►
From YouTube: IETF99-ICE-20170720-1330
Description
ICE meeting session at IETF99
2017/07/20 1330
https://datatracker.ietf.org/meeting/99/proceedings/
A
A
B
A
E
Press
in
the
blue
seats
in
a
minute,
Jonathan
volunteer
subscribe.
Thank
you
and
Nils
will
be
taking
notes,
as
you
saw
when
you,
when
you
speak
and
comment,
please
use
the
microphone
and
please
state
your
name
before
for
the
notes.
Our
agenda
for
today
shall
be
very
short,
introduction.
A
status
update
and
the
rest
meeting
finalized
ice
peace,
whatever
it
takes.
E
So
the
door
is
enter
only
for
the
next
two
hours
yeah.
So
this
is
something
we
do
want
to
finish.
Now,
our
last
remaining
workgroup
item
and
yeah,
we
have
a
proposal
that
we
think
could
be
working
to
get
the
final
issues
clear
in
forward
your
feedback
on
that,
and
then
let's
do
what
it
takes
any
comments
or
questions
on
the
agenda.
E
Okay,
working
group
status
go
sucked.
It
was
a
fairness,
it's
in
quite
some
time
already
with
our
theater.
It
is
waiting
for
ice
peace
which
one
of
the
reasons
why
we
want
to
get
ice
trickle
ice
just
recently,
Bronco
Blasco
ended.
We
got
a
few
comments
and
then
got
an
updated
person.
There's
all
one
outstanding
comment
on
the
mailing
list
that
it
seems
at
least
one
plus
one
was
carried,
so
it
seems
a
reasonable
chains.
E
Please
checkout
them
emailed
on
that
on
the
mailing
list
and
comment
accession
issues,
otherwise
we'll
be
requesting
publication
for
trickle
eyes
shortly.
A
nice
piece
we
will
work
in
group
last
call
after
next
update.
So,
let's
make
sure
all
the
things
we
care
about
are
in
the
person.
That's
gonna
be
published
after
this
meeting.
E
So
the
idea
for
finalizing
eyes,
there
was
a
very
thorough
review
and
quite
a
substantial
PR
from
Peter
Thatcher,
which,
on
the
URL
you
can
see
on
the
on
the
bates
there.
It
was
mostly
editorial,
so
it
did
not
actually
change
the
behavior
of
ice,
but
since
there
were
substantial
editorial
changes,
we
want
to
make
sure
those
changes
didn't
do
any
unintentional
changes
on
the
behavior.
So
that's
why
the
plan
for
today
is
to
review
the
key
parts
of
the
PR.
E
The
PR
is
hundreds
of
lines
of
code,
hundreds
of
lines
of
text,
but
I
did
Mark
twenty-something
parts
that
are
maybe
a
bit
more
substantial
and
the
ideas
go
through
one
by
one:
get
your
feedback.
Okay!
Does
this
look?
Okay?
She
can
we
go
forward
with
this?
Have
it
merged
and
get
this
thing
published
and
then
please
let
us
know
if
there's
any
other
issues
where
that
are
not
in
the
current
PR
there's
a
few
editorial
fixes
that
were
discussed
on
the
on
the
email
list
on
the
next
following
slides.
E
Okay,
so
I
guess
we
gotta
go
then
go
forward
with
with
them
a
few
first
few
issues
crystal
will
be
presenting
those,
and
then
we
go
with
the
PR.
Meanwhile,
please
open
your
laptop
go
to
the
URL
you
see
on
on
that
page
on
the
slide.
It's
in
the
ice,
rink
group,
50,
45
bits
github
is
the
pull
request.
38,
we
will
be
reviewing
that
change
by
change.
So
especially
you
in
the
back
may
not
be
able
to
see
all
the
text
in
the
PR
those
who
want
to
be
actively
contributing
on
this
system.
E
G
Yes,
back
in
April
I
sent
an
email
to
the
list
and
I
had
gone
through
all
these
different
kind
of
states
and
different
kind
of
properties.
We
have
in
ice
there's
a
lot
of
them.
You
have
different
state
machines
and
addition
to
that
you
have
properties
and
so
on,
and
so
on
and
I
was
basically
wondering
whether
we
really
need
all
of
that.
There
were
some
feedback
about
that.
Now
we
could
change
and
change.
G
However,
doing
some
of
those
changes
would
require
quite
a
bit
of
editorial
rework
and
since
nothing
was
really
broken,
that
the
approach
is
really
not
not
to
do
that,
because
you
know
it
would
be
a
kind
of
optimization,
but
it
wouldn't
really
it's
too
late
for
that.
The
ones
that
you
see
here
on
on
on
the
on
them
on
the
slide
are
things
which
are
defined
in
the
draft,
but
at
least
when
I've
been
looking
through,
it
seems
like
it's
not
really
needed
used
anywhere,
so
so
so
so
I'm
gonna
focus
on
that,
for
example.
G
First
we
have
the
ice
state,
the
distaste
of
ice
itself.
We
have
running
completed
in
a
fail,
and
my
question
really
here
is
that:
do
you
really
need
is
what
is
this
state
machine
used
for
I
mean
because,
for
example,
if,
if
ice
fails,
why
should
you
maintain
it
in
a
state
machine,
fail,
I
mean
you're,
just
gonna
quit
doing
ice
and
I.
Don't
didn't
really
see
any
any
other
usage
of
this
I
states?
My
suggestion
was
to
do
it.
Do
we
really
need
it
for
anything.
