►
From YouTube: IETF99-TRILL-20170721-1150
Description
TRILL meeting session at IETF99
2017/07/21 1150
https://datatracker.ietf.org/meeting/99/proceedings/
B
C
B
C
B
B
B
A
A
B
Trill,
oh
you're
in
the
right
place,
trill
good,
good,
I,
don't
know
your
name,
so
I'm
sue.
B
You
you're
welcome
to
sit
up
front
you're
welcome
to
sit
in
back,
but
this
is
a
discussion,
so
you
can
do
whatever
you
want,
I'm,
very
flexible.
We
should
go
back
to
the
note.
Well,
yeah,
yeah,
that's
good!
Okay!
This
is
a
note.
Well,
this
is
a
new
note.
Well,
just
in
case
you
didn't
know,
so
you
should
read
it.
B
There
is
more
included
and
I.
Think
that's
the
best
thing.
I
can
say:
we've
got
it
in
the
notes.
Anyone
who
want
to
ask
any
questions
about
it,
we'll
just
go
in
good
thing
way:
I'm
glad
you're
here
Donald
sent
a
message
to
the
list.
You
should
look
at
it
because
we'll
come
back
to
you,
okay,
so
after
the
trill
note
well
go
ahead.
B
Okay,
17th!
We
are,
we
are
trying
to
get
everybody's
documents
through
trill
in
the
next
six
months,
so
I
need
your
help.
We're
going
to
have
we're
gonna
ask
for
reviews.
Donald
will
help
me
and
John
and
I
will
ask
for
review.
So,
if
you'd
like
your
documents
to
go
through
help
everybody
else's,
please
it's
gonna.
Take
time.
Do
you
want
to
say
something
brilliant,
yeah.
D
Usually,
when
I
do
an
ad
you
review,
if
you
reply
back
and
get
it
fixed
up
really
quickly,
I
can
schedule
it
had
it
last
call
and
on
the
teletype
iesg
tell
a
chat
is
planned
if
you're
not
responsive,
it
could
take
a
lot
longer
and
we're
trying
to
get
these
things
all
done,
while
I'm
still
AJ.
Thank.
B
B
B
So
if
I
tell
you
that
I'll
give
you
my
own
review,
but
Donald
has
graciously
offered
to
help
with
rewrites
and
you've
experienced
that
before
you're
gonna
have
to
help
us
with
the
review
and
last
call
when
I
do
the
last
call
there'll
be
questions
so
that
we
indicated
that
there's
interest
okay,
so
let's
go
through
the
documents.
We're
gonna
work
on
there
is
excuse
me
for
my
phone.
B
Then
there
are
some
there's
a
draft
in
publications
but
needs
to
change,
because
the
LEAs
got
some
really
good
comments.
Now
we
have
drafts
past
the
working
group
last
call,
but
we
had
significant
comments
late
now,
there's
a
good
reason
for
this
and
that
Ally,
if
corrected
I,
think
is
a
politically
correct
way.
The
review
process
in
the
routing
area
trill
usually
gets
at
least
around
QA
review.
It
gets
an
Operations
review
and
make
it
a
security
review.
B
We
weren't
getting
them
quickly.
All
and
thrill
over
IP
got
a
transport
review.
If
these
reviews
get
behind
and
that's
my
job
is
to
press
the
button
and
Donnell
and
I
will
track
it,
it
will
happen
what
happened
with
12
over
IP
drill,
p2
and
pbft
and
resilient
Rios.
If
we
get
the
comments
late,
then
we
have
to
resolve
it
right
and
you're
sort
of
like.
Where
was
this
draft,
so
we're
gonna,
try
to
really
increase
the
cycle
time
by
taking
the
input
and
then
trying
to
track
you
down
to
get
the
response
again.
B
E
B
Think
we're
good
on
that
one
and
I
think
trill
and
the
resilient
real
trees
is
fine,
I'm
being
yeah
yep
we're
both
good
on
that
Donald
and
I
are
trying
to
help
that
go
ahead
for
the
next
status.
That's
it!
Okay,
we're
in
working
group
last
call
and
we've
left.
The
working
group
last
call
to
have
a
discussion
on
the
changes
same
way.
The
EC
and
support
I
haven't
seen
any
crumbs
with
and
the
directory
assisted
in
caps.
B
We
got
some
private
reviews
that
we'll
call
it
so
we're
going
to
leave
those
in
reviews
to
see
if
we
can
clean
it
up
rapidly.
If
not-
and
you
don't
fix
it
rapidly,
then
we've
got
to
stop
and
do
another
two-week
last
fall.
I
really
want
to
try
to
keep
pushing
it.
We
have
two
calls
for
adoption
that
we
talked
about
last
time,
which
is
the
group
King
and
the
Archaea
parent
selection.
Now
for
Donald
as
the
main
author
for
Rakesh
yep,
Rakesh,
good
I,.
A
B
F
C
B
Sorry
I
should
have
said
Rama
instead
of
Rakesh.
My
apologies
wrong
Kumar.
When
we
get
this
adopted.
This
has
to
go
fast
through
the
process,
because
it's
just
being
about
that
donal
already
got
warned.
So
what
you're
going
to
get
is
you're
going
to
get
an
immediate
set
of
QA
reviews
once
we
adopt
it
as
far
as
I
can
tell
this
one's
good
for
adoption,
but
just
to
warn
everybody.
