►
From YouTube: IETF111-HACKATHON-20210723-1900
Description
HACKATHON meeting session at IETF111
2021/07/23 1900
https://datatracker.ietf.org/meeting/111/proceedings/
A
A
A
Okay,
it
looks
like
the
the
numbers
have
sepaled
welcome
everybody
to
the
closing
session
of
the
ietf
111
hackathon.
I
hope
everybody
had
a
great
time
all
the
feedback.
I've
gotten
is
that
it's
been
a
good
hackathon
and
I'm
sorry.
We
had
to
do
another
one
of
these
online
and
I
hope
we
can
do
one
in
person
again
soon,
but
in
the
meantime,
welcome
to
the
iutf
hackathon
closing.
A
So
we
have
we've
done
all
of
this.
The
agenda,
the
note.
Well,
what
we
mostly
have
to
do
today
is
the
project
presentations
so
again
we're
looking
for
short
presentations,
of
which
it
looks
like
we
have
about
10
posted
talking
about
what
problem
you
tried
to
solve
what
you
accomplished
during
the
week
and
some
highlights
of
what
happened.
Lessons
learned:
what
feedback
you
have
to
bring
back
to
working
groups
and
what
impact
on
on
open
source
and
other
sdos
you
have.
A
So
here
is
the
url
for
the
page
to
submit
your
presentations.
If
you
haven't
already,
I
will
try
to
accommodate
you
if
you
post
them
now,
but
we're
going
to
go
through
the
presentations
that
are
there
first
and
then
we'll
see
what
we
have
and
that's
it.
So
let
me
go
to
let
me
change
my
sharing
and
share
the
page
with
that.
Let's
see.
A
A
Okay,
so
that
should
be
it
and
let's
go
to
the
first
presentation:
we're
going
to
start
with
bmwg
container
benchmark.
A
Benchmarking,
let
me
grab
that
okay.
I
guess
I
need
to
share
that
all
right.
A
A
B
Okay,
so
yeah,
so
let's
start
so
hi,
so
my
name
is
kj's
son,
so
from
sunshine
university.
B
So
today,
so
I'll
talk
about
our
hackathon
project,
bmwj
containerized,
infrastructure
benchmarking,
so
our
hackathon
project
have
been
worked
from
the
itef
106
and
we
maintain
our
main
goal
is
to
implement
container
benchmark
environment
with
the
virus,
various
networking
models
and
compare
network
performance
with
various
impacting
factors.
B
B
Yeah,
so
this
is
the
contents
in
the
hackathon
wiki
page,
so
our
team
is
organized
from
the
sunshine
university
in
south
korea
and
then
co-working
with
itunesf
and
ipwave
team
together
at
the
itef
korea
forum.
So
the
second
term
for
benchmarking,
our
main
goal
is
to
verify
packing
lost
pokemons
between
different
acceleration
technologies
which
are
bpp
and
then
dpdk,
obs
and
sriov,
and
we
estimate
that
the
bpp
and
ustpdk
use
different
pre-switch
entities
and
the
packet
processing
mechanism.
B
Actually,
this
work
was
done
at
our
previous
hackathon
actually,
but
at
the
time
our
test
was
not
correct,
because
our
understanding
of
container
talking
model
was
not
good,
so
we
tested
it
again
and
then
there
was
the
numa
testing
report
from
other
committee,
which
is
the
vine
puff
of
the
anuk
projects.
So
we
tested
following
their
test
on
methodology
and
verifying
in
our
testbed
environment.
B
Okay,
so
I
briefly
talked
about
the
draft
update
discussion
and
then
move
to
the
testing,
so
in
our
previous
draft,
so
we
found
that
there
was
some
clear
definition
and
the
figures
for
describing
container
networking
models
also,
first
with
a
survey
and
then
discussion
in
this
hackathon
and
the
really
classified
container
networking
model
with
the
pre-switch
aspect
also
for
describing
performance
impact
factor
in
the
containerized
environment.
B
B
Yeah,
so
the
end
for
your
testing,
so
first
we
made
our
test
bed
with
different
acceleration
technologies,
so
we
used
the
three
accelerations,
which
is
the
overstep
tpdk
bpp
and
the
sriv.
B
So
for
all
case
or
all
cases,
packets
from
the
external
traffic
generator
are
bypassing
corner
space
and
the
old
packet
processing
is
handled
at
the
user's
user
space
level,
so
we're
firstly,
testing
performance
from
baseline
and
then
especially,
we
tried
to
test
packet
loss
between
the
obs,
tpdk
and
then
bpp
for
making
the
multi-interfaced
container.
We
use
the
multi-cni
to
make
the
v-host
user
interface
and
the
srv.
B
We
have
driver
interface
and
the
full
track
traffic
generator.
So
we
used
t-rex
on
the
parameter
server
and
the
between
testing
server.
There
was
no
switch
or
hub
between
the
two
servers.
Okay
next
slide,
please.
B
So
you
can
see
the
four
different
core
allocation
scenarios
in
the
slide,
so
the
older
and
the
oldest
scenario
we
tested
performance
with
the
obs
and
then
vpp.
Okay,
next
slide,
please
yeah.
So
these
tables
are
shown
the
the
specification
and
the
configuration
of
our
testbed
environment.