H
Yeah
I'm,
okay,
with
this
I
mean
you're,
either
doing
ice
or
you've
done
ice
and
either
succeeds
or
fails
that
there
is
these
states
that
might
need
to
be
exposed
in
a
in
api's.
For
instance,
WebRTC
one
is
probably
a
question
for
that.
They
can
resolve
more
than
anything
else.
I
do
think
the
other.
The
other
states
are
like
things
like
frozen.
It
is,
is
kind
of
quarter.
The
whole
thing
yeah.
G
I
come
back
to
this
one
thing
I
forgot
to
mention
regarding
this
I
said
also:
we
had
a
PR
where
we
actually
define
ice
session.
So
now
we
have
a
clear
definition
of
an
an
ice
session
e,
so
we
know
when
it
starts
and
when
it
ends
so
I
think
that's
further
removes
the
need
for
actually
having
this
ice
state
because
we
know
when
it
starts
and
and
how
long
it
goes
on
janux.
I
What
I
recall
and
I
think
investin
I
haven't
read
the
latest
version
of
this,
but
in
5245
there
were
certainly
normative
behaviors,
as
if
something
happens
when
ice
is
still
running.
You
do
this.
If
it's
after
it's
completed,
it's
done
that
I
mean
I,
don't
know
whether
that
needs
to
because
it's
state
or
not,
but
that's
that's
sort
of.
If
we
mean
if
that's
gone,
that's
fine,
but
if
I
mean
used
I
think
that's
being
able
to
say
this
is
when
it
goes
from
running
to
done.
G
I
J
College
X
every
all
the
implementations
I
know
of
whether
it's
old
ice
new
ice,
whatever
have
this
state
variable
that
they
reflect
up
to
higher
level
protocols
that
has
sort
of
these
states
in
it.
And
so,
if
you
remove
this
and
suddenly
it
becomes
no
longer
defined
in
the
spec,
what
the
state
means
or
how
the
implementations
are
supposed
to
set
it
or
whatever,
and
it
just
it
makes
it
much
harder
for
people
I
mean.
J
Should
the
implementations
remove
the
state
out
to
other
applications
or
something
it
makes
it
a
much
more
difficult,
backwards-compatible
plugin
at
a
software
level.
If
you
remove
this,
this
state
variable
and
I.
Imagine
there's
lots
of
other
specifications
that
use
the
value
of
this
state
to
report
in
statistics
and
diagnostics,
information
sure.
H
G
So
anyway,
so
can
we
get
into
a
minute?
So
we
have
a
document.
The
other
thing
was
and
for
for
for
the
checklists,
there
are
actually
two
things.
We
have
a
checklist
state
with
the
running
completed
fail,
and
this
is
something
that
we
do
need
are
not
suggestion
that
we're
gonna
remove
that,
but
we
also
have
something
which
is
checklist
status,
which
is
not
a
state.
It's
a
separate,
you
know
attribute
frozen
and
active
weather,
afraid
checklist
is
frozen
active
and
there
is
a
definition.
G
H
I
I
I
G
So
I
think
we
take
the
approach
to
remove.
It
obviously
will
true
full
of
check
that
we
don't
break
anything
but
again
when
I
did
a
search
for
for
for
leasing
the
draft.
The
only
thing
I
could
find
this
where
we
define
what
it
means,
but
nowhere
else
could
I
find
any.
You
know
what
it's
used
for
so
yeah
I
want
to
further
add.
H
G
And
that's
actually
a
background
of
this,
because
when
we
really
mean-
and
when
you
read
a
spec,
maybe
it's
not
confusing
both
most
of
their
background
is.
This
is
from
feedback
coming
from
from
the
people
actually
implementing
this,
for
example,
and
Taylor,
and
annum
Peter
and
Andy's,
and
these
people
they.
You
know
this
is
confusing.
What
what?
What
do
we
needed?
This
for
I
mean
SSO.
That's
we're
trying
to
clean
up
okay
yeah.
G
This
is
something
which
came
up
Peter,
send
an
email
also
to
the
to
the
list
when
you've
at
some
point
you
you
calculate
this
priority
for
for
or
for
your
candidate
pairs,
and
the
spec
says
that
they
have
to
be
unique.
But
there
is
no
scope.
What
does
this
unique
mess
mean?
And
this
is
something
which
comes
up
Rea
often
it's
something
that,
for
example,
is
Jian
alright.
If
a
last
call
is
gonna
comment
on,
so
we
thought
is,
and
actually
here
I
talked
about
the
the
candidate
pairs.
G
But
what
we
forgot
to
put
here
is
that
there
is
also
this
thing
called
the
local
preference,
which
is
also
said
to
be
unique,
but
we
also
need
to
define
what
that
uniqueness
mean,
but
I
think
it
could
be.
Basically
the
same
thing:
I
think
the
suggestion
that
we've
had
or
on
my
understanding
of
it
is
that
the
unique
west
uniqueness
would
be
within
one
checklist
for
one
media
stream.
So,
for
example,
you
have
audio
and
video.
G
Those
are
two
different
dreams:
it's
okay
to
have
the
same
priority
values
overlapping
in
those
because
you're
gonna
treat
them
differently
anyway,
and
this
would
this
would
this
would
be
during
the
ice
session.
So,
for
example,
if
you
do
a
nice
restart,
you
will
recalculate
everything.
So
it
doesn't
matter
if,
if
you
get
the
same
value
as
you
had
previously
so
I.
I
I
G
And
also
what
Peter
asked
what
was
whatever
we
need
to
some
say
something
about
components,
but
I,
don't
think
you
could
do
it,
because
each
component
is
a
different
candidate
and
and
and
and
the
things
the
fact
that
they
all
have
to
be
unique
covers
components
also.
