►
From YouTube: IETF104-MPLS-20190327-1120
Description
MPLS meeting session at IETF104
2019/03/27 1120
https://datatracker.ietf.org/meeting/104/proceedings/
A
A
So
we
are
starting.
Please
find
your
seats.
We
have
a
very
low
number
of
participant
this
time.
I
think
it's
a
one
of
the
reasons
are
that
I
think
people
are
going
to
launch
this
is
normal
launch
time.
Isn't
it
the?
What
worries
me
a
bit
more
is
that
I
don't
see
some
of
the
presenters,
so
I'm
inclined
to
wait
a
couple
of
minutes.
Can
someone
stick
a
head
out
of
the
door
and
I'll
see
if
Adrienne
is
there,
we
need
him.
A
A
A
C
A
A
You
have
to
add
free
to
the
meet
echo
link
for
now.
Yeah,
you
have
to
add
free,
101
plus
three
is
104
yeah.
Okay,
thank
you
and
you
find
the
the
working
group
data
tracker
page
is
useful.
Gyoon,
look
at
it
now,
none!
Okay!
Next!
So
this
preparing
this
again,
that
has
been
a
little
bit
of
a
roller
coaster.
We
had
six
documents
that
people
asked
to
present
very
early
and
those
six
were
actually
removed.
So
now
we
have
a
little
bit
of
elbow
room
in
the
agenda.
A
A
A
A
A
A
Handled
one
of
them,
so
it
shouldn't
be
out,
should
be
it
should
be
a
I
think
it's
really.
It
was
enacted,
the
one
that
Debra
handle,
and
but
there
are
four
more
in
Deborah
and
in
working
group
chairs-
need
to
discuss
a
bit.
What
you
want
to
do
about
them
next
slide.
Yes,
dilly
rotten.
Okay,
we
have
a
number
of
license.
A
A
A
The
others
are
being
prepared,
the
last
one
we
don't
need
to
respond
to
just
for
information.
Okay.
Next
one
document
status
up
till
five
meetings
ago,
we
had
at
least
one
our
seed
per
meeting
cycles
for
the
last
four
we
haven't
had
any.
The
good
thing
is
now
that
we
actually
had
we
had
documents
in
the
or
associated
askew
and
they
should
actually
be
progressed
quite
soon.
I
wonder
is
Bruno
in
the
room.
A
Okay,
the
question
I
have
for
Bruna
is
about
the
mistress
and
the
status
of
that
document.
The
mistress,
a
mistress
to
a
spring
document,
the
mpls
document
in
the
spring
working
group-
and
it
has
been
misread
for
345
days
or
something
like
that,
but
I
hope
it's
knowing
we
have
documents
with
the
iesg
and.
E
A
A
B
The
the
mr.
F
the
this
is
that
the
one
you're
interested
in
is
that
the
base
sr
mpls
document
yeah,
so
that's
completed
its
second
working
Greg
last
call
it
had
a
working
group
last
call
and
it
went
through
the
process
and
got
major
revisions
and
went
back.
It's
completed.
That
last
call
it's
just
at
the
moment,
waiting
for
the
Shepherd
right
up
to
move
forward,
but
it's
because
it's
already
had
quite
a
lot
of
review.
It
should
be
smooth
from
here
on
in
word.
So
who
was?
Is
the
document
Shepherd.
A
C
A
A
A
A
A
Yep,
you
can
jump
on
most
nine.
We
also
had
a
number
of
new
individual
documents,
as
you
see
only
five
here,
but
the
the
some
of
them
are
in
blue
and
I.
Think
for
those
five
documents
they
are,
they
were
intended
to
know.
Four
of
them
were
intended
to
be
on
the
working
group
list
and
actually
helped
in
we've
drawn
the
one.
That's
still
there
is
the
SPL
terminal.
Illium
will
present
seven.
A
A
A
A
A
A
A
A
Here
is
my
favorite
list,
so
I
list
I
listed
all
the
drafts
that
I
have
in
my
my
mail
folders
that
does
not
are
not
progressing
and
my
feeling
is
I
removed
the
ones
that
people
told
me
they
won't.
They
are
progressing
somewhere
else
so
that
we
have
actually
progress
them
or
we're
not
killing
to
do
anything
about
it.
I'll
remove
those
they're,
still
free
pages
of
the
drafts
that
you
classify
as
a
sleeping
beauty.
You
know
nothing
really
happens.
I,
don't
see
anything
happening
about
this
trap
yeah.
So
there
are
errors
in
this
list.
A
Tell
me
about
the
errors
there
are
duplicates
in
the
list.
So
I
will
remove
that,
but
I
want
help
from
people
that
has
a
draft
that
actually
hasn't
been
progress
for
a
long
time
to
actually
say
no
I,
don't
want
the
progress
anymore
or
if
you
want
to
progress,
is
tell
me
that
you
will
do
God's
now
we
kind
of
start
looking
at
it.
A
B
It's
Adrienne,
I
think
a
huge
number
of
these
are
our
missing
replaced
by
links
in
the
data
tracker.
If
you
were
to
put
these
on
a
wiki
page,
then
people
can
individually
go
in
and
edit.
So
if
you
put
the
column
and
the
what
the
right
resolution
should
be,
people
can
go
and
edit
the
resolutions,
and
that
will
be
much
easier
for
us
to
process
and
trying
to
work
with
email
on
such
a
long
list
at
them.
Okay,.
G
A
E
E
There
it
is
so
here's
some
history.
Ldp
was
first
published
back
in
2001
as
a
proposed
standard.
It
was
replaced
by
RC
503,
six,
which
I
think
is
still
a
draft
standard
in
october,
2007
both
specified
the
use
of
TCP
md5
signature,
option
offered
for
LBP
authenticity
and
integrity,
checking
for
LDP
messages
to
prevent
against
LDP
message
spoofing.
E
As
far
back
as
1998,
the
ITF
was
already
aware
of
problems
with
md5
and,
for
example,
you
can
see
RFC
2385,
for
example,
as
an
early
RFC,
which
talks
about
the
problems
with
md5
and
RSC
59
25
back
in
2010
deprecated.
The
use
of
the
TCP
md5
signature
option,
which
is
what
LDP
still
is
specified
to
use
and
replaced
it
with
the
TCP
authentication
option,
TCP
a
Oh
and
basically
that
our
seed
describes
the
option
itself.