B
So
for
testing
we
use
the
10g
nic
card
with
srv
functionality
and
for
cpu
location,
so
it
was
enabled
to
provide
the
higher
threading
technology
and
then
you
can
see
the
other.
The
configurations
are
in
the
tables.
Okay,
next
slide,
please
yeah.
So
the
as
the
lizards.
These
graphs
are
the
result
of
the
pneuma
affinity
test.
So
the
upper
two
graphs
are
throughput
result
in
the
bpp
case
and
the
lower
two
graphs
are
the
shown
obvious
obvious
pdk
result
so,
and
each
graph
includes
the
four
scenario
with
a
different
number
affinity.
B
So
compared
with
the
two
different
networking
technology,
the
performance
results
show
much
better
than
the
obvious
tpk
result
and
in
both
cases,
results
show
that
allocating
same
pneuma
node
or
we
switch
course
and
then
contain,
of
course,
is
more
effective
for
the
probe.
Improving
the
network
performance.
B
Okay,
next
slide,
please,
okay,
so
the
this
is
for
the
benchmarking
summary
actually
for
packing
loss
measurement.
So
we
cannot
make
any
useful
result
because
the
whenever
we
try
to
measure
with
the
packing
loss
with
110
percent
of
the
throughput
based
on
the
the
methodology.
But
our
results
were
not
the
consistency.
B
So
we
have
some
issues
raising
of
about
the
t-rex
are
having
some
problem
to
draw
packets,
but
at
this
time
we
cannot
say
it
is
exactly
tracks
problems.
So
after
this
hackathon,
so
we
will
keep
our
work
on
these
issues
and
then
and
then
the
pneuma
affinity
testing
was
the
research
that
the
most
re
we
expected
and
it
has
some
performance
degradation
compared
with
the
buying
puff
test
report.
But
it
was
some
reasonable
because
we
tested
in
the
less
resources
that
or
then
there's
environment.
B
However,
the
even
we
consider
some
scaling,
but
the
performance
was
quite
worse,
which
is
some
some
kind
of
terrible
to
use,
especially
in
case
of
the
small
size
packet,
so
the
we
we
will
check
against
our
the
testing
methodology
and
the
environment
and
configuration,
and
then
we
we
improved
our
some
testing
result
so
the
based
on
this
hackathon.
B
So
our
future
work
is
to
bring
that
experience
and
the
discussion
by
updating
our
draft
and
at
the
same
time,
so
we
want
to
make
the
some
opportunity
for
discussing
with
other
the
related
communities
for
the
container
benchmarking.
B
So
from
that,
so
we
hope
that
it
may
bring
the
some
improvements
of
our
bmw
work.
Okay,
next
slide,
please,
okay!
So
this
is
the
last
slide
and
then
these
are
the
participation
of
our
team
and
then
it
is
made
up
the
sushi
university
and
then
thanks
to
the
ita
korea
forum,
to
make
the
joint
hackathon
local
event
on
and
also
or
with
it
nsf
and
ipwave
team.
So
so
the
follow
the
git
link
in
in
the
link.
B
So
you
can
check
the
old
some
environment
configuration
related
files
for
our
hotel
project
so
and
then
we
hope
to
meet
itf
at
the
on-site
hackathon
as
soon
as
possible.
Okay.
So
this
is
the
end
of
my
presentation.
Thank
you
for.
A
Okay,
I,
the
next
step,
is
i2
nsf.
Thank
you,
kj
nice
presentation
and
the
next
one
is
i2
nsf,
which
I'm
going
to
bring
up
now
so
get
ready
and
after
that
will
be
ipwaves,
so
you
can
be
on.
A
C
Okay,
hello
larry,
and
this
is
the
jumper
jump.
Can
you
hear
me.
C
Okay,
let's
start,
let
me
start
okay,
yes,
okay!
This
is
the
hackathon
project
for
i2
nsx
working
group.
So
the
champions
are
me
and
my
phd
student,
patrick
ringa.
Let's
move
on
next
slide,
please.
C
So
our
hackathon
project,
we
want
to
demonstrate
the
latest
version
of
nsf
monitoring,
interface
data
model
and
also
security
policy
translator.
This
time
we
demonstrated
two
features.
The
first
one
is
automatic
data
model
mapper
for
the
itunes
security
policy
translator.
The
second
feature
is
to
support
real-time
monitoring
data
visualization
of
via
web-based
monitoring
next
slide.
Please.
C
This
figure
shows
the
i2
nsf
framework,
the
basically,
the
analyzer
collects
an
asset
for
monitoring
data
from
nsf
below
the
figure
and
then
an
analyzer
visualize,
the
monitoring
data
to
detect
the
security
attack
and
then
that
information
will
be
delivered
over
application
interface
and
the
security
controller
will
update
or
generate
a
new
policy
according
me,
and
then
the
new
policy
will
be
delivered
to
nsf
nsfs
next
slide,
please
the
basically
what
got
done
on
this
custom
project
left-hand
side.
This
is
the
architecture
of
security
policy
translator.
C
C
So
the
first
tech
task
is
automatic
data
model.
Mapper,
like
this
figure
right-hand
side.
We
have
two
data
models.
