►
From YouTube: IETF97-ICE-20161114-1330
Description
ICE meeting session at IETF97
2016/11/14 1330
C
A
D
Exactly
exactly
okay,
so
welcome
everyone
to
the
iceberg
group
session
here
at
88,
97
and
yes,
sorry
about
the
projector
quality.
Since
we
couldn't
do
much
about
that
right
now
and
since
this
an
IPF
meeting,
the
note
will
applies
if
you're
not
familiar
with
it,
please
go
check
it
out,
lose
it
going
on
over
there.
Please
sign
them
and
thanks
guys
for
volunteer
to
do
chá
prescribing
and
note-taking,
and
when
you
speak,
as
always,
please
use
the
microphone
and
state
your
name
before
speaking
on
agenda.
D
D
What
should
we
do
when
we
are
now
done
done
with
the
current
working
group
items
and
if
we
have
time
in
the
end,
there's
also
slot
for
clarifications
on
rice
peas.
So
these
are
not
major
technical
issues
more
on
the
side
of
editorial
things.
So
that's
why
we
have
the
slot
in
the
end,
if
we
have
time
in
there,
today's
session
creature
will
be
asking
for
feedback
on
some
of
those
too
okay,
some
of
them
bit
more
technical
than
the
address.
D
Okay,
I
see
none.
So
we
go
to
drink
a
lot
of
blood
group
status.
So
do
a
sock,
fairness,
craft,
the
iesg
review
is
completed.
Just
day
ago,
however,
now
we're
doing
a
new
idea
last
call
because
I
in
the
previous
I
did
last
call.
There
was
unfortunately
wrong
information
in
the
ID
tracker
on
the
on
the
status
of
indented
status
of
the
craft.
The
draft
is
supposed
to
be
informational.
There
will
be
a
new
Glascow
started
on
that,
but
after
that
is
done,
we
assume
it's
going
to
be
going
today.
D
D
Okay,
then
trickle
I
stages,
so
there
was
a
new
version
submitted
after
the
building
idea,
which
addressed
most
of
the
issues
that
were
discussed
in
Berlin
after
we
got
a
whole
bunch
of
good
reviews
from
a
lot
of
people
thanks
a
lot
for
that.
It's
going
to
make
our
life
in
the
last
call
/
class,
so
much
easier,
and
the
current
plan
for
Treecko
ice
is
to
address
the
remaining
comments
from
Berlin
and
also
additional
reviews
submit
new
draft
towards
the
end
of
the
month
and
start
over
blasco.
D
D
D
D
E
Thanks
yeah,
this
is
going
to
be
quickly
a
one
of
the
things
we
thought
we
had
sold
in
in
Berlin,
but
turned
out
that
we
haven't.
Was
this
ta
issue?
We're
gonna
calculate
that
one
and
action
point
from
Berlin
was
that
some
people,
mostly
they
were
Cullen,
I,
think
Peter.
E
E
There
is
currently
there's
an
open
pull
request,
because
there
are
some
changes
which
are
needed
to
do
in
the
appendix
be
very
minor
is
just
a
couple
of
places.
I
think
we're
where
he
says
that
the
lower
bound
is
20
milliseconds,
but
that
should
be
fired,
because
that's
the
agreement
that
we
came
to
I'm
not
going
to
read
the
text
I'm.
Just
gonna
show
you
the
highlights
about
how
you
now
calculate
the
TA
value.
E
E
Maybe
it's
also
in
the
draft
I
have
to
check
that,
because
I
copy
pasted
to
the
text,
but
anyway
I'll
check
that,
but
anyway,
there
are
some
text
explaining
why,
for
our
for,
in
this
case,
5
millisecond
was
okay,
but
for
transport
protocols
in
general.
It's
it's
it's
it's
not
okay!
So
but
you
can
read
that
in
the
draft
I
don't
have
it
here
and
that
was
it.
D
E
D
F
G
F
Options
we
have
today
by
influencing
the
network
is
RSVP,
which
is
pretty
much
dead
in
the
water,
doesn't
really
work.
We
have
ecn
today
to
get
some
feedback
from
the
network,
at
least
but
at
least
for
UDP
test
its
own
problems,
we're
not
sure
whatever
the
network
would
mark
UDP,
packets,
right
and
DCP
seems
to
have
its
own
set
of
issues,
but
at
least
in
the
confined
realms
of
the
network
it
we're
supposed
to
work.
It
actually
seems
to
work.
So
that's
a
good
thing.
F
F
F
H
Just
likes
I'm
first
answer
that
I
mean
yeah,
there's
a
lot
of
stuff
RTP
that
sort
of
implicitly
you
know,
assumes
a
single
path.
Like
all
the
feed.
There's
only
a
single
feedback,
/
eyes
perspective.
It's
kind
of
an
assumption
that
you're
going
to
converge
on
one.