It
does
not
talk
about
the
actual
cryptographic,
algorithms
that
you
would
use
with
the
option.
E
That's
discussed
it
with
in
our
C
59
26
to
allow
the
the
cryptographic
algorithms
to
be
updated
separately
from
the
option
itself.
So
in
May,
2013
RFC
69
52
recommended
that
all
TCP
based
routing
protocols
move
from
md5
to
TCP.
Ao
that
would
include
l,
DP,
l,
DP
runs
over
tcp
RC,
74/54
and
fifteen
did
that
for
BGP.
E
So
what
we
want
to
be
able
to
do
is
basically
ask
some
questions
of
the
working
group
come
on
there.
We
go
so
the
the
the
goal
of
this
draft-
and
it
was
written
by
the
book
by
the
co-chairs
of
pals
and
MPLS
working
groups-
is
to
update
LDP
to
replace
md5
with
TCP
ao.
But
the
authors
are
really
LDP
experts
and
we
are
not
security
experts,
and
so
we
have
some
questions
to
ask.
E
First,
we
asking
here
the
MPLS
working
group
and
then
I
will
be
presenting
this
in
the
in
the
routing
area
working
group
as
well
to
have
a
larger
discussion
with
the
routing
area
working
group
as
well.
First
of
all,
how
successful
has
TCP
AO
been
for
for
BGP?
You
know
because
if
it's
not
been
successful
for
BGP,
why
bother
doing
this
for
LD
P
as
well?
Is
it
actually
in
use
in
the
field?
And
if
anyone
can
answer
that,
question,
please
come
on
up
to
the
mic
and
then
the
question
is
well.
E
If
we
specify
TCP
AO,
then
we
have
to
choose
a
particular
cryptographic
algorithm
to
use
with
TCP
ao,
and
we
would
want
that
one
one
that
works
reasonably
well
in
routers.
Without
basically,
you
know
slowing
down
the
processing
of
LDP
messages
and
so
on.
So
we
want
something
which
is
reasonably
good
in
terms
of
the
cryptographic
properties,
but
but
something
that
doesn't
degrade
your
performance
either
and
then
the
third
question
that
we
have
and
then
I'll
start
the
line
is:
does
anyone
really
care
about
this
I
mean?
H
E
I
I
The
question
then
I
think
you're
really
asking:
would
they
care
terrifically
75
with
TCP
a
oh,
it's
just
about
to
ask
you
that
question
I
was
reading
your
mind,
I
don't
know
have
I
mean
and
and
if
I
were
to
ask
them
first
question
that
I
would
need
to
have
the
answer
forest
is
it
broken.
I
haven't
seen
any
problem
and
network
with
that.
I
B
B
B
A
B
I
J
Deborah's
Deaconess
ad
I'm,
very
happy
Adrienne
made
the
statement
he
did
because
if
there
had
been
any
security
abs
in
the
room,
when
we
said
hey,
nothing's
ever
happened,
why
we
have
to
fix
it
they
they
would
have
been
lined
up
at
this
mic
in
a
flash.
So
what
they
want
to
see
really
is
that
we
acknowledge
that
there
are
risk
and
that
we
definitely
have
a
plan,
a
target
to
advise
people
using
RFC's.
What
they
can
do
and
as
the
same
clarify
md5
is
not
good
for
security.
J
K
A
Agree
it's
important.
The
the
thing
I'm
wondering
about
is
really
what
are
the
threats
and
are
those
threats
remedied
by
md5,
probably
not
or
the
remedy
by
TCGA?
Oh,
maybe,
but
as
long
as
we
don't
really
have
it
good
threaten
all
analysis.
It's
hard
to
actually
say
to
get,
do
give
the
correct
recommendation.
Well.
J
We
have
the
one
RFC
from
the
we
had
MPLS
security
considerations
or
something
LAN
had
done
it
and
another's
and
I
mean
it's
also.
The
security
ATS
are
asking,
or
they
out
may
be
out
in
left
field-
that
they
don't
understand
our
threats
and
nor
use
models.
Have
those
used
models
changed,
so
they
they
were
actually
asking
for
for
that
type
of
information
and
to
put
in
the
security
considerations
that
they
don't
have
to
every
time
go
through
with
a
fine-tooth
comb
trying
to
figure
out
well.
J
E
E
L
E
A
Yeah,
so
as
a
piece
of
anecdotal
information
was
that
when
we
actually
put
md5
into
the
LDP,
when
we
did
the
document
the
first
time,
we
already
know
that
there
were
doubts
about
md5
and
we
made
a
note
somewhere.
If
you
have
to
look
into
this
it
only
that
we
never
had
any
indication
that
it
was
necessary.
So
that's
why
we
are
where
we
are
where
we
are.
A
B
Sorry
about
that,
if
you're
listening
remotely,
that
was
me
fiddling
with
the
microphone,
my
name
is
TBD
and
I.
Think
on
the
original
draft
agenda,
lower,
put
down
15
minutes
for
this
draft
and
I
made
the
argument
that
we
needed
30
seconds
so
we'll
compromise,
it's
a
really
small
problem
and
a
really
small
draft
to
solve
it.
B
The
the
problem
basically
comes
from
when
we
wrote
70
to
74,
which
was
clarifying
the
use
of
special-purpose
labels,
how
you
allocate
them
how
you
retire
them.
We
thought
we
had
the
terminology
clear
and
consistent
with
the
IANA
registry
since
then,
as
people
have
started
to
write
drafts
that
use
especially
the
extended
special-purpose
labels,
which
are
only
just
coming
into
use,
we've
found
that
people
a
bit
confused
as
to
exactly
what
terminology
they
should
use.
B
So
we've
written
this
tiny
little
draft,
which
tweaks
the
terminology
and
the
idea
is
then
that
we'll
have
something
consistent.
We
can
use
going
forward
and
will
be
stable,
at
least
for
six
months
until
we
discover
that
this
isn't
clear
either
so,
and
there
are
two
registries
in
the
Ayana
maintains
for
this,
and
they
are
located
on
the
same
I
on
a
web
page.
There
are
two
ways
into
that
way:
Paige.
B
So
when
you
go
to
the
the
full
list
of
registries,
you
see
two
entries
separately,
but,
whichever
one
you
click
on
it
takes
you
to
this
page,
and
both
registries
are
there.
