
We
had
some
kind
of
a
meta
configuration
which
actually
generates
a
yellow
configuration
which
runs
the
whole
thing,
so
it
generates
configuration
/
outer
all
kind
of
scripts.
To
start
stop
I
mean
you
you
in
namespaces,
and
you
have
you
know
dozens
of
those
nodes.
So
this
is
non-trivial
how
you
operate
in
such
a
topology.
Then
we
had
no
the
chaos
scripts
that
were
perturbing
the
thing
and
then,
of
course,
a
lot
of
checking
right,
because
you
have
a
lot
of
levels
that
the
protocol
verge.
Do
you
have
the
stuff
tie
broken?
Y
Y
Alright,
so
that's
a
typical
round
some
stage
you
break
a
couple
of
things.
You
fix
a
couple
of
things,
interesting
stuff,
so
that
so
what
did
we
test,
like
pinging
from
all
the
Leafs
to
all
the
Leafs,
where
the
notes
are
up?
Adjacencies
are
up
weather.
All
kinds
of
routes
are
installed
throughout
the
levels
necessary
to
actually
make
forwarding
work
more
ideas.
Are
there
again
lots
of
cool
pictures,
so
this
is
some
protocol
run
where
you
can
SVG
look
at
all
the
packets
flying
between
all
the
notes
and
actually
analyze.
Y
You
know,
what's
what's
coming
up,
what
isn't
so?
What
did
we
do
generate
those
topologies
run
them
flat
it
or
did
you
learn?
We
found
a
couple
of
implementation
issues
very
delicate
stuff,
as
it
mostly
comes
from
Charis
monkey
testing.
You
know
based
on
timing,
stuff,
no
protocol
specifications
issue
but
simple
implementations.
A
for
example.
We
have
v4
and
v6
support
and
we
found
that
when
one
node
was
fly
was
flowing
on
ipv4
in
one
direction.
Y
H
Z
Z
So
they
asked
me
to
go
through
my
use
case
how
DNS
SD
solves
it
and
my
experience
with
the
library
so
my
use
case
is
this
I'm
running
multicast
into
a
home
network,
and
the
idea
here
is
that
you've
got
a
multicast
capable
network
but
you're
within
and
the
wife
I
can't
support
multicast,
because
it's
broadcasting
Wi-Fi
there's
a
draft
about
it.
It's
kind
of
a
pain
so
with
what
we
do
is
discover
the
relay
and
the
relay
can
wrap
the
the
multicast
data
into
a
unicast
tunnel.
Z
So
the
bigger
context
of
this
is
that
I
brought
this
all
these
devices
there's
a
multicast
network
which
is
ingesting
traffic
from
the
internet
with
the
same
idea
with
the
a
unicast
encapsulated
tunnel
and
then
propagating
the
multicast
as
native
multicast
within
the
network.
So
it's
very
important
that
the
home
device
doesn't
reach
out
to
the
internet
when
there
is
a
loca
nearby,
multicast
enabled
network
and
instead
finds
the
local
relay
when
it's
there.
If
it's
not
there,
then
it
should
reach
out
to
the
Internet
to
to
collect
the
data.
Z
So
that's
my
basic
idea
and
there's
three
relevant
documents
that
were
addressed
as
part
of
this
work.
The
DNS
service
discovery,
of
course,
is
RSC
67-63
and
it's
got
great
insurers
that
work
very
well
doctor
who
maybe
used
a
little
bit
of
work
guys,
but
with
Stuart.
There
they're
walked
me
through
it.
It
was
fantastic
and
the
draft
about
the
the
discovering
over
the
Internet,
the
unicast
tunnel
for
a
multicast
source.
Z
There's
an
update
that
I'm
going
to
apply
to
the
draft
based
on
some
things
that
I
learned
during
the
hackathon
about
how
the
discovery
is
going
to
work
and
also
some
interesting
information
for
the
MMD
group
about
a
proof-of-concept
for
the
draft
about
the
Wi-Fi
multicast
problems
that
the
AMT
running
on
the
local
network
on
the
home
network
is
one
of
the
cited
workarounds.
And
so
it's
a
proof-of-concept
that
this
is
a
viable
idea.