The
first
one
is
consumer-facing
interface
data
model,
which
means
the
high-level
young
data
model
and
also
below
that
we
have
nsf
facing
interface
data
model,
which
means
loadable
young
data
model,
the.
Basically
we
want
to
map
between
variables
of
high
level,
young
module
and
loadable
young
module
in
terms
of
young
trees.
C
C
The
second
task
is
a
real-time
monitoring
data
visualization.
So
you
can
see
this
web
browser
displays
real
time.
You
are
a
filter,
filtering
resources,
data.
The
upper
graph
is
the
company
resources
such
as
the
cpu
memory
and
desk
utilization,
and
then
the
below
figure,
the
input
and
the
output
data
traffic.
C
So
what
we
learned
this
time
and
then
we
learned
the
security
policy.
Translator
can
be
exported
by
automatic
data
model
mapper
to
possibility
pre-requisite
setting
for
translation
and
also
the
monitoring
and
data
of
visualization
can
support
the
monitoring
of
multiple
nsfs
in
near
time
in
terms
of
computing
resources
and
traffic.
C
Finally,
with
this
monitoring
data
visualization
feature,
we
can
perform
the
analysis
of
an
nsf
behavior,
such
as
security
attack
detection.
Next
slide,
please
next
step.
We
also
enhance
the
security
policy
translation
transition
later
in
terms
of
context
for
the
grammar
auto
construction.
C
Basically,
this
one
can
partially
aid
the
generation
of
low
level
young
xml
file,
using
this
context-free
grammar
based
on
facing
interface
data
young
model,
and
also
we
can
improve
nsf
i2sf
analyzer,
with
machine
learning
to
detect
the
new
security
attack,
and
then
we
can
update
or
create
a
security
apology.
C
Finally,
we
want
to
migrate
from
fair
interface,
the
virtualization
to
container-based
virtualization
for
itunes.
F
next
slide.
Please.
C
C
So
we
built
the
team
with
sunshield
sunshield
university
and
actually
and
kanga
national
university,
along
with
my
skk
team,
so
our
itunes
have
a
team
work
with
the
ipwave
and
the
benchmarking
working
group
hackathon
project
this
time
next
slide.
So
we
got
a
multiple
supporter,
such
as
the
sponsors
from
an
nrf,
iitf,
tta
and
also
itep
korea
forum
next
slide.
C
A
All
right,
thanks
for
the
presentation
and
the
next
one,
yes,
the
next
one
up
is
ipwave
and
after
that
will
be
bmp
yang,
so
you
can
get
ready
and
I'm
going
to
set
up
the
bpwave
presentation.
Now,
I'm
sorry,
the
i2
of
the
ipwave
presentation.
A
A
A
Okay,
ipwave
is
ready,
go
ahead,
paul.
C
Okay,
thank
you,
lady
okay.
This
is,
if
you
waiver
basic
a
protocol
project.
The
champions
are
me
and
the
doctor,
even
a
christian
next
slide.
C
So,
basically,
the
last
hackathon
project
item
one
one
zero.
So
we
demon
and
directed
the
feasibility
by
udp
based
mobility,
information
sharing.
So,
as
you
can
see,
two
robots
of
vehicles,
which
means
ai
on
robotics
r1
model,
so
we
allowed
the
message
exchange
between
two
rubber
beakers
using
udp,
so
the
ccm
and
the
hdmcm
is
a
corporation
contact
method
for
regular
driving
and
ecm
is
urgent
emergency
contacts,
messages
to
detect
some
obstacles
in
roadway
and
and
then
advertise
their
message
to
neighbor
speakers
to
avoid
accident
next
slide.
Please.
C
This
time
we
want
to
demonstrate
ipv8,
6,
neighbor
discovery
message
for
sharing
a
mobility
information
among
our
beakers.
Using
new
nd
option
label
discovery
option.
It
is
called
the
weaker
mobility
information
option
vmi,
so
it
contains
a
gps
coordinate
such
as
a
latitude
and
longitude
speed
and
direction.
C
So
we
have
several
of
opponents,
including
linux,
loss
of
cat
and
also
rfcs,
and
the
internet
draft,
so
for
the
through
approval
concept,
poc
use
the
law
for
cap
based
icmp
version
6
to
design
and
test
the
new
nd
vmi
option
next
slide.
Please.
C
This
figure
shows
our
new
nd
vmi
vehicle
mobility
information
option,
so
this
option
is
described
in
the
following
two
draft
next
slide.
Please.
C
C
C
What
got
done
we
tested
the
iphone
6
neighbor
discovery
in
the
law,
socket
based,
icm
2006
by
open
source
code,
and
then
we
demonstrated
low
socket,
is
a
feasible
approach
to
test
a
new
iphone
6
nd
option.
In
the
last
time
we
tried
to
use
the
corner
the
modification.
C
C
C
So
what
we
learned
you
can
see.
Two
laptop
computers
are
regarded
as
two
speakers,
so
we
introduced
the
ma
message:
exchange
for
sharing
a
fee
capability
information
so
to
support
this
feature.
The
existing
network
discovery
process
needed
to
be
updated
by
defining
a
new
option
type,
a
different
mechanism
for
sending
labor
advertisement
and
the
neighbor
cash
dedicated
for
this
new
option.
