►
From YouTube: IETF102-CORE-20180719-1810
Description
CORE meeting session at IETF102
2018/07/19 1810
https://datatracker.ietf.org/meeting/102/proceedings/
A
A
B
C
A
So
it's
the
best
adventures
to
do
it
with
either
pet
only
questions
about
ten
times
for
all,
but
you
can
take
great
notes
in
the
rest
of
the
time.
Okay,
so
the
announcement
I
wanted
to
make
senator
is
in
off
for
the
Aged
since
today.
So
we
already
know
the
earth
cinema-
and
we
just
have
to
finish
the
off
for
the
eight
which
is
difficult
when
most
people
on
vacation
but
think
somehow
we
will
do
that
in
the
next
48
hours.
I'm
sure
that's
148
or
forty-eight
means
okay.
A
We
have
been
doing
interactors
to
interrupt
testing
as
a
working
group
a
lot
until
about
2013
when
correct
became,
stable
and
and
it
has
been
approved,
and
now
there
are
various
organizations
doing
that,
and
one
very
interesting
organization
is
Interop,
which
is
an
EU
financed
project
and
they
are
trying
to
build
interrupt
testing
environments
that
actually
already
know
that
there
is
something
called
the
internet.
So
they
build
environments
where
you
can
do
intro
class
testing
over
the
internet.
A
This
is
relatively
easy
for
application
layer
protocols
they,
but
they
actually
do
it
down
to
the
radio
layer
by
shipping
golden
devices
around
that
that
can
be
used
for
that
interesting.
So
everything
else
goes
over
the
internet,
but
you
also
need
this
Gordon
device
to
make
it
change
well,
but
we
don't
need
that
so
for
interrupt.
Placing
please
have
a
look
at
this
side
and
you
have
to
be
a
little
bit
patient.
It
is
an
EU
project,
so
yeah.
Sometimes
the
presentation
is
a
little
bit
dry,
but
there
is
interesting
stuff
happening
there.
A
Okay,
should
we
anyone
from
the
Dutch
people
here
Owens
here,
good
okay,
so
mean
we
actually
do
this
one
first
and
then
jump
back
to
the
cinema
part.
So
the
reason
I
have
asked
the
dots
people
too.
You
can
come
to
the
front.
I
just
talked
boy.
You
come
to
the
front.
The
reason
we
have
asked
the
dog
favorite
to
say
something
is
because
they
are
actually
a
customer
of
collab
and
they
are
an
unconventional
customer
because
they
are
not
interested
in
light
pipes.
D
Yeah
and
thank
you
for
talking
me
agenda
and
I'm
only
gonna
take
a
couple
of
minutes.
We
don't
really
have
time
for
a
kind
of
technical
discussion,
but
I
really
came
here
asking
for
help.
Our
Co
a
protocol
document
is
going
is
going
to
be
starting.
Second
working
group
last
call
so
we're
getting
close
to
the
end,
and
we
could
take
a
look
at
your
expert
review,
given
your
expertise
with
that,
so
the
the
dots
working
group
is
in
the
security
area
and
I
only
actually
have
two
slides.
D
So
I
would
first
point
you
to
the
top
to
orient
you
on
our
use
case.
What
we
are
primarily
trying
to
do
is
standardize
the
signaling
associated
with
denial
of
service
mitigation
and
it's
a
fairly
kind
of
simple
thing.
We're
trying
to
do,
which
is
we
want
to
standardize
the
communication
of
someone.
That's
under
attack
asking
someone
else
to
help
them
do
the
mitigation,
so
it
was
dot
in
this
diagram,
walk
and
point
a
little
bit.
You
know
there
is
some
victim,
that's
getting
attacked
and
then
architectural
II,
maybe
it's
an
upstream
provider.
D
Maybe
it's
some
some
cloud
packet
scrubbing
service
is
going
to
be
able
to
do
some
mitigation
and
then
there's
the
opportunity
to
do
that.
Transitively,
and
so
we
have
created
this
notion
of
a
dots,
client
and
a
dot
server,
and
they
can
be
combined
architectural
II
to
be
a
dots
gateway,
which
is
just
the
back
to
back,
and
the
dots
flying
is
the
thing
that
asks
a
dot
server
for
help.
The
and
add
a
scope
is
how
the
actual
infrastructure,
how
the
actual
infrastructure
communicates
with
a
client,
but
practically
this
is
a
switch.
D
This
may
be
a
specialized
device.
This
may
be
an
application
server,
a
load,
balancer
or
whatnot,
and
then
it
asks
a
dot
server
which
again
it
could
be
any
kind
of
network
infrastructure
and
and
then
that
dot
server
communicates
with
the
the
piece
of
the
infrastructure
that
can
actually
do
the
mitigation,
but
how
it
actually
does
the
tasking
to
kind
of
rate
limit
or
to
do
blocking
again
is
out
of
scope.