They
are
currently
called
special-purpose,
MPLS
label
values
and
extended
special-purpose
MPLS
label,
values,
which
all
looks
perfectly
cool,
but
both
of
those
registries
contain
label
values
that
are
special
purpose.
So
now
everyone
is
confused
when
they
say
special
purpose:
MPLS
label
which,
which
registry?
Are
they
really
pointing
that?
So
we
propose
to
make
this
clear
by
further
qualifying
the
the
base
registry.
B
M
B
B
As
a
base
label,
anything
that
is
larger
than
15
is
a
real
switching
forwarding
label
and
any
attempt
to
make
a
special
reservation
on
that
would
blow
up
in
all
the
deployed
base.
So
we
had
to
steal
one
label
for
the
excel
and
make
the
extended
okay.
Thank
you,
but
I'm
all
in
favor
of
confusing
the
terminology.
If
we
can
find
a
legitimate
way
to
do
it,
I
can.
B
B
B
This
is
the
it's
been
on
the
on
the
mailing
list,
but
Wow
meet
EKKO
is
like
a
week
behind.
Okay
right,
it's
been
on
the
mailing
list
only
for
a
week
or
something
it's
really
really
short.
We
would
like
to
hear
people
tell
us.
We've
got
something
wrong
or
we
don't
need
to
do
it
at
all.
Otherwise,
it
looks
like
it's
so
short
that
I
would
expect
I
guess
low
I
can't
do
it,
but
I
would
expect
the
chairs
just
like
adopt
and
move
through
using
their
process
and
and
get
it
done
quickly.
A
O
O
First
of
all,
let
me
explain
the
SRM
pstp
as
we
are,
as
we
all
know
that
in
SRM
pious
networks,
there
sr
pass
must
be
unidirectional,
but
in
some
transport
networks,
for
example,
and
pstp
networks,
the
SR
bi-directional
path
must
be
established
it
and
the
draft
ITF
spreem
M
has
passed.
Segmented
draft
has
defined
a
path
segments
to
support.
O
S
are
bi-directional
path,
correlation
so
based
on
that,
the
this
document
discussed
in
the
domain
scenarios
in
SRM
pstp
networks
there
in
this
document,
the
SR
bidirectional
end-to-end
Puffy's,
must
be
established
across
multiple
domains
and
the
path
segment
is
used
to
indicate
the
into
the
may
pass
or
end-to-end
paths.
So,
firstly,
there
are
two
models
over
there.
S
are
M
POS
into
the
main.
The
first
one
is
the
stitching
in
a
domain.
Domains
are
isolated
and
thus
teaching
at
the
end.
O
Others,
teaching,
node
or
link
will
push
their
list
of
the
s
ID
to
a
further
Lu
sub
Turner
and
in
nesting
into
domain
model
global
segments
listed
after
their
interests
law.
As
our
load
and
the
juror
is
an
end-to-end
process,
ID,
uniquely
across
all
domains,
and
then
in
this
document
we
also
discussed
there
as
RMP
STP
int
working
with
MPs
TP.
O
This
is
achieved.
Two
shifts
the
to
realize
the
end-to-end
we
pian
surface
by
interlocking
between
the
SR
and
MPs
TP.
Lat
works
with
their
path
segments,
so
first
net
net
look
at
their
SR
mph.
It
is
teaching
into
domain.
There
are
intimates,
as
a
figure
show
in
there
are
just
teaching
models.
Their
end
to
end
parts
can
be
divided
into
several
segments
and
we
defined
or
intimated
path
segments
or
into
may
pass
as
ID
to
indicate
each
segment.
O
O
It
is
a
2x
as
ID
list
exactly
and
there
I
pass
a
2x
and
and
the
payload
following
that,
then,
as
as
we
know
as
our
translate
load
Bo
Porter
top
of
the
labels
under
when
the
trend,
when
the
packets
being
forwarded
to
the
load
X
there
I
pass
a
to
X
will
be
the
first
with
the
Pedro
top
label
of
their
MPs
text.
Then
X
we
all
switched
their
I
pass
a
2x
to
the
Lexus
segment.
Information
like
this
I
I
pass.
O
A
2x
will
switch
to
the
X
to
Y
as
our
te
and
I
pars
X
to
Y
then,
and
in
at
the
reload,
why
it
is
the
same
with
X
and
ER.
When
the
pack
is
free
being
forwarded
to
the
load
of
Z,
it
will
be
owning,
existed
there,
I
pass
Y
to
Z
and
then
I
pass
Y
to
Z
will
correlate
their
bidirectional
tars.
For
example,
I
pass
Y
to
Z
enter
the
load.
O
O
We
called
h,
e+
h,
UT,
to
indicate
the
end-to-end
part,
and
this
is
the
forwarding
parts,
and
there
is
first
part,
is
the
3
2
I
pass,
and
this
is
and
after
the
ingress
load
a
we
were
pushed
a
to
X
srte
and
the
sub
pass
a
to
X
and
the
pounding
si
T
X
X
to
Y,
and
there
upon
e
si
tu
Y
to
Z
and
the
Valerii.
We
will
push
the
epass
a
2d.
O
O
Srte
enter
is
in
photo
may
pass
segment,
for
example,
at
Lord
X,
the
bounteous
ITX.
What
a
X
will
be
switched
to
the
x
y
srte
under
duress,
subparts
x
y
and
there
at
the
load
y.
It
is
the
same
stress
procedure
and
when
the
package
being
forwarded
to
other
low
Z
there
are
will
be
holding
their
epass
hoc
existed,
then
30
will
switch
their
a
2t
e
parts
to
that
z28.
The
reverse
parts
per
segment,
so
they're
in
the
to
end
EEP
ours
will
create
late,
correlate
their
to
the
unidirectional
parts.
O
So
melody
we
discussed
the
SRM
PST
P
and
NP
STP
interworking.
As
the
figures
show,
we
have
as
our
domain
and
mps
TP
domain
and
there
they
are.
They
are
the
SR
term
and
MPs
Turner.
They
are
all
their
bidirectional
trailer
and
we
will
we
will
eat
to
achieve
der
and
to
ender
VPN
service
crossed
SR
and
MP
STP
domains.
Then
we
will
use
the
path
segments
from
eight
one
load,
one
to
load
a
and
teaching
and
under
associated
the
MPS
label
at
the
load.