The
other
thing
I
would
put
a
sort
of
as
the
possible
thing
in
the
building
blocks
is
that,
given
that
we're
doing
sctp
for
deep
for
all
the
data
channel,
stuff
I
think
an
sctp
does
multipath,
but
I
would've
thought
he's
using
it.
F
Create
nice
transitions,
no
need
to
switch
media
addresses
if
I
detects
a
better
path
and
stuff
like
that,
so
and
why
I
am
looking
at
UDP
and
tcp.
It's
always
interesting
to
see
how
is
P,
sometimes
rate
limited
red
team
at
UDP
so
like.
If
this
is
the
UDP
traffic
through
a
nice
V,
sometimes
when
they
have
a
DOS
attack,
they
start
to
cut
traffic
and
obviously
your
ice
session,
which
suffer
and
maybe
gets
even
worse.
E
But
then
one
of
the
things
we
say,
for
example,
is
that
or
we
assume
you
can
switch
from
one
to
another,
because
what
we
are
going
to
suggest.
That
added
text
is
that
if
you
switch
from
UDP
to
tcp
or
from
tcp
to
UDP
that
you
don't
need
to
send
an
updated
offer,
you
can
just
do
it.
I
mean
if
you
have
those
candidate
pairs
in
the
valid
list,
you
can
just
switch
between
them
today.
Already
there
is
nothing
new
needed
for
that.
I
I
I
E
I
K
Like
so
here,
I
tried
these
20
and
here's
the
three
that
work
now
and
oh
by
the
way
to
have
come
up
and
one
has
disappeared,
and
but
that
should
be
booing
in
the
sense
of
saying
it
works
or
doesn't
work.
It
shouldn't
be
any
kind
of
quality
assumption
on
how
how
well
that
would
work.
I
think
that's
something
you
would
want
to
have
layer
above.
F
H
Gentleman
I
sort
of
respond
to
Peter
I
think
you
know
in
terms
of
do
we
have
the
mechanisms
with
ice
bus
renomination
to
do
this
probably,
but
I
think
you
still
need
negotiations
that
yes,
this
really
is
what
I'm
doing
as
opposed
to
I'm
doing
ice
busty
nomination,
but
only
expect
immediate,
a
single
connection
at
a
time,
I'm
and
then
I
think
there's
probably
gonna,
be
a
lot
need
to
be
a
lot
of
work
at
you
know.
H
H
Jonathan,
a
clear
my
mind,
finish
would
be
to
say:
don't
keep
shoveling
new
features
in
dice
this,
let's
get
ice
base
done
with
the
existing
feature
set,
and
this
will
be
an
exception.
It's
a
really
I
think
our
extension
story
is
pretty
solid,
so
I
think,
let's
not
put
this,
as
you
know,
don't
go,
keep
adding
features
to
an
existing
release.
You
know
release
and
then
you
know,
release
the
next
thing
later.
F
D
Okay,
thank
you
Paul
Eric,
so
I'm.
This
whole
discussion
does
relate
to
the
next
point
we
have
on
the
agenda.
It's
like
what
we
want
to
do
in
the
future
will
be
the
ice
working
group.
Is
it
something
something
along
these
lines
or
something
something
else
we
want
to
call
it
a
day.
D
So
if
there
are
no
more
comments
on
this
topic,
okay,
Paul,
you
wanna.
You
already
address
these
points.
Yeah.
I
H
John
flex
I
mean
I,
don't
you
know
favor
major
changes
just
like
for
the
sake
of
major
changes
I
mean
I
would
want
to
see.
You
know
an
individual's
mission
proposal,
for
this
is
how
we
think
I
should
be
completely
rewritten,
and
I
guess
there
have
been
some
in
the
past,
but
I'm
not
sure
how
much
people
believe
my
things
they've
done
so
far
still
and
things
like
that,
so
I
mean
I'm
open
to
it.
I
I
M
Ted
Hardy
I
kind
of
agree
with
Jonathan
putting
a
a
charter
in
front
of
the
ITF
community.
That
says
we're
going
to
change
this
to
say:
make
ice
more
flexible,
robust
and
more
suitable
for
changing
environments
and
allow
major
changes.
You
then
have
to
have
some
milestones
that
listed
what
those
major
changes
were
and
what
they
were
going
to
do
for
the
community,
and
so
in
the
absence
of
a
concrete
proposal
that
would
fit
into
that
kind
of
structure.
Of
giving
you
a
new
milestone,
shoot
for
I
would
say
you
you
keep.
M
The
current
charter
continue
to
make
the
incremental
improvements
that
are
in
mind
for
right
now
and
then,
if
somebody
comes
forward
with
a
proposal
that
you
want
to
take
up
and
the
consensus
is
that
it's
a
major
change.
You
recharter
at
that
point
with
that
is
the
milestone
associated
with
it.
Rather
than
making
this
general
adjustment.
