►
From YouTube: IETF 105: HotRFCs
Description
HotRFCs are lightning talks for the IETF community. A HotRFC session at IETF 105 will be held at 2200 UTC on 21 July 2019.
A
We
have
a
very
full
agenda,
so
my
name
is
Aaron
Falk.
This
is
the
hot
RFC
talk
series
I'm,
happy
that
you
all
have
come.
This
is
I,
think
the
fourth
or
maybe
fifth
time
that
we've
run
this.
The
idea
of
this
session
is
to
have
a
series
of
introductory
talks
to
give
folks
who
are
doing
something
new
in
the
IETF,
a
way
to
get
the
word
out,
and
so
it's
especially
good
for
things
that
cross
areas
or
are
new
to
the
ITF,
maybe
coming
in
from
the
IRT,
F
or
boss.
A
A
So
if
you've
made
dinner
plans,
I
think
you'll
still
be
able
to
to
make
them,
but
the
way
that
we're
going
to
do
that
is
through
very
strict
time
limits,
so
we're
using
the
larger
time
management
method,
and
so,
at
the
end
of
four
minutes,
if
the
speaker
hasn't
concluded,
you
should
hear
this
sound
at
which
point
I'd,
like
everybody,
in
the
room
to
start
clapping
and
thank
our
speaker
very
good
guys,
have
got
the
right
idea.
Now
all
that
comes
down
to
my
ability
to
push
the
button
correctly.
A
Let's
see
on
the
material
site,
you
should
find
all
the
slides
that
you're
going
to
see
tonight
as
well
as
abstracts,
that
summarize
the
talks
and
include
the
coordinates
for
finding
out
more
information.
So
there's
not
gonna
be
any
time
for
a
Q&A.
The
it's
really
intended
to
be
something
that's
very
concise.
Give
you
the
flavor
or
something
new
move
on
to
the
next
thing.
I
hope
you
find
it
interesting
and
so
Brian.
If
you
would
come
up
our
first
speaker
and
I'll,
get
your
slides
up.
B
B
C
B
B
So
what
use
the
service
type
requested
on
IP
address
to
forward
packets
at
the
network
edge
we're
not
talking
about
the
core
of
the
network,
abstract
a
service
as
something
called
a
service
action
type,
which
is
a
small
integer
indicating
the
service
is
a
very
generic
idea
near
the
user.
The
packets
are
routed
by
the
service
action
type
and
not
by
the
address
and,
of
course
before
you
to
ask
the
obvious
question.
Ip
reach
ability
is
one
of
the
available
services,
which
is
how
it
turns
out
to
work
backwards.
B
So
here
are
something
examples
of
service
action,
types,
reach
ability
on
ipv6
reach,
ability
on
ipv4
discovery,
service,
computation
service,
a
storage
service
and
there's
a
list
of
others
in
the
draft.
Packets
are
in
a
format
derived
from
ipv6.
This
is
a
low-level
protocol,
but
there's
a
service
action
type
instead
of
a
destination
address,
because
users
don't
care
about
destination
addresses,
and
there
are
other
goodies
that
look
like
a
lair
violation,
because
actually
this
thing
is
intentionally
a
lair
violation.
There's
some
toy
code
is
probably
not
worth
looking
at
more
important.
B
B
Why
are
we
talking
about
this
in
edge
deployments
of
IOT,
in
particular
the
physical
MTU
and
the
bitrate
Oh?
In
some
cases,
extremely
low?
That's
well
known
so
packet
size
matters
a
lot
he
edge
routers
the
edge
routers
may
be
constrained
themselves.
That
has
other
implications.
Besides
the
problem
of
bandwidth
and
bitrate
header,
decompression
and
header
compression
header,
decompression
use
resources.
B
128
bit
addresses
use
memory
so
for
a
certain
class
advertised
there
is
a
strong
case
for
both
reducing
packet
size
and
reducing
the
header,
decompression
and
compression
overhead.
What
why
are
you
doing
by
using
shorter
addresses
routing
on
the
shorter
addresses,
not
transmitting
unnecessary
bytes
and
avoiding
any
fancy
compression
decompression
algorithms
at
the
level
of
processing
the
IP
header?
B
How
you
define
an
address
length
in
within
a
domain
all
addresses
within
the
domain
are
assumed
to
have
a
prefix
in
front
of
that
of
128
minus
n
bits
and
there's
lots
of
bit
bytes
and
headers
that
you
can
get
rid
of.
If
you
think
about
it,
and
we
use
a
flexible
header,
encoding
octet
to
tell
the
lowest
layer
of
software
how
the
header
has
been
encoded.
B
So
you
might
end
up
with
a
complete
ipv6
header
that
consists
of
a
low
length,
NextEra
hop
limit
and
a
truncated
destination
address,
because
it
turns
out
in
the
right
conditions,
all
the
other
bits
and
pieces
in
the
header
intact.
Any
use-
and
those
are
the
bits
that
you
expect
the
this
constrained
router
to
look
at
where
six
low
working
group
on
tomorrow
and
same
side,
meeting
discussion
in
not
Adam
on
Wednesday.
The
end.
A
D
So,
what's
a
main
target
problem:
first,
the
traffic
in
real
networks,
especially
metro
network,
is
always
heavily
unbalanced.
Although
we
have
ecmp
or
you
can
build
such
kind
of
technologies,
surely
they
don't
really
work
very
well
and
second,
the
traffic
is
always
changing
so
fast
and
traditional
traffic
engineering
technology
cannot
adapt
the
traffic
in
real-time.
So,
with
the
ICANN
technology,
we
can
provide
guaranteed
networ
load
balance
in
real-time
millisecond
level
and
as
well
as
SLE
assurance
and
high
availability.
D
E
Hi
so
Adam
and
Bob
and
I
are
working
with
various
international
standard
setting
bodies
and
government
agencies
on
a
really
critical
problem.
Go
next
slide
yeah.
So
it's
just
this.
A
small
unmanned
air
vehicle
can
very
rapidly
get
close
to
a
crowd
or
a
piece
of
critical
infrastructure
without
being
observed
on
route
and
some
public
safety
person
armed
with
only
something
that
he's
likely
to
be
carrying
needs
to
be
able
to
quickly
identify
that
thing.
Is
it
a
friend?
Is
it
a
fall?
Can
we
look
up
its
flight
plan?
E
Maybe
can
we
establish
communications
with
its
operator
and
know
for
sure
that
it's
the
operator
of
that
aircraft
that
we're
talking
with
this
problem
of
mapping
a
physical
location
to
a
trustworthy
identity,
seemed
to
me
related
to
the
identifier
locator
split
in
the
internet
and
where
we're
trying
to
go
between
a
logical
location
and
an
identity,
and
so
we're
looking
at
the
use
of
hip
to
do
this
and
I'm
gonna?
Let
a
hip
expert.
Take
the
rest
of
the
time.
F
Speaking
with
Stu
and
Adam,
we've
worked
out
value
of
hits
as
the
remote
ID
for
UAE's.
It
provides
a
trustworthy
identity
to
pair
with
physical,
logical
location
data,
as
you
know,
hits
our
value.
Ipv6
addresses
can
be
used
directly
over
the
broadcast
of
broadcast
media.