Z
B
AB
Thanks
hello,
there
Rio
and
I
are
here
from
the
University
of
San
Andrews
we've
been
working
on
an
implementation
of
the
identifier,
locator
network
protocol,
which
we've
been
debugging
and
testing
and
will
be
released
in
in
May,
thanks
to
our
store
for
helping
us
be
here
and
have
our
demo
kit
just
a
little
bit
of
background
on
Island
P,
it's
a
change,
the
way
that
addressing
works
for
IP.
So
this
is
what
we
might
see
today.
We
have
a
Linux
box
and
we
see
that
an
interface
has
an
IP
address
attached
to
it.
AB
That
means
effectively
a
transport
session
is
bound
to
an
interface
which
can
be
a
little
bit
inconvenient
if
you
want
mobility
or
if
you
want
multi,
powerful
multihoming,
so
in
Ireland
P
we
try
to
change
the
way
that
addressing
is
done.
We
introduce
new
namespaces,
so
there's
a
specific
identifier
at
the
transport
layer.
That's
only
used
end
to
end
and
there's
another
namespace
allocator
at
the
network
layer.
That's
used
just
for
routing,
and
then
we
have
dynamic
bindings
between
them.
AB
So
a
node
identifier
can
be
dynamically
bound
to
one
or
more
locators
and
the
locator
can
be
dynamically
bound
to
an
interface
and
by
playing
around
with
those,
you
can
get
different
things
like
multihoming
and
mobility
and
a
mixture
of
those
things
if
you
want
to.
This
is
just
a
summary
of
the
the
prop
the
new
namespace
and
the
details
are
in
RFC
67,
42
48,
which
are
experimental
and
also
they
form
the
basis
of
the
spec
R
on
which
we're
building
and
testing
the
implementation
in
Linux.
AB
So
it
has
the
same
syntax
and
semantics
as
an
ipv6
routing
prefix,
which
means
ipv6
routers,
should
just
treat
an
island
P
packet
as
an
ipv6
packet
and
then
the
node
identifier,
which
is
the
lower
64
bits
of
the
128-bit
ipv6
address
space
that
is
used
at
the
end
system.
So
the
N
system
stack
has
to
be
updated
to
recognize
this
bit
and
to
deal
with
the
new
semantics.
AB
So
we
have
a
demo
and
it
will
be
showing
how
mobility
and
multihoming
can
be
achieved
together
by
binding
a
node
to
multiple
networks
through
the
dynamic
binding
between
a
need
and
an
l,
64
value,
and
we
have
mobility
by
having
a
node
move
between
networks.
This
is
the
little
main
testbed
we
have,
which
is
currently
at
the
back
and
it's
been
torn
down,
but
we'll
be
at
the
hack
demo
session
tomorrow
and
the
blue
nodes.
AB
The
correspondent
node
in
the
mobile
node,
our
Island
P,
enabled
and
they're
communicating
across
across
a
set
of
routers
that
are
ipv6.
Only
they
don't
know
about
Island
P
at
all,
but
what's
happening
is
that
the
mobile
node
is
moving
across
those
networks.
We
we
emulate
that
by
taking
interfaces
up
and
down
and
at
the
same
time,
there's
a
flow
running
using
iperf
between
the
CN
and
the
MN,
and
we
show
what
happens
that
the
flow
remains
fairly
stable.
AB
The
communication
continues
and
the
IPF
happens
to
be
an
ipv6
binary
hasn't
been
recompile
for
and
P.
Oh
okay,
the
grass
didn't
come
up.
You'll
just
have
to
come
to
the
dinner
tomorrow
to
see
what
happens
and
as
I
said,
we
are
at
the
demo
tomorrow
and
more
information
about
island
P
is
available
at
the
website,
and
that's
also
going
to
have
the
link
to
the
code
release
in
Linux
in
May.
Thank
you
very
much.
D
AC
Hi,
so
we
made
a
hackathon
for
the
low
loss,
low,
latency,
scalable,
throughput
architecture,
and
so
this
is
mainly
against
fighting
the
so
teeth
that
you
observe
on
your
bottlenecks.
Links
when
you
have
capacities
seeking
TCP
fills.
So
why
is
that
bad?