O
A-
and
this
is
the
forward
forwarding
paths
and
there
in
the
reverse,
pass
at
load
a
we
will
make
mapping
their
MPs
labeled
z28
you
the
path
segments
from
A
to
Z.
Then
we
are
achieved
arenda
to
enter
a
VPN
service.
So
we
in
this
document
we
discussed
the
several
years
into
the
main
use
cases.
So
last
step
we
were
searched
or
solutions
for
as
our
MPs
TP
into
mein
and
their
comments
and
discussions
are
very
welcome.
Thank
you.
A
I'm
looking
sorry
I'm
looking
at
this
spring
shorter
and
the
spring
shorter
says
that
SR
typically
works
within
one
single
trust
domain,
and
then
it
goes
on
and
make
some
statement
about.
Some
cases
might
be
different,
so
this
type
of
work
is
actually
across.
What
I
understand
are
it?
Trust
amounts
so
Debra,
okay-
and
we
have
it
comment
here,
but
Debra
do
you
have
an
idea
what
we
need
if
we
were
to
do
this
work?
A
J
I
I
would
say
without
looking
at
the
document,
of
course,
to
be
very
careful
on
the
the
word
there
domain
in.
If
how
about
what
that
implies
and
and
I
do
know,
that
segment
browning
has
been
very
careful
in
their
definition
of
domain,
because
we
went
back
and
forth
on
that.
Quite
a
bit
in
is
G,
so
I
would
say
you
know
just
work
with
the
chairs
and
of
segment
routing
and
and
and
just
save
it,
as
this
is
still
just
an
early.
J
Very
early
document,
I
think
just
cross
your
T's
and
make
sure
you've
defined
domain
as
they
do,
and
and
and
that
everybody
can
understand
what
it
is
that
you,
if
you
really
mean
intradomain
as
regards
psyche,
same
m
rounding
right,
so
just
work.
You
know
the
charity
lower.
If
you
know
you
filled
those
interests
here
to
do
this
work
with
the
other
chairs
to
see
if
there's
a
clash
or
something
what
whatever
and.
A
Yeah
we
will
discuss
it
I,
just
kind
of
wanted.
A
immediate
reaction,
I
point
out
that
there
is
something
that
needs
to
be
addressed
here.
I
think
that
this
document
would
probably
need
some
discussion
on
the
relationship
between
trust
domains
and
the
domains
that
you
have
in
this
document.
It's
not
it's
it's
for
me,
it's
kind
of
reasonable
to
say
that
those
the
SR
domain
and
the
SRM
fellows
TP
domain
or
within
the
same
trust
domain,
but
you
need
to
define
that
clearly
to
get
around
later
surprises.
Omega2,
for
example,
I
see
your
Directorate
reviews.
L
L
L
H
Joel
Halpern
again
from
Erickson,
let
me
suggest
looking
at
it
slightly
differently.
Getting
at
the
point
you
were
making
lower
the
point
as
I
understood
the
presentation
to
inter
work
between
SR
mpls
and
conventional
MPLS
when
you're
doing
past
sensitive
information.
Well,
that's
a
perfectly
valid
thing
to
run
into
in
a
perfectly
ordinary
deployment
scenario.
I'm
doing
that
over
here.
C
A
Thank
you
well,
I,
think
that
is
a
reasonable
look
at
it
I.
My
my
my
take
is
at
the
top
of
is
Trust
or
administrative
domain.
That's
the
the
big
thing,
and
that
could
be
slice
and
dice
in
different
ways
internally,
but
then
we
are
not
really
in
the
domain.
We
are
going
across
different
different
technologies.
D
Okay,
I
was
say
differently
again,
I
understand,
Deborah's
suggestion
is
to
clarify
how
the
main
is
used
in
the
context
of
this
proposal,
so
we
don't
have
to
boil
the
ocean
and
provide
explanation
of
all
possible
interpretations,
but
just
discuss
and
what
we
will
agree
on
how
we
interpret
the
main
in
the
context
of
this
work.
Yes,
only
this
work
yes
and
I
take
clue
that
been
suggested.
B
Perfect,
this
is
Adrian
yeah,
so
I
think
what
spring
did
was
they
used
a
term
called
SR
domain
to
mean
an
SR
to
him,
and
that
enabled
them
to
say
that
an
SR
domain
has
trust
requirements
and
also
that
SR
does
not
leave
that
domain.
That's
actually
the
definition
of
an
SR
domain
that,
when
an
SR
packet
gets
to
the
edge
of
the
domain,
it
ceases
to
be
SR
and
if
there's
another
SR
domain,
you
have
to
restart.
You
can't
get
this
hour
across.
B
So
looking
at
this
document
very
briefly,
just
over
a
hundred
uses
of
the
word
domain
and
the
document
it
starts
off
talking
about
SR
domain,
but
then
I
think
it
moves
on
to
talking
about
different
types
of
domain.
Where,
yes,
our
domain
is
big
and
within
it
there
are
different
domains.
So
that's
kind
of.
D
Yes,
you're
correct
there.
There
is
this
unintended
change
of
terminology
and
that
might
be
explained,
but
at
the
same
time,
okay,
so
just
to
say
well,
we
have
to
be
more
disciplined
with
the
terminology
and
that's
what
we're
gonna
do
we'll
review
the
document
who
appreciate
your
comments.
Please
do
it
on
the
mailing
list
and
help
us
to
make
it
better
document.
But,
yes,
we
will
need
to
clarify
their
mapping
between
I
start
the
main-
and
this
use
case.
A
A
D
A
A
We
found
that
there
are
things
to
discuss,
so
we
don't
normally
want
to
take
that
type
of
discussion,
especially
if
we
aim
to
do
an
update
during
your
working
group.
Adoption
Paul,
so
I
close,
the
working
group,
adoption,
Paul
I,
told
Greg
to
go
and
talk
to
the
people
making
comments
and
actually
try
to
post
a
new
version
or
the
document.
And
then
we
restart
the
working
group
adoption
Paul
later.
D
D
Of
discussion,
but
some
comments,
we
agreed
with
ASEAN
Rakesh
that
would
be
certainly
addressed
in
the
following
work,
and
that
can
happen
after
working
group
adoption
as
we
received
as
the
working
group
adoption
pool
was
officially
opened.