B
I
am
before
as
a
chair,
I'm
pulling
this
in
I'm
warning
everybody
we're
on
the
last
six
months
of
all
these
documents,
we're
trying
to
get
all
the
documents
through
because
a
lot
of
our
workers
through
and
into
product,
and
then
we're
trying
to
wrap
up
all
the
documentation
again.
The
reason
why
we
have
all
this
back
load
is
we
weren't
getting
QA
reviews.
Now
we
thanks
to
alia,
we
fixed
that
process,
I'm
going
to
depend
on
you
to
help
me,
get
it
through.
Okay.
B
Next,
all
of
these
other
drafts,
you
notice
a
despot,
Rev
DCI.
We
need
to
know
if
there's
real
usage
for
DCI
or
whether
it's
going
away.
If
it's
some,
some
of
the
guys,
said
the
use
case
in
the
need,
it
might
be
going
away
again.
Some
of
these
use
cases
come
out
of
India
and
some
of
their
networks
are
changing
the
topology,
the
multi
SiC
single
level
and
unique
multi
level.
B
Nicknames
now
focus
on
those
ones,
mainly
on
sand
and
he's
out
these
two
you
may
be
receiving
on
the
single
and
unique
some
comments
from
Donald
I.
Don't
know,
I
met
early
this
week
and
walked
through
all
the
documents.
Please
pay
attention
when
we
go
through
the
revision,
because
we'll
do
one
more
revision
into
working
group
last
call
I,
don't
think,
there's
much
technically
but
Donald
alive.
The
drafts
needed
a
little
bit
of
fixing.
B
Yep
so
that
one
we're
probably
just
gonna
drop
one
off
the
list:
oh
yeah,
trill
yang,
om
I,
have
a
lot
to
say
on
so
I'll
put
that
on,
but
we're
caught
there
between
the
revised
data
stores.
I
mean
we
ease
up.
You
know
if
the
trill
yang
is
been
implemented
in
Huawei
I
talked
to
Deepak
about
Cisco.
You
know
these
implemented.
The
reason
is
the
yang
debate.
We've
got
the
revised
data
store
anything
that
I
do
I'm
having
to
do
like
I
did
with
the
BGP
stuff
I
have
to.
B
If
it's
out
there,
I
need
to
publish
what's
there
and
then
I
need
to
move
at
the
same
time
to
the
revise
Davis
North.
That's
why
I
need
to
know
if
somebody's
got
it
out
there
I
will
put
the
revised
draft
out
with
the
BGP
mem
for
admin
with
the
BGP
yang,
modelled
there
were
15
people
and
I
know
that
cisco
has
the
yang
in
the
OEM
and
so
I'm
tending
to
put
it
out
there.
B
But
I
just
need
to
know
if
there's
more
than
one
okay,
so
you
don't
have
to
answer
on
list
unless
you
I'm,
not
the
young
people,
but
I'd
like
to
go
back
to
check
whether
it
if
you
would
do
that.
That
would
be
very
helpful
and
I'll
check
with
Cisco
to
see
if
they're
still
Des
Moines.
B
G
F
B
B
Yes,
so
we've
completed
the
milestones
and
we're
doing
well
on
that.
Let's
go
to
the
next
one.
We
have
overdue
ones,
but
again
those
are
based
on
the
reviews
that
we
didn't
get
into
late.
So
we're
gonna
try
to
focus
on
trill
over
IP
focus
on
T,
multi-level
and
the
our
bridge
stuff.
So
I
think
we've
got
drafts
for
everything
and
we'll
go
from
there.
Okay,.
H
C
B
I
I
The
draft
specifies
the
packet
format
and
the
security
it
tries
to
cover
transport
considerations,
but
we'll
find
when
we
bring
up
the
issues
that
we
missed,
some
of
them
MTU
fat
flows,
quality
of
service
and
middle
boxes.
Next,
the
existing
draft
specifies
three
formats
for
trill
and
it's
extensible
so
that
there
can
be
more
formats
in
the
future
if
they're
needed.
I
There
are
this
is
trying
to
give
us
a
sense
of
what
they
are
on
the
wire
there's
trill
encapsulated
in
UDP
over
the
link,
there's
trill
encapsulated
in
Ethernet
in
VX,
LAN
and
UDP
over
the
link
and
there's
trill
encapsulated
in
TCP
over
the
link.
Those
are
the
three
formats
that
are
defined
next
page.
I
I
I
This
is
allowed
now,
even
in
ipv4
for
RFC
69
36
and
it's
allowed
when
there's
gonna
be
IP
inside
the
tunnel.
So
you
actually
don't
need
the
external
checksum
to
find
out
if
the
correct
end
destinations
like
the
correct
end
destination
was
met,
and
so
we
need
to
make
the
draft
clearer
about-
and
we
mentioned
it
in
the
draft
that
we
could
use
zero
check
sums.
But
we
need
to
make
it
clearer
about
when
drill
over
IP
meets
the
conditions
for
our
C
69
36
and
when
it
doesn't.
That
was
the
feedback.
I
I
mean
I,
guess
it
could
be
conceived
of
as
a
textual
as
a
technical
change.
If
someone
had
turned
off
checksums
in
a
case
where
they
shouldn't
have,
but
it
it's
mostly
a
clarification,
because
I
don't
think
that
would
that
people
would
probably
have
done
that
congestion
control.
The
current
draft
has
a
good
start,
uncongested
control.