Well,
basically
build
up
queues,
and
so
it
increases
delay,
and
so,
if
you
do
want
to
do,
for
example,
ERP
our
workloads,
it
doesn't
work.
AC
Actually.
Last
year,
our
C
82
11
came
out
which
will
act
which
relax
it's
the
easy
on
meaning.
So
we
could
experiment
with
it
and
then
there's
a
bunch
of
drafts
that
leverage
all
of
this
to
define
architecture,
how
to
communicate
easy
and
feedback
to
senders,
how
to
build
a
panic
that
does
did
that.
Does
this
and
stays
fair
with
other
TCP
flows?
AC
And
then
there
are
there's
a
final
draft
that
discusses
what
we
call
the
DC
TCP
/
requirements,
which
were
specified
here
four
years
ago
and
that's
a
bit
the
sad
side
of
the
story,
so
DC
TCP
has
been
in
Linux
from
us
forever.
Now
the
RSC
82
11
is
there.
So
that's
nice
jacket
lecturer
as
mostly
notable
for
a
while,
and
so
this
is
static
is
well
known.
Understood,
equity
Chien
has
been
prototype
in
an
older
version
of
Linux.
AC
Gqm
was
released
publicly
in
July
2016
and
while
the
requirements
are
text,
so
what
did
we
do
for
the
hackathon
from
there?
So
we
wanted
to
take
all
these
mini
blocks
and
we
get
to
a
point
where
we
had
environments
so
that
open
source
people
Goods
X
Vivat,
not
only
behind
closed
doors.
So
we
got
a
bunch
of
scrip
to
set
up
a
VM
bdub
the
custom
kernel
in
extensions
with
sample
apps
that
showcase
how
to
use
it.
AC
If
you
want
to
use
TCP,
Prag
or
DC
TCP
or
something
else,
we've
upgrade
a
data
acquisition
to
the
letters.
Letters,
revision
of
the
draft
and
stammered
across
a
couple
of
issue
if
generate
the
generic
receiver
flow,
then
segmentation
of
route,
which
are
extremely
important.
If
you
want
to
get
high
performance,
what
we
started
working
on
the
implementing
I
filled
out
for
requirements
for
TCP,
and
we
actually
have
a
small
prototype
for
quick
that
is
compared
with
al
first.
So
that's
that
showcase
of
January
kiddies,
so
I
said
before
accurate.
AC
Is
he
and
his
annex
is
a
small
modification
to
the
tcp
wire
protocol,
and
so
it
modifies
bits
in
the
headers
and
it
turns
out
that
the
meaning
of
those
bits
is
extremely
ty
goes
extremely
deep
in
the
sack.
So
we
had
to
make
fixes
in
the
lower
level
of
the
kernel,
and
we
actually
expect
expect
hardware
to
have
some
issue
with
it.
Not
all
tcp,
probably
not
a
requirement.
We
need
it
and
is,
as
part
of
this
is
well
for
quick.
AC
For
example,
you
have
an
embedded,
isn't
feedback
Condors,
that's
propagated
in
the
leg
frame,
so
implementing
implement
needs
for
trick
is
way
easier.
And
finally,
once
you
get
low
latency,
while
you
can,
for
example,
for
quick
start
to
do
smart
thing
with
the
built-in
streams,
so
there
were
five
of
us
here
from
different
places.
We
had
we
pushed
most
of
the
code
and
get
up
scripts
to
be
the
VM
kernel
changes
sake,
some
test
codes
that
might
need
to
be
cleaned
up
a
bit
and
then
there's
the
right
project
for
the
usual
reference.
AD
AD
So
our
target
during
the
hackathon
is
a
Verity
in
the
performance
of
Satish.
So
for
someone
who
don't
know
Satish
sect
a
she
is.
The
protocol
stack
is
running
on
the
low-power
wireless
constrained
device,
which
is
used
in
the
industrial
scenario
and
and
a
player
is
running,
a
music
stack
trying
to
connect
it
to
the
internet
and
so
sick
tisha
is
a
active
working
group
and
it
already
published
several
or
RFC's
like
RC.
Eighty
one,
eighty
and
eighty
four
eighty,
and
also
several
drops
that
we
are
right
now
is
trying
to
evaluate
the
performance.