If
that's
what
you're
transmitting
like
Bluetooth
Bluetooth,
the
five
is
one
of
the
buys
for
this
with
prove
ownership
using
the
high
for
the
signature,
full
mobility
and
multihoming,
so
in
case
you're,
using
multiple
files
on
this
particular
UA,
we're
supported,
you
can
use
the
hip
face
IPSec
for
secure
communication.
F
If
that's
what
you
desire
need-
and
we
can
come
up
to
a
secure
registration
of
the
full
identity.
Bootstrap
on
a
first
come
first
owned
for
the
ID.
That's
what's
available
there,
but
it's
not
exactly
what's
needed
for
this
case.
There's
some
work
needs
to
be
done.
The
original
design
of
hip
talked
about
hierarchical,
hips,
I've
done
a
draft
on
this
since
then,
and
because
of
ownership
and
other
registry
issues,
hierarchal
hits
would
definitely
be
a
value
for
this.
We
need
to
have
an
expanded
registration
process
of
what's
in
the
current
documents.
F
In
terms
of
a
federated
registrated
registration
authorities
to
tie
in
the
various
metadata,
the
actual
physical
location
and
ownership
and
other
information
about
the
device,
some
of
the
new
crypto
support,
which
is
out
there
taking
advantage
of
seaboard
where
we're
appropriate
and
since
they
keep
on
in
the
work
talking.
Oh
look
at
hip
as
a
whole,
Olaf
method-
that's
we've
seen
so
far.
There
would
be
other
work
to
be
determined
as
we
go
through
this
process,
but
we
feel
that
this
is
a
very
good
fit
of
technology
and
a
requirement
situation.
F
A
G
Sharon
from
the
X
I
want
to
talk
to
you
about
a
cool
problem
and
RFC
draft
that
solves
it.
It
has
to
do
with
the
sharing
of
physical
information
of
a
on
the
road
between
cars
and
it
makes
use
of
a
grid
of
the
earth
actually
gonna
go
rid
of
the
earth
called
h3
and
an
overlay
network
which
can
make
those
tiles
of
the
earth
addressable
called
Lisp.
G
So,
just
as
a
background
on
the
problem,
we
distribute
hundreds
or
thousands
of
cameras
each
week
to
drivers
for
them
to
drive
around.
They
are
paired
with
their
phones.
They
do
it
because
of
insurance
in
motivations
and
the
insurers
want
them
to
do
that.
But,
as
a
result,
we
actually
have
tens
of
thousands
of
eyes
crawling
the
physical
space
and
the
public
domain
in
every
major
US
city
and
because
of
the
pairing
to
the
phone.
It's
not
just
cameras,
they
actually
understand
what
they
see
with
the
eye
and
they
are
connected
using
a
network.
G
G
G
G
This
addressing
scheme
allows
me
to
share
information
between,
let's,
say:
mobility:
car
going
here:
mobility
car
going
there.
These
cars
is
this
guy
cars
future
and
vice
versa.
They
can
publish
and
subscribe
to
it
in
tests.
We
were
able
to
communicate
a
red
light
bridge
in
other
four
milliseconds,
a
bit
of
a
lab
conditions
between
a
camera
on
the
junctions
and
all
the
cars
approaching
and
affected
by
it.
So
if
you're
interested,
it's
discussed
in
the
list
working
group
tomorrow
and
happy
to
have
more
participants.
H
The
motivation
for
this
work
is
for
data
center
that
the
high
link
speeds
that
are
making
network
transfer
complete
faster
in
fewer
out.
--'tis
short
data
burst
require
low
latency,
while
longer
data
transfer
requires
high
throughput.
So
these
are
two
requirement
there.
The
RTA
is
the
common
protocol
in
data
center
network.
However,
the
congestion
control
is
not
optimized
for
different
usage
and
lack
of
interoperability.
H
Allowing
for
flexibility
in
running
an
optimized
congestion
method
in
the
network
interface
card
in
having
fast
congestion
notification
to
the
sender
can
improve
the
our
DMA.
They
have
a
trench
transfer,
it's
common
in
the
deficit
and
network.
The
TCP
traffic
is
mixed
with
the
our
DMA
traffic
and
the
causes
conflicts
between
priorities
between
them
and
what
we're
saying?
Is
there
a
better
method
that
could
solve
the
problems
mentioned
above
while
addressing
interoperability
so
that
our
DMA
traffic
can
be
treated
more
efficiently?
H
So
we
have
two
documents
on
the
open
control,
architectures,
the
requirement
document
and
the
architectural
document.
The
requirement
discussed
the
problems
of
the
currently
mode,
direct
memory,
access
fabric,
congestion
and
link
technology
and
the
requirement
for
better
performance
and
the
architecture
proposed
an
open
control
architecture
of
host
networks
for
the
high-performance,
our
DMA
fabric,
to
provide
better
congestion
and
the
link
for
HPC
and
distributed
storage
application.
H
The
architecture
itself
is
seen
in
this
slide.
There
are
there's
the
center
the
reaction
point
and
then
the
receiver
is
the
notification
point,
but
the
switch
can
serve
bosses.
The
congestion
point
that
also
the
congestion
point,
but
also
can
provide
identification.
So
there's
can
be
a
net
only
control
channel
and,
of
course,
they
Nick
Nick
control
channel.
H
That
will
provide
the
information
to
allow
the
sender
to
respond
better
and
faster,
so
the
design
configuration
is
provide
better
information
about
the
congestion,
stay,
stay
to
the
sender
faster
and
more
accurate
from
the
network
and
notify
the
sender
the
reaction
point
from
their
network
support.
It
can
support
proactive
response
from
the
notification
point
and
support
our
DMA
transport
different,
our
deviate
transport.
So
this,
what
we
are
talking
about
is
something
that
would
be
transport
agnostic
and
support,
multiplex,
the
traffic
of
our
DMA
and
TCP.
H
I
Hi
I'm
Paul
Congdon
and
if
we
had
coordinated
better
I,
probably
would
have
gone
just
before
Ronnie,
because
this
is
very
similar
topic,
but
I'm
kind
of
approaching
it
from
a
little
bit
of
a
more
higher
level,
not
a
specific
solution.
Ronnie
has
some
some
drafts
that
that
describe
that
solution.
Someone
talked
about
strat
strategies
to
dramatically
improve
congestion,
control
and
high
performance
data
centers.
So,
first
of
all
data
center
congestion
is
different
than
the
internet.
Congestion.
I
think
we
all
know.
I
That's
a
few
handful
of
transactions
can
cause
a
lot
coming
in
from
the
web
can
cause
a
lot
of
activity
inside
a
data
center,
so
data
centers
have
much
different
environment.
There's
different
bandwidth
delay.
The
switches
are
implemented,
different
and
small
buffers
high
speed
links.
You
know,
then
different,
Internet
routers,
there's
a
lot
more
homogeneity
in
the
network
design.
It's
not
so
random.
The
traffic
is
very
concentrated.
Servers
and
storage
are
all
in
close
proximity
and
there's
a
lot
of
different.
I
The
traffic
profiles
are
kind
of
well-known,
highly
coordinated
correlated
and
typically
they're
managed
with
a
lot
fewer
people
under
a
single
management
domain.