So
we
don't
need
to
say
anything
about
that.
Okay,.
G
Yeah
I
think
one
more
thing:
we
don't
have
a
slide
on
that
and
and
I
think
it's
more
editorial,
but
I
want
to
bring
it
up
anyway,
just
to
make
sure
because
I
think
it's
it's
important,
we
did
add
some
time
ago.
We
have
text
now
saying
that
before
I
mean
this
was
regarding
this
removal
of
aggressive
and
everything.
But
what
we
now
say
is
that
until
you
do
nomination,
you
can
send
media
on
any
candidate
pair
I
mean
that's
something
we
agreed
on
and
that
and
once
you've
done
the
nomination.
G
You
only
use
that
pair
to
to
send
media
what
I
think
it
was.
Peter
said
that
the
spec
is
a
little
unclear
it
now
the
way
he
reads
it.
That
means
that
you
need
to
use
this
nomination
for
for
the
whole
session.
He
wanted
a
clarification,
saying
sure
you
can
do
REME
ination
and
then
you,
obviously
you
will
end
up
using
another
pair
so
that
just
adding
one
or
two
sentences
to
to
the
to
clarify
that
if
you
do
a
Reno
mination,
you
don't
need
to
do
our
ice
restart.
K
All
my
on
the
same
topic:
do
we
have
any
text
which
basically
says
what
to
do
with
the
candidate
pairs
which
you
like
after
you
nominated
I.
K
Don't
think
I've
a
glee
remember
that
I
was
like
going
through
that
and
I
didn't
find
anything
and
it
seems
they're
like
implementations
which
basically
drop
anything
after
they
nominated,
and
there
are
other
other
implementations
which
keep
that
stuff
around
and
like
another
interesting
question
is:
what
do
you
keep,
for
example,
consent
running
on
these?
In
case
you
want
to
switch
in
case.
You
want
to
re,
nominate.
G
Consent
is,
of
course,
if
you,
if
you
talk
about
this
continuous
consent,
that's
a
course
out
of
scope
to
answer
your
question.
I,
don't
think
the
draft
says
anything
what
you
do
with
the
other
candidates,
for
example,
you
you
must
discard
them
or
anything.
So
maybe
that's
a
further
clarification.
We
should
say
that
if
you
think
or
if
you
you
know,
want
to
be
able
to
do
a
real
amination
at
some
point,
you
you
should,
you
know,
keep
them
alive
by
by
sending
this.
E
Okay
exercise,
we
seem
to
have
a
functional
mic
now
it
was
in
the
front.
So
we
have
a
closer
look
at
that
issue,
but
otherwise
I
guess
we
have
these
no
closed.
We
have
a
way
forward,
so
next
step
would
be
to
go.
Some
key
parts
of
the
massive
PR
and
I
said
earlier.
I
recommend
coming
a
few
seats
forward,
we're
actually
able
to
see
the
screen,
or
at
least
open
the
pool
request
number
38
on
your
on
your
laptop
on
your
browser.
So
you
see
what
we're
going
through
here.
E
E
It's
also
be
at
least
we're
not
changing
in
behavior,
just
making
it
more
clear,
the
first
one
being
there,
the
definition
of
selected
pair
and
select
a
candidate.
You
can
see
the
old
version
in
the
top
and
the
new
version
in
the
bottom,
and
the
key
part
there
is
is
now
tend
says:
sending
media
instead
of
sending
and
receiving
media.
E
So
that
was
the
how
we
actually
changed
the
behavior.
What
you
can
do
with
your
valid
candidates
already
much
earlier,
but
seem
to
be
something
else
left
left
over
here,
but
is
this
something
that
looks
okay
and
for
all
of
these
I'd
like
to
get
a
few
notes
in
the
audience?
When
you
have
ready,
you
think
it
makes
sense.
No
comment.
I
Exam
I
mean
it's
undecided
about
this.
I
mean
certainly
I,
believe
52:45
normative
behavior,
saying
that
you
actually,
especially
you,
should
throw
away
media
received
on
the
non
nominated
pair
after
after
a
nomination
and
I
believe
that
this
was
intended
as
to
some
degree,
you
know
a
weak
security
measure
before
you
had.
I
You
know,
probably,
if
you
didn't
have
proper,
you
know
SRTP
running
or
you
know,
possibly
a
denial
of
service
prevention,
so
I
mean
so
I
mean
the
question
is:
should
you
take
any
media
regardless?
You
take
any
media
from
unknown
non
selected
candidate
I.
Think
one
of
the
goals
here
is
to
be
able
to
latch
on
to
the
proper
path
and
throw
away
anything.
That's
you
know
somebody
spamming,
you
from
you
know
just
randomly
probing
courts
from
an
unknown
port.
E
E
I
E
So
that's
probably
a
bit
different,
because
this
is
only
about
the
selected
pair
and
selected
candidate,
only
definition
of
that
and
and
it
used
to
be
sending
a
receiving,
but
it
can
be
in
the
beginning
different
also,
you
would
be
receiving.
I
J
Think
this
is
the
type
of
thing
that
is
like
when
I
was
reviewing.
It's
really
hard
to
decide
if
this
is
a
normative
change
to
the
spec
or
not
without
going
through
the
whole
spec.
You
can't
just
read
ten
lines
here
on
a
slide
inside
what
this
means.
It's
really
difficult
on
this
one
I
looked
at
it
and
I
thought:
okay,
when
you're
doing
DTLS,
SRTP
I,
don't
I,
don't
see
any
changes
on
this,
which
I
think
is
probably
what
what
the
people
had
in
their
mind
when
they're
thinking
about
this.
