►
From YouTube: IETF109-IPPM-20201116-0500
Description
IPPM meeting session at IETF109
2020/11/16 0500
https://datatracker.ietf.org/meeting/109/proceedings/
A
A
A
C
A
Oh
andrew
mcgregor
has
offered
to
take
notes.
That
would
be
fantastic,
and
if
anyone
wants
to
help,
please
go
to
cody
md
as
well.
A
All
right,
let's
get
started
ian,
do
you
want
to
try
to
talk
again
just
make
sure
we
can
hear
you.
C
A
Cool
great,
so
we're
gonna
start
now
I'll
be
driving
the
slides
ian.
If
you
could
keep
an
eye
on
the
chat
and
then
also
keep
an
eye
on
the
time,
that'd
be
great,
sounds
perfect.
I
have
it.
A
A
A
This
defines
the
contribution
policies
to
the
ietf
and
you
can
find
it
if
you
just
do
a
google
search
for
iotf,
don't
well
here's
some
of
the
links
that
are
relevant
if
you're
here
you're
on
meat
echo,
the
jabber
room
is
very
conveniently
accessible
in
the
chat
panel
and
our
ether
pad
for
notes
using
cody
md,
and
we
do
have
a
note,
taker
and
just
watch
chat
and
be
our
scribe.
A
All
right,
so
we
have
a
pretty
full
agenda
like
normal
we're
going
to
cover
first,
some
of
our
existing
working
group
items.
There
aren't
too
many
of
these.
A
lot
of
them
are
pretty
mature.
Mainly
we
have
a
bunch
of
iom
documents
that
are
progressing.
A
After
that,
we
have
several
other
documents
that
have
received
a
lot
of
discussion
and
would
be
candidates
for
calls
for
adoption
soon
and
then
some
lightning
talks
to
round
us
out.
A
We've
got
an
update
on
the
status
of
documents.
We've
been
a
busy
working
group.
We
have
a
lot
of
products
that
we've
come
out
with,
so
we
have
four
documents
in
the
rfc
editor
queue
currently,
and
we
have
three
others
that
are
going
through
iesg
and
ad
evaluation.
A
A
All
right
and
then
briefly,
we
did
want
to
have
a
the
one
document
that
we're
not
going
to
have
a
formal
presentation
on
is
the
ippm
stamp
yang
document,
which,
if
greg,
is
here?
If
you
just
briefly
comment
on
kind
of
the
status
and
where
we
see
that
going
in
the
future
months.
E
So
just
the
last
minute
update.
E
Stamp
option
till
we
it's
clearer
to
us
discuss,
and
so
we
have
enough
support
from
isg
and
it's
moving
further
and
the
plan
is
that
now
start
working
actively
on
stamp
yang
and
bring
it
updated
version
to
the
next
meeting.
A
F
See
if
this
works
great
hi
yeah,
I
just
wanted
to
to
kind
of
one
comment
about
the
stuff
in
iesg
evaluation.
So,
as
you
know,
tommy
I
just
pushed
the
button
a
stamp
option
to
uob.
So
that's
off
the
rc
editor.
So
thank
you.
Greg
for
staying
on
that
and
iom
data
has
been.
I've
been
waiting
on
a
revision
on
that
for
about
three
months
now.
F
A
All
right
frank,
I
can't
hear
you
quite
yet
if
you're
speaking,
but
I
think
you're
also
up
for
the
documents
cannot
hear
you
right
now.
I
cannot
hear
you
I'm
not
sure.
If
I'm
the
only.
A
B
Okay,
excellent,
sorry
for
that
yeah
and
then
martin
yeah,
thanks
for
the
hats
up,
we've
been,
I
think,
ping-ponging
the
ball
between
us
and
but
I
think
we
we
have
a
plan
forward
and
tal
can
speak
to
that
in
a
minute,
because
I
talked
to
tell
a
couple
of
days
ago
and
you'll
push
out
a
revision,
including
the
security
remarks
that
you
have
so
that
we
can
go
and
hopefully
also
progress.
B
The
data
draft.
This
one
is
around
the
three
active
working
groups
are
so
we
have
and
we
have
split
the
presentation
into
two.