Next
slide,
please
next
step,
we
will
apply
our
implementation
of
nd
vmi
option
to
logo
vehicle
that
is
the
ai
on
robotics
r1
model.
C
C
We
also
demonstrated
our
implementation,
so
you
can
see
our
demonstration
video
clip
at
youtube
next
slide,
please.
C
So
this
hackathon
project
is
a
collaboration
work
with
the
sunshield
university
and
the
national
university
and
congo
national
university.
This
is
the
portal
of
our
active
class
members,
our
ip
wave
custom
team.
I
work
with
the
itunes
theft
and
pm
wg
and
hackathon
teams
this
time
next
slide.
Please.
C
This
is
our
teams
for
ipua,
i2nsf,
bmwg,
hackathon
teams.
Thank
you
for
your
attention.
A
A
Any
questions
for
for
paul
about
ipwave,
okay,.
D
It
looks
like
nick
yeah
you're
in
the
queue.
E
Hi
there
thank
thanks
so
much
for
sharing
the
work.
I'm
I'm
not
familiar
at
all
with
the
vehicle
information
sharing,
but
it
seems
pretty
exciting.
I
I
was
just
curious
if
you
either
this
week
or
in
general,
have
done
some
work
on
the
security
and
privacy
considerations,
it
seems
like
potentially
sensitive
revealing
data
that
might
be
shared
and
could
potentially
be
misused
in
different
ways.
I
was
just
curious
if
you
had
some
initial
work
on
that.
C
Okay,
all
right:
that
is
a
good
question,
so
the
information
of
the
location
speaker
may
be
a
privacy
issue,
but
our
goal
is
to
share
the
location
vehicle
to
support
collision
avoidance
for
driving
safety,
so
also
we
can
consider
authentication
mechanism
to
protect
this
message
in
near
future.
Thank
you.
A
F
Yeah,
no,
I
attempted
to-
and
it
took
a
moment
so
this
is
essentially
going
to
construct
the
same
issue
as
we
have
with
automated
number
plate
readers
so
anywhere
that
you
would
consider
an
automated
number
plate
reader
to
be
an
invasion
of
privacy.
F
Anyone
with
an
appropriate
antenna
will
have
the
same
thing
and,
unfortunately,
authentication
on
its
own
is
going
to
be
difficult
to
make
adequate,
because
anything
that
needs
to
do
collision
avoidance
will
need
to
have
that
authentication
information,
and
I
I
see
this
as
a
very
difficult
mutual
authentication
problem
to
to
actually
get
right
without
having
some
substantial
cryptographic
setup
before
you
can
actually
exchange
data
and
in
a
collision
avoidance
scenario,
the
latency
required
to
do
that
sounds
problematic.
A
Okay,
okay,
so
thomas,
is
this
more
questions
for
this,
or
are
you
the
bmp
and
yang
guy.
G
Yep,
can
you
hear
me?
Yes,
we
can
okay
excellent,
so
I'm
thomas
from
the
bmp
and
yang
hackathon
group
next
clip
please.
G
Great
thanks
a
lot
so
this
time
we
were
focused
installing
on
testing
the
performance
on
bmp,
the
the
impact
the
bgp
monitoring
protocol
has
on
on
bgp
and
on
cpu
memory
usage
on
the
route
processors
there,
especially
in
in
context
of
bmp
arches
cited
in
out
and
the
local
with
path
marking
and
additional
tlvs.
G
We
were
testing
if
there
is
a
difference
when
the
bmp
session
towards
the
collector
is
stable,
unstable
and
also
while
it
makes
a
difference
if
the
peers
are
flapping
or
not,
and
also
we
were
curious
about
the
bmp
route
monitoring
messages.
If
you
can
actually
measure
the
delay
across
the
bgp
propagation
and
it's
if
it's
always
ensured
that
we
have
the
complete
view
of
the
bgp
rip
state
next
time.
Please.
G
So
we
were
using
pmact
as
a
as
a
for
data
collection,
cough
car
as
a
message
broke
through
it
for
the
time
series
db
and
pivo
for
actually
accessing
the
data
and
wireshark
for
the
bmpd
sector
and
exobgb
for
generating
the
layouts.
There
is
a
link
with
the
small
tutorial
how
you
can
set
up
your
own
environment
next
slide,
please.
G
So
this
is
the
the
network
we
used,
so
we
have
on
the
very
left
hand,
side.
The
exhibit
should
be
generating
the
vpn
v4
routes,
where
the
the
juniper
route
is
basically
the
route
reflector
and
we
were
meshing,
adjacency
rip
out.
G
So
basically
the
ultra
advertisement
towards
the
worldwide
route
reflector,
where
we
are
measuring
adjacency
in
pre-policy
and
adjacency,
rip
out
post
policy
and
then
importing
the
routes
on
two
different
and
pspe
router
on
ios
on
cisco,
s60
and
ios
xr,
the
testbed
was
generating
1
million
vpn
v4
out
as
fast
as
possible
and
on
the
route
reflector.
We
were
supporting
adjacency
hip
in
out
as
well,
also
that
the
path
marking
and
tlv
drafts
next
slide,
please.
G
So
what
we
achieved
this
time
is
that
the
the
automation
continued.