Okay,.
I
So
let
me
see
if
I
understood
correctly,
the
right
ITF
process
in
this
situation
is
to
propose
something
outside
the
Charter
in
this
group,
but
then
that
would
be
like
a
possible
topic
of
recharter
when
we
say
okay.
This
is
something
specific
we
can't
do
here,
but
we
would
right
now,
but
we
would
reach
arter,
but.
I
M
Ted
Hardy,
there's
always
more
than
one
way
to
do
it
within
the
ITF
process,
and
you
could
take
it
to
dispatch
and
dispatch,
could
tell
you
to
go
over
here,
because
there's
already
a
working
group
or
you
could
do
a
whole
bunch
of
other
things,
but
that
certainly
works
within
the
IETF
process
as
I
understand
it,
and
so
that
that
is
a
way
to
do
it.
That
certainly
well
understood.
Okay,
let's
good
feedback
thanks.
H
Drop
that
sort
of
to
maybe
you
know,
phrase
that
a
little
more
positively
I
think
previously
we've
been
focusing
on
getting
ice
mated
things
done
so
we've
sort
of
been
saying
things
that
are
not
directly
than
charter.
You
know
sort
of
getting
a
lot
of
time,
I
think
we're
now
moving
towards
lets.
You
know
give
meeting
time
and
less
time
to
new
innovative
work,
so
we
can
discuss
it
right.
J
Campbell
and
thought
I'll
preface
this
with
I'm
saying
this
without
really
knowing
what
the
concrete
proposals
are
likely
to
be
as
it
was
just
mentioned.
You
know
we
have
both
the
ability
to
bring
things
here
to
the
working
group
or
to
take
things
to
dispatch.
What
I
would
suggest
is
if
these
are
things
that
are
very
self-contained
and
ice.
It
makes
perfect
sense
to
do
it
here.
D
Anyone
else,
okay
seems
we
don't
have
anyone
in
the
room
out
of
people
who
are
have
been
proposing
it
earlier
major
changes,
so
maybe
they
me
has
gone
away
and
that's
that's
completely.
Okay,
I
mean,
I
think,
it's
a
it's
a
success
to
say:
okay,
we're
done
and
we
can
close
the
crew
that
that's
not
a
that's
a
success
and
it's
good
thing
to
do
so.
We
shouldn't
be
inventing
new
work.
Definitely
don't
don't
don't
get
it
don't
get
that
way.
D
What
we
want
to
do
now
figure
out
easter
someone
who
has
a
substantial
drive
to
find
the
do
new
stuff
eating
the
charter
of
the
current
clue
that
we
should
be
doing.
If
not,
let's
declare
success
and
be
done
with
it
as
an
option
C.
Maybe
all
that
would
be
t
OD
or
maybe
that
was
then
I
was
gonna.
Let
us
be
one
of
the
letters,
okay,
so
okay,
at
least
in
this
room
we
don't
have
a.
J
D
E
Yeah,
it
says
here
clarifications
they
are.
There
are
a
few
technical
things
here
which
is
coming
up
also
originally
I
I
didn't
have
too
much
time
for
this,
and
when
you
see
some
of
the
things
I'm
gonna
present
to
you,
you're
probably
wondering
why
am
I
wasting
everyone's
time
here
you
know
doing
showing
this.
E
Why
don't
I
just
do
a
pull
request
or
you
know,
suggest
a
fix
on
the
list
and-
and
you
know
we're
done
and
and
my
we
don't
really
know
we
have
time
but
I
mean
my
intention
wasn't
really
to
go
in
and
start
discussing
all
this
issue.
Well,
the
thing
I
want
to
get
out
of
waste
is
that,
in
my
opinion,
there
are
still
some
editorial
work
that
we
need
to
do
in
a
draft.
E
It
is
difficult
to
read
and
that's
not
only
my
opinion,
there's
been
some
people
who
have
our
new
two
eyes
who
have
read
it
them.
You
know,
please
read
it
and
see
what
you
think
any
it's
a
complex
document
to
read.
The
intention
is
also
not
too
too,
when
I,
when
I
wrote
these
things.
My
my
my
my
tea
was
it's
not
to
rewrite
the
whole.
You
know
spec,
to
make
it
more
readable.
I
think
these
are
things
fairly
small
things
that
can
be
dude
done.
E
You
know
replacing
some
words,
removing
some
sentence,
adding
some
sentence
and
those
kind
of
things
to
make
the
dis
pic
more
readable.
Actually
before
we
publish
it
but,
like
I
said,
there's
also
some
some
some
technical
issues
somewhere
with
one
which
we
also
already
discussed
in
in
Berlin,
which
was
raised
by
by
email.
Unfortunately
he's
not
there,
I,
don't
know
if
he's
on
there.
Oh
yeah
he's
not
there
either
so
but
anyway,
so
we'll
come
to
that.