It's
really
just
the
communication
of
saying,
hey,
I'm
under
attack.
The
server
says:
got
it
I'm
gonna,
do
something
about
it.
D
The
clients
saying
hey
I'm
still
under
attack
the
server
says
yeah
I'm,
trying
to
do
something
about
it.
The
clients
saying
hey,
please,
stop
I,
think
I'm
good
or
the
server
saying
hey
I'm,
not
seeing
that
attack
anymore.
Are
you
good?
Can
I
stop
and
so
to
facilitate
that
that
communication
we
have
developed
two
protocols,
there
is
a
signal
channel
and
then
there
is
a
data
channel.
So
let
me
first
start
with
the
data
channel
in
this
workflow
that
I
described
up
here.
D
There's
a
concept
of
there
is
peacetime
and
then
there
is
time
under
attack,
so
the
peacetime
is
when
the
victim
is
not
being
attacked
and
the
attack
time
is
when
the
victim
is
experiencing
a
denial
service
attack.
So
the
data
channel
built
over
s
cough
is
really
the
communication
between
the
client
and
the
server
to
do
various
kind
of
provisioning
again.
This
is
this
is
envisioned
to
be
done
at
peacetime
and
then
under
attack
circumstances.
D
There
is
the
signal
channel,
which
is
the
co-op
protocol
that
I'd
love
your
review
on,
which
is
used
to
kind
of
signal.
Hey
the
attack
has
kind
of
started.
Please
stop
the
mitigation
into
exchange
that
information
about
what
is
occurring
all
right
makes
sense.
Any
quick
questions
about
that
setup.
Okay,
so
I
would
point
you
to
and
I
have
a
list
of
references
to
the
the
dot
signal
channel.
D
We
are
about
to
start
working
group
last
call
this
coming
Monday
we're
gonna
open
it
for
kind
of
two
weeks,
and
we
would
love
your
eyes
on
that
particular
kind
of
draft
I
have
various
section
numbers
that
point
to
various
properties
of
how
we
have
configured
kind
of
that
configured
co-op
and
that
communication.
What
drew
us
to
co-op
initially
was
we
expected
packet
loss?
No
surprise
under
that
attack.
We
wanted
gone
confirmable
messages.
We
wanted
small
messages,
efficient
message
sizes.
D
D
Okay,
so
I
welcome
you
to
please
take
a
look
at
it
as
our
coop
kind
of
experts
and
the
draft
you
want
to
take
a
look
at
is
dot
signal
channel
and
we
will
start
a
working
group
last
call
on
Monday
I'll
send
a
mile
sent
a
pointer
to
the
man
list
spare
and
please
track
me
down.
If
you
have
any
individual
questions
about
this
I
appreciate
your
time
here.
A
E
But
what
has
changed
since
last
time
is
that
we
are
now
reusing
the
base
10
ml
media
types,
so
the
original
version
was
proposing
the
register,
new
media
types
to
be
used
with
touch
and
fetch,
but
it
turned
out
that
that
is
actually
not
necessary.
We
can
just
define
new
semantics
for
the
same
media
types.
I
would
still
get
the
same
functionality
of
being
able
to
update
and
fetch
parts
of
a
Sentinel
document
efficiently.
The
world
conceived
a
feature
we
left
out
as
discussed
in
the
London
meeting.
E
That
was
maybe
honestly
complex
before
the
kind
of
gains
that
you
are
getting
with
that,
and
it's
always
nicer
to
have
a
simple
document.
One
comment
we
got
relatively
recently
is
that
eyepatch
is
actually
our
most
of
time
there
better
way
of
using
patch
the
item
voted
person
of
so
now
the
government
is
focusing
throughout
the
text
on
iPads
and
just
mentions
that
the
same
semantics
also
work
with
a
batch
method.
E
We
also
update
security
considerations,
so
there
is
a
risk
with
veteran
paths
that
single
fetch
and
batch
operation
may
impact
multiple
resources
and
if
multiple
resources
have
an
access
code
rules,
of
course
you
have
to
take
the
aggregate
of
their
operation
into
account
when
you
are
evaluating
those.
So
that's
not
explicit
in
the
security
considerations.
E
Finally,
it
was
discussed
if
we
need
operation
codes
for
patching
and
in
particular
it's
simple
to
do
replacing
and
adding,
but
if
also
like,
appending
and
deleting
operations
or
or
needed.
So
for
that
we
do
think
about
it
quite
a
quite
a
bit
and
it
seems
that
we
can
actually
get
away
without
operation
codes.
E
So
a
lot
of
the
logic
is
still
similar
as
it
used
to
be,
but
we
had
a
bit
more
text
on
what
it
means
to
appending
new
record
store
to
a
patch
and
how
you
can
potentially
delete
records
from
a
patch
so
how
it
works.