AD
So
what
we
plan
to
do
first,
to
implement
a
TMS
drop,
that
we
are
using
open,
publishing
projects
and
open
diversity
is
a
open
source,
implantation
of
a
sick
dish
and
it
has
already
implemented
the
whole
sector
stack.
So
we
uses
this
deck
to
replaces
the
Mac
layer
by
the
latest
ms
dropped
and
to
for
the
benchmark
projects
is
a
open
source
as
well,
and
it
contains
who
party
the
front-end.
AD
AD
If
you
open
the
link,
it
will
look
like
this
and
on
the
first,
the
chart
is
a
the
end-to-end
average,
the
latency
and
the
second
one
is
the
end-to-end,
the
reliability
and
the
third
one.
Is
the
the
cell
mess
a
massive
cell
usage,
which
is
a
indicator
for
the
traffic
elodea
network
and
for
our
so
you
probably
looked
a
difference
if
you're
open
the
link
right
now,
but
which
is
fun,
so
we
are
really
more
targeting
the
and
reliability
rather
than
latency
yeah.
This
was
a
sick
t-shirt
targeting.
So
what
are
you
learned?
AD
Is
we
found
the
latest?
A
massive
draft
version
is
better
than
the
first
one
in
three
aspects
network
forming
time
and
as
I
sell
location
efficiency
and
the
anti
on
releve,
reliability
is
also
increased
and
we
also
found
a
problem.
Is
the
synchronization
takes
long?
That's
because
channel
hopping
features.
AD
AE
Either
everyone,
my
name,
is
Ken
and
I'm
going
to
talk
about
integration
of
six
dish
which
tank
they
was
just
talking
about
thanks
for
that
with
the
riot
operating
system.
So
our
specific
goal
was
to
tighten
the
integration
of
open,
WS
n
as
an
implementation,
the
reference
implementation
into
riot
on
some
useful
hardware,
and
so
this
is
occasional.
Look
for
us.
We
said
some
restarting
work
from
earlier
Sprint's,
that's
valuable
nonetheless.
AE
Nonetheless,
in
particular,
we're
trying
to
get
it
to
run
on
open
mote
B
hardware,
which
is
the
reference
platform
that
also
eventually
will
get
us
to
fifteen
f/4g
in
long
range
and
also
to
improve
the
integration
at
layer
four,
so
that
we
can
run
our
applications
on
top
of
it.
So
let
me
explain
that
a
little
bit
better
so
on
the
left,
you
can
see
how
how
things
are
fit
together
on
the
top
part
of
the
work.
AE
So
what
we've?
What
we've
been
able
to
do
on
the
top
end,
is
to
use
the
fit
IOT
lab,
which
is
very
convenient
for
us
to
be
able
to
just
run
experiments.
And
so
you
can
see
here
on
this
Wireshark
output.
We
have
the
co-op
output,
a
co-op
request
and
response,
as
well
as
the
frames,
so
we
kind
of
sort
of
had
this
working
in
the
run-up
to
to
the
IETF,
but
we
were
able
to
confirm
that
things
were
working.
So
this
is
this
is
very
good
and
so,
in
terms
of
top-down
results.
AE
From
from
working
at
the
sock,
API
layer
and
and
one
thing,
I
should
also
say
about
the
sock
API
is
that
already
it's
implemented
implemented
for
riots,
natives,
a
stack
as
well
as
lightweight
IP
and
EMV
six.
So
it's
really
cool
that
we
were
able
to
graft
into
open
WSM
to
be
able
to
to
work
with
it
at
that
level.
So
I
was
able
to
rebase
the
work
that
I've
done
and
resolve
some
compiler
issues
and
prove
that
things
are
still
working
on
the
IOT
lab
modes.
AE
So
this
will
be
valuable
to
us
in
the
future
and
also
is
great
for
me.
I
look
far
away
for
most
of
the
other
contributors,
so
to
have
some
face-to-face
time
is
great
in
terms
of
bottom-up
results.
We've
discovered,
so
this
is
running
on
TI
hardware
that
we
need
to
learn
more
about
TI
hardware,
and
so
we
start
from
an
abstract
interface.