G
Krister
this
is
actually
one
of
the
things
I
mentioned
during
the
slides
that
that
there
have
been
a
comment
saying
that
we
should
make
it
clear
that
you
can
re
nominate
during
this
session
without
doing
a
nice
restart.
If
you
think
that's
wrong,
then
then
we
need
to
have
a
discussion
about
that,
but
but
but
but
regarding
this
specific
text,
I
mean
I
agree.
G
E
I
I
E
I
I'm
sorry,
the
other
point
I
would
make.
Is
that
what
we
know
said,
one
of
the
points
of
of
selection
is
once
we've
done
selection,
but
once
or
rather
what
selection
has
happened.
Both
sides
are
then
entitled
modulo,
you
know
leaving
them
up,
modulo
packet,
reordering
and
the
wire
to
shut
down
all
the
non
selected
candidates,
and
you
want
to
be
able
to
do
that
very
you
know
often
very
quickly.
You
know
to
stop
using
resources
on
the
turn
or
turn
off
one
of
your
radios
or
whatever.
F
E
E
A
very
good
idea:
no
I
mean
person
to
actually
do
the
fix,
so
I
guess
what
it's
from
Peter
Thatcher
the
pull
request:
origin
I,
don't
see
Peter
remotely
so
okay,
Chris
Turner
is
gonna.
Take
this
together
with
Peter
and
I,
got
okay.
E
L
E
So
this
may
take
a
bit
longer
than
we
expected,
but
other
ones
hopefully
are
a
bit
faster.
So
this
one
8-inch.
M
E
That's
oh
yeah,
I
mean
all
of
these.
Are
editorial
I
accept
the
previous
one
is
supposed
to
be
editorial,
but
it
turned
out
to
be
bit
more.
So
this
is
about
the
text
you
see,
we
used
to
say
you
have
a
single
turn
turn
relay,
and
now
we're
saying
can
be
can
be
multiple,
so
all
of
this
text
goes
away,
but
it's
super
hard
to
actually
so
probably
easier
to
open
on
your
laptop
and
take
the
details
there.
This.
F
E
Yeah
I
mean
the
girls
already
did
look
at
this.
They
look
okay,
but
there's
a
risk.
Just
like
the
previous
one.
We
thought
it
was
completely
editorial,
but
it
turns
out
there
are
more
implications,
so
there's
something
about
200
other
changes
which
are
okay,
fine,
fine,
fine,
fine,
fine.
These
20
were
like.
L
G
Christa
this
ash
is
the
one
comment
I
gave
to
Peter
when
I
saw
this
Peter
I
said
you
know,
you
should
give
some
explanation.
Why
why
you
have
done
this
and
that
I
mean
if
it's
just
a
spelling
error?
That's
fine,
but
but
in
this
case,
where
there's
a
lot
of
text
removed,
I
think
there
should
be
a
little
more
in
explanation
exactly
why
it's
done,
because.
G
C
E
E
Yeah,
actually,
sorry
about
that,
it
was
the
next
one.
Were
the
multiple
turn
servers,
so
this
is
the
one
I
mentioned,
so
we
are.
Actually
these
chains
is
moving,
removing
a
lot
of
text.
So
how
many,
if
you
had
a
chance
to
have
a
look
at
this
before
the
session?
Okay,
come
okay!
So
what's
your
reflection
on
this
particle
one?
My.
J
Reflection
I'm
reading
this
whole
thing.
It
was
completely
indecipherable
and
I
know
that
that's
useless,
like
I,
feel,
like
you,
wouldn't
did
all
this
good
work
and
I
should
you
know
I
mean
all
the
auditory.
Everything
is
actually
our
tutorial
and
writing.
It's,
like
it's
great
I'm
sure
like
no
complaints
about
what
the
any
authors
do,
but
having
these
all
mix
together
like
this,
it's
really
hard
to
pull
apart
and
figure
out
what
to
do
with
them.
So
every.
J
I
was
like
you
know
what,
everywhere
that
twenty
lines
of
text
was
deleted.
I
was,
like
you
know,
it's
going
to
take
me
an
hour
to
see
what
the
impact
of
about
twenty
line
of
text
delete
and,
by
the
way,
I've
always
thought.
This
document
had
way
too
much
text
in
two
thirds
of
it
should
be
deleted,
so
I
should
be
doing
a
little
dance
that
that's
finally
happening
that
someone's
doing
it.
J
L
J
E
So
this
PR
is
already
the
improved
one.
This
one
there
was.
They
used
to
have
also
technical
changes
here.
Now
they
will
remove.
So
this
would
be
not
purely
editorial
and
then
I
did
another
round
where
I
tried
to
flag
okay,
like
through
your
changes
and
more
substantial,
more
substantial
ones
are
the
ones
we
do
now
going
through.
So
we're
not
gonna
go
through
every
step
change
but
I
do
agree.
It's
a
it
took
me
two
afternoons
to
do.
J
J
E
J
F
This
has
been
part
of
the
reason.
I
think
we're
doing
this
painful
exercise
in
here
is
because
it
wasn't
happening
right.
We
we
not
only
do
we
have
a
huge
thing
of
a
huge,
huge
PR,
that's
hard
to
go
through,
but
we've
got
all
sorts
of
people
that
are
busy
on
everything
and
are
having
a
lot
of
energy
for
this,
so
whatever
it
takes
to
get
that
energy
going
and
just
get
this
done.
Everyone
remember
that
done.