So
we
we
continued
to
improve
the
testbed
so
now,
including
also
different
scenarios,
where
we
have
like
a
stable
in
stable,
pgp
environment
versus
also
initiating
the
the
the
peerings
again.
So
we
can
compare
what's
the
difference
between
the
two
we
are
measuring
through
young
push
the
cpn
memory,
consumption
on
the
pgp
process
and
have
now
also
a
comparison
between
the
four
different
operating
systems.
G
Next
slide,
please
speaking
about
time,
stamping.
So
these
time
stampings
have
been
improved
in
in
the
wire.
So
this
time
also
ios
6e
is
able
to
expose
metrics
with
seconds
and
milliseconds
time.
Stamping,
and
we
are
looking
forward
on
the
huawei
via
beside
that.
Millisecond
is
supported
as
well.
That's
important
because
we
want
to
measure
the
end-to-end
propagation
delay
impact
when
bmp
is
enabled
next
type.
Please,
we
were
mostly
using
young
push.
Different
young
models,
depending
on
the
operating
system,
were
available
on
some
operating
systems.
G
It's
not
yet
available
to
measure
that
on
a
process
level,
so
we
are
used
screenscaping
cli
to
get
the
data
next
slide.
Please!
So
here
just
a
comparison
on
the
left
side.
Basically,
when
on
the
central
route
reflected
either
pmp
is
turned
on
or
turned
off,
you
see
a
slight
difference
in
the
bgp
propagation
delay,
so
the
scale
is
how
much
prefixes
within
a
second
are
being
propagated
across
the
network.
G
On
the
right
hand,
side
you
can
see
basically
the
same
the
comparison
bmp
on
and
off
on
the
central
route
reflector,
but
compared
to
the
left
side
on
the
right
side,
we
were
basically
flapping
all
piers
on
the
out
reflector
next
slide.
Please
now
same
thing,
but
now
collapsed
on
on
one
graph,
so
the
difference
between
bmp
on
off
and
flapping
next
slide.
G
Here
we
are
looking
at
the
impact
on
the
route
reflector
in
in
the
middle
in
terms
of
memory
and
cpu
cpu,
so
you
can
see
in
the
middle.
So
at
thousand
we
were
basically
flapping,
the
pier,
so
you
can
see
basically
before
and
after
the
flapping
the
difference
if
pmp
is
turned
on
or
turned
off
in
terms
of
cpu
and
memory
utilization.
So
you
see
there
is
not
much
of
a
difference
next
slide,
please,
since
we
were
measuring
the
the
end
to
end
delay
across
the
route
reflector.
G
This
is
basically
on
the
juniper
router.
We
can
see
how
much
the
impact
is,
or
basically
when
we
leverage
that
bmp
turned
off
turned
on
on
the
central
route,
reflector
what
impact
it
has
on
on
the
remote
routers,
and
there
is
virtually
no
difference
in
terms
of
memory
and
cpu
next
site.
Please
also
the
same
on
the
emperor's
pe
routers.
G
We
don't
see
much
of
a
difference.
It
all
depends
very
much
on
when
you're
on,
in
a
congestion.
How
quick
the
the
data
collection
can
keep
up
next
slide,
please!
G
So
what
we
learned
again,
it's
not
the
first
time
that
good
preparation
and
test
automation
was
very
important
on
the
test
paid
side.
We
are,
let's
say
we
we
noticed
we
need
to
improve
in
terms
of
scalability.
A
A
E
Hi,
everyone
good
to
see
all
of
your
projects
I'll
be
talking
about
big
bang.
Today,
we've
we've
talked
about
this
at
a
few
hackathons
in
the
past.
So
if
you
guys
go
to
the
next
slide
I'll
talk
about
what
it
actually
is,
this
is
an
open
source
project,
a
sort
of
scientific
toolkit
used
by
a
lot
of
researchers
for
studying,
collaborative
communities,
open
source
projects
and
a
lot
of
internet
governance
groups.
E
If
we
go
to
the
next
slide,
we
talk
about
the
plan
this
time
around,
we
had
a
slightly
smaller
group,
but
we
were
trying
to
add
new
data
sources,
getting
mailing
lists
from
a
couple
other
fora,
3gpp
and
igf,
which
were
of
interest
to
a
couple
of
researchers
and
then
we're
trying
to
connect
the
data
tracker
and
maybe
github
as
new
data
sources
and
that's
to
get
at
the
sort
of
research
question.
We
want
to
look
at
more
this
time
around
about
organizational
affiliations
of
participants
and
the
email
domains.
E
So
if
you
go
to
the
next
slide,
we
can
start
looking
at
some
graphs
yeah,
and
this
is
two
ietf
mailing
lists-
do
and
the
itf
privacy
discussion
list
where
the
x-axis
is
the
number
of
participants
and
the
y-axis
is
the
number
of
messages
and
that
each
point
is
is
an
organization.
We've
inferred
the
affiliation
through
the
data
tracker
and
then
you
can
see,
and
I've
tried
to
classify
them
roughly
into
corporations,
academics
and
nonprofits,
and
so
you
can
see.
E
Maybe
what
is
not
too
surprising
that
many
corporations
are
sending
more
participants,
but
also
that
occasionally,
a
non-profit
or
an
academic
group
might
only
send
one
or
two
people,
but
send
quite
a
few
messages.