But
to
show
what
I
mean
this
is
something
this
is
actually
somewhat.
E
Someone
said
in
Berlin
when
we
discussed
mlc
she
and
I,
don't
remember
who
it
was,
but,
but
you
know,
I
think
this
this.
It's
not.
You
know
it's
maybe
a
little
overrated,
but
I
think
it.
You
know
shows
that
it
is
difficult
to
understand
some
of
the
parts
and
I
think
it's
more
I
think
the
discussions
we
had
it's
very
much
related
to
to
what
I
call
checklist
interactions.
E
For
example,
when
you
do
when
you
do
checks
on
on
on
in
one
check
list,
how
does
it
interact
with
with
other
check
later,
when
are
you
gonna
unfreeze
candidates
in
in
another
checklist
and
which
ones
and
so
on?
So
that's
coming
up
here
later
also,
this
is
something
which
was
came
up
on
the
list
a
while
ago.
E
So
that's
what
we
prune
knows,
but
a
thing
which
came
up
on
a
list
was
that
or
the
question
was
that:
doesn't
this
also
apply
to
2p
reflexive
candidates?
And
obviously,
when
you
have
your
initial
checklist,
you're
not
going
to
have
any
pair
reflexive
candidates
because
they
will
come
later
once
you
start
with.
E
Also,
I
don't
know
if
the
draft
explicit
it
says
that,
but
I
assume
you're
going
to
do
it
because
you
used
you
know
doing
it
because
you're
gonna
have
your
server,
reflects
the
candidates
and
everything
in
there,
so
you
will
prune
it
again,
so
so
so
so
so
the
idea
here
was
that
should
we'd
also
then
prune.
These
appear
reflexive
candidates
and
their
basic
two
ways.
We
could
do
this
and
first
is
that
just
add
to
the
explicit
text
which
talks
about
server
flexi
candidates.
E
We
also,
you
know,
add
you
know,
the
pair
of
lexi
wants
the
second
one,
and-
and
this
is
one
of
the
things
I
talked
about
earlier.
This
may
seem
like
an
easy
fix,
but
we
need
to
make
sure
it
doesn't
have
any
any
side
effects
is
that
we
have
a
more
generic
statement
saying
that
you
always
replace
the
local
buy
it
by
the
base.
I
mean
in
some
cases
they're
going
to
be
identical
to
begin
with,
but
I
mean
that's
fair
enough,
but
we,
but
if
we
always
replace
it,
our
contacts.
H
Just
editorial
because
server
of
Lexington
pure
foxes
are
the
only
ones
where
the
base
is
not
equal
to
the
candidate
because
for
host
and
relayed
they're
equal
yeah.
So
how
are?
How
are
these
two
statements
effectively
different
other
than
editorial
and
how
you
presented
editorial
I
agree
is
important
because
the
documents
confusing
but
I,
don't
think
it's
the
tactical
decision.
I.
E
Talked
to
it
actually
Watari
about
this,
and-
and
he
mentioned
that-
or
he
asked
me
that
I
should
look
into
the
TCP
draft
and
look
at
this.
What
do
they
call
her
NAT
NAT,
NAT
assistant
and
a
sister
candidates
and
and
I
did
that,
but
I
don't
think
it
would,
because
my
understanding
is
that,
even
when
you
use
not
assisted
candidates,
the
the
local
candidate
is
going
to
point
to
the
you
know
whatever.
Then
that
gives
you,
but
the
base
is
still
going
to
be.
E
H
Yeah,
I
mean
again
thought
about
the
tcp
cases,
so
I
mean
it
sounds
like
you
know,
probably
the
describing
it
in
terms
of
two
okay
for
UDP
they're
identical
this
is
editorial,
but
I
agree
that
I
think
to
probably
makes
sense.
If
there's
you
know
extensions
in
future,
though
you
don't
know,
I
mean
I
think
in
participe,
you
don't
signal
them
as
net
as
that
assisted
right.
You
still
will
have
to
use
one
of
these
four
names.
I.
D
And
also
a
lot
of
these
things
in
general
that
are
critical.
Presenting
has
a
bit
about
editorial
nature,
but
the
thing
is
there,
because
the
spec
is
relatively
complicated.
We
don't
want
to
do
big
editorial
changes
without
checking
with
wider
group
of
people.
Is
this
the
right
thing
to
do,
because
it's
easy
to
miss
some
corner
case
that
only
those
two
people
are
aware
and
those
two
people
to
change
who
they
are
so
depending
on
the
part
of
the
ice
out?
D
E
Yeah
I
think
I
mean
my
suggestion
would
be
that
way
we
move
ahead
with
with
alternative
to,
but
you
know
III
make
a
pull
request
so
so
before
we
you
know,
people
don't
have
to
try
to
read
this
in
your
head
and
and
and
people
can
look
at
it
before
we
before
we
submit
a
new
version.