So
all
this
means
that
these
data
center
congesting,
the
environment
in
which
it
runs,
is
much
different
than
the
internet,
and
we
know
we
do
internet
standards
here,
but
but
we
also
do
have
focused
on
data
center
standards,
DC
TCP
as
an
example,
so
the
data
center
needs
low,
latency,
low
overhead,
high
efficiency
and
high
throughput.
I
So
one
thing
the
data
center
kind
of
has
in
common
with
the
Internet,
is
there's
this
trend
to
do
more
over
UDP,
something
like
we've,
seen
a
lot
of
activity
with
quic,
for
example.
So
what
if
we
you
know
could
do
something
that
that
was
more
sort
of
quick
like,
but
not
quick.
You
know
as
heavy,
perhaps
for
specific
for
the
data
center,
and
maybe
we
did
something
that
was
Datagram
congestion
control
like
but
again
focused
on
the
congestion
problems
of
the
Internet
of
the
data
center.
I
It
would
need
to
be
Hardware
off
loadable,
but
maybe
with
less
emphasis
on
security
and
crypto
and
all
the
threading
kind
of
stuff
that
have
it
would
need
to
be
really
common
congestion
control,
as
described
in
the
previous
talk
and
ideally
would
need
to
have
the
network,
visibility,
marking
and
signaling
abilities
from
the
network
itself.
So
the
ITF
has
this
expertise.
So
let's
leverage
and
not
leave
it
to
the
UDP
application
riders.
Let's
see
if
we
can
do
something
we
know
datacenter
congestion,
is
different.
I
The
authors
of
these
papers
have
done
a
lot
of
work
on
congestion,
trees,
so
there's
kind
of
two
types
you
know
in
network
and
in
congestion,
and
it's
constantly
moving
moving
around
in
cast
or
in
network
we've
got
solutions
today,
such
as
ecmp
ecn,
even
even
in
a
lossless
environment,
ecn
with
priority
flow
control.
All
of
these
have
various
pros
and
cons,
but
they're
not
completely
addressing
the
needs
that
we
see
so
some
ideas
of
augmenting
ecn,
with
with
a
data
center
focus
for
a
UDP
layer.
I
You
know
providing
more
feedback
from
the
switches
in
the
packet
headers.
Perhaps
you
know
marking
delay
inside
packets.
So
that
we
can
get
really
fine
grain
view
of
what's
going
on
being
able
to
figure
out,
is
this
congestion
we're
experiencing
right
now?
Is
it
in
network
or
is
it
in
cast?
You
know,
because
maybe
we
would
take
different
approaches
being
able
to
speed
up
notifications
to
the
source,
so
they
can
quench.
You
know
in
a
way
that
addresses
it
and
then
implementing
fast
mechanisms
in
the
switch
to
respond
immediately.
I
So
so
what
like
to
do
is
discuss
the
technical
approach
and
feasibility
of
these
ideas
in
a
side
meeting.
So
we
have
a
side
meeting
tomorrow,
Notre
Dame,
it's
the
same
one
that
was
announced
and
we're
also
providing
remote
participation
for
that,
and
then
we're
going
to
be
requesting
a
ETF
mailing
list
as
well,
so
I
think
I
made
it.
We
also
have
a
whole
bunch
of
references
in
this
deck.
If
you
want
to
read
them.
J
K
Hi
I
will
represent
a
walk
showing
that
we
can
leverage
an
existing
network
intelligence
to
build
service
from
Zhang
Sheng
in
most
of
existing
deployment
service
functions.
Even
trains
are
controlled
by
a
central
control
point
which
computable
ideal
placement
of
virtual
functions,
such
as
IDs
for
your
world
and
the
flow
path
that
goes
for
them
raising
some
program,
since
it
creates
failure
and
the
network's
scalability
issues
it's
oddly,
interoperable
with
existing
networks
and
it
under
exploit
existing
protocols
or
negative
a
network,
for
instance,
service
function.
K
Training
program
is
mainly
a
routing
program
because
you
have
to
route
traffic
for
a
set
of
white
one
between
two
before
which
in
the
destination.
So
what
your
purpose
is
to
augment
classical
entire
entire
gate
for
protocol
and
make
them
a
function
aware.
You
know,
in
order
to
build
service
function
path,
I
will
quickly
remind
how
integral
from
functions
you
have
a
gateway
which
are
connecting
networks,
sharing
information
and
building
a
network
view
and
based
on
what
it
it
will
compute
routing
tables.
K
L
L
So
ITF
is
quite
a
unique.
We
have
many
contribution
from
universities
and
we
even
have
a
research
arm.
I
at
EF,
but
in
my
personal
views,
there's
still
an
gap
in
our
networking
area
as
a
whole,
so
our
innovation
and
the
design
highly
rely
on
experience.
I,
don't
find
a
well-defined
body
of
knowledge
or
a
set
of
approaches
that
can
help
us
to
make
design
choice
and
understand
the
trade-off
behind
those
decisions.
So
I'm,
not
the
only
one
have
this
feeling.
So
these
are
very
interesting
paper
from
Jennifer
works
for
the
Princeton
Princeton
University.
L
Here
she
coated
some
doubts
from
people
inside
and
outside
networking
community.
So
is
networking
a
set
of
protocol
acronyms
or
a
heap
of
head
formats,
or
we
are
just
a
big,
a
bunch
of
boxes.
Of
course
Jennifer
defeated
as
though
all
these
questions
elegantly,
but
at
the
end
of
the
paper
she
also
suggested
that
we
can
make
the
question
we
ask
more
precise
and
the
way
we
answer
them
more
Legolas
and
the
week.
Maybe
we
can
not
so
much
value
new
problems
over
deep
answers
to
existing
questions.
L
We
need
to
encourage
a
more
thorough
and
complete
and
the
deeper
research.
So
why
IDF?
So
in
one
way,
this
theory
can
help
IDF
to
explain
and
predict
the
outcomes
from
new
design
and
facing
your
problems
and
also
it
can
help
us
for
clinical
help
new
generation
to
inherit
and
make
their
own
contributions
so
on
the
other
way
when
IDF,
who
can
can
help.
So
we
have
both
rich
and
very
successful
experience
and
expertise.
L
In
theory,
we
have
so
many
academic
participants,
so
why
now
now
we
are
facing
some
new
requirements
such
as
deterministic
and
bounded
delay
services.
So
we
may
require
more
theoretical
analysis
than
before.
So
this
is
the
other
example
in
Ige's
the
proposed
to
apply
a
network
calculus,
a
kind
of
queueing
theory
to
the
time
sensitive
networking
problem.
L
So
there
are
several
useful
reference.
Of
course,
this
in
one
way,
is
in
warranty
as
the
mathematical
foundations
and
to
us
and
on
the
other.
Yes,
the
theory
of
the
zan
hope
the
semen
proposed
a
framework
long
time
ago.
I
don't
think
I
have
time
to
adjust
the
details,
but
if
you
are
interested
contact,
us
me
and
my
friend
can
Leon
from
China
Mobile.
Thank
you.
M
Okay,
this
is
about
software
management.
If
you
don't
use
these
standards,
which
is
really
the
entire
software
industry
outside
networking.