So
that's
just
a
couple
of
mailing
lists
and,
and
hopefully
we
can
apply
that
to
a
larger
set
of
data
on
the
next
slide.
E
We
can
see
some
other
work
done
on
3gbp,
so
this
is
a
similar
kind
of
analysis,
except
just
using
the
email
domain,
rather
than
using
the
data
tracker
and
document
authorship
to
figure
out
affiliations
and
and
they've
also
looked
at
time
zones
kristoff
maps
the
time
zone
of
the
mail
center
to
see
where
participants
are
sending
from
that
might
be
something
we
should
try
to
apply
to
ietf
data
as
well,
and
then
the
next
slide
has
another
graph.
E
This
is
looking
at
participants
based
on
email
domains
and
to
look
at
the
sort
of
standard
deviation.
How
how
much
of
a
range
there
is
within
a
single
domain
about
how
many
messages
are
sent
to
a
particular
mailing
list,
and
so
said,
benthal
is
working
on
this,
hoping
that
we
can
maybe
do
some
machine
learning
to
do
some
inferences
based
on
different
domains
and
how
they,
how
they
send
different
patterns
of
messages
to
a
particular
mailing
list.
E
So
those
are
our
graphs,
hopefully
we'll
be
able
to
build
on
that
in
the
in
the
coming
months.
On
the
next
slide,
we
have
a
few
lessons
learned,
yeah
or
relearned.
I
appreciate
that
was
one
of
the
other
presentations.
All
data
is
messy.
We
always
see
this.
It's
always
a
surprise,
even
though
we
see
it
every
time
on
on
things
like
organization
and
affiliation.
Those
are
very
messy
concepts.
E
The
data
tracker
thanks
to
everyone
who
helps
run
the
data
tracker
that
has
some
great
data.
The
api
is
providing
us
useful
things
about
affiliation
and
document
authorship.
We
also
connect
this
at
github,
but
that
seems
a
little
less
promising
lots
of
people's
affiliations
are
changing
over
time,
particularly
at
ietf,
and
so
that
would
be
useful
to
look
into
and
then
one
challenge
about
the
remote
hackathon.
We
were
all
different
countries,
it's
much
harder
to
debug
someone's
python
issue
from
a
long
way
away.
E
Finally,
on
the
last
slide,
these
are
team
members
and
you
can
follow
this
link
about
the
research
project
and
we
welcome
both
contributions
and
also
research
questions.
If
there's
something
you've
wanted
to
know
about
itf
participation,
we
have
some
tools
and
one
and
one
look
at
that
data.
A
Okay,
thanks
thanks
nick
any
questions
for
nick.
D
Yeah,
I
actually
had
a
question.
This
is
charles.
You
know
the
question
I
had
was
in
regard
to
github
and
I
don't
know
if
you
saw
any
instances
where
people
were
having
automated
messages
from
like
when
you
do.
D
E
I
I
haven't
run
into
that
problem,
yet,
although
I
am,
I
am
prepared
for
it
because
yeah
there
there
is
often
that
case
where
we're,
assuming
that
an
email
address
always
refers
to
a
person,
but
yeah,
like
updates
from
updates
from
github,
might
look
like
they're
from
a
person
but
actually
be
from
a
robot,
and
we
also
see
that,
like
there's
often
an
outlier
in
many
of
the
larger
analyses
for
sort
of
like
administrative
email
addresses
from
the
ietf
right
whenever
they
send
an
announcement
of
a
new,
a
new
draft
publication
or
a
meeting
request.
E
Something
like
that.
There's
often
an
outlier
where
it
looks
like
the
most
active
person
is
internetdrafts.itf.org.
So
that's
that's
one
of
those
messy
things.
Do
we
have
to
clean
up
one
of
those
outliers.
B
A
All
right,
thanks
nick
appreciate
that,
and
next
we
have
id
visualizations
somebody
up
to
present
for
that.
Okay,
here
we
go,
go
go
for
it
just
say
next,
when
you
want
the
next
slide.
I
Cool,
thank
you.
My
name
is
kasira
and
ietf
newcomer.
Can
you
go
to
next
slide?
Please
so
problem
I
tried
to
tackle
was
to
visualize
how
people
have
contributed
through
internet
drops
over
the
time
and
how
different
working
group
has
evolved
over
the
time.
For
that
I
used
a
program
called
ghost,
which
is
used
to
visualize
how
people
have
worked
on
a
github
repository,
and
what
I
try
to
do
here
is
to
simulate
a
git
log
from
internet
draft
data.
Can
you
go
to
next
slide.
I
And
there
should
be
a
nice
image,
but
it's
missing,
but
that's
all
right,
so
I
used
bbxml3
reference
data,
so
that
has
all
this
information
graph
dates
and
the
draft
name.
So
I
derived
the
all
the
information
as
commit
authors
as
commit
committers
and
then
the
from
the
id
name.
I
put
those
drafts
into
a
directory
structure.
That's,
for
example,
for
an
ieta
for
working
roof
draft.
It
will
be
go
under
ietf.
I
I
And
can
you
click
on
the
play
button?
Please.
I
From
the
videos
you
can
see
like
one
one
big
cluster
occurring
for
individual
submissions
and
then
based
on
a
number
of
graphs
for
a
working
group,
there
are
different
colors
assigned
for
different
working
groups
and
they
are
forming
different.