I
mean
it's
a
very
simple
change.
E
Editorial
wise
is
just
in
one
place
in
the
document
where
we
have
so
it's
not
going
to
be
a
big
change
is
its
so
from
from
that
perspective,
it's
easy
and
related
to
this
I
didn't
do
a
slide
on
that,
but
related
to
this
is
I
also
plan
to
align
terminology
about
base,
because
sometimes
the
documents
a
base,
sometimes
it
says,
base
candidate,
which
was
a
little
confusing.
I
think
so.
I'm
prob
probably
going
to
use,
base
or
I
think
candidate
base
it's
better,
but
I'm
gonna
align
that
terminology
anyways,
so
columns
are.
N
Switching
topics
here,
I
I'm,
very,
very
sympathetic
to
how
much
we
tell
you
to
do
one
thing
when
yankee
to
somebody
else
editorial
and
keep
moving
things
back,
I
apologize
in
advance
for
all
of
that,
but
I
mean
it
seems
like
we
have
much
more
important
things
to
be
discussing
here,
like
congestion
control
that
we
don't
really
have
one
that
works,
I
mean
we're
going
to
get
to
that
type
of
stuff.
This
is
editorial.
D
D
N
D
E
Think
this
will
be
good
and,
as
I
also
said,
I
guess
before
he
came
in
here
is
that
my
intention
was
really
not
to
spend
a
lot
of
this
face-to-face
time.
With
these
issues,
I
mean
they
will
be,
I
will
do
pull
request
and
so
on,
but
I
just
want
to
make
people
aware
of
it
that
there
are
some
still
some
work.
We'd
have
to
do
it
before
we
publish
it,
but
I
think
the
idea
is
that
we
move
forward
with
alternative
to.
A
E
Yeah
well,
I
send
it
to
listen
and
yeah
the
new
versus
Amy
shisha
I'm
not
going
to
rebreathe.
That
up
here
is
that,
but
basically
what
he
says
that
there
are
cases
where,
where
you're
going
to
unfreeze
the
whole
checklist,
you
have
a
frozen
checklist
and
their
case
is
where
you're
gonna
unfreeze
the
whole
checklist
I
think
there
was
already
in
Berlin
someone
who
gave
a
comment
that
he
couldn't
find
anything
in
the
spec
saying
that
email
said
that
you
know
yeah.
E
There
is
definitely
these
texts
and
I
guess:
okay,
I
couldn't
find
it
either
the
text
there.
There
is
some
text
and
I'm
going
to
come
back
to
that
actually
later,
where
you
will
unfreeze
for
each
foundation
one
pair
in
the
checklist.
So
maybe
that
was
what
email
was
was
thinking
about,
so
so
so
we
see
about
that,
he
was
actually
going
and
we
had
his
table,
which
you
used,
which
was
I,
think
very
useful,
which
he
used
used
to.
You
know
demonstrate
this
problem.
E
E
I,
don't
think
that
will
happen
either
based
on
where
I
read
and
I
sent
you
an
email
about
that,
but
but
anyway,
so
we
need
to
figure
that
out.
Then
there
are
a
few
aggressive
when
we
need
the
removal
of
the
Gracie
nomination
whenever
we
did
have
a
pull
request
and
we
remove
the
steak.
But
then
I
think
Bernard
find
out
that
there
is
still
some
some
some
some
text
that
needs
to
be
modified
or
removed.
E
E
This
is
something
else
which
came
up
on
on
the
list
and
I'm
not
going
to
present
it
in
a
little
different
ways.
As
you
know
we
have
to
I
mean
we
have
all
these
states,
and
then
we
have
a
checklist
can
be
two
types
it
can
be
active
when
at
least
one
other
candidate
pair
is
in
a
waiting
state
and
frozen
if
all
the
candidate
pairs
are
frozen.
But
then
this
question
is
that
ok,
but
what
if
no
candidate
is
in
the
wedding
state,
but
one
or
more
candidates
are
in
the
in-progress
state.
E
E
You
know,
fix
it's
a
small
editorial,
that
that
was
another
thing,
and
this
is
the
thing
that
came
up
on
at
least
do
we
need
to
talk
about
these
lists
types
to
the
begin
with
I
mean
can't
we
say
at
least
where
you
know
the
pairs
are
frozen,
or
at
least
we're
if
we're
one
or
more
pairs
aren't
frozen.
I,
don't
know,
but
those
kind
of
things
it's
a
little.
When
you
read
this
because
you
have
all
this,
you
have
a
state,
you
have
a
candidate
pair
States,
you,
you
have
the
checklist
state.
E
You
have
the
ice
session
state.
Then
you
have
these
types
of
checklists
and
then
you
have
this
ques
trigger
check.
You
and
I
mean
it's
kind
of
messy,
but
but
that's
really
not.
This
was
more
just
about
this
definition
of
that.