I'm
just
gonna
go
briefly
talk
about
the
v6
options,
which.
C
B
Combination
of
the
v6
deployment
and
the
euler
v6
options
and
then
telus
going
to
go
and
talk
about
the
other
two.
So
on
the
v6
options
draft,
if
you
flip
to
the
next
slide,
there
is
two
updates
that
we've
done
so
you've
seen
two
new
revisions:
the
o3
and
the
o4
version.
The
o3
version
was
a
reflection
of
the
last
working
group
meeting
where
we
at
length
debated
well
how
we
should
go
and
address
the
fact
that
well,
there
was
a
a
section
on
deployment
considerations
which
was
outdated.
B
Well,
that
was
a
stale
paragraph,
because
we
had
all
the
deployment
considerations
moved
to
the
quote-unquote
deployment
draft
and
so
based
on
the
chair's
recommendations.
We
decided
to
go
and
merge
the
two
documents,
as
opposed
to
progress
them
independently,
and
so
that
happened
and
the
o3
version
was
just
an
editorial
merge
with
kind
of
nuances
to
go
and
resolve
circular
references,
otherwise
right
because
the
the
documents
were
referring
to
each
other.
B
So
that's
the
the
all
three
thing
that
that
happened
along
with
obviously
removing
the
the
paragraph
in
question
and
if
you
flip
forward
one
additional
slide,
then
you'll
see
that
after
the
merger
tell
me
if
you
can
go
and
go
to
the
next
slide.
Okay,
thank
you.
Then
tommy
spent
a
little
bit
of
time.
That
was
one
too
fair.
B
Far
then,
thanks
to
tommy,
he
spent
a
fair
amount
of
time
on
cleaning
up
the
language,
especially
in
the
the
sections
and
came
in
from
the
deployment
draft,
so
that
we
have
a
fairly
solid
document
by
now.
B
That
includes
extensive
coverage
of
deployment
considerations,
along
with
the
ask
for
code
points
for
v6
hot,
by
hop
option
types
and
the
destination
option
type,
and
the
only
thing
that
we
haven't
really
resolved
was
because
well,
I
think
we
hadn't
really
discussed
that
at
length
and
tommy
just
came
up
with
a
suggestion.
B
Was
this,
like
the
author
list?
Is
too
long,
so
the
typical
thing
that
we
have
and
so
suggestion
is
to
again
move
to
an
editor
model
like
we
had
it
with
the
the
data
draft,
which
would
mean
like,
given
that
sweater
authored,
pretty
much
the
v6
option
document
and
wrote
it
and
I
wrote
the
deployment
draft
to
a
large
extent.
B
The
suggestion
is
just
to
go
and
have
these
two
pieces
with
people
as
editors,
and
if
there
is
nothing
other
major
coming
up,
I
think
we
can
go
and
aim
for
maybe
another
revision,
maybe
none
and
well.
We
need
one
revision
to
go
and
clean
up
the
author
list,
but
maybe
we
can
also
go
and
do
additional
nuances
if
somebody
has
ones,
but
I
believe
what
we're
very
close
to
being
able
also
to
go
and
go
to
work
new
gospel,
that's
pretty
much.
It.
A
Yeah,
I
don't
see
anyone
in
the
queue
yeah.
Thank
you
for
that
update
and
just
a
comment
in
general
on
our
author
lists.
You
know
this
is
kind
of
the
same
comment
we're
going
to
have
on
any
document
that
has
the
warning
within
the
rf
within
data
tracker.
So
please,
if
you
have
a
lot
of
authors,
let's
do
an
editor
model,
there's
nothing
wrong
with
doing
that.
So
thank
you
for
being
willing
to
move
to
that
authors.
A
I
H
H
H
So,
basically,
we've
updated
the
draft
based
on
our
thoughts
from
the
design
team
meetings
and
just
a
few
words
about
the
solution.
The
current
proposal
is
basically
on
the
forward
path.
The
iom
packet
has
its
loopback
flag
set
and
then
once
an
ion,
node
decides
to
loop
back
that
packet.
H
It
clears
that
low
back
flag.