So
you
you,
you
know
every
every
company
starts
the
same
way.
You
write
a
little
service.
You
start
including
more
services
to
do
your
function,
and
then
you
learn
about
micro
services.
It
becomes
this
and
of
course
you
get
this
and
of
course
you
need
a
management
plan
to
to
make
sense
of
this
all
so
for
each
one
of
those
little
boxes.
M
M
They
forgot
something
so
clearly.
This
is
a
lot
of
work
and
the
standards
for
yang
Andres
conf
solve
this
entirely
with
the
industry
as
a
whole
and
to
give
you
an
idea,
the
scope
of
this
problem
puppet
labs.
They
estimate
there
are
now
more
people
in
the
management
plane
than
there
are
in
the
data
plane.
So
more
people
writing
software
about
software
than
the
software
itself
solving
problems.
M
M
Rest
comp
really
is
designed
really
well
to
make
it
palatable
to
the
industry
as
a
whole,
which
is
what
his
design
was.
So
any
I
really
like
how
rest
comp
has
different
modules.
So
if
you
know
amazon
published
a
yang
for
alexa
skill,
I
could
add
that
in
and
all
of
a
sudden,
you're
saying
Alexa
said,
make
me
a
mai
tai.
So
since
then,
I've
used
this
in
lots
of
different
industries.
Iot
schools,
malware
detection,
the
standards
work
really
really
well,
but
of
course,
that's
only
so
good.
M
So
when
we've
arrived
is
when
I
can
tell
say,
pager
duty,
which
is
a
system
for
alerting
you
to
say:
hey,
post
grads
when
there's
a
slow
query,
paged
me
about
it
and
I
didn't
have
to
involve
DevOps
I
didn't
have
to
program
anything,
and
this
is
all
possible
with
the
standards.
But
none
of
these
tools
know
about
these
standards
or
implementing
those.
So
I'm
really
here
to
talk
about
this.
This
problem
in
industry
as
a
whole
see
if
this
problem
resonates
with
anybody.
M
I
realized
ITF
is
concerned
about
networking,
but
there's
definitely
a
bigger
hole
and
I
estimate.
This
is.
There
are
actually
10
million
developers
this.
This
could
actually
help
them
out
so
and
I'm
one
of
those,
because
without
these
standards,
I
find
using
these
standards
in
outside
networking.
It
really
doesn't
doesn't
really
solve
the
bigger
problems.
So,
like
I
said,
if
this
is
resonates
with
you
reach
out
to
me,
there's
no
working
groups
on
this,
but
maybe
next
ITF
there
could
that's
it.
N
Hi
everyone
I'm
here
to
talk
about
like
aetherium,
improvement
proposals
and
there's
no
time
for
introductions
we
just
in
jumping.
So,
as
you
may
know,
or
may
not,
a
theorem
is
the
second
largest
cryptocurrency
just
based
on
the
market
cap.
There
is
a
possibly
the
number
of
developers
and
tools
that
are
based
on
that.
N
Alright,
the
first
and
hold
blockchain
DLTS
theorem
is
a
home
for
ENS,
which
is
a
theorem
naming
service
which
they're
trying
to
do
that,
eat
ETL
D
and
also
do
DNS
SEC
in
the
blockchain,
also
for
EFS
or
D
central
file
system
applications.
There
are
a
lot
of
more
apps
there.
So
I'm
mostly
argument
looking
processes
here.
So
etherium
is
one
of
the
healthiest
essential
dev
cycles,
ethan
amongst
the
blockchain
and
cryptocurrency
projects,
or
is
it
far
from
the
perfect
perfect?
N
Some
of
the
historical
and
challenges
like
the
central
decentralization
requires
a
non-hierarchical
development
cycle
and
it's
really
hard.
So
they
have
this
e
IP,
which
is
in
a
theorem
improvement
proposals
which
was
inspired
by
Bitcoin
improver
proposals
and
that
was
inspired
by
pet
Python
enhancement
proposals
which
that
also
in
was
enjoyed
by
IETF
request
for
comment,
and
there
are
some
aspects
of
it
dropped
on
this
path.
N
We
don't
know
where
exactly
but
a
security
consideration
section
on
the
IP,
it's
not
there,
I'm
really
trying
to
put
add
it
back
and
it's
gonna
be
added
in
the
next
month.
Probably
so
there
is
some
other
process.
This
is
like
the
process
today,
I'm
gonna
go
quickly
on
on
these,
and
the
slides
are
there
for
more
details.
So
after
discussing
in
the
forum's
and
getting
the
community
supports,
the
EIP
author
would
add
the
CIP
there,
so
security
concession
is
missing
there,
we're
trying
to
add
it
there
and
then
goes
to
a
selected
track.
N
So
the
track,
the
core
track
is
actually
the
consensus
protocol
and
the
protocol.
It's
a
really
critical
one.
It
has
resulted
in
forks
in
the
network
and
a
lot
of
controversy.
There
are
some
others
that
look
ERC.
If
you
have
heard
of
tokens
security
tokens,
you
are
c20
they're
all
in
this
track,
and
there
are
some
other
things.
So
the
process
itself
has
like
four.
You
could
say
five
steps.
N
So
when
it
gets
a
community
support,
it
gets
the
work
in
progress
draft
and
basically
they
write
these
usually
major
major
and
if
it's
a
career,
Pete
needs
implementation.
The
basic
talk
about
it
talks
about
the
proof-of-concept,
after
that
it
has
in
the
EIP
number
and
goes
to
a
draft
mode
here.
The
European
and
further
make
sure
the
draft
to
have
in
a
way
that
actually
makes
sense,
and
it
has
some
more
support.
And
after
that
it
goes
to
the
last
call
this
process.
N
Every
iteration
takes
more
than
around
two
weeks
and
it's
actually
the
ends
up
with
a
call
like
a
conference
call.
So
here
mainly
they
talk
about.
If
there
are
there
any
security
implementations
and
that
needed
to
be
checked
here,
mainly
on
test
tests
and
it's
more
on
a
non
official
one
and
after
that,
if
it's
a
quarry,
IP
gets
accepted
and
all
the
clients
at
least
three
of
the
clients.
N
N
So
this
is
overall
what
I
just
talked
about,
and
this
is
the
first
attempt,
as
far
as
we
know,
to
visualize
this
and
have
something
to
show
that
this
is
the
process
usually
and
still,
even
with
this
process,
it
happens
on
github
proposals,
people
request
everything
happens
there,
there's
this
forum,
if
they're
your
magicians,
that
people
discuss
there
and
magicians
there
for
the
optimization
and
also
this
is
the
call
this
is
like
a
chord
F
called.
Actually
the
decisions
happen
there.
So
why
am
I
talking
about
this?
N
O
Hello:
everyone,
my
name,
is
chingu
I'm
from
Poway
I'm
here
to
discuss.
What's
the
next
step
of
young
unite
have
actually
a
little
bit
of
background
for
me.
I
used
to
chair
area
same
area
same
in
Holmes
area
who
deliver
young,
be
the
model
for
era
VPN
and
actually
witnessed
the
young
take
off
and
followership,
but
we
still
feel
there's
some
problem.