Assuming
we're
gonna
go.
We're
gonna,
keep
the
active
checklist
that
it
should
also
cover
when
a
candidate
bear
is
in
progress.
Alex.
H
I
mean
I,
guess
things:
where
are
these
states
used
and
where
the
document
is
to
say
if
it's
in
this
site
do
this?
If
it's
not
state,
do
that?
That's
the
thing
to
look
at
to
see
what
happens?
You
know
figure
out
what
should
happen
if
you
know
there's
in
progress,
but
nothing
is
waiting
or
everything's
completed
or
failed
or
so
and
so
forth.
You
know
what
is
you
know,
ya
mean
even
if
there
isn't
anywhere
where
this
is
used,
then
you
can
remove
it.
You
know
it's
the
usual.
E
H
Saying
that
I
don't
know
what
the
ants
I
mean.
I.
Think
probably
you
know
if
there's
in
progress,
not
waiting,
but
you
know,
if
it's
you
know,
if
we're
this
is
used,
is,
do
you
pull
one
off
to
start
it
then?
Obviously,
if
you
don't
have
anything
waiting
is
not
relevant.
If
it's
something
else,
then
that's.
Why
that's
why
I'd
want
to?
I
need
to
review
the
spec
or
hopefully
have
you
review
this
back
and
tell
me
to
say
where
this
is
used,
to
figure
out
what
the
right
answers
and.
E
A
similar
question,
I
get
is:
why
do
we
need
is
valid
least,
why
don't
we
just
define
a
new
candidate
pair
state
which
is
valid,
and
then
you
know,
but
idea
again,
maybe
that's
a
little
more
than
what
we
want
to
do
at
this
stage
show.
So
that's
why
I
didn't
add
it
to
it
to
the
slides,
then
we
have
is
sending
from
local
candidate
were
sending
from
base.
There
is
some
few
places
in
in
the
document
which
talks
about
you
know
you
send
a.
E
The
point
is
that
if
you
look
at
the
definition
so
over
or
what
local
candidate
and
what
base
means
you're
always
going
to
send
it
the
connectivity
checks
from
the
base,
so
the
suggestion
basically
would
be
here
is
that
you
know
we
do
that
modification.
Also,
it's
not
in
too
many
places
in
the
document,
so
I
guess
this.
This
would
not
be
a
big
rewrite
so,
but
but
check
that
or
or
I
mean
because
only
thing
the
local
candidate
is
going
to
be
used
for
its
when
the
other
one
send
stuff
to
you
surprise.
E
E
Does
it
need
to
have
some
specific
characteristics?
Is
it
the
most?
You
know
you
know
what
is
it
I
mean
in
and
now
in
someone
can
say
it's
the
first
M
line
in
sdp,
but
keep
in
mind.
This
is
not
this
teepee,
so,
for
example,
in
in
in
the
music
spec,
we
may
say
that
you
know
when
you
use
this
with
cpt
p,
then
first
first
media
stream.
It
means
the
first
M
line
in
stp,
but
I
think
we
should
say
something
here.
E
H
I
agree
that
it's
yeah,
the
ordering
I'm
not
just
which
one
is
first,
but
the
total
ordering,
because
there's
a
lot
of
things
where
it's
not
just
the
first
one,
but
the
next
one
is
also
used
in
the
number
of
places
and
I.
Think
that
has
to
be
defined
by
the
using
respects
or
what
is
the
order?
The
total
order
of
the
media
streams
and.
E
I
was
thinking,
for
example,
when
I
talk
about
these
characteristics,
I
mean
I,
mean
I,
guess
in
most
cases,
if
you
have
multiple
streams
in
a
normal
case,
you're
gonna
have
an
equal
amount
of
candidates
for
each
of
them,
but
in
theory
you
could
have
I
mean.
Let's
say
that
you
have
some
kind
of
media
stream
that
you,
for
whatever
reason
you
always
want
to
send
it
through
a
a
turn
server.
So
you're
not
gonna,
collect
other
candidates
for
it.
E
Let
me
finish:
this
I
have
it
in
my
head,
so
that
saw
them,
but
then
for
your
ad
or
other
media
streams,
you
can
have
all
this
host
local
or
whatever
so
I
thought
that
first
media
stream,
the
characteristics
should
probably
be
a
media
stream.
Where
you
have
a
lot
big
large
set
of
candidates,
you
don't
think.
H
Two
are
entirely
disjoint
yeah
because
you
know
they're,
actually,
two
different
boxes
that
you're
controlling
which
we
have
to
handle
I,
think
this
gets
into
the
email
table
issue
where
it's
like
it's
not
it's
the
first
stream
that
has
this
foundation
that
caught
that
you
know,
and
then
things
like
that
and
that's
possibly
the
whole
description
of
the
algorithm
needs
to
be
I.
Think
so.