We
received
comments
from
Tom
and
it
was
absolutely
appropriate
comments
and
I
accept
all
of
them,
but
we
couldn't
communicate
fast
enough
to
separate
the
comments
that
he
believes.
That
needs
to
be
addressed
before
document
adopted
by
the
working
group
and
then
can
be
worked
later
on.
D
D
D
Waiting
for
Tom
to
respond
on
his
opinion,
what
he
thinks
as
a
mandatory
changes
before
another
working
group
adoption,
and
what,
with
our
agreement
and
commitment
to
follow
through
all
the
comments
that
we
received
so
far.
What
he
agrees
that
would
be
acceptable
to
address.
After
working
group
adoption
and.
D
So
that's
priorities
for
work
that
possibly
after
working
group
adoption,
but
it
might
change
based
on
Tom
input
and
the
last
one.
It
expands
go
beyond
eighty
twenty
nine.
It
was
pointed
as
optional
by
ACN
Rakesh,
because,
yes,
there
have
been
several
extensions
to
MPLS
LSP
to
other
cases
like
point-to-multipoint
LSP,.
D
D
A
D
Well
again,
if
again,
if
I
remember
correctly,
that's
the
terminology
used
in
43
79
and
then
consequently
in
80,
29
so
but
I'll
take
another
look
at
the
terminology
is
used
because
again
the
TLD
is
called
target
effectively.
So
it
might
be
that
it's
more
like
in
lingo
that
we're
talking
about
as
this
is
a
FAQ
whereas
yeah
it
might
be
not
really
a
fact,
but
it's
basically
we're
describing
for
which
LSP
this
label
was
allocated,
and
that's
we're
verifying
that
this
OSP
echo
request
arrived
on
the
LSP
that
this
label
been
allocated
for.
D
So
basically,
we
one
of
their
main
om
or
goals
that
our
speaking
achieves
is
to
verify
their
consistency
between
the
control
plane
and
the
data
plane
so
that
their
endpoint
receives
their
LSP
being
from
the
Elco
request
from
their
OSP
that
the
label
being
assigned
for
so
basically
there.
There
is
a
correlation
between
the
control
plane
when
the
label
was
assigned
to
the
data
plane
status.
Where
this
echo
request
arrived
from.
M
A
P
Q
D
P
G
D
A
G
Yeah
I
think
I
mean
again
it's
like
LBP
ipv4
and
RSVP
ipv4.
So
it's
basically
uses
the
protocols
that
we
used
to
I
know
instantiate
the
LSP
path,
so
I'm
not
sure
if
we
really
need
to
define
if
it
is
for
traffic
engineering,
because
I'm
not
aware
of
any
other
application
of
RSVP
to
instantiate
the
LS
piece.
G
G
This
was
identified
as
part
of
Interop
testing
with
the
NTC,
so
RFC
8287
basically
defines
the
you
know.
They
targeted
FEC
stacks
of
TLV
for
segment,
routing,
IGP
prefix
said,
and
not
just
since
he
said
so.
One
off
for
the
thing
that
was
most
in
RFC
8287
is
to
you
know,
clarify
if
or
how
the
length
of
the
sub
TLB
should
be.
You
know
computer,
so
this
is
an
example
of
one
off
for
the
targeted
FEC
stacks
up
tlv,
which
is
for
y
GP
prefix
segment
ID.
G
So,
as
you
could
see,
one
of
the
field
is
a
result
field.
So
there
are,
you
know,
implementation
where
some
implementation
considers
the
result
field
as
part
of
the
length
while
defining
the
TLV,
but
another
implementation
does
not
include
this,
which
basically
created
some
Interop
issue.
So
this
draft
basically
clarifies
that
to
address
the
center
of
issues,
so
we
basically
defined
that
the
resort
field
must
be
considered.
G
While
you
know
computing
the
length
field,
so
it
explains,
are
the
different,
sub
tlvs
and
identifies
you
know
what
would
be
the
length
along
with
a
couple
of
you
know
examples.
So
it's
a
pretty
straightforward
done,
a
simple
draft
that,
basically,
we
believe,
is
a
logical
fix
for
one
of
the
intra
problem.
Like
Kalos
mentioning
I
know.
This
is
recently
adopted
as
a
workgroup
document
and
this
being
a
straightforward,
Dom
or
fix.
We
believe
this
is
ready
for
workgroup
loss,
but
any
comments
or
feedback
so
welcome.
B
R
This
is
question
from
Howie
and
this
is
a
new
trust
named
RSP,
it's
time
for
cyclic,
queuing
and
forwarding,
or
pundit
or
agency.
So,
let's
begin
with
the
background
of
these
large-scale
deterministic
Network,
which
we
are
doing
in
the
dead
tonight.
Working
group
and
LDN
aims
to
achieve
fondant,
latency,
forwarding
on
layers
three
networks
and
the
contain
long
distance
links
and
large
numbers
of
nodes
and
flows
and
out
in
record
sites.
R
Therefore,
in
Audion
its
request
at
dates,
prime
animals
is
used
to
indicate
which
optional
cycle
packets
belongs
to,
so
that
a
downstream
node
could
use
this
indication
to
put
the
packet
into
the
right
receiving
hue.
So
if
we
consider
a
specific
package
once
the
cycle
for
the
sending
packets
in
the
ingress,
node
is
determined.
The
one
in
the
us
node
is
determined
to
such
achieving
into
in
only
the
latency.
R
So
this
is
the
airspeed
is
crying
for
a
secure
life.
The
idea
is
very
single:
the
downstream
node
allocates
three
different
empires
labels
for
each
ESP
and
advertises
information
to
the
upstream
node
by
using
signaling
protocols
such
as
r2
Apte.
So
each
of
these
levels
is
associated
with
a
specific
queue
in
the
abdominals
as
throwing
the
right
figure.
R
We
assume
that
if
the
package
is
sent
from
the
upstream
node
q1,
it
should
be
labeled
waste
1003,
and
if
the
packet
is
informed
q2,
it
should
be
leveled
waste,
one
solid
and
one
so
about
the
FIB
installation.
We
assume
that
a
package
sent
from
the
upstream
node,
q1,
q2
and
q3
should
be
put
into
the
downstream
nodes,
q3
T,
1
and
Q
2
respectively.
R
Note
that
how
to
is
establish
such
making
relationship
will
be
described
in
the
future
version
of
this
document
and
therefore
the
package
since
form
downstream
North
Q
1
will
be
put
into
the
time
frame,
node,
q3,
so
on
and
so
forth.