F
That's
right,
probably
the
most
important
bug
in
this
right
now
is
it's
not
done,
but
I'll
push
a
little
bit
back
on
that
last
one
when
you
go
through
iesg
review
and
naked
shoulds
are
often
questioned
so
even
and
and
I'm
not
saying
that
every
should
should
have
a
long,
detailed
explanation
of
why
it's
there,
but
it's
generally
good
at
least
keeping
Center
Institute.
That
says
why
Hawaii
you
would
not,
when
you
might
contemplate
not
following
this.
Otherwise
you'll
get
a
whole
bunch
of
ways,
and
this
a
must
questions
and
I
asked
you
review.
N
So
maybe
another
question
for
consideration
for
the
group:
I
haven't
followed
all
the
updates
and
changes
that
have
been
done
to
for
30
to
45
days
over
the
last
I.
Don't
know
how
many
years
at
this
point,
probably
but
I,
think
it's
worthwhile.
Considering
I
mean
how
comfortable
are
people
putting
in
a
bunch
of
editorial
changes
at
this
very
late
stage,
I
mean
how
well
motivated
is
it?
Is
it
really
necessary?
Is
it
worth
the
risk?
Do
people
actually
want
to
do
this?
Is
there
a
need
for
doing
this.
K
Little
man
I
guess
it
depends
on
whether
you
want
to
have
like
new
implementations
on
the
market
or
not,
because
quite
honestly,
like
I,
mean
reading
eyes
is
like
without
any
code
base.
It's
like
super
confusing.
The
only
reason
I
can
I
can
make
it
through.
The
thing
is
because
I
can
compare
like
running
code
with
50,
actual
text
and
I
know
it's
like
working
code,
so
I
think
there
is
a
need
to
to
to
clean
it
up.
Yes,
and
do
you
think
that
the
changes.
K
N
K
N
E
Okay,
good
and
I
could
second,
that
I
mean
I
think
we
should
keep
the
bar
high
for
this
late
changes.
But
if
someone
is
gonna,
do
a
PR
I
think
it's
a
it's
free
then
makes
it's
much
easier
for
us
to
us
and
do
the
changes
and
everything.
So
those
I
would
definitely
consider
a
random
comment
from
the
bushes
saying:
hey,
please,
but
improve
this
I
would
be
scarred
at
this
stage.
J
Sort
of
supporting
that
whole
argument,
too
I
think
that
if
we
do
manage
to
make
the
document
success
significantly
simpler
to
understand
and
comprehend
our
odds
of
getting
reasonable
real
review
on
the
real
technical
issues
as
we
go
through
this
of
all
the
changes
we've
made
over
the
past
four
years
are
much
higher,
so
there
might
be
some,
even
though
we're
going
to
do
a
bunch
of
pain
and
suffering
to
make
it
more
comprehensible.
Here
there
may
be
some
real
good.
That's
about
you!
Maybe.
E
Okay
good
point:
our
way
forward,
shall
we
go
to
the
next
issue
or
do
you
think
we
have
some
better
way
of
tackling
this.
E
I
E
B
I
Wondering
I
mean
one
thing:
I'm
wondering
is
the
dropping
of
the
last
sentence.
Saying
terms
should
be
well
that
I
gave
the
wrong
term
credential.
Maybe
that's
more
normative
behavior
of
turn
not
normative
behavior
of
ice,
so
I
shouldn't
be
specifying
it
I'm,
just
I
feel
like
that's
different
than
the
most
of
the
thrust
of
it,
which
is
only
use
one
turn
server.
Yeah.
J
J
J
E
I
I
K
Newtonia,
so
reality
is
at
least
from
from
WebRTC
is
that
people
tend
to
put
in
like
a
whole
list
of
turns
servers.
Typically,
people
who
don't
know
what
they're
doing
they
basically
will
provide
like
15
turn
servers,
and
then
they
they
think
that
makes
net
reversal
better
than
if
they
provide
only
a
single
turn
server.
I
Yeah
so
Jonathan
I
mean
I,
guess
it
should
say
it's
a
only
use
multiples
of
these.
If
you
have
some
meaningful
reason
to
believe
they
have
different
network
characteristics,
you
said
I
mean,
like
you
know.
If
you
believe
there
and
you
know
different,
you
know
like
ones
in
your
DMZ
and
ones,
on
the
public
network,
or
you
know,
one
is
TCP,
one
is
UDP
or
whatever,
and
then
the
other
thing
we
do
still
want
to
say
is
once
you've
picked.
What
a
turn
server
makes
the
same.
I
You
know
well
once
you've
picked
a
set
of
turn.
Servers
use
that
same
set
of
turn,
servers
across
all
your
checklists,
don't
make
different
choices
for
different
checklists,
and
you
know
when
you
do
DNS
right.
We
do
SRB
and
DNS
resolution
yes
or
VN.
A
resolution
pick
the
same
actual
IP
address
across
all
of
them.
Don't
spread
them
across
multiple,
but
don't
spread
them
across
the
DNS
load.
C
E
I
E
J
Just
one
thing:
I
was
gonna
mention
on
my
belief
of
why
people
and
WebRTC
are
putting
in
these
twenty
a
list
of
20
or
whatever.
Is
they
don't
really
own
20
or
have
credentials
to
use
the
20
per
se
they're
run
by
other
people
and
there's
basically
using
other
people's
bandwidth
and
often
those
get
turned
off,
so
they
just
keep
adding
more
that's
on
and
off.
So
it
might
be
that
really.
This
is
probably
outside
of
the
ice
spec,
but
maybe
WebRTC
spec
that
it's
like
look.