We're
trying
to
understand
the
data
sheets,
so
working
with
the
timer
and
the
radio
API
is
prove
to
us
that
we
have
more
work
to
do.
AE
We
need
to
improve
the
quality
of
the
driver,
so
not
as
positive
as
we
might
hope,
but
still
there's
there's
a
lot
to
be
said
for
learning
and
the
face-to-face
time
so
just
to
wrap
it
up.
Here's
the
the
four
members
of
the
group
and
we
do
have
some
work-in-progress
PRS,
that
you
can
take
a
look
at
on
the
riot
website.
A
AF
Hello,
everyone,
I'm,
Hauge
and,
and
me
and
Schenk.
We
have
been
working
on
ripple
over
ble
during
the
last
30-some
hours,
so
our
motivation
was
a
little
bit
looking
into
Bluetooth,
Low
Energy
and
the
integration
with
the
internet,
and
there
is,
of
course,
some
standardization
from
the
Bluetooth
SIG
and
also
from
there
art.
AF
That
can
be
scanned
and-
and
so
all
we
did
is
take
some
ripple
metadata
information
like
rank
and
dodeca
ID
and
so
on
and
implement
it
or
put
it
into
this
advertisement.
Data
so
nodes
can
actually
based
on
the
advertisement
data
they
receive
and
do
decisions
on
who
to
connect
to.
In
our
case,
we
decided
to
have
the
strategy
that
notes
simply
decide
their
parents,
so
they
only
connect
to
their
parent
actively.
AF
So
our
results
and
we
got
it
more
or
less
stable.
In
solution
than
our
table,
there
was
six
nodes
and
a
we
configure
them
in
a
way
that
they
only
accept
one
connection.
So
basically
in
topology
with
the
6-up
thing,
it
was
dead
before
10
seconds,
maybe
and
during
this
talk
of
all
the
torts
and
we
improve
performance
by
about
600
percent,
so
we're
at
a
minute
right
now
and
yeah.
So
all
that's
left
is
a
lot
of
debugging
and
what
we
learned
and
the
general
concept
is
actually
quite
straightforward
and
simple.
AF
It's
for
advertisement
data
do
some
metric
calculation
and
do
a
connection
decision
and,
of
course,
there's
some
problems
we
accounted
and
for
one
is
the
neural
net
Bluetooth
stack
that
we
use.
Basically,
we
have
a
multi-layered
state
machines,
because
you
have
the
general
connection
between
notes
and
then
you
open
up
channels,
and
on
top
of
that
you
basically
have
the
ITP
traffic
and
so
to
synchronize
all
these
state
machines.
There
seem
to
be
a
lot
of
facts
on
the
code
and
it's
very
hard
to
to
get
a
consistent
state
over
multiple
notes.
AF
But
of
course
I
guess
everybody
in
the
room
agrees
open
source
software
is
very
nice.
In
this
regard,
because
we
have
the
full
sources
of
nimble's,
we
can
really
go
down
in
the
detail
of
the
glucose
deck
and
also,
of
course,
of
riot
and
the
network
stack
to
make
sure
that,
where
they
step
in
consistency
coming
from
and
yeah
of
course,
one
another
problem
and
sniffing
for
ble
advertisements
in
this
room,
particularly
it's
a
quite
fun,
but
not
really
Apple
yep.
That
was
it.
So,
thanks
for
your
audience,.
A
G
Y
Yeah,
thank
you
so
welcome
everybody,
I'm
Vincent's
a
few
words
about
the
sweet
connect
project.
It
deals
with
packet
loss
recovery
using
FEC
codes
for
whatever
collects
the
key
benefits
of
this
approach.
Respect
to
also
block
coding
techniques
like
which
woman
wrapped
off
incline,
that
is
to
reduce
the
added
latins.
That's
the
key
points
and,
of
course,
this
work
has
a
strong
relationship
with
the
Korean
efficient
network
communication,
so
true,
and
of
also
energy,
as
well
as
some
work
worth
doing
on
this
oil
CFX,
his
team
at
vwg,
so
the
Academy
all
worth
twofold.