They
are
from
in
their
own
clusters,
and
can
you
go
to
the
next
slide?
Please.
I
I
Thank
you,
and
if
you
go
to
that
github
project,
then
there's
a
list
of
links
for
different
work
groups
from
from
the
beginning
of
date,
from
bpxml
data
on
how
the
those
were
group
areas
has
evolved
all
the
time
and
there's
another
video
I
generated
for
2021
for
all
the
drop
submissions,
including
individual
ones,
and
the
other
working
group
based
submissions
to
see
how
that
evolves
and
how
people
have
contributed
over
the
time
and
that's
it.
Thank.
A
You,
okay,
okay.
Sarah
thank
you
for
thank
you
for
this
and
and
welcome
to
ietf
participation.
Nick
you
have
a
comment.
E
Yeah,
I'm
I'm
just
really
excited
to
see
other
researchers
working
on
the
itf
data
itself
on
on
the
drafts.
That's
that's
great.
Have
you
looked
at
the
data
tracker
api
or
can
we
can
we
help
connect
you
to
that.
A
A
Thanks
again
quesara
and
I'm
sorry,
I
forgot
to
say
that
the
next
step
is
asdf,
and
here
is
the
af
asdf
presentation
already
for
you.
So
carsten
go
for
it.
H
A
A
A
H
Okay,
sorry,
that's
okay,
yeah!
So
one
of
the
formats,
one
of
the
modeling
formats
that
people
here
in
the
itf
are
probably
familiar
with,
is
yang.
So
that's
one
target
for
for
the
converter
work.
There
is
a
w3c
activity
called
web
of
things.
There
is
a
microsoft
thing
called
dtdl
and
so
on.
H
So
the
the
point
of
sdf
is
to
be
useful
as
a
hub
between
all
these
formats,
so
that
you
can
take
a
wot
description
converted
into
sdf
and
then
convert
it
into
a
microsoft
description
and
so
on,
so
that
that's
the
background
slide
two
please
and
the
the
objective
this.
H
The
objective
for
this
week
was
to
to
bring
together
the
people
working
on
the
various
converters
and,
of
course,
all
these
converters
have
a
common,
have
a
set
of
common
or
similar
components
that
have
to
do
with
ingesting
and
processing
and
generating
sdf.
So
it's
useful
for
people
to
talk
to
each
other,
and
we
also
wanted
to
produce
some
tangible
results.
H
H
So
that's
our
target
organization
that
that
we're
trying
to
do
things
for
the
the
main
problems
this
week
were
number
one.
We
have
way
too
many
programming
languages
and
platforms.
Actually,
we
are
not
aware
of
two
projects
that
use
the
same
language.
H
So
it's
not
like
everything
here
is
written
in
python.
Somebody
is
actually
using
python,
but
everybody
else
is
using
something
else,
and
that's
of
course,
also
related
to
the
fact
that
we
have
some
some
constraint
devices
that
may
want
to
make
use
of
that,
and
that
may
be
hard
in
python,
and
the
other
issue
was
that
everybody
is
on
vacation
this
week.
So
this
was
a
particularly
hard
week
to
get
people
to
join
slide.
Three,
please!
H
So,
despite
all
this,
what
we
actually
got
done
was
we.
We
now
have
a
better
shared
understanding
of
how
sdf
actually
works,
and
we
have
various
improvements
in
the
individual
converters
that
were
brought
to
to
the
headphone
and,
most
importantly,
we
have
those
prototypes
of
the
continuous
integration
models
modules
and
those
will
allow
using
github
actions
for
the
ci.
So
you
can
clone
or
fork
the
1dm
repository
and
throw
your
own
stuff
in
there
and
see
what
happens,
and
I
think
that
that's
a
really
useful
thing
to
have.
H
In
the
long
run,
we
have
learned
to
look
more
at
round
tripping
as
a
test
strategy,
and
one
issue
with
round
tripping
is
that
after
a
round
trip
you,
it's
really
hard
to
decide
whether
you
still
have
the
same
thing,
so
you
need
some
comparison
function
as
well.
We
also
had
some
interaction
with
the
the
w3c.
H
They
have
a
meeting
this
week
and
probably
will
have
another
meeting
next
week
and
and
we're
trying
to
integrate
some
some
of
their
work
with
our
work
slide.
4,
please.
H
And
the
next
thing
we
found
is
that
there
are
three
things
that
that
are
difficult
in
computer
science,
and
one
of
that
is
naming
things
and-
and
we
hit
by
that
as
well.
So
we
we
have
to
do
a
little
bit
more
work
about
name
spacing
and
using
json
pointers
for
that
and,
finally,
a
piece
of
new
work
that
we
probably
will
take
to
the
working
group
is
this
sdf
designed
for
dynamic
action
objects.
H
We
already
can
do
this
that
that,
in
the
ace
working
groups,
authorization
information
format,
but
we
have
to
make
sure
that
that
similar
things
are
possible
in,
in
the
other
components
working
together
and
in
particular
we
also
have
to
talk
about
errors,
because
if
you
have
physical
interaction,
you
have
unexpected
things
happening
slide
five.