So
in
this
way
that
into
in
the
pounding,
the
latest
thing
could
be
achieved.
As
for
the
LD
interrupt,
so
next
steps
about
the
straps,
we
will
describe
how
to
establish
the
mapping
relationship.
We
have
a
self-learning,
McNamee
indicates,
prime,
and
we
need
more
suggestions
and
comments
on
this
new
work.
Thanks.
I
R
Our
solution,
it's
like
it
doesn't
matter
which
lie
both
extremely
I,
would
associate
it
to
a
specific
you,
because
today
we
have
a
self-learning
magnin's
of
mechanism
in
the
downstream
node.
To
learn
this
information,
for
example,
if
I
receive
a
package,
Labrador
is
one
solid
and
one.
In
the
downstream
note,
it
can
automatically
send
this
packet
in
a
specific
queue
in
a
local
computation,
I.
I
Can
animate
the
loss
if
the
node
one
is
sending
the
packet
with
one
label
1001
to
node
2
node
2
locally
is
somehow
changing
1001
into
1,002
1,003
1,004,
depending
on
the
queues
internally
yeah.
If
I
follow
that
right,
then,
what's
then
it's
local
to
the
node,
then
why
does
it
need
to
be,
or
does
it
not
mean
to
be
conveyed
to
the
external
set
of
neighbors
like
node
1,
and
if,
if
there's
no
need,
then
we.
R
I
A
Ready
we
can
I
try
to
well
clarify,
at
least
for
me,
your
question
so
sending
a
packet
to
a
node.
Does
that
has
a
specific
label
depending
on
which
Q
should
go
into
that's
what
you
mean
with
external,
or
is
it
just
inside
the
node?
And
when
it
comes
out
again
on
the
outgoing
interface,
it
has
a
just
one
single
label
for
him.
I
R
I
can
clarify
this
if
the
downstream
node
received
a
package.
This
is
the
first
packet
from
this
afternoon.
Node
it's
received
and
this
level
is
1001
and
it
will
if
it
receives
this
packet
in
psycho
X.
So
it
will
map
this
rabble
to
psycho,
X
plus
2,
so
it
will
establish
the
meta-narrative
relationship
size
each
which
level
should
be
mapped
to
which
kill
and
use
this
mapping
relationship
to
instruct
the
subsequent
packet.
I
If
it's
a
local
computation
and
behavior
change
mean
most
implementations
today,
do
something
similar
from
the
queuing
point
of
view
by
looking
at
an
MPLS
TC
bits
as
an
example
of
previously
known
as
exp,
and
then
what
are
the
value
might
be
there
map
it
to
one
of
those,
many
cues
that
the
implementation
on
the
platform
may
have.
That's
all
the
local
behavior
right
I
mean
we
don't
really
specify
how
that
gets
done.
Internal
inside
the
implementation
say
if
the
the
intent
of
the
graph
is
to
specify.
D
Actually
a
Greg
nurse
kzt,
especially
for
the
dead
net,
if
even
if
it's
a
local
behavior
I,
think
that
might
be
helpful
and
useful
to
do
the
document.
That
explains
it,
because
there
are
a
lot
of
needs
to
be
required.
Coordination
and
everything
has
to
act
in
accord
in
order
to
achieve
their
objectives
of
that
net.
So,
yes,
it
might
look
like
it's
low,
chimera
and
implementation,
but
it
might
be
important-
and
let's
look
at
this
more
because
it
should
it
might
be-
just
need
to
be
advertised
as
capability.
N
N
G
R
N
Think
that's
a
different
approach,
so
it's
misleading
to
the
oscqr
okay,
and
for
that
I
would
like
to
mention
just
a
previous
session
in
this
room
in
that
not
that
there
was
a
lengthy
discussion
on
the
basic
proposal
you
refer
to,
and
it's
not
clear.
What's
the
next
step
forward
and
what
what
the
next
step
we
are
taking,
what
is
needed?
It
has
been
shown
that
the
mathematical
proof
is
incorrect.
So
it's
up
it's
open,
what's
happening
in
that
net
and
what
would
be
the
next
step?
Thank
you.
Q
A
H
K
Jenny
Kumar
Broadcom
before
I
go
in
this
drafts.
Discussion.
I'll
talk
about
the
background
of
the
draft
that
how
it
came
into
the
existence.
So
there
was
a
work
which
was
done
for
customers
in
the
IP
network,
where
they
wanted
to
collect
the
data
on
a
live
packet
which,
as
we
call
as
in
band
Network
telemetry,
that
work
is
being
deployed
into
large.
Msd
sees
quite
a
few
deployments.
K
Are
there
and
that's
an
v4
TCP
UDP
GRE
IP
and
IP
h,
ESP,
it's
a
pretty
large
and
cap
set,
which
is
being
deployed
so
once
that
deployment
was
done,
the
requests
were
coming.
That
can
we
do
the
same
thing
into
MPLS
and
then
based
on
that
the
evolution
of
this
raft
happened,
which
is
in
band
Network,
telemetry
and
MPLS
domain,
and
hence
that
has
the
acronym
mint
great
you
are
there
before
I
go.
You
have
a
question:
okay,.
D
I
can
wait
and
just
want
to
point
that
we
had
the
discussion
when
original
proposal
for
IOM,
as
in
band
om,
came
about
around
and
I.
Think
it's
not
accurate
and
it's
misleading,
because
in
band
means
that
om
packet
follows
the
data
packet
and
active
om
can
be
in
band
and
actually,
if
it's
not
in
band,
then
the
value
of
this
active
om
is
very
limited,
so
in
particular
in
MPLS
and
ecmp
environment.
D
D
You
know
if
in
MPOs,
if
we're
talking
about
active
either
it's
BFD
or
info
sls
beeping.
So
if
we
be
disciplined
and
use
entropy
label
on
data
flow
and
or
active
or
a.m.
that's
supposed
to
monitor
at
the
data
flow,
then
they
are
in
then
because
in
band
again
my
interpretation
is
that
you
follow
exactly
the
same
path,
same
notes,
interfaces
and
links
so
use
of
entropy
label
and
the
same
income
label
encapsulation
and
it's
applicable
both
to
MPLS
and
SRM
alloys.
I,
see.