So
on
the
reverse
path,
the
luba
flag
is
not
set,
and
now
iom
transit
nodes
along
the
path
just
push
iom
data
and
once
that
packet
reaches
the
iom
edge
at
the
edge
of
the
domain.
Basically,
the
encapsulating
node,
which
originated
the
lubeck
packet,
receives
that
looped
back
packet
and
detects
the
fact
that
this
was
a
loopback
packet
based
on
comparing
its
node
id
to
the
id
of
the
current
iom
node.
H
H
So
since
we
haven't
had
any
comments
about
the
solution,
we're
assuming
that
it's
a
reasonable
solution,
of
course,
if
there
are
any
other
comments,
we'd
be
happy
to
hear
them,
but
this
was
actually
the
last
open
issue
we've
had
about
this
draft
and
assuming
we've
resolved
this,
we
would
like
to
recommend
to
proceed
to
working
group
last
call
comments.
H
D
We
have
two
people
in
the
queue
greg.
Do
you
want
to
go?
First,
let's
see
if
I
can
mention
okay,
well,
I'd
like
to
meet
you,
so,
hopefully
tommy
can
figure.
Okay!
Thank
you.
E
E
H
Yeah
we've
had
this
question
in
the
past
and
again
it's
it's
somewhat
of
a
semantical
question,
but
direct
exporting
is,
in
my
view,
management
plan
entity.
Basically,
the
result
is
reporting
to
report
it
to
a
collector,
whereas
blue
back
is
used
by
the
data
plane.
Okay
or
it's
a
control
plane
entity,
but
it's
used
along
the
path
of
the
data
plane.
H
E
But
in
direct
export,
as
I
understand
their
destination
of
this
exported
packet
could
be
their
edge
of
iom
trace
so
right
good
effectively,
it
can
be
anything
including
the
node
which
is
receiving
in
the
loopback.
So
from
that
point
of
view
it
does
look
like
the
loopback
is
a
special
case
of
the
direct
export.
H
Right,
yes
again,
I
think
we've
discussed
this
in
the
past,
so
you
could
implement
loopback
by
direct
export,
but
you
would
get
a
slightly
different
result
because
again
in
loopback,
you
get
the
ion
data
from
each
of
the
iom
nodes
along
the
path
and
in
direct
exporting.
You
only
get
the
information
from
a
single
iom,
node
at
the
time
so
buy
routine.
E
E
So
to
me,
the
direct
export
does
look
like
a
more
general
and
more
flexible
method
of
tracing
and
getting
collecting
their
information
from
the
trace
of
the
package.
So
I
am
information
from
the
pack.
E
I
I
think
that
we
can
take
it
to
the
list,
but
I
still
have
this
question
and
I
would
like
to
clarify
it
before
we
move
forward
with
the
next
step
progressing
this
document.
Thank
you.
I
F
On
the
list,
sorry
to
not
confirm
so
I
I
did
take
a
look
at
these
drafts
a
couple
days
ago
and
thank
you
for
addressing
the
the
like
the
basic
amplification
issue,
which
I
think
we've
discussed
ad
nauseam.
I
I
was
attempting
to
think
this
through
and
I'm
not
that
for
both
this
and
direct
export.
F
I'm
somewhat
concerned
about
corner
cases
where
multiple
namespaces
are
interacting
in
ways
that
cause
geometric
or
even
infinite
loops
of
of
of
of
iom
packets
going
back
and
forth.
I
will
take
that
to
the
list,
because
it
would
be
impossible
for
me
to
explain
verbally
right
now,
but
just
please
look
out
for
that
and
I'd
be
interested
in
your
thoughts
on
this.
If
you
would
like
me
to
talk
through,
I
can,
but
I
have
a
feeling
it
just
gets
too.
J
Yeah,
I'm
actually
still
worried
about
the
amplification
attack
because
it's
different
than
with
trace
routers
tracer.
You
have
to
send
one
packet
for
every
packet
you
receive
back
and
that's
not
the
case
here.
You
have
to
send
one
pack
and
get
like
bunch
packets
back
depending
on
how
long
it
passes.