H
So
this
was
a
small
group
of
four
people:
first,
two
first
timers
among
them,
but
we
also
could
stand
on
the
shoulders
of
of
the
people
of
at
least
two
people
who
were
on
vacation,
arie
and
jana,
whose
tools
were
used
and
which
are
linked
in
in
the
various
notes
we
are
taking.
And
in
this
slide,
so
we
were
six
people,
four
of
them
present
and
made
a
little
bit
of
progress.
A
All
right,
thanks,
carsten
and
again
pleased
to
see
a
couple
of
first
timers
there.
A
Okay,
that's
all
the
presentations
I
think
we
have.
Does
anyone
think
you
posted
a
presentation
that
I
did
not
bring
up
brendan.
A
I
don't
see
it
here,
but.
H
A
Okay,
I
I
don't
see
it
in
the
hackathon
111
project
presentations,
yet.
F
Okay,
it
was
listed
on
the
on
the
closing.
I
guess
it
didn't
get
uploaded.
A
Okay,
do
you
want
to
be
feel
free
to
talk
about
what
you
guys
did?
If
you
have
slides
you
can
email
me,
I
can
get
them
up
quickly
and
if
not,
you
can
just
talk.
F
Yeah
I
can,
I
can
get
those
slides
emailed
over
to
you
right
away
in
just
a
moment.
A
Okay,
I
can
do
that
refresh
my
mail,
you
send
them
to
the
hackathon
list
or
what.
F
Oh,
I
sent
them
to
barrylibacomputer.org.
F
A
F
Get
going
with
that,
so
it
was
software
and
firmware
updates
for
iot
devices.
The
hackathon
plan
that
we
started
with
was
to
get
software
and
firmware
to
the
well
item.
One
get
software
and
firmware
to
a
device.
Now
for
constrained
iot
devices.
The
plan
is
to
use
lightweight
m2m,
with
oscore
security
for
co-app
and
for
regular
iot
devices,
use
teep
for
trusted
app
update
on
tee.
A
F
Then
the
idea
was
to
secure
the
update
using
the
suit
manifest
and
then
use
the
new
software
firmware
next
slide.
Please.
F
So
we
identified
a
few
issues
with
the
t
protocol
which
have
been
posted
to
the
tp
github.
F
We
started
work
on
integration
of
oz
core
into
lachan,
for
which
is
a
lightweight
m2m
server
and
which
is
a
lightweight
m2m,
client
and
and
that's
still
in
progress
tested
registration
with
the
wakama
client
to
the
lashan
server
using
oscore
and
there's
a
couple
of
links
to
the
github
for
for
getting
this
done,
and
then
we
also
started
suit
integration
in
mcu
boot
and
that's
still
in
progress.
F
So
there's
a
few
bits
that
are
are
going
on
there
and
you
can
see
the
different
parts
that
we've
worked
on
in
the
in
the
the
repos
that
are
listed
there.
F
Next,
please,
so
we
learned
a
lot.
We
had
great
interactions
despite
the
online
nature
of
the
event,
but
of
course,
regular
work
disturbs
the
flow
and
our
our
plan
was
too
ambitious
as
always
and
secure
boot
loaders
turn
out
to
be
substantially
more
complex
than
you
might
think,
and
the
testing
of
boot
loaders,
as
it
turns
out,
is
substantially
different
than
protocol
testing.
F
And
then
we
want
to
thank
matt
gamor
for
suggesting
lightweight
m2m.
As
a
theme,
it
was
a
great
way
to
sort
of
get
more
people
involved
and
work
with
different
people
and
so
for
a
wrap
up.
We've
got
the
the
various
team
members
and
there
were
some
additional
people
helping
out
from
a
variety
of
other
working
groups
and
yeah.
Well,
I
guess
in
the
case
of
lightweight
mtm,
more
than
just
other
working
groups
and
that's
about
it.
A
Yeah,
how
are
they
so
any
questions
for.
A
Brendan
okay,
I
guess
that
does
anybody
else
have
a
presentation
that
I
did
not
show.
A
D
Yeah,
I
just
wanted
to
get
on
to
say
thanks
to
everyone
for
for
drawing
the
hackathon
and
and
very
fantastic
job
running
it.
Thanks
really
appreciate
that
I
I
was
had
another
standards
meeting
going
on
the
same
week,
so
I
wasn't
able
to
be
as
involved
in
the
hackathon,
as
I
typically
like
to
be,
but
it's
it
was
great
to
be
able
to
see
so
much
great
stuff
happen,
and
I
wanted
to
show
off
my
hackathon
shirt
too.
A
Very
cool
and-
and
I
guess
we
will-
we
should
thank
icann
again
for
their
support
in
sponsoring
and
thank
all
of
you
who
participated
then
you've.
You
really
made
the
hackathon
work
and
and
I'll
say
once
again
that
charles
and
I
are
both
listed
as
co-chairs
of
this
hackathon,
but
charles
is
the
one
who
does
all
the
work
on
this.
So
thanks
so
much
charles
for
putting
these
things
together
and
having
done
so
from
the
beginning.
A
D
Certainly
my
pleasure-
and
hopefully
I
don't
know
we'll
see
we'll
find
out
pretty
soon.
I
guess
what's
going
to
happen
for
112,
but
we'll
have
a
hackathon
one
way
or
the
other
so.