Now.
Is
that
when
you
do
an
ad,
you
do
a
patch
operation
with
a
sentimental
record
that
has
a
name
that
doesn't
exist
in
them
in
the
pack.
Originally,
then
it
means
you
are
adding
a
new
record.
E
Also,
there
was,
the
current
text
has
says:
maybe
gives
a
bit
wrong
in
press.
The
time
is
always
required.
We're
gonna
be
updating
that
text
shortly,
then
append
operation.
So
if
you
have
multiple
records
with
the
same
name,
if
you
put
there
a
different
time
than
what
exists
in
any
of
the
existing
records,
this
will
then
become
an
append
operation.
E
Replace
works,
just
like
it
used
to
be,
but
now
we
verify
that
you
match
the
name.
If
the
past
record
has
a
name,
if
I
had
to
say
another
record
on
the
target
pack,
with
the
same
name
as
a
replace
operation,
or
if
your,
if
your
patch
record
has
a
name
and
time
both
of
them
need
to
match
on
the
target
pack,
and
then
it
becomes
a
replace
operation,
so
those
are
actually
were
already
in
the
previous
version.
But
now
it's
clarified
how
they
are
exactly
working.
A
new
feature
in
this
person.
E
E
A
few
consideration
on
this
mechanism
are.
The
good
thing
is
that
seems
that
we
don't
need
the
operation
codes.
So
what
we're
considering?
Or
it's
not,
that
we
would
reserve
one
field
for
future
acceleration
codes.
Now
it
seems
that
it's
not
necessary
to
even
reserve
that
one
field
for
it
and
if
you
later
feel
that
okay,
maybe
we
need
some
more
extensions
well,
we
can
always
define
a
new
media
type
that
is
very
similar
at
this
media
type
and
has
potentially
this
operation.
So
we
still
do
have
an
extension
point.
E
If
we
really
need
it,
one
down
side
of
this
way
of
patching
is
that
you
cannot
add
time
to
an
existing
record
with
a
single
patch
operation.
So
if
you
have
a
record
without
time,
you
just
cannot
put
their
time,
because
now
it
becomes
an
append.
However,
you
can
delete
that
record
and
put
the
record
or
replace
the
whole
pack,
so
it
just
you
need
to
do
two
operations
instead
of
one,
but
this
seems
like
a
reasonable
trade
of
adding
time.
A
A
Next
I
have
two
slides
that
I
will
try
to
do
really
quickly
or
do
you
want
to
do
them,
so
we
did
something,
maybe
not
so
bright
when
we
did
the
IANA
considerations
for
first
anime
or
maybe
the
the
fact
that
we
try
to
cater
for
EXCI
was
not
so
bright
anyway.
Each
time
a
new
fear
is
added,
new
field
name
is
added
to
senem.
A
The
XML
schema
needs
to
be
changed
and
the
the
changed
XML
schema
needs
to
get
a
new
name.
The
schema
in
the
document
is
called
a,
and
maybe
the
next
one
will
be
called
B,
and
the
idea
is
that
an
exit
document
can
reference
the
schema
by
name
at
the
start,
because
changing
the
schema
means
some
of
the
serialization
properties
change.
So
you
have
to
know
what
schema
actually
has
been
used
for
for
encoding
or
something
now.
A
The
problem
is
that
most
registrants
of
field
names
don't
care
for
exit,
so
they
don't
have
an
expert
who
would
be
able
to
update
that
XML
schema
and
it
already
happened,
lightweight
m2m
registered
VL,
oh
and
of
course
they
didn't
care
about
the
the
schema.
So
what
the
draft
says
is
you
don't
have
to
update
the
schema?
You
keep
it
blank
and
when
finally,
a
registry
of
Haris
friend
comes
along
that
does
a
new,
feared
and
cares
about.
A
Actually
they
can
do
as
EEMA
that
does
all
the
existing
new
fields
and
yeah
and
and
registers
a
new
version
for
the
schema
which
what
we'd
be
together
with.
Definitely
there
is
a
slight
weirdness,
because
that
right
now
don't
really
have
a
place
where
we
can
record
that
schema.
The
IANA
registration
process
doesn't
have
a
conventional
place
to
keep
updated
documents
like
that.
So
the
reason
we
are
saying
this
is
that
right
after
cinema
was
approved.
A
Of
course,
like
whatever
came
along
and
said,
we
want
this
new
field
and
we
started
to
scratch
our
heads
because
they
didn't
understand
the
procedure
either
and
did
something
something
was
wrong:
I
think
they
wrote
B
in
the
registration
and
and
didn't
provide
a
schema
or
something
like
that.
That's
all
yeah.