It
doesn't
just
ignore
the
topic
among
other
things,
but
there
are
a
couple
of
things
that
we
need
to
do
we
look
at
what
the
effect
is
of
using
serial
unicast
instead
of
multicast
on
congestion,
and
what
that
is.
I
Is
that
normally
you
multicast
okay
and
you
multicast
on
a
link,
and
it
goes
everywhere
I'm
in
some
cases,
when
either
link
doesn't
support
multicast
or
you
have
some
other
reason
to
do
so.
You
serialize
multicast,
so
you
send
a
message
to
every
one
of
the
other
trill
nodes
on
the
link
that
obviously
is
going
to
create
more
traffic
and
we
need
to
at
least
write
text
that
explains
the
impact
and
I
suppose
in
doing
that,
we
might
find
there's
a
problem,
but
I
don't
anticipate
that
I
just
think
we're
gonna.
I
The
other
thing
is
that
we
always
have
a
problem
in
the
IETF
with
anything
that
tunnels
drill
over
IP
is
effectively
tunneling
right,
we're
tunneling
tril
in
IP,
and
whenever
we
tunnel
things
we
get
the
question.
How
do
we
know
that
the
traffic
inside
the
tunnel
is
congestion,
control?
There's
a
rule
in
the
IETF
that
anything
that
runs
over
IP
needs
to
provide
congestion
control.
I
Okay,
TCP
provides
congestion
control,
MMP
TCP
provides
congestion,
control
quic
provides
congestion,
control
HTTP
because
it's
over
TCP
does
but
UDP
does
not,
and
applications
running
above
UDP
arp
required
to
provide
it.
Okay,
like
like
all
the
ietf
applications
running
above
UDP,
there
are
two
cases
where
we
can
run
into
traffic,
that
isn't:
congestion,
controlled.
I
Okay,
that's
UDP
applications
developed
by
someone
else
where
they
didn't
bother
to
do
congestion,
control
or
non
IP
traffic,
and
so
this
has
come
up
in
a
lot
of
tunneling
situations
before
the
question
is:
how
can
we
tell
if
it
is
congestion,
controlled
and
typically
in
the
IETF,
we've
kind
of
waved
our
hands
and
said?
Well,
if
it's
IP,
it's
all
congestion
controlled,
even
though
we
know
that
isn't
entirely
true.
I
Okay
and
the
answer
is
you
can't
necessarily
tell
and
I
don't
know
we'll
have
to
figure
out
exactly
what
to
say
about
this
in
the
spec
right,
because
you
really
can't
tell
if
a
UDP
application,
you
know
nothing
about.
Has
congestion
control
I
mean
that
you'd
have
to
be
like
a
mind.
Reader
I
mean
so
it's
unclear,
I
think
we'll
just
have
to
say
that's
a
risk
or
something,
but
the.
I
Right
but
it
because
we're
tunneling:
okay,
if
that's
just
on
your
local
link,
it's
okay
right,
because
well,
it's
not
okay,
but
you're
gonna
figure
it
out.
You're
gonna
be
able
to
figure
out
what
node
is
doing
it
you're
gonna
be
able
to
walk
into
their
office,
but
when
you
start
putting
stuff
in
a
tunnel
and
sending
it
across
the
internet,
then
you're
possibly
congesting
the
Internet,
not
just
your
own
office,
and
so
it
is.
I
It's
controlling
provisioning
of
wireless
access
points,
okay
and
it
breaks
an
AP
into
two
parts.
It
allows
an
IP
tunnel
between
them
and
we
had
to
deal
with
the
same
issue
and
I
think
we
should
go
look
at
what
the
wording
was
in
that
draft
and
see
if
it's
still
considered
sufficient
I
I,
don't
know.
I
I
F
I
D
I
We
really
we
really
want
to
finish,
and
most
of
these
are,
although
in
some
cases
they're
gonna
require
some
working
out
they're,
not
really
spec
changes.
They're
they're,
like
doing
your
homework,
checking
that
things
are
act,
that
there
is
no
real
concern
and
documenting
that
I
understand
next
side.
I
Man
and
it's
a
real
thing-
I
mean
it's
a
fairly
recent
RFC.
We
say
nothing
about
it.
We
need
to
actually
say-
and-
and
in
this
case
it
is
a
spec
change,
because
we
need
to
do
what
RFC
60-40
says
we're
supposed
to
do.
It's
a
little
trickier
that
then
just
follow
the
RC
6040.
So
explain.
The
plan
such
as
it
is
so
a
trill
over
IP
link
is
an
IP
tile.
It
should
support
ECM
when
the
trill
campus
supports
ECM.
I
The
trill
over
IP
header
should
be
easy
and
marked
based
on
EC
and
marking
in
the
trill
header
of
the
packet
being
transport.
Those
pictures
for
this,
so
you
don't
have
to
just
get
this
the
first
time
packet
being
transported
with
IP.
If
ecn
is
not
supported
in
the
trill
campus,
it
does
not
seem
worth
while
to
dig
past
the
trill
layer
to
get
ecn
from
into
the
inner
IP
traffic.
So
we're
gonna
propose
this
next
slide.
I
If
you
got
in
our
bridge
and
some
IP
routers,
because
this
is
your
tunnel-
the
green
line
to
another,
our
bridge,
so
you're
running
trill
over
IP
in
this
middle
green
line.
Right
and
over
here
on
the
sending
our
bridge,
the
the
ingress
of
the
tunnel
for
this
particular
packet,
you
have
a
link,
header
and
a
trill
header,
and
that
trill
header
has
a
vcn
mark
in
it
and
payload.