O
So
that's
why
I
bring
this
topic
here,
and
so
young
data
model
get
a
lot
of
traction
actually
operator
began
to
plan
to
deploy
a
young
beta
model
and
but
the
the
reason
for
that.
Actually
young
actually
not
only
happy
to
automate
as
a
networker,
but
also
allow
operator
to
build
the
more
agile
service.
Young
actually
can
model
the
service
from
the
top
and
also
allow
you
to
model
the
collaboration
of
raising
of
the
device.
Allow
you
to
configure
the
product
on
the
device
get
a
net
was
set
hopper,
so
usually
the
young
beta
model.
O
You
know
we
can
reference
the
intact
bender
in
the
operated
test
report.
We
can
see
actually
that
como
has
widely
adopted
it,
but
for
young
the
model
still
an
early
in
adoption
stages.
So
what's
the
reason
behind,
so
we
see
actually
because
the
many
operate.
Actually,
you
know
not
engaging
ideal
for
this
young
data
model
development
actually
for
some
operator
who
already
deployed
these.
O
Can
technology
may
not
aware
I
developed
this
young
Tatum
model
even
for
the
operator
who
actually
know
what
I
do
to
him,
but
they
don't
know
how
how
this
young
would
put
together
to
deliver
service
to
fulfill
the
service?
Actually,
so
there's
a
critical
gap.
Actually
they
were
leader
that
I'd
have
young
pay
the
model
actually
like
a
sufficient
input
from
operators.
So
to
address
this,
actually
one
of
the
approach
a
way
proposed.
O
Actually
we
can
define
young
they
the
model
framework
actually
to
have
operator
to
how
to
integrate
the
young
data
model
in
the
same
namespace,
and
so
we
have
a
chapter
actually
posted
in
OBS
technologies
that
will
be
discussed
in
OBS
a
publishing
on
one
Wednesday
and
try
we
actually
we
work
together
with
operator.
Try
try
to
fill
this
gap
to
help
operator
to
provide
such
guideline.
How
to
you
know,
figure
out
how
even
a
layer
model
can
put
together
and
in
addition,
we
we
think
you
know.
Standard
walk
is
not
enough.
P
Okay,
I'm
full
hon
Baker,
and
this
is
the
mathematical
mesh.
Oh,
so
internet
security
is
broken.
It's
broken
because
users
are
finding
security,
it's
just
too
much
effort
and
you
can't
solve
that
by
asking
them
to
try
harder.
It's
broken
because
applications
are
not
solving
the
real
problem.
I
mean
like
look
at
all
the
data
breaches
their
data
at
rest
breaches.
P
So
the
measure
dress
is
three
core
problems:
device
management,
so
you
can
glue
all
your
devices
together,
so
they're
worn,
gestalt
contact
management,
so
you
can
connect
all
the
people
that
you
connect
to
together
and
have
access
to
all
their
public
keys
on
all
your
devices
and
a
secure
control,
plane
messaging,
which
is
end-to-end,
secure
and
traffic
analysis
resistant,
Oh
wrong.
So
it's
based
on
the
principle
of
I'm
using
more
advanced
cryptography
than
in
Bruce
Schneier
book.
One
key
cryptography
is
great
des.
P
You
know
you
can
do
a
lot
with
des,
but
two
key
is
better
because
you
can
separate
out
the
roles
of
encryption
and
decryption
if
one
is
good
and
two
is
better,
why
don't
we
try
three
four
five
separate
the
roles
out
more
and
we
can
do
more
cryptography
and
more
cryptography
means
more
security.
Yes,
there's
a
bunch
of
small,
powerful
concepts
here.
P
I
can't
go
through
these
at
all,
but
basically
there's
five
basic
technologies
here
and
it's
a
grab
bag
and
they're
all
designed
to
work
together,
but
you
can
also
use
them
a
la
carte
in
your
projects
going
to
be
talking
about
this
effects
dispatched
tomorrow
at
1:30,
and
this
is
the
wrong
set
of
slides.
It
seems
a
she
sent
in
I'm
about
to
start
releasing
this.
So
if
you've
got
a
strong
opinions
on
it,
please
talk
to
me
before
I
release
it,
because
the
minute
I
have
users
I'm
going
to
start
protecting
legacy
and
code.
P
If
you
want
to
change
anything
talk
before
I
release
the
code
there's
a
whole
bunch
of
drafts
and
the
drafts
are
based
on
running
code
that
comes
from
the
reference
library.
So
all
in
MIT
license
all
c-sharp
all
that
good,
sir,
and
the
objective
here
is
to
provide
people
with
kind
of
like
a
psychic
learn
for
cryptography.
If
you
look
at
the
advanced
in
AI,
that's
happened
over
the
past
ten
years.
That
happened
after
the
AI
people
made.
P
It
really
easy
for
people
to
add
AI
into
application
with
things
like
pandas
and
scikit-learn,
so
the
reference
codes,
an
attempt
to
do
that's
the
same
thing
for
cryptography
and
you've
got
all
the
same
goodness
of
blockchain,
pkcs7,
PGP,
x.509
and
so
on,
but
in
a
much
smaller
library,
because
if
you
make
all
the
systems
use
the
same
unified
approach,
you
can
get
rid
of
a
lot
of
code.
I've
got
rid
of
2/3
of
the
code
of
the
past
two
years,
so
meetup.
So
we're
going
to
be
talking
about
sex
dispatch
tomorrow
at
1:30.
A
Q
I'm
either
Luba
chef
and
I
really
want
to
talk
to
you
about
law,
packet,
loss,
detection
in
encrypted
protocols,
I'm.
Looking
at
quick,
alright
I'm,
a
transport
guy,
I
love
to
think
of
my
networks
as
just
dumb
pipes,
but
really
for
them
to
be
dumb
pipes.
There
could
be
somebody
inside
that
pipe
who
is
looking
for
leaks
and
can
find
and
page
them
quickly
if
it's
senior
TCP
flow
in
that
pipe,
it's
pretty
good.
She
can
look
at
sequence,
numbers
at
numbers
figure
out
if
there
is
leak,
see
it
if
it's
upstream
downstream.
Q
Q
One
bit
is
a
cube.
It
stands
for
square
signal,
it's
very
simple
and
that's
in
essence,
64
when
you
send
in
packet,
64
of
them
comes
zero,
cubed,
64,
one
64
0
so
forth.
The
other
packet,
the
other
surahs
a
little
bit,
is
an
else
for
loss
event
a
bit
and
it
is
set
when
there's
a
special,
unreported
loss
counter,
which
is
greater
than
zero
and
unreported
loss.
Counter
is
maintained
by
the
sender
when
there
is
a
packet
that
the
sender
sent
and
depend
sender
deems
it
lost
it
increments.
Q
The
counter
when
the
packet
with
l1
is
sent,
the
counter
is
decremented.
So
if
you
can
see
the
picture,
there
are
two
packets
in
red,
the
lost
about
one
or
tt-to
one
or
two
later
Center
determined,
they've
been
lost
and
it
sent
two
packets
with
l1.
Okay.
How
can
we
use
that?
How
is
it
useful?
So
end-to-end
loss
is
pretty
simple.