There's
like
I
think
I
pointed
this
out
to
JDR
latent
in
the
development
of
5245
and
I.
H
Think
there's
some
word
some
text
in
there,
which
is
already,
which
is
kind
of
flaky
about
that.
So
we
might
want
to
even
real
make
sure
we
actually
have
this
algorithm
well
specified.
So
it's
like
you
know
the
thing
you
want.
You
know
you
start
check
started,
know
the
first
media
stream
that
has
this
foundation
or
that
I'm
not
described
I'm,
not
guy
works
with
of
the
mic,
but
basically
that's
what
starts
out
as
unfrozen.
You
know
the
first,
the
first
check
with
this
foundation
or
something
I
mean
you
I,
hopefully
understand
what
I
mean.
H
This
is
the
meal
table,
but
this
is
the.
This
is
the
area
where
we
need
to
make
sure
that
this
is
right
for
these
like
distributed
media
cases
or
you
know,
like
you,
said,
one
legal
assistants
return
or
you
know
once
one
media
stream,
the
turn
server
failed.
So
you
don't
have
a
turn
candidate
for
that
stream,
or
something
like
that.
That's
bunch
of
cases
like
this
and
it
also
interacts
with
a
lot
of
the
trickle
cases.
Yeah.
E
E
We,
the
I've
I,
don't
well
as
far
as
5245
beasts
is
concerned,
only
the
first
one
matters,
because
then,
when
you
when
you,
when
you
find
when,
when
I
six,
so
when
I
check,
succeeds
you're
going
to
go
through
all
your
other
checklist
and
see
if
you
have
matching
foundation
in
those,
there
is
no
word.
There
is
no
text
about
saying.
Then
you
go
to
the
second
checklist.
H
And
then
is
there,
but
but
there's
the
issue
that
I
think
I
just
mentioning.
That
is
again
the
distribute
media
case,
where
you
know
if
a
certain
foundation
only
appears
in
the
second
one
is
it's
like
in
the
second
and
the
third,
then
you
go
and
start
that
check
on
the
second
I
know,
which
one
is
second,
you
know
you
need
to
know
have
a
total
order,
so
we
can
find,
which
is
the
first
one
for
each
foundation.
I.
E
E
D
A
little
have
a
look
at
that,
but
that
Jonathan
do
we
have
a
us
a
volunteer
to
make
a
pull
request
on
clarifying
text.
Oh.
H
D
So,
let's
figure
out
the
call
on
that
Jonathan
Emil
christer
more
sometime
soon
after
this
meeting
excellent.
Ok,
we
had
notes
note
takers.
Take
that
note
of.
E
Yeah
and
then
this
is
just
a
small
check,
but
again
a
question.
I
got
a
stalker
connectivity
checks
in
in
in
the
definition
for
waiting.
The
definition
waiting
for
says
is
that
check
has
not
been
performed
for
this
pair
and
then
comes,
and
it
can
be
formed
as
soon
as
it
is
the
highest
priority
waiting
pair
on
the
checklist.
But
that's
not
really
true,
because
you're
always
gonna
send
trigger
checks.
E
First,
no
matter
what
the
priorities
at
least
I
couldn't
find
any
text
saying
that
when
you
when,
when
when
you
have
a
trigger,
when
you
trigger
checks,
suddenly
you
know
you
raise
the
priority.
I
mean
the
priority
has
been
calculated
before
a
pair
becomes
is
added
to
that
trigger
queue.
So
so
so
I.
H
E
H
Wow,
oh,
but
his
point
is
that
if
the
trigger
queue
is
not
empty,
then
even
though
something
is
the
highest
party
waiting
for
the
checklist,
it
still
can't
be
performed
immediately.
I
think
is
the
argument.
So
this
is
the
wording.
There
is
not
quite
right.
That's
what
he's
claiming
which
I
guess
I
kind
of
see.
Yeah.
E
And
I
mean
my
point:
is
that
I
don't
really
think
in
the
definition
for
waiting?
We
don't
really
need
to
say
when,
when
when,
when,
when,
when,
when
something
is
gonna
be
sent,
I
mean
I,
don't
think
it
was
so
I
think
basically
I
think
we
could
could
could
could
remove
this
and
just
says
that
that
a
check
has
not
been
performed
for
this,
but
will
be
it's
ready
to
be
performed
when
I
played
this
some
some
word.
But
my
point
is:
we
don't
really
need
to
talk
about
this,
because
this
belongs
elsewhere.
E
Yeah
yeah,
something
like
that
yeah
to
make
sure
this
is
now
in
a
state
where
it's
going
to
be
sent
or
OSHA
and
related
to
that
is
that
in
the
same
sections,
the
definition
for
frozen
says
that
are
checking
a
check
for
this
pair
hasn't
been
performed
and
it
can
can't
be
performed,
and
then
you
have
involved
until
some
other
chick
succeeds,
but
pairs
can
actually
be.