I
Then
we
will
copy
the
ecn
mark
from
that
trill
header
into
the
outer
IP
header,
so
that
ECM
can
be
tracked
across
the
tunnel
and
then
we
will
combine
them
back
into
the
trill
header
ecn
mark
okay,
because
while
it's
pasta
passing
through
the
IP
routers,
if
there's
congestion,
they're
not
going
to
put
the
ECM
mark
in
the
trill
header
to
them-
that's
just
payload.
They're
gonna
put
it
in
the
IP
header,
the
outer
IP
header
right.
So
we
want
to
recombine
them
so
that
if
either
of
them
is
showing
congestion,
we
show
congestion.
I
Okay,
that
way
the
further
traffic
can
act
on.
You
know
that
the
destination
whatever
in
the
other
side
of
the
trail
campus
can
act
on
that
congestion.
Now
next
slide.
If
the
original
our
bridge
is
doing
trill
and
inside
the
the
trill
header,
the
IP
header
has
ecn
marks
and
the
ECM
mark
can
just
be
I
support.
I
Acn
or
it
can
be
that
there
is
congestion,
I,
don't
remember
what
the
two
codes
are
for
that,
but
they're
actually
three
codes,
two
of
them
are
I
support,
ECM
and
one
of
lutely
has
there's
congestion,
but
but
that's
just
for
historical
reasons.
If
the
IP
payload
there
has
a
nice
en
mark,
okay,
we
never
parse
into
that
IP
payload
they're,
like
some
of
the
hardware
that
supports
trill,
doesn't
like
look
at
that
IP
payload
at
all
to
decide
what
to
do
with
this
packet.
I
This
is
to
do
something
and
nobody
does
it
right,
and
so
we
need
to
talk
to
the
transport
area
and
see
if
they
can
live
with
that
that
level
of
support
for
EC
yet-
and
he
said
it
was
okay,
so
probably
it'll
be
okay,
because
Bob
Briscoes,
the
one
who
who
wrote
RFC
6040
or
whatever
it
was
so
so
anyway.
So
that's
our
proposed,
fix,
Frisian,
I.
Think
we'd,
like
probably
to
know.
If
anybody
here
has
any
legally
strong
thoughts
about
that
one
way
or
the
other.
B
I
G
I
I
This
is
a
common,
that's
a
differentiated
services
code,
point
I,
think
I
I,
don't
like
to
use
acronyms
without
saying
what
they
are.
Trill
over
IP
provides
a
mapping
from
trill
QoS,
so
the
sort
of
four
bits
to
the
dscp
bit
that
mapping
needs
to
be
reviewed
based
on
the
latest
are
sees.
Apparently
we
may
not
have
gotten
it
entirely
right,
based
on
the
latest
are
sees,
or
maybe
we've
been
overtaken
by
events.
I
I
But
also,
we
need
to
say
in
our
draft
now.
This
is
kind
of
this
is
one
of
these
things.
I
sometimes
object
to
in
the
ITF.
Actually
we're
gonna
say
in
a
draft
that
no
one
who
is
running
this
service
is
likely
to
read
that
they
should
be
aware
that
if
they
use
troll
over
IP,
the
actual
QoS
they
get
over,
the
Internet
may
be
different
than
the
kind
of
QoS
they
get
over
trip.
D
Sometimes
things
that
seem
obvious
to
us,
not
as
obvious
to
the
readers
of
the
documents
after
you've
been
I
reminded
so
I.
Just
when
I
first
took
electrical
engineer,
you
know,
like
magnetism
stuff,
my
TA
claimed
that
if
you
just
looked
at
the
charge
distribution,
you
could
totally
see
what
the
field
lines
were
and
by
the
end
of
the
class
we
got
there
I
think
there's
a
lot
of
IETF
work
like
that
where
we
think
all
of
the
implications
are
obvious.
Yeah.
I
But
the
thing
is
the
person
who's
gonna
be
reading.
Our
spec
is
the
one
who
is
implementing
the
bit
mapping
all
right
and
they
should
not
get
confused
by
the
sentence.
They
should
do
what
we
said.
Okay,
it's
a
problem
is
in
operations
right
in
operations.
If
you
drill
over
IP
the
depending
on
how
your
routers
that
we
don't
even
know
the
brand
of
interpret
dscp
colored
points,
you
may
get
different
quality
of
service
than
you
expect,
because
yeah.
J
I
think
the
problem
is
that
some
providers
interpret
them
differently
and
we
do
say
that
you
that
you
have
to
be
able
to
configure
this,
giving
you
a
hint
that,
if
that's
your
problem,
maybe
what
you
want
to
do
is
configure
the
mapping
to
be
different
from
what
we
give
is
the
default.
But
the
other
problem
is
that
if
you
go
through
multiple
providers,
kudos.
I
Yes
longer
and
not
better,
but
it
will
not
hurt
the
spec,
so
we
will
say
the
spec
in
the
case
of
transport,
in
a
case
of
transport
using
TCP,
so
that
case,
where
trill
over
IP
is
using
TCP.
So
it's
drill
over
TCP,
/
IP,
a
separate
TCP
connection,
/
provided
QoS
level
is
needed
in
order
for
the
QoS
stork.
I
This
is
because
a
typical
one
connection
is
going
to
be
cached
with
the
QoS
that
first
presents
with,
and
so,
if
you
don't
do
separate
connections
for
the
separate
QoS
values,
all
the
traffic
will
get.