K
P
D
E
J
Mostly
out,
but
what
I'm
saying
is
that
the
thing
that's
causing
us
there's
something
happening
in
the
real
world:
do
the
implementers
that
is
causing
us
grief
here,
and
so
we
might
want
to
try
and
put
some
text
in
that
tries
to
reduce
that
grief.
Maybe
that's
not
possible.
Maybe
this
is
wrong.
J
Q
So,
just
to
clarify
what
Colin
and
Jonathan
said
before
so
a
lot
of
people
put
URLs
and
some
of
them
have
the
same
characteristics.
I,
don't
know
if
any
document
outside
is
gonna
say
like
you
have
20
URLs
they're
different
apart
from
looking
at
them
and
saying
they're
different,
but
if
some
provide
turn
TCP
Canada
or
run
over
443
in
some
situations,
they
would
basically
be
better
if
you
can't
find
a
UDP
route
of
it
and
so
definitely
need
more
text
here
for
solving
those
kind
of
situations
and.
K
J
L
I
O
E
Okay
and
I
guess
that's
a
general
question:
yeah
the
I
suspect
yeah.
It
is
too
long.
Whatever
you
can
do.
Shorten
makes
a
lot
of
sense,
but
it's
parsley
long,
because
there
is
helpful
in
quotes
informative
text.
When
you
would
you
use
a
VPN
and
such
I
guess
it's
about.
Where
do
you
want
to
put
the
balance.
J
I
have
no
idea.
I
can
live
with
this
other
way,
but
I
mean
it.
You
know
someone
previous
is
geez.
I,
don't
know
about
the
current
one
would
have
said.
You
said
you
effectively
have
a
foot
here.
That
tells
you
which
value
you
should
use
with
no
advice
or
reasons
about
why
you
might
use
a
different
like
we're
deleting
all
the
advice
that
covers
why
it's
a
should
and
why
we
leave
flexibility
to
people
and
how
they
might
choose
really
I
can
live
with
deleting
all
that
device.
E
J
J
But
it
comes
a
question
of
do
you
think
people
need
to
understand
why
they
should
set
these
numbers
that
way
or
do
you
think
they
will?
You
know,
should
always
just
do
them
exactly
like
that,
in
which
case
we
can
change
them.
I
should
too
a
must,
but
if
you're
gonna
leave
it
as
a
should,
maybe
you
should
have
some
advice
there
about
how
you
change
the
numbers.
I,
don't
think
so.
N
F
Has
been
famous
before
right,
I
mean
if
we're
gonna
say
shouldn't
something
like
this
you're
going
to
get
the
oil
they're.
Not
a
must
question
that
doesn't
necessarily
you
mean
you
know
10
sentences
or
whatever
over
here,
unless
you
think
you
do,
but
it's
helpful
to
have
something
that
describes
what
sort
of
variation
from
that
should
you
are
being
is
being
contemplated.
I
J
J
We
have,
including
a
reference
to
dual
stack,
fairness
that
you
know
is
normative
and
at
the
same
time,
we
take
all
the
other
text
and
we
stuff
it
into
an
appendix,
and
we
put
a
little
note
here,
other
considerations
for
thinking
about
setting
these
numbers
can
be
found
in
appendix
of
foo,
and
we
never
have
to
read
the
appendix
again
and
people
won't
get
confused
with
it.
When
they're
reading
the
real
thing
I.
E
J
C
I
I
E
G
Chris
I
probably
need
a
text
but
I
wonder
whether
we
we
need
it
there
I
mean
if
we're
gonna
talk
about
this
real
amination
and
stuff,
like
that.
Maybe
we
should
have
a
separate
section
for
that
and
there
we
can
describe
why
one
would
do
this
and
there
we
can
talk
about.
You
know
that
candidates
no
longer
work
to
have
it
in
one
place,
because
I
think.
G
G
I
I
P
I
Right
yeah,
this
isn't
clear
if
this
is,
if
that
Clause
applies
to
set
of
genomic
set
it
to
set
it
our
said
it
later,
that's
very
much
not
clear
at
all
either
so
I
think
yeah.
This
party
should
be
media
handling
and
I
mean
yeah
I
mean
sort
of
the
your
peers
with
aggressive
domination.
Your
checklist,
your
check,
failed
case
I,
guess
we
don't
have
got
gressive
anymore.
Well,
I
accept
it
you're
still
talking
before
you
could
still
be
talking
to
25.
E
L
I
I
J
Agree
what
Chris
was
saying
this
would
be
better
somewhere
else
in
this
back
maps.
Hillary
was
here,
he
may
say:
I
only
removed
this
because
it's
duplicated
somewhere
else
in
the
spec.
So
it's
fine
right,
so
this
change
may
be
fine,
but
this
case
does
happen
on
a
regular
basis
where
you
have
a
firewall
that
allows
stun
through
but
blocks
RTP.
That's
I
mean
there's
other
case.
That
can
happen
too,
but
it's
going
to
continue
to
happen
with
things
that
are
doing
everything
new.
E
J
K
G
Again,
this
is
related
to
what
we
talked
about
the
the
the
real
emanation,
because,
even
if
it
would
keep
the
existing
text
okay.
So
what
does
this?
What
happens
if
this
happens,
because
there
it
doesn't
describe
he
said-
should
be
used
for
media
unless
this
and
that
that
the
pair
doesn't
work
okay,
so
the
pair
doesn't
work?
What
I
do
then?
So
we
still
need
to
to
have
that
text
and
I'm,
but
I
think
that
texture
shouldn't
be
here.