Y
First
of
all,
we
want
to
design
open
source
free
and
reference
correct
for
signing
window
cards
in
order
to
facilitate
one
tests
and
expectation
of
those
technologies,
and
the
second
goal
is
to
change
our
generic
API
for
such
FEC
clicks
and
that's
also
very
important
to
take
advantage
of
this
åkesson
project
to
try
to
see
problems
in
this
internal
draft.
So
this
is
only
our
second
åkesson.
Y
We
start
in
last
time,
so
Bangkok
we-we-we
I
think
we
had
great
treatments
because
we
managed
to
fix
the
few
problems
mistakes
or
in
Precision's
in
this
generic
APA
internet
drafts.
We
almost
finish
the
encoder
part,
which
is
a
modest
of
the
work
we
are
still
going
on
with
tests,
but
basically
it
starts
working.
We
also
managed
to
do
this
reference
demo
applications
and
let
you
test
this
and
test
and
stress
the
nikolic.
We
started
a
Python
wrapper
and
we
also
start
in
some
more
or
less
the
decoding
parts.
So
that's
great.
Y
D
J
Good,
thank
you.
Okay.
This
is
report
from
the
wishi
hackathon
activity
and
we
see
stands
for
work
on
IOT
semantic
and
hyper
media
interoperability.
This
work
originates
from
a
tingling
research.
Crew
works
up.
We
had
two
years
ago,
axel
here
in
Prague,
and
we've
been
in
these
five
hackathons
so
far.
Turning
some
of
those
great
research
ideas
into
practice.
J
J
This
time,
we've
focused
quite
a
bit
on
coral,
the
new
constraint,
restful
application
language
and
how
that
can
be
used
to
bring
IOT
and
hypermedia
closer
together.
Also,
we
were
exploring
on
the
semantic
space.
What
is
the
right
level
and
what
is
the
right
place
for
expressing
semantics
in
you
know
in
IOT,
hyper
medium
and
also
in
particular,
work
and
improving
the
semantics
in
Europe
and
resource
description
for
iOS.
J
Some
of
our
achievements
from
the
weekend
there
is
now
new
open
source
parser
from
text
form
of
coral
select
published,
but
it
is
going
to
be
out
there
soon.
Also
for
the
riot
OS,
there
is
basic
choral,
encoder
and
parser,
and
also
a
module
for
the
constrained
ir
is
available
and
open
source
because
everybody
likes
use
case
and
examples
there's
a
set
of
those
described
in
in
our
ether
pad.
J
J
Also,
there
is
a
Python
coral
implementation,
which
was
now
updated
to
the
latest
version
of
coral,
and
also
now
has
a
nice
visualization
hinting.
So
you
can
make
coral,
perhaps
more
human
readable
in
some
cases
of
there
is
some
work
on
improving
resource
directly
implementations
and
also
making
the
life
addendum
TD
generator
that
we
have
been
using
in
the
previous
hackathons.
You
use
the
latest
version
of
thing
description
and
be
more
compatible
with
the
rest
of
the
ecosystem.
J
We
were
about
a
dozen
people
working
on
this
and
we
thinking
probably
next
time
we
don't
need
a
bigger
table.
It
was
sometimes
bit
hard
to
squeeze
in
and
running
in
the
same
table.
Luckily,
we
wanted
to
do
that.
I
would
also
have
a
three
newcomers.
So
thank
you,
everyone
for
joining
and
if
you
want
to
have
more
information,
there's
the
entry
point
just
follow
the
links
and
semantics
and
you'll
find
a
lot
of
information.
Thank
you.
AG
AG
There
is
still
a
little
part
of
it
that
is
not
finished
and
he's
delaying
a
little
bit
the
standard,
and
that
part
is
about
simulcast,
which
is
the
capacity
to
not
only
send
several
streams
since
I'm
single
connection
bundled,
but
with
different
resolution
of
the
of
the
video
in
the.
In
our
case,
it's
only
on
the
sender
side,
where
we
try
to
send
different
resolution.
The
receiving
side
is
only
receiving
one
of
the
stream,
depending
on
the
bandwidth
condition
any
necessitates
to
have
a
media
server
in
the
middle.
AG
So
it's
a
very
specific
case
where
you
actually
not
only
have
a
browser,
but
you
have
the
protocols,
but
you
also
need
to
have
the
media
server
people
to
come
together.