If
you
put
it
all
in
one,
TCP
connection
will
get
the
same
QoS
and
it
will
be
whichever
one.
The
router
first
happened
to
notice
and
associate
with
this
flow,
which
is
not
what
you
want.
I
Is
that
a
that's
about
it
to
say
on
this?
So
next
slide
TCP
transport?
We
offer
the
TCP
transport
as
one
of
the
format
options.
They
said
that
they
would
prefer
that
we
call
it
a
transport
when
we
run
it
in
TCP
instead
of
an
encapsulation,
since
there
is
no
one
to
one
mapping
between
tril,
packets
and
TCP
packets,
and
that
is
very
interesting
because
they
demanded
in
banana
that
I
call
it
an
encapsulation
when
we
put
it
in
mp
PCP,
because
we
aren't
the
transport
MP.
Tcp
is.
I
Joe
and
and
Mary
I
probably
ought
to
talk
to
each
other,
but
we
will
call
it
a
transport
and
not
an
encapsulation,
because
it's
not
actually
each
packet.
Each
trill
packet
isn't
encapsulated
in
one
TCP
packet
that
isn't
how
it
works.
Tcp
is
flow
based,
and
so
there
won't
be
a
one-to-one
mapping,
and
so
he
is
correct.
It
is
not
properly
an
encapsulation.
I
We
will
I
need
a
third
word
for
this,
because
I
need
it
for
banana.
So
if
anyone
has
one,
please
see
me
later,
we
add
framing
with
the
length
field
so
that
the
incoming
TCP
packets
can
be
parsed
and
reassembled,
if
necessary,
to
back
to
the
original
trilogy,
and
you
have
to
do
this
over
TCP,
because
TCP
does
not
provide
any
packetization.
So
3/4
of
your
packet
might
be
in
one
TCP
packet,
the
other
quarter
and
the
first
half
of
the
next
one
might
be
in
the
next
TCP
packet.
I
And
so,
if
you
don't
put
in
some
kind
of
length
or
or
delineator
demarcation,
you
don't
know
where
your
original
packet
started
and
stopped.
So
we
will
do
that
and-
and
there
apparently
some
things
we
can
do
to
achieve
better
performance
and
we
will
take
those
on
her
visor
and
try
to
see
what
we
should
put
in
the
and
the
draft
about
that.
So,
as
I
just
said
when,
when
a
tcp
is
used
to
trill
packet
can
be
split
across
multiple
TCP
packets.
J
We
have
this
empty,
you
thing
and
if
you're
doing
UDP
or
other
things,
then
you
really
do
the
this
this
campus-wide
minimum.
Do
you
do,
though
you
want
the
fact
you
want
to
be
sure
that
your
link-state
packets
get
through
or
else
I
Esaias
won't
work,
okay,
but
the
problem
is:
if
you
have
a
TCP
link,
at
least
in
principle,
everything
I
mean
you've
configured
it
to
calculate
everything
in
TCP,
then
even
your
control
claim
packets
over
that
one
link
can
be
chopped
up
into
a
smaller
TCP
packet.
J
J
J
I
Need
some
wording
about
that?
It
will
probably
still
keep
sort
of
the
MTU
for
the
reassemble
trill
packet,
the
same
as
it
is
now
the
causality
in
in
what
we
say
here.
What
we
said,
what
you
just
said
is
could
be
turned
around
right.
It
could
be
that
if
you
have
a
lower
MTU
link
between
your
sites,
you
should
encapsulate
in
PCP,
because
they'll
be
reassembled
into
the
full-size
packets
and
you
don't
have
to.
E
I
I
There
are
a
lot
of
reasons
why
fragmentation
is
bad.
They
include
middleboxes
where
there's
state
and
they
don't
always
make
sure
that
the
two
fragments
sometimes
they're
blocked.
Sometimes
you
get
issues
where
the
two
fragments
don't
take
the
same
path.
There
can
be
a
lot
of
problems
with
fragmentation
in
the
network.
There
also
are
less
efficient.
They
you
can
get
into
these
silly
packets,
where,
if
your
MTU
is
like
five
larger
than
will
fit
every
time,
you
do
something
you
get
full-sized
packet
and
a
five
byte
packet
and
so
they're
inefficient.
I
Within
a
data
center,
use
of
UDP
with
fragmentation
is
generally
consistent
with
a
trill
philosophy
of
mostly
worrying
about
the
routing
messages
and
not
worrying
about
the
lower
layer
messages,
but
TCP
is
actually
a
better
way
to
avoid
fragmentation
on
lower
empty
links
and
and
also
will
then
allow
the
fragments
front
like
that.
We're
gonna
reassemble
into
full
trail
packets
to
get
through
Nats
and
firewalls.
Next
now,.
K
I
You
need
to
use
static
bindings
for
that,
and
we
need
to
write
up
what
you
have
to
do
in
order
to
get
trill
to
work
over
a
net
neighbor.
Our
bridge
addresses,
as
reported
in
the
source
IP
addresses,
need
to
be
mapped
using
the
static
binding
into
actual
remote
IP
addresses
before
being
listed
in
the
is
is
neighbor
TLD.
So,
basically
we
we
can't
count
on
a
NAT
having
an
Al
g4
trill,
where
it
would
translate
these
inter
addresses.
We
need
to
make
sure
we're
using
the
correct
dinner
addresses
across
a
net.