J
J
So
I
mean
like
the
minimum
would
be
to
talk
about
rate
limiting
these
kind
of
things
you
can
reflect,
but
I
think
if
you,
if
you,
if
you
consider
a
different
design
for
getting
the
same
information,
that
would
be
worth
it
like,
instead
of
trying
to
reflect
from
every
router
one
single
packet,
you
could
just
send
one
packet
back
on
the
on
the
reverse
part
and
accumulate
information
there,
or
something
like
that.
Because
that's
a
serious
issue
and
I
think
it
needs
to
be
further
addressed.
H
So,
regarding
amplification
attacks,
again,
of
course,
maria
we've-
we've
received
this
feedback
in
the
past
and
what
we've
done
in
the
draft
is
a
few
things.
First
of
all,
we
point
out
the
problem.
We
say
we
have
a
problem.
We
know
we
have
a
problem,
so
it's
it's
out
there
in
the
draft.
H
Okay,
so
the
magnitude
of
the
amplification
is
limited
and
we
also
discussed
rate
limiting.
Like
you
mentioned,
that's
that's
also
one
of
the
things
that
is
described
in
the
draft,
so
I
think
we
at
least
tried
to
point
out
the
problem
and
to
suggest
solutions,
and
I
think
the
your
your
last
suggestion
to
basically
limit
the
number
of
responses
you
get
from
each
message.
H
That
would
just
reduce
the
performance
of
the
mechanism,
so
there
was
there's
always
a
trade-off
here
between
how
bad
the
security
attack
is
and
what
you
can
gain
from
the
mechanism.
J
Yeah,
so
the
problem
is,
there
is
a
tradeoff
right
and
this
is
a
serious
attack
and
the
problem
is.
This
attack
is
not
really
limited
in
a
protocol
way
that
you
can
say
you
know
it's
never
more
than
this
much
packet
right.
It
can
be
like
a
huge
number
of
packets
and-
and
I
think,
that's
still
not
really
satisfying-
I
did
I
did
somehow
over
at
the
rate
limiting,
because
I
was
looking
at
the
security
section
and
it
doesn't
say
anything
there.
J
So
maybe
you
have
to
put
some
more
text,
but
I
think
more
consideration
would
still
be
needed.
D
Okay,
thank
you.
I
appreciate,
what's
maybe
follow
up
on
the
list
and
see
if
we
can
come
to
a
good
conclusion
on
these
issues
so
sure
back
to
you,
tommy
thanks
help.
H
So
we
had
a
long
discussion
about
this,
both
in
design
team
meetings
and
on
the
mailing
list,
and
we've
recently
submitted
an
updated
draft.
We
have
two
pending
open
issues
which
we're
going
to
discuss
in
the
next
few
slides,
so
bear
with
me.
The
main
change
in
the
current
version
of
the
draft,
which
is
version
zero
two
compared
to
the
previous
version,
is
that
we
moved
the
discussion
about
the
hub
count
to
an
appendix
so
right
now
in
terms
of
the
functionality
of
the
draft.
H
It
does
not
define
a
hop
count
field,
but
it
does
have
an
appendix
that
discusses
the
need
for
hub
count
how
it
can
be.
It
can
be
implemented
by
using
the
ttl
data
field,
the
node
id
hub
limit
data
field
and
the
trade-off
between
having
a
dedicated,
hop
count
field
and
using
the
existing
data
fields.
H
A
few
words
about
this
open
issue,
so,
regarding
the
hubcount
field,
we've
discussed
this
a
few
times
in
the
past.
So
just
a
few
words
to
recap
here,
so
it's
either
we
don't
use
a
spec,
dedicated
hop
count
field
or
we
do
use
it.
So
if
we
don't
use
a
dedicated,
hop
count
field,
the
main
advantage
is
that
iom
nodes
along
the
path
do
not
need
to
update
the
direct
exporting
option
header
they
just
need
to
forward
it
to
process
it,
which
simplifies
what
ion
nodes
need
to
do
in
terms
of
direct
exporting.