G
J
Given
this
is
one
of
the
things
that
probably
lots
of
implementations
do
and
I
don't
think
the
chrome
implementation
actually
does
like
I'm,
okay,
with
the
urging
this
as
long
as
it
gets
put
somewhere
else.
But
what
I
don't
want
to
have
happen?
Is
this
gets
removed
and
then
people
decide
not
to
add
anything
else
later
like
this
text
needs?
You
know
what
you
said
yeah
for
sure,
but
if
that
translates
to
we
don't
add
any
text
anywhere
else
later
then
I
don't
think
it
should
be
merged.
G
E
E
O
E
M
K
I
On
the
Paris
set,
when
you
receive
a
yeah
basically,
so
the
idea
is
that
you
have
to
order
something
to
have
the
nominated
flag
in
for
the
control
that
you
have
to
receive.
Lee
use
candidate
and
successfully
compete
the
track
I'm,
not
there.
So
that's
I
think
that's
what
this
is
hangs.
Yeah
flag
is,
you
know
it
is
the
it
is
selected
pair,
not
every
call.
G
G
E
E
E
Q
E
J
So
I
was
just
looking
back
at
the
previous
one.
I
I
do
not
think
its
editorial
I,
don't
think
it's
wrong
change,
its
necessarily,
but
I
think
it
changes
the
way
you
select.
The
IP
addresses
I,
don't
I,
probably
you
know
it
might
be
worth
pulling
that
out
in
a
PR
itself
and
trying
to
break
it
into
some
bite-sized
chunks
or
something
like
that,
but
it
seems
to
be
I,
think
I,
think
the
old
texts
and
the
new
texts
would
in
many
canes
at
least
some
cases
result
in
different
traffic
over
the
wire.
B
E
K
It
says,
like
a
full
agent,
should
not
stop
sending
not
stop
sending
checks
and
responses
from
a
candidate
until
three
seconds.
After
all,
media
streams
using
that
Kennedy
are
completed.
Is
that
the
reference
to
the
to
one
of
the
ice
states,
or
is
that
like?
Does
that
mean
three
seconds
after
my
application?
On
top
of
it
has
stopped
using
it
I.
J
So
I
think
this
might
be
a
little
bit
confusion
between
how
WebRTC
uses
ice
and
how
other
things
might
use
ice.
So
in
WebRTC
the
consent
checks
would
keep
going
until
three
seconds
after
all,
media
had
stopped
using,
but
I
think
the
ice
back
in
other
uses
of
it.
We're
just
going
to
use
ice
to
set
this
up
and
you're
not
doing
consent
checks.
Then
it
would
be
once
you've
got
it,
set
up,
you'd
stop
and
when
ice
is
completed,
not
when
the
media
is
completed.
J
G
G
Yeah,
oh
yeah,
sure,
okay,
okay,
yeah,
I,
I
smear
the
stream
here
so
I
think
this
takes
this
a
little
confusing.
Maybe
he
meant
something
else.
Maybe
him
and
I
mean
there
is
no
ice.
There
is
no.
There
is
no
pare
state
called
completed
right.
So
he's
probably
talking
about
the
ice
state
here
right
but
but
yeah,
so
so
so
and
I
guess
these
three
seconds
has
something
to
do
with.
I
Part
of
the
three
seconds
was
to
in
handle
the
to
handle
aggressive
part
of
it
was
just
to
make
sure
that
you
know
other
checks
get.
You
know
responded
to
rather
than
having
to
wait
for
the
timeout,
but
yeah
I.
Think
this.
The
all
media
stream
is,
unless
you
mean
you're,
doing
something
clever
like
we're
using
candidates
across
you
know
ice
sessions.
That's
fine!
That's
your
problem,
but
yeah
I.
Think
that
all
media
streams
there
is
confusing
so
other
than
that.
I
I
I
E
Yeah,
that's
probably
what's
very
easily
missing
here.
Yes,.
K
D
I
I
K
I
Using
a
previous
reason
for
using
the
previous
sessions,
okay
now,
the
tricky
thing
here,
of
course,
is
that
it's
distinguishing
previous
session
versus
a
new
media
session,
which
is
might
be
a
terminology
issue.
We
distinguish
you
guys
session
from
media
session
I
would
have
to
check
because
we
definitely
need
to
investment,
make
sure
we
do.
We
definitely
had
a
word
for
you
know
of
you
know
the
entire.
You
know
sequence
of
ice
sessions,
which
I
think
is
what
he's
using
media
session
here
for
it's
a
little
unfortunate
if
that's
both
session,
but
I.
G
Yes,
cursor
remember
we
had
a
discussion
about
this
at
some
point
because
we
wanted
to
have
it
mainly
need
a
little
more
because
I
think
using
previous
session
sees
is
confusing.
If
you
talk
about
previous,
you
talk
about
the
previous
I
session,
it
should
be
using
the
you
know,
characteristics
of
the
previous
ice
session
or
something
now
say,
especially
since
the
text
talks
about
both
ice
session
and
media
session.
So
so,
when
you
say
previous
session,
what
do
you
mean?
Do
you
mean
the
media
session
decison?
So
maybe
these
this
leads
needs
a
little.
E
E
G
G
I
Think
what
it's
trying
to
get
it
both
words
are
trying
to
get
across,
which
I
don't
think
we
quite
event
word
for.
Is
you
know
the
only
way
you
met
the
only
way
you
can
do
this
is
this.
You
cannot
do
this
other
than
by
doing
a
nice
crease
tart,
so
I
mean
that's.
Why
the
if
you
want
to
you
must
you
know,
I
mean
I,
agree,
it's
a
little
bit
of
a
weird
formulation
but
I'm
not
sure.