I
We
generally
need
keepalive
messages
which
we
haven't
needed
in
order
to
make
sure
that
the
NAT
state
stays
active
in.
If
there's
a
long
interval
between
trill
packets.
We
also
see
fragmentation
and
there's
an
interaction
with
security
as
well,
which
is
that
I
have
I
piece.
Aqui
SP
packets
need
to
be
encapsulated
in
UDP
IP
to
go
through
an
ACK.
B
B
I
I
mean
there's
going
to
have
to
be
I,
wouldn't
say
it's
new
technology,
but
it
is
going
to
change
implementations
because
there's
gonna
be
a
question
as
to
whether
they
get
configured
to
be
not
compatible
or
not,
or
do
they
all
run
that
compatible?
We
have
to
define
these
keepalive
messages
and
say
how
often
to
send
them
right.
I,
don't
know.
I
J
J
I
There's
some
I
mean
I,
suppose,
there's
a
possibility
of
saying
something
like
it
doesn't
work
over
now.
But
that
probably
is
not
what
we
want
to
say,
because
if
we
want
to
be
able
to
do
like
the
remote
office
scenario,
those
might
be
behind
a
DSL
link
with
an
app
right.
I
mean
you.
So
probably
we
want
to
support
it.
Working
over
Nats.
D
D
I
I
D
J
D
Right
so
a
nice,
my
perspective
would
be
really
good
to
get
feedback
on
the
mailing
list
as
to
whether
the
NAT
use
case
is
critical
for
folks
who
are
looking
at
deployment
or
implementation.
If
that's
something
that
they're
targeting
or
if
the
focus
is
more
on
v6
networks
and
such
where
NAT
is
not
an
issue,
do.
I
I
could
ask
around
that
if
we
find
that
because
it's
DSL
where
you're
stuck
with
an
app
because
it's
actually
done
by
the
operator
and
so
I,
could
ask
around
and
see
if
DSL
providers
adding
ipv6
or
not
right
and
if
not,
we
could
say
our
answer
to
NAT
in
the
remote
office.
Space
is
if
your
ipv4
services
and
added
use
ipv6.
D
I
B
G
B
I
J
Hopefully,
this
will
be
a
really
short
presentation
of
a
simple
feature,
especially
since
we're
so
winding
down
activity
in
the
working
group.
This
provides
an
escape
channel
whereby
vendors
can
define
their
own
control
messages,
uses
the
arbitral
facility,
which
is
a
way
you
can
send
control
messages
between
our
bridges
or
between
the
end
stations,
and
there
are
bridge
that's
either
they're,
both
or
when
they're,
both
on
the
same
link
and
there's
an
RFC
specifying
how
it
works
and
everything
there's
a
couple
examples
on
this
slide
that
do
make
use
of
it.
J
So
this
is
what
they
look
like:
there's
a
an
ether
type
that
identifies
our
bure
channel
message
and
then
there's
like
a
sub
channel
protocol
numbers
like
a
sub
protocol
within
our
bridge
channel,
and
that
is
what
specifies
what
the
channel
protocol
specific
data
is.
You
could
have
anything
over
there,
so
the
idea
is
very
simply
to
let
vendors
extend
to
this
by
providing
a
way
for
them
to
define
their
own
sorry
messages
and
they
can
send
their
own
control
messages
as
they
desire.
J
The
vendors
can
do
this
in
other
ways
and
if
our
bridges
all
have
IP
addresses
their
MAC
addresses
are
something
that
you
can
imagine
other
ways,
those
actually
not
required
and
in
principle
and
our
bridge
could
have
only
a
nickname.
They
have
all
PPP
ports
that
has
no
MAC
addresses.
You
know
not
very
likely,
but
it
could
be
so
anyway.
This
was
a
universal
method.
It's
guaranteed
to
be
able
to
communicate
to
an
AR
bridge
based
on
the
nickname
or
you
can
also
send
multicast
to
a
label
data
label
scope
set.
J
So
the
idea
is
you
basically
pick
up
a
vendor
channel
ie,
an
affix
a
vendor
Channel
and
in
what
was
the
data
portal
protocol
specific
thing?
We
put
a
little
vendor
channel
header
and
there
could
be
an
organizational
unique
identifier
or
a
company
ID
I
Triple
E,
now
cells
company
IDs
cheaper
than
oh
you
eyes,
though
you
eye
reserves,
the
whole
block
of
MAC
addresses
and
it's
24
bits.
The
only
the
CID
is
48
bits,
but
so
that's
much
cheaper
and
doesn't
give
you
any
MAC
addresses.
J
So
it
should
be
really
a
sort
of
a
simple
format.
Thing
speaks
it
gives
this
facility
to
vendors,
actually
everything
after
the
ouy
or
CID
is
really
completely
under
the
control
of
the
vendor,
but
there's
actually
an
expired
graft
on
this,
which
has
a
recommended
vendor
sub
protocol
and
and
version
for
that
sub
protocol
for
the
vendor
and
I
easy
way
to
report
errors,
and
this
seems
like
a
good
idea.
You
certainly
want
it.
Don't
want
to
have
vendors
have
to
use
up
multiple
Oh
UI
is
just
because
they
want
multiple
vendor
channel
protocols.
B
Donald,
your
mom
after
you
revise
the
draft.
Will
you
make
sure
Indian
announced
after
it
goes
to
the
automatic
announcement?