We
probably
don't
want
to
try
to
fix
this
in
all
48,
but
I
think
we
still
have
to
think
a
little
bit
about
how
we
handle
this
Jim
Jim.
B
G
Considering
the
document
is
enough:
48,
as
of
today,
do
you
actually
want
to
publish
existing
XE
media
types,
or
do
you
want
to
remove
it
from
the
document
and
just
move
it
to
a
separate
document,
so
don't
publish
it
in
the
current
one?
They
just
I.
Don't
remember
the
length
I'm
just
offering
you
an
option.
H
E
A
Maybe
it's
mostly
a
matter
of
creating
awareness,
so
all
players
know
that,
and
we
know
that
when
somebody
approaches
us
and
says
I
want
to
do
a
new
sentimental
tag.
We
not
just
say
that's
easy,
but
we
also
say:
do
you
need
XE
to
find
out
whether
they
need
to
exercise
the
second
half
or
not?
So
maybe
that's
all
that
that's
actually
needed,
but
librarian
terms,
screwed.
This
up
slightly
and
I
think
is.
E
It
and
again
yeah
and
by
what
I'm
screwing
WI
was
me
screwing
it
up.
Actually,
so
it
was
a
wasn't,
didn't
quite
realize
how
much
work
it
is
register
EXI
schema.
But
yes,
so
the
latest
approved
person
should
not
have
that
in
place.
So
the
requests
coming
towards
IANA
will
not
request
any
access
key.
If
everything
went
as
planned.
A
E
G
Can
ask
ignorant
question
because
this
is
the
first
time
I've
seen
it
so
I'm
still
trying
to
process
it?
Will
you
need
new
media
type
for
EXCI
stuff?
Or
is
it
just
you
will?
Oh
you
will
not.
If
you
have
a
new
schema
like
this,
yeah
actually
will
be
in
it.
You
know
the
will.
You
need
new
media
type.
We.
G
A
A
G
A
place
for
that
yeah
I
do
remember,
looking
at
it
and
I
wasn't
entirely
confident
about
it,
but
I
didn't
know
enough
about
well.
I
still,
don't
know
enough
about
taxi.
All
right.
I
know
we're
all
about
exceeding
I
ever
wanted
to
at
this
point.
Okay,
so
will
you
have
some
text
or
some
draft
about
this
describing
this
shortly?
A
Statement
in
the
document,
it's
just
the
practical
consequences
are
a
little
but
bit
more
involved.
Then
then
we
thought
we'd
failed
on
the
first
try,
but
it
was
a
communication
failure.
It
wasn't
on
the
failure
was
a
procedure
in
the
draft.
It's
just
that
the
lightweight
and
people
just
didn't
know
that
there
was
a
question
they
to
think
about.
A
A
G
G
A
E
A
A
Hold
it,
we
have
heard
about
the
stateless
forward
proxy
option
a
few
times
in
working
meetings
here
and
as
with
many
things
when
you
think
about
it.
For
a
year
or
two,
you
get
a
few
new
ideas.
What
it
actually
is
about,
and
what
happened
is
that,
since
the
document
that
is
using
this
is
now
in
workgroup
lost
last
caller
has
just
finished
finished
working
on
last
car,
we
were
starting
to
think
again
and
this
thinking,
unfortunately,
has
had
some
results.
But
let
me
first
explain
what
what
the
point
is.
A
The
point
is,
when
you
have
a
proxy,
normally
the
proxy
forwards,
a
request
to
the
server
to
the
origin
server,
and
while
it's
waiting
for
the
response
has
to
keep
some
stage
to
be
able
to
send
back
the
response
to
the
original
flight.
So
that's
the
normal
operation
of
a
proxy
and
the
problem
in
the
six
ish
minimal
draft
us
that
this
guy
here
actually
is
a
very
constrained
system.
So
this
is
a
jointer
that
wants
to
talk
to
the
join
authorization
server.
Whatever
it's
called
today.
A
Jrc
already
know
what
the
operation
means,
and
this
guy
just
happens
to
be
a
random
right,
but
that
happens
to
be
close
to
the
China.
So
we
don't
want
to
burden
this
light
bulb
with
keeping
that
state,
in
particular
after
a
power
failure.
I
think
we
all
think
about
powerful.
Yesterday.
The
situation
may
be
that
there
are
quite
a
few
of
these
joiners
coming
in
at
the
same
time.
So
you
you,
don't
have
deterministic
deterministic
limit
to
the
amount
of
state
that
might
be
needed
for
a
very,
very
constrained
system.
A
A
So
we
had
to
really
find
this
option
and
because
we
are
all
thinking
in
terms
of
tokens
now
we
are
calling
it
the
second
token
option
and
it's
a
critical
option
like
they're
talking
us
and
it's
not
safe
to
forward
like
the
tokens
and
yeah
that
that
works.