It's
just
a
fraction
of
packets
with
l
equals
one
bits
that
observer
observed
upstream
loss
is
basically
how
many
bits
with
a
particular
Q
value
are
missing
from
a
block.
Q
Please
see
us
we'll
be
a
here
Monday
tomorrow,
8:30
to
9:30
for
a
side
meeting
and
we'll
talk
about
it
in
TS
vwg
on
Thursday,
that's
going
to
be
focused
on
protocol
details
as
well
as
privacy
and
ossification
risk
and
then
on
Friday
onmyoji
will
actually
present
lots
and
lots
of
data
and
talk
about
it.
And,
of
course
you
can
talk
to
us.
Thanks.
R
R
R
If
they're
this
is,
this
is
gonna,
be
a
one
minute
talk
if
there's
any
future
discussion.
If
people
want
to
try
and
talk
about,
let's
codify
this,
let's
make
an
RFC
out
of
it.
I'm
available
I'll,
be
here
all
week.
There's
my
email
address.
I've
got
a
couple
minutes.
If
anybody
wants
to
throw
anything
out
about
it
right
now,
we
can.
R
S
Hello,
everyone,
my
name
is
Stuart
Cheshire,
as
some
of
you
may
know,
I'm
the
creator
of
zero
configuration
networking
and
in
a
surf
discovery
and
what
Apple
calls
Bonjour
I'm
sure
everybody
knows
it's
used
widely
in
Apple
products.
What
a
lot
of
people
don't
realize
is
it's
now
in
most
Linux
distributions,
it's
used
by
chromecast
it's
on
Android
and
it's
starting
Windows
10,
it's
even
in
Microsoft
Windows.
S
Anywhere
in
your
software
that
you
have
somewhere
for
a
user
to
type
in
an
IP
address
or
a
host
name,
you
can
use
DNS
service
discovery
to
give
them
a
list
of
options
to
choose
from.
Instead,
if
you
have
iPhones
here,
you
can
tap
on
AirPrint,
and
you
pick
the
printer.
You
want
there's
nowhere
to
type
an
IP
address.
If
the
thing
you
want
doesn't
show
up,
you
fix
that
by
fixing
the
network,
not
by
typing
in
IP,
addresses
on
the
client
devices.
Traditionally
this
is
used.
S
Peer-To-Peer
multicast,
what's
great
about
that
is
it
requires
no
infrastructure.
You
just
hook
up
two
laptops
with
an
Ethernet
cable,
no
switch.
They
can
discover
each
other
two
phones
using
peer-to-peer
Wi-Fi,
no
access
points.
They
can
discover
each
other
services,
that's
great
for
small
networks,
but
it's
inefficient
on
large
networks
because
it
impacts
every
device
on
the
network,
even
the
ones
you're
not
talking
to
right.
Now
it's
slow
on
Wi-Fi,
because
multicast
is
sent
to
low
rate
and
multicast
SAR
batched
with
the
beacon.
S
So
we
want
to
discover
things
that
are
multiple
hops
away
without
flooding
the
multicast.
And
if
you
do
this
with
your
iPhone
today,
you
send
a
multicast
but
the
printers
not
nearby.
We
solve
that
by
adding
a
discovery
proxy.
Your
device
now
makes
a
TCP
can
action
by
sending
IP
packets,
multiple
hops
through
that
network.
The
discovery
proxy
can
then
do
the
multicast
on
your
behalf
and
send
the
answer
back.
You
discover
the
printer.
You
can
now
make
an
end-to-end
TCP
connection
and
use
that
we'd
love
you
to
get
involved.
S
You
can
join
the
DNS
SD
mailing
list.
You
can
check
out
the
code
from
the
IETF
hackathon.
We
have
an
open,
wrt
package
for
this
little
open,
wrt
router.
In
about
five
minutes.
You
can
add
a
discovery
proxy
to
this
and
have
one
running
on
your
own
home
network,
we'll
be
showing
this
at
the
hack
demo
happy
hour
tomorrow,
just
across
over
there.
So
you
can
come
and
see
it
working
for
yourself
and,
of
course,
come
to
our
dns
SD
working
group
meeting
on
thursday.
Thank
you.
T
So
first
about
the
motivation
as
internet
he
involves,
we
could
say
we
have
already
experienced
Oh
IP,
the
first
generation,
but
there
are
still
some
challenges:
for
example,
the
isolated
the
network,
Islands
or
the
limited
program
program,
program,
ability
and,
most
importantly,
and
the
network
is
still
on
its
own
and
application,
and
the
network
isolated
at
the
coupled
and
now
we
have
ice
our
v6.
We
could
say
we
are
entering
a
new
generation,
all
IP,
the
second
generation
and
ice.
Our
v6
has
its
own
mission
and
first
it
is
a
based
on
ipv6.
T
So
it
has
the
affinity
to
IP.
So
it
could
be
much
easier
to
integrate
the
network
domains
and
also
it
has
the
programmability.
So
it
has
the
programmer
programmable
fields,
so
you
could
convey
more
information
from
the
applications
to
the
network,
so
in
that
case
you
could
integrate
the
application
and
the
network
make
them
to
work
together.
So
here
are
the
ipv6
extension
hiders.
We
have
the
hop
by
hop
and
destination
options,
hiders
and
also
the
routing
header.
T
I
sorry
choose
one
of
these,
so
we
have
the
arguments,
views
and
also
that
here
we
you
we
could
use
to
convey
some
information
from
the
applications
and
that
are
the
foundations
for
the
application.
Aware.
Ipv6,
all
eyes
are
basics,
networking
we
call
it
AP
and
6.
So
basically,
we
are
actually
make
use
of
the
ipv6
extension
headers
to
convey
some
informations
from
the
applications
into
the
network.
So
the
network
could
make
the
flying
granular
traffic
operations
to
do
the.
T
For
example,
the
resort
natural
resources
meant
to
guarantee
the
SLA,
so
it
could
be
applied
to,
for
example,
the
fixed
mobile
broadband
like
the
to
be
to
see
scenarios
and
also
the
mobile
broadband,
and
for
the
solution
side
we
could
have
the
host
side
solution.
That
is,
the
application
directly
put
the
information
into
the
ipv6
hiders.
All
it
could
be.
The
network
site
solutions
that
is
the
network
edge,
could
attack
the
application
and
provide
the
information
about
the
application
and
can
wing
it
into
the
network,
and
we
are.
T
U
Yep
great,
my
name
is
Joseph
Potvin
I'm
executive
director
for
excel
rhythms
foundation.
Some
colleagues
and
I
have
initiated
a
project
called
an
Internet
of
rules.
I'll,
say
right
away.
If
you
think
it's
a
terrible
name
that
we
should
not
be
calling
what
we're
doing
that.
Please,
let
us
know
I'll
restrict
this
to
the.
Why
and
the
what
of
what
we're
doing
I
could
explain
how
there's
running
code
behind
this
we're
doing.
U
Alpha
testing
right
now,
but
I've
had
a
very
interesting
conversations
over
the
last
two
days
as
part
of
the
hackathon,
and
there's
lots
of
different
ideas
about
and
good
ideas
about
how
to
do
this
next
so
very
happy
to
see
what
Rodney
was
talking
about
a
few
sessions
here
ago.