K
A
point
I
get
yeah
I
agree
with
that.
So
if
you
use
the
entropy
level
in
the
NS
beeping
now
it
can
also
follow
the
same
path.
So
if
there
is
a
suggestion
to
change
the
name,
we
are
open
to
that.
It's
a
first
draft
so
coming
back
to
the
draft,
so
as
I
was
giving
the
background,
this
has
been
deployed
in
IP
networks.
There
are
conversations
for
deploying
in
in
the
MPLS
Network
I,
don't
know
if
I'll
be
able
to
go
in
all
the
details.
K
But
the
key
point
is
the
usage
of
the
special
label
composite
special
label.
As
was
talked
earlier,
this
raft
makes
use
of
that
I.
Don't
know
it's
the
first
stuff
to
make
use
of
it,
but
I
realized.
There
are
authors
on
this
raft
and
the
other
SPL
draft
which
are
common.
So
that's.
Why?
Probably
they
lean
to
use
this
all
the
suggestion
to
use
the
CSB.
K
So
motivation
is
again
that
the
protocol
itself
should
be
very
simple:
it
should
be
in
band
and
we
should
be
able
to
trace
the
flow
of
the
packet,
which
is
exactly
same
as
the
original
live
packet
flow.
There
are
a
few
enhancements
which
we
can
do
in
this,
which
is
captured
in
the
details,
which
is
a
clone
mode
or
in
band
mode
in
man
mode,
meaning
that
all
the
data
which
is
being
collected
by
inserting
the
metadata
in
the
packet
itself
is
done
for
the
live
packets
and
there
is
a
clone
mode.
K
We
make
a
copy
of
the
packet,
and
that
can
be
done
for
various
reasons,
because
the
NTU,
the
PMT
you
is,
is
less.
The
part
of
the
packet
may
exceed
the
peon
to
you,
or
there
are
other
reasons
to
do
that.
In
that
case,
there
is
ability
to
create
a
cloning
of
the
cup
of
the
packet,
and
then
the
in
bank
collection
of
the
data
is
done
for
the
clone
mode.
Great.
D
D
D
Hybrid
two
step
proposes
to
collection
of
data
in
the
follow-up
packet
that
has
the
same
transport
layer
encapsulation
as
a
data
packet.
So
the
data
packet
effectively
is
only
a
trigger
all
telemeter
information
is
in
the
follow-up
vagin,
and
actually
the
HCC
defines
that
if
this
packet
is
full,
then
another
packet
can
be
generated
so
basically
will
arrive
with
the
train
of
telemetry
information
filled.
Packets
and
postcard
is
that
when
telemetry
is
exported
of
each
note,
that
traversed
by
the
trigger
packet
right.
K
So
thank
you
for
that.
So
I
can
give
a
brief
background
on
that
conversation
which
Greg
just
talked
about.
So
if
you
look
carefully,
there
are
three
parts
to
the
proposal
and
these
proposal
are
also
moving
in
the
IP
PM
group.
There
is
a
protocol
and
cap
part,
especially
in
the
IP.
Then
there
is
a
data
header
part
that
how
do
you
represent
the
data
in
a
in
a
way
where
you
can
figure
out
either
to
check
some
of
the
data,
the
length
of
the
data
or
put
some
control
information
and
then
the
data
itself?
K
So
we
have
a
proof
point
that
it
works
right,
I'm
talking
to
various
folks
in
area
where
we
can
have
some
agreement
on
the
data
header,
as
well
as
the
data
types,
so
data
types
become
a
registry
and
then,
based
on
the
registry,
we
can
define
namespaces,
which
is
essentially
the
set
of
data
types
which
uses
the
common
data
header,
and
then
they
can
be
carried
over
in
different
protocols.
Protocols
could
be
a
postcard,
could
be,
IOM
could
be
IFA
so
and
so
forth,
and
the
same
proposal
follows
here
in
this
case.
K
K
This
is
the
terminology
the
way
we
have
defined
and
I
believe
based
on
the
input.
If
mint
or
in-band
network
telemetry
is
not
the
right
acronym,
then
then
we
have
to
change
the
labels.
So
I'm.
Okay
with
that,
but
just
to
give
you
an
idea
what
essentially
it
says
that
mint
is
a
domain
where
all
the
nodes
are
participating
in
mint,
essentially
collecting
the
data
on
a
per
packet
basis.
K
The
node
in
a
domain
is
called
as
a
mint
node
and
the
mint
label,
which
is
which
is
the
CSP,
which
is
the
extended
label
using
the
label.
Extension
15
is
an
are
located
in
the
draft.
We
have
taken
a
Liberty
to
use
the
first
label,
which
is
16
since
nothing
was
used
so
far,
but
that's
just
an
attempt.
If
that's
not
the
right
thing,
we
can
always
change
that
mint
metadata
is
the
information
or
is
the
data
set
which
is
collected
on
a
per
node
per
packet
basis?
K
There
are
functions
on
the
node,
and
these
are
the
these
are
the
functions
which
either
sit
on
the
edge
of
the
domain
or
sit
within
the
middle
of
the
domain.
The
edge
of
the
domain
is
defined
by
the
node,
which
performs
the
initiation
function
or
the
termination
function.
Initiation
means
that
when
the
packet
comes
in,
there
is
an
insertion
of
mint
header
and
also
the
mint
metadata
by
that
node
and
then
the
transit
nodes.
All
they
do.
They
do
the
detection
part
that,
yes,
this
is
a
mint
packet.
K
It
has
a
presence
of
the
CSP
label
and
then
based
on
that
it
inserts
the
method.
It
adds
some
predefined
location
and
then
the
last
node
at
the
edge,
which
is
the
termination
node,
does
the
same
thing
that
it
detects
that
it's
a
mint
packet
and
then
based
on
it's
a
copy
of
the
packet.
Then
it
drops
it.
If
it's
a
live
packet
or
in-band
packet,
then
it
strips
the
foreign
data
creates
a
report
and
report
could
be
anything.
It
could
be
postcard
for
exact
example
and
then
sends
it
out
to
the
collector.
K
D
K
D
D
1588
propagation,
over
MPLS,
so
in
this
RFC
residence
time
measurement,
we
proposed
to
transport
1588
through
the
domain,
is
associated
channel
a
new
type
and
then
collect
on
each
node
residence
time
in
a
scratch
field,
and
that
can
be
used
not
only
it
can
be
used
either
to
measure
residence
time
across
the
path
or
it
can
be
used
as
well.