It's
fine.
There
are
two
tokens
in
a
message
that
the
one
that
is
always
used
for
the
request,
response
measuring
and
the
second
token-
and
you
actually
have
to
produce
a
little
bit
of
redundancy
here
to
to
make
that
happen.
A
So
yeah,
just
just
to
show
how
this
looks
like
in
the
60k
s--.
This
chosen
us
call
option
just
because
in
six
days
you
would
have
an
else
call
option,
but
this
option
here
is
separate
independent
of
the
usage
of
osco.
This
is
just
an
example
showing
it
in
the
six
dash
context,
so
you
would
have
a
total
of
one
that
is
up
to
eight
bytes
long
and
everything
else
that
doesn't
fit.
There
would
go
into
this
token
two
option,
which
would
have
the
number
19
correct.
A
Options
are
differentially
encoded,
so
9
plus
10
is
19,
and
that
is
the
information
that
you
hope,
the
the
origin
server
or
the
next
proxy
sends
back
to
you.
So
this
is
this
is
now
all
not
new,
except
that
it
has
a
new
option,
but
having
a
new
option
is
something
that
is
very
normal
routine.
For,
however,
we
make
it
easy
to
define
new
options.
So
that's
one
way.
A
So
this
is
kind
of
I
mean
if
you
think
about
it
for
a
while.
This
is
this
is
the
only
design
that
makes
sense
so
that
that
would
be
trying
to
repair
the
the
mistake
of
limiting
the
talking
length
to
age.
So
the
the
problem
with
this,
of
course,
is
first
of
all,
it's
a
change
in
a
place
where
we
didn't
deliberately
place
an
extension
point.
So
implementers
were
not
aware
that
this
change
was
looming.
A
They
are
very
aware
that
they
have
to
handle
new
options,
so
every
type
implementation
has
a
way
to
handle
the
options,
but
this
is
new
stuff.
This
has
to
be
actually
changed
and
that's
not
a
problem
in
the
sixty's
environment
because
they
have.
These
are
very
constrained
systems.
They
have
full
control
over
their
implementations.
They
are
not
just
slapping
a
random
library
on
top
of
their
stuff.
They
really
have
to
do
this
in
a
very
tight
way.
So
then
for
the
sixers
people.
This
works
very
well.
A
A
In
the
Sixers
case,
we
also
have
to
check
whether
there
is
a
random
forwarding
node
on
the
way
that
maybe
actually
doesn't
support,
extended
tokens
and
that
that's
actually
much
more
so
for
anybody
else.
Who
is
trying
to
use
this
extension,
because
we
have
no
way
to
signal
the
presence
of
this
extension
for
options.
We
have
this
critical
versus
elective
thing
and
we
have
the
four
hundred
four
four
zero,
zero
anti
and
so
on,
actually
violating
that.
The
proclamation
would
also
give
you
a
four
zero
zero
answer.
So
it's
not
that
much
different.
A
But
if
itself
it's
doing
the
right
thing
way
too
late
or
doing
something.
That's
not
quite
right,
but
is
with
in
the
the
expectations
that
implementers
might
have.
So
if
we
go
forward
with
doing
the
right
thing,
we
would
right
around
it
update
7
2,
5,
2
and
H
3,
2
or
3,
because
that's
the
true
documents
that
we
find
how
our
talking
rose
into
that
had
our
one-on-one
for
you
to
p1
for
TCP
for
TCP.
J
So
I
think
this
is
a
decent
idea.
Hesitance
a
good,
but
you
know
there's
no
good
way
to
solve
our
eyes
and
I
followed
everything
up
until
slide.
9,
where
you
lost
me,
which
is
the
one
right
before,
though
yeah
that
one-
and
so
here
is
the
here.
I'd
certifying
question
is
when
you're
talking
about
a
client,
your
intermediary
needs
to
perform
a
stateful
request.
J
Are
you
assuming
that
this
is
a
device
that
has
the
capability
of
meaning
that
the
e
client
or
intermediary
in
your
role
of
a
client
that
that
thing
has
that
capability,
theirst,
equal
or
stateless
both?
And
so
what
tries
one
and
it
can
fall
back
in
something
if
it's
o
but
notice
that
the
next
top
does
not
support
stateful
and
it's
happy
digits.
It
does
not
support
stateless
and
it's
happy
to
fall
back
and
just
use
a
state
for
one
I
keep
state.
Is
that
the
assumption
here
yeah.
A
So
the
way
this
would
work
in
practice
is
that
an
implementation
would
keep
exactly
one
state
one
place
for
state.