H
H
H
Header
and
the
other
is
to
have
a
16
byte
option
header
and
in
the
second
mode
you
also
have
two
fields
which
are
sequence
number
and
the
flow
id,
so
these
two
fields
are
optional
and
the
way
the
draft
defines
it
right
now
is
that
the
length
of
the
option
header
is
implied
by
the
lower
layer
headers.
So
you
know
the
length
by
reading
the
header
of
the
previous
layer,
so
the
open
issue-
and
we've
discussed
this
in
the
past.
H
What
happens
if,
in
the
future,
we
want
to
add
more
fields
to
the
direct
exporting
option?
That
would
mean
we
would
have
to
somehow
indicate
which
which
mode
is
used
which
fields
are
used.
That
would
be
a
problem.
So
how
do
we
handle
this
open
issue?
One
solution
would
be
to
use
the
reserved
flags
right
now.
We
have
quite
a
few
reserved
flags.
We
can
use
some
of
them
to
indicate
whether
we
have
the
optional
fields
or
not.
H
H
A
A
D
B
Okay,
excellent,
I
wanted
to
go
and
just
echo
that
the
second
option
is
the
second
solution.
Option
is
just
one
that
came
out
of
the
iom,
design
team
as
a
option
that
I
believe,
is
very
clean
because-
and
I
think
that's
already
indicated
on
slide
number
nine.
C
B
Count
limit
or
hop
count
in
index
is
something
that
is
very
similar
to
the
hop
limit
that
we
already
have
for
tracing.
B
We
have
a
sequence
number
defined
for
end
to
end
in
the
current
data
draft,
which
would
redefine
index,
and
I
don't
really
see
a
logic
where
we
would
need
to
go
and
make
dax
a
replication
of
what
we
have
in
other
fields,
as
opposed
to.
We
use
decks
as
a
very,
very
simple
one,
fixed
header,
no
options,
and
then
everything
else
that
should
go
and
have
additional
flexible
fields
could
go
into
other
options
that
are
already
built
for
tracing
for
for
for
flexibility
right.
B
So
if
we
make
a
clear
cut
there,
I
think
we
resolve
a
lot
of
things
by
keeping
deck
simple
and
by
reusing
what
we
already
have
in
splatter
instead
of
re-specifying,
and
so
we
could
go
and
have
another
ref
of
the
data
graph
at
some
point,
and
there
is
already
in
dish,
discussion
and
a
couple
of
new
drafts
coming
in.
B
I
think
we
have
one
in
the
lightning
talks
where
people
argue
for
new
fields
and
we
can
have
a
new
field
that
says
well
flow
id,
for
instance,
what
we
have
in
the
data
in
the
dex
option
right
now
and
add
that
maybe
to
e2e,
so
that
would
cut
things
and
make
it
relatively
simple.
We
can
go
and
close
stacks
and
then
resolve
the
issues
that
we
all
have
here.
B
E
Good
update,
thank
you
and
definitely
would
like
to
contribute
to
the
discussion.
I
would
come
reading
and
send
a
note
on
the
mailing
list
about
their
options.
I'm
strong
believer
in
tov
and
I
think
that
that's
the
most
flexible
and
that's
what
we've
done
for
stamp
extensions
so
might
be.
We
can
reuse
this.
A
A
Ago,
all
right
we're
not
hearing
anything.
Maybe
you
could
put
your
comments
in
the
chat
if
that's
okay,
because
we
should
probably
move
on
for
time
reasons
yeah.
So
thank
you
to
all
for
presenting
this.
I
think
we've
gotten
a
lot
of
good
input
here
for
people
who
did
comment.
Please
send
an
email
to
the
list
about
this.
A
I
think
it'd
be
good
to
have
more
of
this
kind
of
discussion
going
on
with
the
whole
working
group
there
and
as
far
as
progressing
things,
I
think
it
sounds
like
there
needs
to
be
work
on
both
of
these
continuing
and
I'd
also
like
to
see
and
ensure
that
we
rationalize
the
relationship
between
the
two
different
documents
dex
and
the
loopback
flag.
A
So
I
imagine
them
kind
of
progressing
together
rather
than
separately.
At
this
moment,.