Will
you
provide
to
the
list
a
description
worth
summation
of
this
description
will
be
used
and
how
it's
happening
so
that
people
can
make
a
good
decision
and
will
option
sure.
J
B
G
L
G
Target
3
review
an
ad
review,
so
there
it
is,
it
has
been
quite
a
long
time
already.
So
the
us.
We
have
some
outstanding
comments
from
the
general
review
in
the
ops
in
the
securities
I'm
going
to
look
at
them
one
by
one
next
page,
so
the
the
general
review
T
it's
a
fairly
lengthy
review,
but
most
of
them
editorial
the
clarification
the
there.
G
G
This
mechanism
is
out
of
scope
in
our
document,
but
the
review
thinks
that
implies
that
some
mechanism
of
this
to
do
to
do
the
key
key
exchange
must
be
existed
so,
but
they
also
think
saying:
there's
something
is
out
of
scope
for
document,
doesn't
I
anything
about
whether
this
mechanism
exists
or
does
not
exists.
So
that's
the
these
are
one
disagreement.
I
B
J
D
So
the
issue
with
this
document
is
the
concern
about
the
security
considerations,
and
that
is
what
there
is
a
discuss
on.
Of
course,
we
always
really
greatly
appreciate
the
reviews
and
comments
that
are
received
during
the
Directorate,
but
it
is
up
to
the
area
directors
who
have
those
directorates
to
decide
whether
or
not
they
are
discussed
worthy
that
is
not
discussed
worthy
and
is
not
blocking.
Thank
you.
B
G
G
Yeah
Sosa
from
the
operational
point
of
view
now,
impact
on
failure
to
optimize
should
be
no
worse
than
their
existing
case
without
optimization.
This
is
I
guess
this
is
what
we
have
done
is
is,
let
me
see
right,
III,
I,
think,
for
from
this
document
we
have
already
specified
right,
yeah,
it's
no
worse
than
without
my
optimization.
So
this
they
should
be
okay.
G
So
therefore,
the
steady
stick
or
counter
both
from
the
operational
in
that
management
view
they
talks
about
the
statistical
counters,
I
guess
so
for
this
one
we
we
think
we
can
actually
it's
up
to
the
implementation,
because
it's
not
really
the
statistical
country's
try
to
remember.
Okay,
for
example,
how
many,
how
many
ARP
Oh
address
requests
had
been
received,
something
like
that,
but
it's
really
up
to
the
implementation,
but
we,
we
can
add
some
text
to
to
give
more
clear
description
on
this
and
okay
move
on
to
the
management
it
talks
about
the
yam
model.
G
How
to
configure
between
different
policy
should
be
covered
because,
okay,
this
talks
about
because
we
have,
we
can
use
different
policies,
for
example,
whether
to
reply
based
on
the
local
cache
or
or
let
the
address
request
passing
to
the
real
destination
in
order
to
do
the
double
confirmation,
so
there
are
different
policies.
So
this
comment
basically,
is
whether
okay,
ideally
as
a
young
model
I,
we
have
a
gain.
B
D
B
B
G
The
the
last
one
okay,
the
last,
is
the
security
from
from
security
I
suggest
to
need
a
clearer
comparison
of
using
or
not
using
the
optimizations.
So
what
we
are
thinking
generally,
this
okay,
we
do
not
think
we
impose
ending
anything
worse
or
better
by
using
the
optimization
from
from
this
Oh.
Actually,
unless
we
are
using
a
very
reliable
directory
to
to
push
and
pull
all
this
address,
mappings,
otherwise
is
we
don't
we?
We
don't
have
any
change
to
current
security
considerations,
so
probably
I
guess
we
are.
We
just
give
some
text
explain.
B
J
I
mean
there's
further
points
here
and
in
general,
some
of
his
points,
which
we
put
on
the
next
slide.
Actually,
it
seemed
like
he
was
asking
for
a
thorough
threat
analysis
of
layer,
2
security,
which
I'm
not
sure
we
need
to
do
like.
He
said
what,
if
the
you,
what
if
the
attacker
creates
a
fake
DHCP
server,
well
sure
that
things
get
screwed
up
it
lets
you,
you
know,
I
mean
there's
what
we're
not
trying
to
solve
the
problem
of
fake
DHCP
servers.
J
D
Think
we
should
talk
offline
and
have
the
you
know
do
a
reference
to
that
and
so
on.
I
will
you
know.
I
have
not
talked
to
Eric
about
this,
yet
if
that
looks
like
it
would
be,
productive
I
can
certainly
do
so,
but
I
would
like
to
have
more
in
hand
on
which
aspects
and
how
they're
being
addressed
before
doing
so.
M
G
Duplicate
IP
address,
so
let
me
see
yeah
hand
on
it,
yeah.
Basically,
what
we're
thinking
is.
Unless
we
have
a
very
complete,
reliable
directory,
then
we
am
then
we,
we
are
very
sure.
Okay,
this
add
up
to
authorize.
These
appear.
Ip
address
should
be
used
by
which
one
otherwise
there
is.
There
is
no
much
thing
we
can
do,
because
the
duplicate
address
I
mean
waster,
no
mention
on
ARP
and
and
either
there.
This
is
the
case,
but
we
don't
do
anything
worse
than
that.
G
So
so
so,
unless
we
have
this
real
a
bottle
tree,
then
we
should
trust
it.
Otherwise
we
have
got
nothing
to
do
with
it.
That's
that's
my
that's
my
thought
anything!