So
when
it
comes
up
and
it
prepares
for
being
a
stateful,
stateless,
forwarding
proxy,
it
tries
doing
the
the
stateless
thing
once
it
exercises
the
protocol
machinery
for
status,
but
keeps
the
stage
to
be
able
to
handle
errors.
So.
J
That's
the
part
that
I'm
a
little
bit
more
skeptical
about
I,
still
think
the
rest
of
the
stuff
that
you
talked
about
may
still
be
worth
doing.
But
let's
take
a
case
of
an
intermediary
in
the
role
of
a
client
so
and
so
you're
saying
it's
not
just
the
client
an
intermediary
needs
to.
If
it's
an
intermediary,
then
only
keeping
one
means
that
there
may
be
other
ones
that
come
in
because
you're
the
intermediary
to
come
in,
and
so
that
means
you're,
gonna
drop
them
anyway.
Right
because
you're,
an.
A
J
J
Be
right
right,
so
I,
don't
like
this
and
said
that
my
point
was:
the
slide
was
a
little
bit
under
specified
and
so
I
had
to
ask
you
to
elaborate
on
what
you
meant
the
design
would
be,
regardless
of
all
these
different
design,
approaches
and
stuff.
This
may
still
be
the
least
bad
option,
no
matter
which
way
there
we
go
there.
It's
just
I'm.
You
have
to
deal
with
the
cases
where
you
know
the
path
changes
in
the
middle
of
things,
and
so
you
know
meaning
your
next
hop.
A
H
A
H
A
H
H
He
just
can't
get
his
message
through.
If
somebody
in
between
is
not
stateful,
so
as
much
as
I
love,
your
your
layering
of
the
additional
token
in
there
takes
up
room,
it's
big,
it's
extra
stuffing
the
package,
but
it
works.
Doesn't
it
depend
it
works
in
all
cases,
the
client
puts
it
in
there
or
you
know
they
know
not
being
an
original
client,
but
they
Dino
the
proxy.
The
proxy
puts
it
in
their
work.
H
A
Forward
means
safe
to
forward
without
knowing
what
it
means,
that's
a
very
short
well
so
for
proxies.
We
have
critical
and
elective
options,
and
originally
we
thought
that
would
be
all
we
need,
but
the
problem
is
that
there
are
options
that
the
proxy
doesn't
have
to
understand,
but
the
end
has
to
understand
and
if
we
only
have
a
critical
bridge
and
set
that
then
the
first
proxy
who
doesn't
understand
the
option
sends
back
an
error.
A
H
Okay,
it's
it's
only
I
like
your
point.
There
I
don't
know
how
to
solve
it,
but
it
it
appears
that
that
layering
is
perhaps
a
safer
thing
to
do,
even
though
it
takes
up
more
bits
on
the
wire
and
it's
ironic,
because
I
just
got
off
a
conference
call
where
we
did
the
same
thing.
We
had
three
bits
that
we've
reserved
and
we
didn't
want
to.
We
wanted
to
extend
them
since
1995
and
we
realized
there's
so
many
implementations
that
were
barf.
We
decided
not
to
extend
that
into
embed
an
actual
length.
H
A
I
think
we
have
a
typical
cognitive
dissonance
problem
here
that
we
know
this
is
the
the
right
way
to
do
it
and
we
rather
do
it,
but
we
know
that
there
are
at
least
communication
problems
here.
I
think
in
the
end,
be
the
actual
problems
on
the
wire
are
not
going
to
be
very
much
different,
but
there
are
communication
problems
that
we
have
to
explain
to
people
that
we
are
actually
changing.
The
structure
of
the
four
byte
header
little
bit
and
allowing
numbers
there
9
to
15
or
9
to
14.
Actually
that
haven't
been
allowed.
A
A
Yeah,
so
how
do
we
decide
this?
So
what?
What
Klaus
proposals
here
is
to
actually
write
this
up
and
maybe
in
particular,
write
up
slide
9,
because
this
is
really
confusing
on
the
slide.
You
have
to
have
text
explaining
that
and
the
the
six
which
people
have
given
us
a
little
bit
of
time
here
to
prove
this,
because
they
essentially
have
to
have
the
final
solution
before
I
ATF
103.
H
H
That's
not
a
that's,
not
a
degraded
state,
that's
an
impossible
state,
so
my
my
aha
perhaps
is
that
you
have
one
or
two
slots
for
state
and
if
you,
if
you
can't,
if
you,
if
you
can't
do
the
stateless,
because
the
guy
ahead
of
you
doesn't,
then
you
start
dropping
stuff,
but
maybe
five
minutes.
Ten
minutes
go
by
things
will
recover
and
actually
your
message
will
get
through
right.
So
that's
a
degraded
state,
but
it's
one
that
ultimately
works.