So,
for
the
first
time
here
we
said:
okay,
let's
have
everybody
around
the
table,
the
client
sites
or
the
browser,
vendors
and
the
SFU
developers.
So
we
can
fix
the
bug
all
together
and
have
a
real
interoperability
test
and
write.
AG
You
know
fix
bug
all
together
right,
so
we
wanted
to
create
bugs
fix
them
and
generate
some
tests
that
could
be
run
later
on
to
make
sure
that
there
is
no
regression
in
the
different
tests
which
the
WPT
for
the
w3c
and
a
new
test
suite
called
kites
that
we
using
to
test
the
protocol
and
interoperability
for
the
IETF
parts.
This
was
the
biggest
hackathon
for
IETF.
That
was
the
biggest
hackathon
for
the
WebRTC
group
as
well.
AG
We
were
19
people,
including
13
people,
listing
only
web
RTC
as
the
reason
why
joining
the
hackathon,
we
had
all
the
browser,
vendors
actually
coming
and
joining
developer
of
the
web,
RTC
group
and
so
on
and
so
forth.
We
had
three
of
the
tech
lead
of
the
media
server
open
source
in
the
world
out
of
the
six
or
seven
areas
of
there.
So
it
was
quite
a
says:
everybody
had
a
different
goal.
Yes,
a
few
vendors
wanted
air
safety
to
be
the
best.
AG
AG
There
were
some
really
nasty
bugs
where,
instead
of
allowing
bandwidth
to
be
used
for
the
high
resolution,
you
would
actually
reserved
bandwidth
for
the
low
resolution,
which
is
not
exactly
what
you
would
like
to
do
intuitively
some
very,
very
nasty
corner
cases
that
were
only
found
because
both
the
server
people
and
the
browser
were
here.
We
ended
up
with
a
document
showing
the
status
of
all
the
browsers
as
well
of
the
status
of
all
the
SFU.
That
will
allow
us
to
make
a
road
map
for
the
next
meeting
until
morale.
AG
A
G
AH
So
we
the
target
of
the
platform
with
the
BPP
Fido,
the
project
and
also
the
p4.
In
addition
to
that,
the
target
fact
out,
we
have
actually
several
function
to
do
that,
so
the
the
target
would
be
limited
to
about
this
limited
time.
The
the
we
already
understand
the
the
current
draft
not
mentioned
clearly
so
them
through
the
hackathon.
We
are
trying
to
figure
out
what
a
missing
and
what
we
need
to
do
in
the
future.
So
this
is
what
we
got
done
and
we
have.
AH
The
list
of
the
function
should
be
target,
but
we
focus
on
this
too.
It's
a
basic
to
GTP
and
gdp
to
this
a
v6.
So
this
ugly
handwriting
illustration,
is
they
only
the
guidance
for
our
colleague,
so
I'm
40
for
that,
but
our
team
successfully
implemented
this
idea
great.
So
thank
you
very
much.
So
what
we
hara
here
is
something
the
coding
or
their
service.
Ik
will
found
to
find
the
application
inside
of
the
segment
ID,
even
for
the
other
side
of
the
standardization
body.
So
it
is
a
possible.
AH
AH
AA
A
AA
AA
AA
AA
So
what
we
learned,
maybe
OpenSSL
needs
to
rewrite
API
for
percent
shade
with
education
next
scientists,
so
the
team
duffel
invest,
could
Iran
Rahul
you
rush
myself
and
West,
really
one
young
guy,
Kishan
and
also
we've
got
one
person
who
has
been
working
on
and
I
just
mentioned
this
briefly.
His
name
is
finish
together
with
GEICO
keifa
they've
been
working
on
improving
the
HTTP
four
five
one
module
for
that.
It's
in
revealed
by
the
Drupal
developers,
so
we
had
one
young
person.
L
So
luckily
we
have
tools
such
as
wreck,
gem
and
p.m.
and
others.
We
also
had
to
take
into
consideration
the
version
of
Ruby
and
node
running
in
the
background
entering
Travis
testing
and
deployment
also
unit
testing,
which
we
should
be
compatible
with
OS
running
at
VM
level.