You
want!
No,
no!
No!
Okay!
That's
the
last
one:
oh
there
were
most
likely
logging.
The
duplicate
address
this.
What
what
okay?
This
is
to
suggest
whenever
there
is
a
duplicate
rest
being
detected,
probably
wouldn't
need
to
do.
The
logging,
but
I
kind
of
think
is
still.
G
G
I
thought
about
the
fake
DHCP
server
yeah.
This
is
all
I,
guess
yeah.
So
currently
the
status
is
publication
requested.
We
solve
some
comments,
but
we
have
some
I
guess
two
from
two
rivers
we're
going
to
resolve
them
when
we
resolve
them
and
eighties
have
signed
off
then,
okay,
we
hope
it
won't
be
released.
A
I've
said:
editor
yeah.
Okay,
thank
you.
C
B
From
the
top
okay,
this
is
what
I
mentioned
before
and
I
all
of
them
yang
models
that
haven't
gone
through
the
is
G
are
being
requested
to
do
revise
that
the
revised
data
storm
models
for
yang
I
need
the
input
I
mentioned
earlier.
In
order
to
go
on,
did
you
find
it?
Okay,
yeah
there?
It
is
okay,
just
go
down.
It's
just
one
slide.
It
was
real,
quick,
that's
that's!
B
This
is
my
funny
joke
for
the
thing
this
is
the
pre
and
do
in
the
end
config,
and
that's
how
long
they're
debating
I
thought
it
was
a
cute
picture.
It's
it's
a
dr.
Seuss,
if
you're
a
US
person,
and
then
we
think
it's
resolving
actually
into
me.
The
revised
data
store
the
network
management
I,
don't
know
what
an
MD
a
sense
for,
but
it's
something,
but
the
revised
datastore
draft
is
actually
looks
like
it's
doing
well
and
about
completed
set
some
more
nits.
K
B
Other
thing:
that's
very
I'm,
more
interested
in
talking
about
thin
waist
up.
I
just
need
to
tell
you
that
these
are
all
free,
revive
straps,
the
connection
or
the
mention
of
strap
is
actually
making
progress
through
the
line.
So
I've
gotta
find
out.
Could
you
pack
what
whether
he
wants
the
trip
om
to
go
through
the
connection
with
stuff,
which
you
know
it
is
oh.
B
I'm,
sorry,
I'm,
sorry,
John!
Let
me
go
through
this
again
for
you
guys
there
the
trip
all
the
yang
models.
Whoa,
all
the
yang
models
are
being
revised
because
the
yang
model
structure
is
changing
to
what's
called
a
revised
data
store.
Unfortunately,
it's
at
us
another
one
of
the
working
groups
I
chair,
that
cause
that,
by
adding
dynamic
data
stores
in
cause
it
okay,
you
want
to
tell
me
what
what
a
correctly
there
sure.
D
Oh
yes,
yeah!
No,
it
wasn't
that
it's
the
issue
of
operational
and
intended
state
versus
configuration.
Bias
on
to
the
point
of
the
yang
I
would
add
a
third
option
to
what
you
have
just
to
jump
in
yes
slightly,
which
is
to
take
what's
implemented,
and
you
know,
do
the
presentation,
layer,
transformation
to
the
data
store
and
then
have
what's
actually
implemented
as
an
appendix
in
the
same.
B
D
B
G
L
B
So
but
okay
I
could
check
yeah
if
you
just
double
check,
I
assume,
not
okay
and
does
anybody
have
any
feeling?
No
feeling
is
fine,
we'll
just
do
what's
right.
Okay,
I
see
no
strong
feeling
jumping
up
okay
Donald.
Will
you
bring
up
from
ways
I
think
we've
get
just
enough
time
to
talk
about
the
smart
end,
notes,
problem.
M
B
B
B
N
J
N
N
J
L
L
B
Think
those
are
all
taken
care
of
I.
Think
we've
done
in
notes.
Okay,
I'm
again,
my
whole
focus
here
is
to
get
your
drafts
out
and
in
to
my
intent,
is
to
send
all
of
these
within
two
working
groups.
Last
fall.
Is
there
anyone
that
you
feel
I
mean
other
than
the
Trillo
am
because
I
have
to
deal
with
the
revised
data
stores?
Is
it
and
I'm
not
intending
to
send
trophy
CI
for
all
the
rest?
B
N
B
N
J
J
B
B
B
Will
be
the
announcement
that
it
has
been
adopted?
All
the
drafts
that
we
mentioned
go
down
to
the
adoption
draft,
so
I
can
say
that
again
those
two
were.
These
two
are
adopted:
I
just
I'm,
because
we're
in
a
different
type
of
season
of
just
getting
everything
done,
I
wanted
to
make
sure
you
Donald,
and
the
working
group
knew
that
this
has
got
to
go
fast.
B
As
long
as
there's
no
objection,
all
these
things
will
go
to
working
group
last
call
if
you've
been
a
drill,
graft
and
all
the
adoption
stuff
will
go
in
I
will
do
a
personal
review
and
then
I
will
ask
to
go
to
working
group
last
call.
This
is
this
is
a
different
season.
Okay,
please
respond.
Your
same
great
response.
I'll.
Ask
you
a
question:
I'll:
ask
you
things
about
the
document.
My
last
few
things
about
deployment
you've
been
doing
great
about
responding
any
other
questions,
any
other
thoughts.
Well,
I'll
return.