A
Much
all
right
so
next,
I
think
we
are
getting
into
some
of
the
documents
that
we
have
calls
for
adoption
out,
starting
with
the
ones
that
are
related
to
the
spring
working
group
documents,
where
we
have
two
proposed
extensions
to
stamp
and
two
amp
that
have
been
spun
off.
So
we
could
discuss
those
here
in
ippm,
so
rockets
do
you
wanna?
A
And
as
a
time
track
check
we're
running
a
little
longer,
the
last
one,
but
that's
okay,
because
it's
you
know
still
very
important
to
our
core
work,
so
yeah
we're
gonna
reserve
the
same
amount
of
time
that
we're
planning
for
these
documents.
Thank
you.
A
I
All
right
you
hear
me:
yes,
we
can
hear
you,
okay,
thank
you,
so
my
name
is
raques
gandhi
and
I'm
presenting
the
two
drafts
that
tommy
mentioned
for
the
bump
light
extensions
and
the
stem
extensions
for
segment
routing
networks,
this
where
part
of
the
spring
drops
and
they
will
spun
off
two
two
ippm
drops
for
the
extensions
for
these
protocols.
I
I
So
this
is
addressing
two
requirements:
one
is
being
the
in-band
delay
and
the
synthetic
loss,
as
well
as
a
standalone,
direct
mode
loss
measurement
and
the
scope
of
this
draft
is
t1
lite.
That's,
basically
the
messages
that's
defined
in
five,
three:
five:
seven
test
messages
with
a
user
configure
ipvp
paths
which
are
very
similar
to
the
one
that's
rfc
8762
stamp.
That
became
a
standard
recently
next
slide.
I
Please
so,
as
we
mentioned,
that
there
is
a
spring
draft,
corresponding
drop
for
this
ippm
extensions
and
in
this
string
spring
drops
defines
procedures
for
end-to-end
sr
paths
for
srm
pls,
as
well
as
srv6
data
planes,
as
well
as
for
links,
and
here
the
there
is
basically
the
protocol,
extensions
that
in
theo
applied.
There
are
two
of
them
and
next
slide.
Please.
I
I
I
I
F
I
Please
so,
as
tommy
mentioned,
that
these
drafts
are
in
working
group
adoption
or
right
now
so
far,
we
have
received
review
comments
on
this
one,
the
first
one
being
on
the
status
of
the
draft.
Currently
it
defined
extensions
for
the
applied
messages.
I
We
think
it
updates
the
5357
because
of
the
message
format,
change,
adding
a
new
field.
It
is
informational
right
now,
but
because
of
the
messages,
probably
it
could
be
a
proposed
standard.
We
look
forward
to
the
input
from
the
working
group
on
the
on
the
status
of
this.
I
There
was
a
review
comment
that
maybe
it's
not
specific
to
sr
it
could
be
generalized
or
could
be
renamed.
That
sounds
like
a
good
idea
for
this
drop.
There
were
a
few
other
editorial
comments,
abbreviations
or
using
a
correct
terminology
from
the
rfc
5357,
which
you
agree.
I
There
is
also
showing
us
entire
packet
or
loss.
We
should
be
very
clear
about
this-
is
that
word
loss
measurement,
some
editing
to
move
some
text
from
one
section
to
another,
so
yeah
we
would
be
making
those
editorial
changes
as
well.
There
was
a
proposed
well,
maybe
we
can
use
icmp
for
direct
loss
measurement.
We
think
that's
out
of
scope
for
this.
This.
The
focus
of
the
draft
is
the
t1
base
messages
next
slide.
Please.
I
Yeah,
so
these
were
the
comments
very
specific
to
this
draft.
There
are
comments,
some
comments
as
well
for
the
spring
drafts,
and
there
are
some
comments
as
well
for
the
scan
and
apps
that
we
will
take
a
look
the
next.
So
many
thanks
for
all
your
great
support
on
the
mailing
list
and
the
review
comments
that
you
have
provided.
I
This
is
in
yppm
working
group,
adoption
poll
and
or
actually
it's
it
ends
on
the
16th.
So
there
is
few
more
hours
left
for
this
one,
and
I
appreciate
your
review
comments
for
the
for
those
drops
today
or
anytime
later,
as
well.