For
correction
of
1588
clock.
D
K
D
D
K
B
Adrian
can
I
make
a
plea
to
you
to
take
the
explicit
ion
a
value
out
of
your
draft
and
replace
it
with
the
letters
TBD
I
will
do
that
to
get
Marty
vio.
You
have
to
be
anything
you
like,
because
it's
really
dangerous
putting
values
in
drafts.
You've
picked
one
that
has
been
allocated
already
by
Ayane
for
something
else
realize
that
yeah.
K
So,
example,
of
mint
deployment,
where
an
MPLS
domain
is
actively
be
actively
participating
in
maintainer
collection,
as
you
can
see
in
ER
has
been
designated
with
a
mint
initiation
and
the
termination
function
and
by
the
way
there
are
logical
functions,
same
physical
node
may
be
performing
both
the
functions.
That
confusion
happens
whenever
we
talk
about
it,
but
there
is
just
a
logical
representation
of
the
topology
and
transit
routers.
The
LS
R's
are
the
one
which
are
acting
as
the
mint
transit
nodes
or
mint
nodes,
inserting
the
metadata
and
on
the
penultimate
hop.
K
Both
the
behaviors
are
supported
where
we
can
do
the
PHP,
and
in
that
case
the
last
hops
metadata
actually
is
not
available,
some
deployments,
it
might
be.
Okay,
some
deployments.
It's
not.
Okay
and
the
reason
this
is
being
considered
is
ability
to
put
less
stress
stress
on
the
hardware.
If
hardware
has
to
do
the
label
pops
and
then
the
second
set
of
lookups,
it
becomes
very
expensive
to
do
the
hardware
implementation.
K
K
K
The
tunnel
mood
is
supported
again.
This
draft
does
not
go
in
the
details
of
the
tunnel
mode,
but
the
basic
concept
is
propagation.
Those
edits
will
come
in
the
next
set
of
draft,
which
is
the
propagation
that,
whenever
mint
packet
enters
into
an
another
tunnel,
just
like
TTL
propagation,
we
can
have
a
configuration
which
says
that
propagate
the
mint
information
when
it
enters
into
the
tunnel.
What?
K
Essentially
it
means
that
we
can
unable
still
a
poor
packet
flow
and
we
will
collect
the
data
before
it
enters
into
the
tunnel
mode
in
the
tunnel
itself
and
when
it
transits
the
tunnel,
then
also
the
data
is
collected.
So
that's
the
tunnel
mode
in
the
transport
mode.
There
is
only
a
single
stack
which
is
present
and
again.
This
is
to
ease
the
implementation
that
if
there
are
some
hardware
limitations,
where
it's
not
it's
very
expensive,
to
support
two
modes,
then
the
tunnel
mode,
then
transport
mode
may
be
a
good
idea.
K
I'll
just
quickly
go
over
the
header
before
at
this,
so
that
my
time
is
close.
This
is
the
header
and
in
this,
if
you
see
by
the
way
here,
I
use
TBA,
but
in
the
draft
we
ended
up
using
actually
a
label.
So
there
are
two
headers
one
is
a
mint
header
which
has
the
control
information,
and
this
header
influences
the
behavior
of
the
packet
in
the
data
path.
It
tells
what
to
do
what
not
to
do.
Then
there
is
a
metadata
header.
K
It
has
no
control
information,
it
does
not
control
the
behavior
of
the
packet.
All
it
has.
That
was
the
length
of
the
metadata.
What
kind
of
metadata
to
collect?
Is
there
a
need
to
take
an
action
on
the
metadata
itself?
Let's
say
we
do
some
kind
of
thresholding
and
what
is
the
current
length?
There
are
some
implicit
checks
which
can
be
done,
use
this
information
with
sort
of
functions
like
a
control
function,
but
they're,
really
not
a
control
function
and
the
reason
I
am
bringing
it
in
the
IPP
and
working
group.
C
A
K
So
next
protocol
feel
is
the
one
which
can
be
used
to
create
a
chain,
and
this
is
something
very,
very
interesting,
because
when
you
put
the
mint
header,
then
you
can
actually
indicate
what
kind
of
payload
is
carrying
I'm
going
to
stop
on
stop
here.
The
next
steps
is
that
we
are
looking
for
review
and
comments
on
the
graph.
G
So
a
little
couple
of
slides
back
to
the
tunnel
mode
yeah,
so
I
understand
the
ingress.
You
add
or
copy
the
the
mint
label,
the
overlay.
So
what
happens
in
the
egress
and
there's
a
transit
tunnel
and
if
you
want
to
recap
slate
and
send
it
out
kind,
if
you
have
data
in
the
mint
at
the
outer
layer,
what
do
you
intend
to
do
so.
K
In
this,
so
that's
a
very
good
question
yeah.
So
there
are
two
ways
we
can
handle
it.
I
can
with.
We
can
reflect
it
back
into
the
inner
mint
stack
and
that's
how
that
mint
data
can
be
carried
and
it
may
may
be
OK
for
the
flexible,
NP
use,
which
can
do
much
more
flexible
processing
of
the
data,
but
fixed
pipelines.
It's
not
ok,
there
is
a
mode
which
is
supported,
which
is
called
as
a
fragment.
D
D
A
S
S
Ok,
an
intention
of
this
job
is
Jeff,
the
defiance
to
the
encapsulation
for
amperes
in
band
performance
measurement.
This
method
is
based
on
FC,
8321
Audrey,
the
marking
method.
This
method
requires
amperes
flow
identification,
which
is
a
disgusting
RFC,
83
72
and
the
Salone
synonymous
flow
label
is
a
way
to
do
this
flow
identification.
But
it's
a
bit
complex
to
manage
and
it
seems
not
applicable
to
nested
flow
ID.
So
this
chapter
PI's
a
special-purpose
label.
S
S
S
A
A
So
we
concluded
the
meeting
next
time
is
where
monster
all
going
back
to.
We
start
this
kind
of
thing
where
we
give
him
back
to
the
same
set
before
a
second
or
third
or
the
fourth
time
so
Montreal
is
the
place
been
before
I.
Don't
know
if
it's
the
same
motel
but
I.
Remember
some
good
place
so
see
you
there
safe
trips.