Wireshark
was
used
for
pre
check
and
post
check
and
also
of
the
code
deployment.
As
usual.
We
also
have
to
debug
a
lot
on
phaaze.
L
We
have
to
debrief
ourselves
a
lot
for
the
idea
graph
so
that
we
can
catch
up
with
the
new
change
which
has
been
going
on
recently.
So
so
can
you
switch
on
the
author
so
also
for
to
the
Hazara
weaknesses.
We
should
take
into
account
the
possibility
to
decrypt
user
sessions.
Usually
oz4
is
injecting
typography.
This
is
still
widely
used
during
post
handshake
on
key
exchange.
We
have
not
seen
that
cause
for
still
being
advertised,
such
as
in
net
SSH
and
node.
Q
L
Well,
a
new
switch,
so
as
mentioned
there
are
too
many
deployment
of
as
such
still,
shipping
with
asti
for
and
I
believe.
We
still
have
lots
because
I
invite
open
source
there
was
to
support
this
to
move
forward
in
the
future.
So
thanks
to
a
race
Cola
for
needful
at
iesg
and
camera
Falls,
the
IDF
draft.
Okay,
your
slider
up.
L
That
we
have
a
Bruno
panel
who
also
work
for
the
ITF
mobile
app
working
group,
so
we
had
move
on
the
ITF
mobile
app
to
make
more
improvement.
We
also
had
Jackie
a
looky
who
worked
a
lot
on
the
node
to
deprecate
Asif.
All
the
team
is
now
a
group
of
twenty
guys
working
on
different
working
groups,
all
from
Marshalls.
A
Okay,
anyone
else
with
the
presentation,
anyone
we
missed
all
right,
well,
great,
a
huge
thanks
to
everyone
who
did
present
here
if
you
didn't
present,
but
you
have
some
results
you
want
to
upload
or
if
you
want
to
you
know,
reupload
your
presentations.
If
there's
some
stuff
you
you
didn't
put
in
or
whatever
you
know,
that's
perfectly
fine,
it's
great
to
have
that
as
a
resource
and
as
a
record
of
what
happened
here.
Also
having
the
recording
from
the
meet
echo
session
is
is
great.
A
One
of
the
things
I'm
trying
to
think
about
now
is
a
lot
of
you
shared
links
to
github
repos
and
various
open-source
projects
that
you
were
working
with.
I
think
it'd
be
good.
If
we
thought
about
how
to
you
know,
make
that
easier
for
people
to
find,
and
so,
if
you
have
some
ideas
about,
you
know
how
to
make
it.
You
know,
of
course
you
can
go
to
the
wikis
and
you
can
sort
do
everything,
but
maybe
ways
we
can
tie
the
work.
A
You
did
back
to
the
drafts
and
you
know
just
make
all
that
easier
for
people
to
navigate.
Let
me
know,
because
that's
something
I'm
starting
to
think
about
and
talking
with
the
iesg
about
one
again
thanks,
everyone
I
know
we
ran
over
time,
but
I
really
did
want
to
give
everyone
a
chance
to
share
their
results.
A
I'm
also
looking
for
suggestions
about
what
we
change,
so
that
this
event
continues
to
go
well,
I
assume
it
will
continue
to
grow,
I
sure
hope
it
does,
and
so
we
need
to
think
about
scaling
it
and
things
that
work
better,
so
I'm
open
to
suggestions
there
to
catch
me.
Now,
after
this
or
anytime
I'll
be
around
til
Friday
with
that
I
just
want
to
thank
again
Oracle
Oracle
cloud
infrastructure.
It
was
fantastic
that
they
sponsored
the
hackathon
this
time
that
really
helped.
A
This
is
getting
more
expensive
because
we
take
up
more
basse
need
more
tables
need
more
food.
We
certainly
could
have
used
more
beer,
so
you
know,
if
any
of
you
have
a
the
ability
to
help
sponsor
I.
Welcome
that,
let
me
know
and
or
talk
to
Alyssa
and
we'll
figure
out
how
to
try
to
make
that
happen.
But
thanks
to
Oracle
for
that,
and
thanks
to
me,
tech,
oh,
they
did
a
great
job
and
alright
that's
it
from
from
my
side
thanks.