I
O
G
J
E
I
E
I
Johnson
onyx
I
think,
if
I,
if
Peter,
is
correct,
that
that
is
indeed
that
this
is
indeed
a
redundant
definition
of
sort
of
selected
pair.
Then
yes,
except
that
there's
a
verb
missing
in
the
second
second
sentence.
This
blog
over
that
candidate
media
is
forwarded.
G
We
can
do
that
is
Chris.
We
can
do
that
Peter.
But
one
thing
to
remember
also
is
that
you
can't
just
to
emerge
because
I
think
he
has
done
the
PR
based
on
not
the
latest
version.
So
there
are
some.
There
are
some
merge
conflicts
that
we
will
need
to
take
care
of.
So
so,
but
of
course,
if
theater
can
do
it,
fine,
but
actually.
E
J
I
Jones
lacks
I,
mean
I.
Think
that
I
agree.
Maybe
you
don't
want
iced
arch,
beast
bait,
specific
stuff
explicitly,
but
I
think
your
might
need
something
like
remor
generic.
You
know,
you
know
any
you
know,
but
whatever
for
TOEFL
you're
running
over
this
needs
to
reconsider
its.
You
know,
like
your
transport
level,
so
like
obviously
your
congestion
control
and
jitter
or
dedication
things
like
that,
because
you
know
SCTP
should
you
know
needs
to
you
know
you
know,
needs
to
redo
its
congestion
control
and
you
know
whatever,
whatever
protocol
you're
doing
over
here.
I
E
I
J
We
really
need
the
this
advice
is
very
important
to
people
that
are
implementing
the
overall
system.
Yes,
it's
unfortunate
it's
in
this
spec,
but
unless
there's
another
spec,
we
can
reference
from
this
one.
It
needs
to
stay
in
this
spec.
You
know-
and
you
know,
ice
was
developed
and
is
primarily
used
for
RTP,
and
this
is
a
very
complicated
part
of
how
ice
screws
up
RTP
and
you
need
to
interact
with
the
two.
J
E
J
O
J
E
J
E
J
I'm
fine
with
the
current
one
but
I'm
one
of
these
people
who
believes
that
if
I
write
text
like
it's
super
important
that
you
set
the
X
bit
to
Y
and
the
protocol
will
totally
fail,
if
you
don't
set
it
to
Y
that
that
is
normative
text
and
does
not
require
the
word
must
in
caps
for
things
to
be
normative
text.
So
I'm
fine
with
this
as
it
is,
but
many
people
seem
to
feel
that
if
the
word
must
isn't
there,
it's
not
normative.
Q
So
I'm
on
the
fence-
and
this
paragraph,
partly
because
whatever
at
the
top
of
the
paragraph,
is
something
that
each
system
needs
to
each
our
DP
system
needs
to
cater
for
that,
irrespective
wise
so
and
the
second
bit,
which
is
about
the
marker
bit
I,
think
some
of
the
text
already
exists
in
RFC
3550
about
the
talkSPORT.
The
question
is:
is
this
guidance
really
needed
in
ice
for
it.
L
Q
E
E
J
Well,
I
think
maybe
we
should
ask
whether
there
was
a
real
reason
motivating
this
change.
It's
definitely
an
not
an
editorial
change
and
it
definitely
does
not
work
with
old
versions
of
ice.
I
mean
I.
Don't
think
we
can
really
do
this,
because
if
some
old
version
of
ice
changes
it
on
you
and
you're,
like
whoa,
I'm
gonna
throw
you
out
now
because
you
changed
something
that
says:
must
not
it's
not
really
compatible.
I
J
G
A
I
I
Think
that's,
although
I
think,
that's
what
that's
trying
to
say.
So
you
mean
basically,
if
so,
your
default
assumption
is
if
I
was
controlling
before
I'm
still
controlling,
but
still
have
to
be
the
tiebreaker,
if
the
other
guys
Dixie's
control
them
anyway,
because
it
might
be
three
PCC
or
something.
I
E
J
I,
don't
see
how
we
could
take
ask
the
is
GM
to
approve
a
document
yeah
and
have
crypto
agility
to
shot
to
36.
At
this
point
in
time,
I
mean
Fleming.
You
you
know
more
about.
This
may
be
a
new
same
page
like
okay,
I
was
my
first
reaction
was
like
I,
never
want
all
of
our
stuff
to
always
have
to
upgrade
to
the
latest
version
of
everything.
If
we
don't
need
it,
we
don't
want,
don't
use
it,
but
on
this
one
it
seems
unfortunately,
is
G.
Is
gonna
punt
it
back
to
us.
E
F
F
E
F
L
F
Graphs
get
through
that
would
say,
yeah
we're
using
sha-1,
but
we're
using
it
in
this
kind
of
okay
way
in
and
it's
too
much
trouble
to
change
right
now,
we'll
change
the
time
of
the
future
and
and
that's
okay.
As
long
as
it's
clear
and
the
security
considerations,
maybe
Chris
be
even
better,
be
agile.
P
C
E
D
J
N
J
J
Idea
so
I
I'm
not
raising
an
issue
here,
but
I
have
I
I
am
concerned
about
the
timing,
changing
how
that
the
ordering
and
timing
changes
and
whether
that
will
actually
allow
it
to
find
connectivity
in
a
bunch
of
use
cases.
I've
not
gone.
What
Dan
York
did
long
ago
of
work
through
the
exact
flows,
to
figure
out
if
there's
a
problem
or
not
but
I
sort
of
I'm,
not
volunteering,
to
go.
Do
that
but
sure
wish
somebody
had.