H
A
With
the
amendment
to
actually
keep
one
state,
you
know
there
is
fall,
approvers,
of
course,
in
a
sixty
environment
you
would
expect
to
see
more
than
one
light
bulb
and
it's
all
you
would
just
if
you
don't
get
an
answer
from
the
first
one.
You
you
don't
retransmit
to
that
guy.
You
try
another
one.
First.
J
Take
Taylor
one
other
questions
about
this
slide.
Is
there
actually
a
case
where
the
client
needs
a
token
larger
than
eight
I
understand
the
issues
about
the
intermediary
I?
Just
this
is
error
on
the
client-side
and
I
understand
the
intermediary
case,
which
is
why
I
put
them
that
one
before
or
I,
just
can't
think
of
a
use
case
for
clients?
Can
you
well.
A
A
A
J
But
not
necessarily
most
clients
right
if
you're
a
a
human
interface
thing
like
a
remote
controlled
device
like
your
slight
clicker
or
something
like
that,
there's
no
need
okay.
You're
proxying
is
some
other
protocol,
we're
trying
to
keep
States.
You
have
to
respond
to
you
that
I
understand
that.
So
when
writing
this
up
in
more
detail,
then
yeah
make
sure
that's
kind
of
a.
C
Monkey
a
school
which
I
just
have
to
think
about
in
start
equals
one.
So
if
you
have
this
constraint
device
that
does
it
support,
co-co-co
or
whatever,
why
would
it
start
sending
out
multiple
requests
and
just
not
able
to
save
this
state
because
it
must
to
have
implemented
Coco
or
some
other
mechanism
anyway?
And
this
is
a
kind
of
stronger
device.
The.
K
C
A
Good
question
but
I
think
the
answer
is
we
do
have
this
problem:
okay,
so
to
meet
since
there
is
sufficient
sympathy
for
for
this
approach
in
the
room
that
that
is
worth
that
the
office
actually
write
this
up,
and
we
look
at
this
within
a
limited
time
and
I'm
not
going
to
try
to
describe
the
timeline
here
from
scratch,
but
yeah.
This
has
to
be
done
quickly.
Who
would
be
able
to
review?
Who
would
be
willing
to
review
a
draft
even
if
it
comes
at
an
unpleasant
time
in
August,
Peter,
Jim.
E
E
So
this
was
first
presented
in
the
previous
IDF.
Did
a
few
changes
since
it
and
was
also
very
good,
last
call
I,
think
the
last
call
and
about
today.
So
the
changes
on
the
classical
comments
were
that
we
had
at
a
hint
on
action.
Payload,
so
also
an
error
code
can
have
a
payload
telling
the
client
okay.
What
would
be
the
right
way
of
accents
forward?
We
are
not
defining
in
this
draft
any
party
or
payload
format,
but
just
hinted
even
use
one.
E
Then
there
was
a
slightly
bigger
thing
that,
whereas
the
original
version
was
saying
that
you
cannot
do
the
same
request
again,
if
you
do
this,
get
this
error
code.
There's
a
point
taken
that
may
be
also
similar
request
should
be
able
to
be
suppressed
with
this
error
code,
so
bit
more
details
on
this
slide.
This
was
originally
comment
from
a
peon
on
the
mailing
list.
For
example,
if
you
do
something
like
stream
pattern
use
cases,
it
is
a
draft
into
the
party.
E
You
wouldn't
be
actually
doing
the
same
request
that
is
causing
those
subtle
variations
that
the
Reapers
you
arise
may
not
be
identical,
but
they
are
still
related,
and
it
seems
to
be
useful
that
this
same
error
code
could
also
be
used
in
that
kind
of
cases.
Of
course,
now
the
challenge
comes.
So
how
do
you
know
what
is
same
request
and
and
and
what
is
a
similar
request?
A
Okay,
so
this
is
another
one
of
those
customer
and
used
drafts,
because
we
have
other
people
who
would
like
to
use
this,
including
other
stos.
So
we
should
not
delay
this
unnecessary,
so
yeah,
my
question
would
be:
can
we
work?
Let's
call
this
now
you
did
last
call
and
it
today
we
did
do
we
have
to
look.
Let's
call
it
again.
No,
you
change,
didn't
change
anything.
A
Come
from
an
architectural
point,
what
happens
here?
This
is
actually
a
message
to
the
application
that
it
should
be
nicer
to
the
server
and
it's
the
application.
That
also
would
know
what
similar
might
mean.
This
is
not
something
that
happened
that
happens
in
the
coop
stack,
the
actual
reaction
to
this
response.
What
would
happen
in
the
application?
So
architectural
I'm
happy
with
using
something
that's
very
visible
from
the
point
of
view
of
the
co-op
site,
the
Homestake
has
no
idea,
but
it
doesn't
have
to.
E
And,
of
course,
I
mean
before
we
hit
the
publish
button.
All
reviews
are
very
welcome.