Unfrozen
also
do
to
address,
and
it's
not
all
always
when
something
succeeds.
E
For
example,
if
they,
if
they,
if
the
timer
expires-
and
they
are
no
waiting,
no
pairs
in
the
waiting
state,
then
you
are
going
to
start
unfreezing
the
frozen
ones.
But
that's
done
because
of
that
not
because
of
another
check
being
done,
but
again,
I'm
not
sure
how
much.
Where
do
we
need
to
really
specifies
here?
You
know
when,
when
and
if
they're
going
to
be
a
baby
unfrozen,
the
important
thing
here
is
that
saying
that
they
are
frozen
and
and
and
not
they
will
not
be
checked
until
they
are
unfrozen.
E
E
I
think
you
shouldn't
really
when
oh,
when
the
only
thing
you
have
left
is
unfrozen
checklists.
When
you
don't
have
anything
else,
that's
going
to
trigger
the
text,
then
you
start
unfreezing
them.
If
you
still
have
them
and
I
guess
this
could
happen.
If
in
this
you
know
that
we
talked
about
earlier,
but
I,
don't
think
just
because
one
checklist
is
done,
you
should
go
and
unfreeze
everything
else.
I.
E
E
D
E
Yeah,
so
what
I
was
saying
so
what
I
was
saying
is
that
it
was
this
discussion
with
with
the
transport
people.
You
were
involved
in
that,
and
there
was
a
pull
request,
which
was
merchant,
and
it's
now
in
the
latest
version
of
the
beast
draft,
and
there
is
currently
a
pull
request
to
align
Appendix
B,
very
small
change.
Then
I
said
hey.
This
is
basic.
What
the
change
is
and
I
see
the
researcher
arrow.
It
says:
50
milliseconds
there
it
should
be
5
milliseconds
and
that
actually
buggy
existing
in
the
in
the
draft.
E
Also,
I
checked
that
so
so
it
was
not
mine,
but
basically
what
it
says
here
is
the
tea.
The
tea
a
value
is
5
milliseconds,
but
you
can
use
either
values.
Then
it
talks
about
this
global
T
value
to
take
that
into
consideration
and
then
read
some
text
talking
about
why
5
millisecond
is
good
for
ice,
but
it's
not
good
for
transport
protocol
in
general.
So.
N
I
N
N
I
I
I
N
D
P
I
Again,
you
can
go
check
the
email
thread,
but
I
mean
I
was
very
clear,
saying:
okay,
this
is
this
is
what
you
want
and
then
write.
The
Rope
wrote
the
poll
request
with
all
the
explanation.
They
proof
read
the
pull
request.
They
specified
here,
the
ITF
RFC's
that
you
that
we
need
a
reference
and
everything
so.
O
So
cold
feet
about
Colin
pokin,
so
I
mean
I'm
not
disagreeing
with
any
of
that
I
believe
you
did
all
of
those
things,
but
as
a
transport
person
myself
and
not
someone
who's
commented
on
this
I
suspect.
If
you
go
to
the
transport
and
say
it's
a
hundred
packets
outstanding
without
any
feedback,
you'll
get
her,
oh
really
so,
as
I
say,
I
would
recommend
and
I.
If
you
dont
les
entirely
up
to
you,
but
I
would
recommend
you
double
check
with
transferred
a
tease
I,
don't.
P
Yeah
Alyssa
Cooper
I,
don't
so
yeah
I
mean
there.
You
can
send
it
to
tsv
art,
but
I
really
feel
that
it's
unfair
to
you
to
be
badgered
about
what
you
did,
because
you
did
exactly
the
right
thing
and
there's
a
long
thread
about
this
on
the
ice
mailing
list.
That
includes
a
lot
of
input
from
Jenna
and
David
black,
which
were
the
instigators
of
this
originally
so
I
I
think
you
did
a
good
job
with
this.
So
just.
I
One
more
to
think
the
back-up
plan
is
basically
to
you
know.
If,
if
somebody
says
you
know
we,
we
can't
go
forward
with
just
rule
3
out
of
the
three
because
reduced
down
to
the
third.
We
could
just
back
up
and
expand
it
back
to
the
three
rules
and
the
three
rules
are
still
basically
in
there,
but
there
as
part
of
an
explanation
and
if
it
can't
the
worst-case
scenario
is
that
we
take
the
reduced
rule
and
expand
it
back
to
the
three
rules
and
make
it
not
normative
instead
of
vicks
explanatory.
N
J
Is
ben
campbell?
One
comment
on
history:
how
we
got
here,
I
realized
that
there's
always
a
concern
that
the
tsv
a
DS
may
surprise
this,
but
we
got
here
by
going
to
the
TS
of
the
80s
and
they
sent
us
to
these
people
to
talk
about
it.
So
if
we
come
back
to
the
tsv,
a
DS
we're
just
going
through
the
circle.