We're
actually
working
on
the
generalization
of
how
words
like
must,
must
not
required
shell
and
so
forth,
actually
get
deployed.
U
So
if
you
can
imagine
any
IOT
implementation
and,
let's
say,
there's
some
Court
decision
in
some
province
or
state
and
that's
gonna
affect
privacy
rules
in
terms
of
how
data
moves
around
from
and
to
all
sorts
of
IOT
devices
exactly.
How
is
that
new
rule
going
to
get
deployed?
You
could
even
say:
oh,
it's
going
to
get
deployed
through
the
equipment
providers.
How
is
even
that
rule
expressed
and
say,
English
or
or
spanish,
going
to
get
deployed
to
the
45,000
suppliers
of
these
IOT
devices?
U
This
affects
issues
in
money,
finance,
commerce,
tax,
international
trade,
machine
control
systems,
internet-based
networking
whatever
it
is.
The
gap
in
this
space
is
so
huge.
It's
hard
to
see.
So
it's
hard
to
communicate
this
sometimes
because
most
people
are
looking
for
little
gaps.
This
is
an
enormous
gap
next,
so
the
problem
is
enormous.
U
Ly
wasteful
error-prone
redundancy,
each
of
you
who
is
not
from
Canada
when
you
pay
your
hotel
bill
right
now,
you're
actually
supposed
to
be
zero
rated
on
the
value-added
tax,
but
you're
going
to
pay
it,
and
if
you
were
to
try
to
claim
it
back
on
the
way
out.
Good
luck
to
you.
Little
rules
like
that
come
in
and
they
were
just
simply
not
deployable.
Today,
that's
tax,
but
any
of
these
other
domains.
You
have
same
kind
of
problem.
It
requires
a
domain-specific
language
which
we've
prototyped
right
now
and
not
prototype.
U
We've
actually
implemented
an
alpha
and
an
optimized
algorithm
search,
we're
not
dealing
with
if-then
statements.
We're
dealing
with
given
X
facts.
Con
has
a
context
when
there's
an
input
fact
for
circumstance.
Then
Zed
ought
to
be
the
case.
So
these
are
just
three
facts.
Next,
the
actual
software
we
have
deployed
begins
by
interacting
with
an
API
there's
a
small
component
piece
and
auxiliary
piece
called
lichen.
You
can
use
it
or
not,
use
it.
It's
a
it's!
A
reference.
Implementation
create
your
own.
If
you
like,
roll
your
own,
that
provides
the
context
FAQ
in
the
input.
U
In
fact,
out
to
the
server
in
the
server
is
a
distributed
set
of
servers
with
the
collection
of
algorithms
which
implement
the
rules,
and
these
algorithms
are
simple,
declarative,
tabular
or
tupple
oriented
programs.
We
deliver
those
back
as
output
facts.
We
do
not
inject
them.
We
stop
at
the
point
of
delivering
back
the
fact
there
that
there
is
a
rule
and
here's
what
it
is
and
next
please
so
Excel
go
the
domain-specific
language,
it's
think
of
programming
in
terms
of
control,
tables,
decision
tables
and
so
forth,
with
a
few
reserved
words
very,
very
minimalist.
U
It's
deliberately
not
turing-complete
and
it
uses
all
any
standard
schema
when
it
comes
to
identifying
what
industry.
It
is
what
what
jurisdiction?
It
is,
what
product
type
it
is
next,
please,
inter
Libre,
sorry
about
it,
I'm,
okay,
that's
the
search
engine
and
then
next
slide
I'm
presenting
in
this
room
on
Tuesday
8:30.
C
Hi,
so
I'm
gonna
talk
a
little
bit
about
multiple
provisioning
domains,
a
term
that
you've
probably
heard
and
maybe
know
what
it
means,
but
I
think
a
lot
of
people.
Don't
so.
Hence
the
talk.
So
the
point
of
this
is
back
in
the
1990s.
You
had
a
host.
It
was
probably
sitting
on
your
desk.
It's
connected
to
a
network.
It
got
DNS
and
IP
addresses
somehow
from
the
network
and
everything
just
worked.
That
was
great
good
times.
C
Times
have
changed,
and
so
recently-
and
you
know,
certainly
in
the
20
teens
and
even
before
that
it
was
not
uncommon
to
have
a
host
connected
to
more
than
one
network.
Typical
scenarios.
Vpn
is
a
classic
typical
scenario,
but
a
less
well
less
well
understood
scenario
or
is
when
you
have
a
host
is
connected
to
say
a
phone.
For
example,
that's
connected
to
your
cell
service
provider
and
to
a
wireless
network
which
may
have
a
captive
portal
on
it.
C
C
Anyway,
so
so
the
host,
so
if
a
host
looks
up
a
service
using
using
using
the
DNS
server
that
it
got
from
from
network
a,
it
is
and
then
tries
to
use
the
answer
it
got
on
network
B.
If
that
service
is
actually
only
reachable
on
network
a
it's
gonna
fail,
so
we
need
to
be
able
to
associate
networks.
We
need
to
be
able
to
associate
things
like
DNS
servers
and
IP
prefixes
together
and,
of
course,
in
the
2020s.
Things
are
even
more
exciting.
This
is
starting
now,
but,
but
you
know
you
can
see
there.
C
C
The
answer
is
with
provisioning
domains.
So
a
provisioning
domain
is
a
connect
collection
of
configuration
information,
that's
known,
to
have
come
from
the
same
source.
So
in
the
previous
example,
where
we
had
network
a
and
network
B
networking
and
network
B
were
both
different
provisioning
domains
and
similarly,
in
the
2020s
example,
where
we've
got
provider
a
and
provider
B,
those
are
both
also
different
provisioning
domains.
C
So
how
do
we
communicate
with
this
to
the
hosts
in
the
2010s?
Usually,
we
would
just
notice
that
we
had
two
network
interfaces,
and
so
we
would
be
able
to
set
up
an
ad-hoc
provisioning
domain.
We
would
just
assume
that
information
that
we
got
on
an
interface,
a
and
interface
information,
that
we
got
an
interface
B
are
not
interchangeable
and-
and
we
would
just
do
that
automatically
without
being
told
that
doesn't
work
so
well,
if
you
have
a
border
router
and
you're
behind
the
border
router
and
the
border
router
is
connected
to
two
different
homes.
C
So
in
that
case
we
need
something
called
an
explicit
provisioning
domain,
and
so
work
has
been
going
on
in
the
ITF
recently
to
make
explicit
provisioning
domains
happen.
So
the
original
work
was
done
in
the
myth
working
group.
We
produced
a
document
called
the
multiple,
multiple
provisioning
domain
architecture,
which
explained
the
problem
that
I
just
explained
to
you
in
quite
a
bit
more
detail
and
more
accuracy,
but
didn't
actually
solve
it.
C
And
then,
just
recently,
a
number
of
folks
in
the
interior
working
group
been
working
on
a
document
called
draft
end
area
providing
domains
discovering
provisioning
domain
names
and
data.
So
this
is
using
an
RA
option
to
explicitly
state
that
this
prefix
and
all
the
information
about
this
prefix
is
in
its
own
provisioning
domain,
which
has
this
name,
and
so
that
work
is
almost
done.