Like
it's
a
very
strong
craft,
it
has
a
couple
of
pages
so
we'll
take.
You
got
two
minutes
to
read,
but
if
there's
something
that
it's
worth
clarifying,
it's
good
to
verify
now,
rather
than
in
that
with
iesg
discussing.
L
Peter,
fantastic
I'm
sure,
but
they're
related,
it's
not
very
clear
to
me
and
how
are
you
going
to
define
it
Phil
this
being
table?
Those
will
be
in
table
format
specifier
and
you
leave
things
open,
for
example,
resources
part
of
some,
a
collection
that
rock
I
might
find
how
I
could
define
it.
But
what
can
saying
so
many
things
it
simply
fake
yeah.
E
Exactly
and
that's
the
point
that
it
would
beat
up
to
the
application
to
have
that
knowledge.
Alternatively,
we
could
say
that
this
method
only
applies
if
the
request
is
identical,
that
was
step
or
each
no
and
that's
of
course,
one
way
we
could
go
forward,
but
the
dance
I
mean.
So
you
would
not
have
to
use
this
similar
method
at
all.
E
It's
only
for
the
application
where
it
actually
makes
sense,
and
in
that
case
the
application
would
need
to
have
some
way
of
telling,
for
example,
it's
part
of
a
bigger
specification
that
okay
in
these
cases,
you
would
also
send
this
response
code,
but
you
are
right
out
of
who
the
Cork's
that
would
not
be
able
to
know
it
would
need
to
have
additional
application
layer
information,
but
that
would
be.
We
cannot
at
least
with
the
current
mechanisms.
E
But
here's,
of
course
the
alternative
is
that
we
only
keep
the
identical
one
and
go
forward
with
that,
but
that
may
be
unfortunate
kind
of
restricts
the
use
cases
for
this
code.
Okay,
so
I
would
go
if
there's
no
big
concerns
go
forward
with
is
similar.
At
least
you
read
that
description
and
if
it
needs
clarification,
is
clarified.
A
So
to
me
it
seems
that
the
right
way
forward
is
for
you
to
fix
the
typo
and
for
the
chairs
to
ship.
This
to
the
is
three
and
there
is
an
ITF
last
call.
So
if
you
find
another
problem
that
there
is
still
a
way
to
raise
that
so,
but
it's
leaving
the
working
group
so
that
that
would
be
my
proposal.
Anybody
here
having
a
big
problem
with
that:
okay,
wonderful.
A
So
there
are
two
more
housekeeping
type
drafts.
One
is
about
media
type,
the
content
format
that
allows
to
aggregate
several
payloads
into
one,
including
the
case
that
you,
a
curate
zero
payloads
into
one,
because
you
want
to
say
you
don't
have
anything
at
this
point
in
time.
So
this
this
document
was
adopted.
A
It
replaces
a
number
of
proposals,
including
the
maybe
thing,
and
it
resurrects
nicely
resurrects
something
we
originally
wanted
to
do
in
2012,
but
at
the
time
we
didn't
really
have
sea
ball
widely
available.
So
we
couldn't
do
that.
So
is
it
time
for
working-class
Collier,
any
objections
to
learning
to
work
in
close
call?
That's
not
my
decision,
then
its
finest
decisions
or
talk
to
him
about
that.
L
Again,
Peter
I'm
still
very
much
in
favor
of
working
group
last
call,
but
there's
another
thing:
we
have
contact
formats
which
I
defined
in
HD
corpus
and
have
asked
from
preliminary
publication,
which
it
has
been,
don't
be
very
grateful
for
that,
but
also
this
one.
We
would
like
to
have
an
preliminary
request
for
this
registration.
Also
of
this
content
form
it
Emma
clear.
Okay,.
A
A
B
B
A
B
A
A
A
A
There
are
about
1400
media
types,
definitely
less
than
2000.
We
have
6500
content,
format,
numbers
and
right
now.
Anybody
who
wants
to
use
co-op
with
a
new
media
type
has
to
do
all
this
registry
stuff,
and
the
idea
is
to
just
go
ahead
and
waste
some
code
points
and
say:
okay,
all
those
media
tribes
that
that
theoretically
can
be
registered,
that
there
are
some
for
which
it
doesn't
work.
It
doesn't
make
sense
because
they
had
mandatory
parameters,
but
all
other
media
types.
A
A
E
F
A
A
So
I
think
it's
not
a
disaster
that
we
don't
get
with
these
things.
Now
we
can
pick
it
up
in
the
next
federal
injury
and
we
have
to
decide
the
details
for
that.
We
used
to
have
virtual
interims
quite
a
few
early
on
and
even
a
physical
one,
but
we
haven't
done
this
and
I
think
with
this
backlog.
We
have
to
start
doing
it
again.