I'm
mostly
telling
you
about
this,
because
I
think
it's
really
useful
and
you
want
to
know
please
come
to
intern.
A
A
V
V
This
is
a
Rube
Goldberg
machine
which
actually
turns
out
to
be
a
pretty
good
implementation
of
what
I
think
a
lot
of
network
paths
look
like
these
days,
so
how
we
got
here.
We
did
a
follow
on
to
the
tcp
over
satellite
working
group
in
1997
to
2000
talking
about
very
long
RTT
interaction
with
slow
lost
recovery.
V
We
I
brought
a
proposal
for
TCP
over
cellular
to
the
IETF
and
the
ad
suggested
a
structured
approach
instead
of
doing
TCP
over
foo
links,
Aaron
and
I
co-chaired
the
pilk
working
group
1999
to
2004,
and
we
had
specific
recommendations
that
we
issued
it's
BCPs,
most
of
them
for
specific
link
characteristics
and
that
was
adopted
actually
by
web
forum
version.
2
was
life
was
good
for
a
while
things
have
probably
changed
since
2004.
V
A
lot
of
stuff
is
on
here,
there's
actually
probably
more
things
that
have
changed
that
then
I
didn't
put
on
the
slide.
Things
like
CD
ends
and
stuff
like
that.
But
you
know
a
lot
has
been
going
on.
That'd
be
just
to
pick
one
of
them.
We
weren't
thinking
about
multipath
at
all.
We
were
thrilled.
We
could
get
one
path
to
work.
V
So
what
I'd
like
to
do?
At
IETF
105,
we
had
a
side
meeting
and
went
out
at
ITF,
one
104
in
the
code
lounge,
and
that
was
helpful.
So
I
was
going
to
another
one
of
those
meetings,
but
I've
also
been
told
active
in
the
panarchy
research
group
and
they've
been
thinking
about
path,
aware
networking
I'd
like
to
talk
to
them
about
what
is
engineering
and
what
is
research
so,
what's
ready
for
the
ITF
tournament
record
nations
for
and
was
not
ready
for
the
ietf
to
make
recommendations
for
independent.
V
That's
the
wind
energy
is
meeting
the
thing
you
know
so
like
I,
say
the
thing
that
I'm
saying
here
that
I
need
to
make
pretty
clear.
Is
that
I'm
talking
about
I'm
talking
about
producing
BCPs
for
protocol
designers,
because
we
still
get
a
lot
of
product?
We
still
get
a
lot
of
proposals
from
people
who
are
trying
to
solve
specific
problems
with
well-meaning
ideas
that
the
transport
community
has
identified
problems
with
a
long
time
ago,
but
we
never
write
those
things
down.
V
My
favorite
one
was
you
know
if
you
read
my
PhD
dissertation
from
ten
years
ago,
you
would
clearly
see
that,
and
that
was
the
best.
That
was
the
best
recommendation.
We
had
I
think
the
ITF
has
done
better
in
the
past.
I
think
we
could
do
better
hope
to
see
you
on
Wednesday
morning
and
Thursday
afternoon.
A
A
W
W
We
should
finish
with
much
lesser
than
50
minutes.
I
only
have
a
single
page
here
and
I'm
gonna
give
a
simple,
okay
announcement
on
the
loops
path,
because
this
is
one
of
the
very
first
sessions
of
ITF,
so
I
want
you
guys
who
were
possibly
just
here
and
with
your
jetlag
aware
that
this
time
the
session
starts
from
10
okay.
So
tomorrow
morning,
10
o'clock,
we
are
going
to
have
the
pub
loops
which
stands
for
the
local,
optimization
and
path
segments.
W
Okay,
the
okay
give
a
very
simple
description.
The
path
of
a
long-haul
Network
can
be
naturally
or
purposely
partitioned
into
multiple
segments,
via
possibly
like
tunnel
stitching,
so
loops
aims
to
provide
loss,
recovery
locally
over
some
specific
segments.
So
that's
the
basic
idea
of
loops.
So
here
are
two
pointers
of
the
drafts.
So
please
read
them
and
please
join
us
tomorrow
morning:
10:00,
okay,
thanks.
X
So
I'm
team
April
I'm
here
to
talk
quickly
about
the
DNS
transparency
project.
Many
of
you
are
probably
familiar
about
how
DNS
works.
It's
pulled
a
system
where
there's
no
way
to
currently
get
a
push
based
notification
to
arbitrary
end-users
without
having
to
arrange
something
with
the
DNS
provider.
X
It
could
be
a
user
or
all
sorts
of
things
like
that,
and
this
would
get
hopefully
real-time
updates
from
as
many
TL
DS
and
then
possibly
later
on.
Second
and
third
level
domains
as
possible
and
then
it'll
aggregate
that
and
ship
it
out
to
whoever
wants
to
get
it
in,
hopefully
real-time
or
near
real-time
formats.
X
X
If
you
just
want
to
see,
we
have
a
very
simple
web
page
up
there,
and
we
also
have
a
form
where
you
can
fill
it.
If
you
want
to
get
more
information
as
we
get
more
data
than
ships
out
we're
in
the
process
of
trying
to
stand
up
or
merge
with
a
be
taken
under
some
other
existing
nonprofit,
an
independent
agency
or
independent
organization,
so
we
can
have
a
independent
status
and
build
something,
that's
hope
fully
available
to
the
community
at
large
and
we're
hoping
to
make
it
free
to
end-users.
X
A
All
right,
thank
you.
So,
the
last
session
there
less
talk
is
mine,
so
I
know
some
of
you
know
about
this
have
been
participating
in
it,
but
not
everybody.
Maybe
so
this
is
kind
of
a
hobby.
Little
side
project
I've
been
doing
at
ITF
meetings,
which
is
to
try
to
do
a
little
social
event.
That's
modeled
on
the
sitcom
outrageous
opinion
session,
and
it
was
the
ideas
to
do
some
satirical
talks.
I
was
calling
it
bad
idea,
Pecha
Kucha,
for
a
while
trying
to
generalize
it
to
not
strictly
be
Pecha
Kucha
format.
A
We've
got
several
folks
signed
up
now
and
I
saw
a
few
come
into
my
inbox
during
the
session.
So
I
think
the
list
is
growing.
It's
Thursday
night
at
9:30
in
Duluth,
which
I
think
is
the
room
right
next
door
to
hear.
If
you're
interested
in
giving
a
talk,
you
can
do
a
Pecha
Kucha,
which
is
a
very
structured
20
slides
each
slide
is
up
for
20
seconds
should
be
mostly
text
free,
yeah
or
you
can
just
do
a
short
lightning
talk.
A
It
has
to
be
less
than
seven
minutes
and
the
way
to
get
time
on
the
agenda
is
to
send
me
an
email
with
a
talk
title
and
then
I
need
slides
by
six
o'clock
on
Thursday,
so
everybody's
invited
there
will
be
some
beer
and
I
hope
to
see
you
all
there
and
thank
you
for
coming
here.
If
you
have
feedback
about
this
or
the
Pecha
Kucha
talks,
please
send
me
an
email,
I'm,
always
happy
to
get
feedback.
So
thank
anybody.
Go
have
dinner
and
enjoy
the
week.