A
Yeah,
and
do
we
want
to
go,
we
want
to
talk
about
the
other
draft
before
we
get
to
the
queue
or
do
you
want
to
take
questions
on
this
one
independently.
I
The
both
drafts
are
very
similar.
Discussion
could
be
very
similar
as
well
might
as
well
go
to
the
second
draft
and
then
but.
A
Let's
do
that,
can
we
just
kind
of
get
to
the
differences,
maybe.
I
I
I
There's
another
tl
we
defined
that
identifies
the
destination
node
address.
So
when
the
query
is
sent
for
127
address
this,
this
would
carry
the
actual
node
address
to
verify
that
when
the
response
comes
is
coming
from
the
right,
node
and
next
slide.
I
Please
so
yeah,
so
this
is
also
very
similar
to
the
trump
light.
Only
difference
here
is
there
is
ssid,
but
otherwise
it's
very
similar
to
the
both
are
similar
to
light
and
next
lightness.
I
Again,
the
review
comments
are
very
similar,
so
draft
status.
I
think
the
previous
comment
doesn't
apply.
This
one
is
already
proposed
standard,
so
we're
just
updating,
rfc
8762,
pretty
much.
Everything
else
is
the
same
in
terms
of
review
so
yeah.
This
is
also
in
adoption
for
many.
Many
thanks
to
your
review
comments
and
suggestions.
Yeah
thanks.
E
E
Yeah
yeah.
Thank
you.
So
no
three
slide.
Three.
Let's
slide
one
back!
No,
this
yeah!
This
is
three
at
least
it's
number
three.
Yes,
so
t1
white
already
provides
delay
and
synthetic
loss
measurement.
E
I
agree
that
direct
mode
loss
measurement
is
useful,
but
I
think
it
will
be
helpful
if
the
working
group
discusses
how
to
do
it,
because
what
appears
that
you
propose
two
different
solutions
for
one
problem,
so
you
propose
some
extension
to
t1
white,
which
is
a
non-standard
and
it's
only
informational
and
you
propose
something
to
stamp
when
all
that
can
be
addressed
with
one
mechanism.
E
As
I
pointed
out
in
my
comments
using
multi-part
message,
extension
to
icmp.
E
E
And
just
to
save
time,
that
will
be
it.
So
I
would
like
to
move
this
discussion
of
my
comments
to
the
mailing
list.
I
I
E
Okay,
please
send
them
so
that
ippm
working
group,
because
again
because
this
two
documents
actually
like
two
pairs
of
documents
being
discussed
on
both
spring
and
ippm
lists.
I'm
not
sure
that
everybody
in
ippm
list
group
are
reading
a
spring
working
group
discussion.
I
A
All
on
the
list-
and
we
also
have
martin
in
the
queue
martin,
do
you
want
to
speak
up.
F
Thank
you
rakesh
for
the
presentation.
I
I
read
this
for
the
first
time
a
few
days
ago,
and
I
was
really
trying
to
understand
the
motivation
for
this
work
and
honestly
I
had
a
lot
of
trouble
getting
that
out
of
the
the
introduction,
and
some
of
that
is
what
greg
was
saying
about.
F
Okay,
that's
interesting!
It's
it's
very
much
conceptually
different
from
what
I
perceive
a
lot
of
iptm
stuff
to
be,
which
is
like
measurement
techniques,
to
like,
say
that
you
know
to
say
that
there's
already
a
device
doing
this
measurement,
we're
just
trying
to
like
extract
it
from
the
device
is
like
why
not
just
use
http
or
something
to
get
that.
I
Yeah,
so
we
already
have
the
t-wamp
like
machinery
which
collects
timestamps
today
and
and
hoping
the
whole
thing
just
to
collect
the
counters
from
tx
and
rxi.
This
has
been
done
in
like
y1732
or
rfc6374,
so
there
are
precedence
of
this
hardware
capabilities
that
exist
for
the
counters,
the
hardware
capabilities
that
exist
for
t1.
I
So
there
is
a
lot
of
motivation
to
add
these
small
extensions
and
for
segment
routing
use
cases
where
we
are
measuring
the
actual
customer
traffic.
This.
This
is
a
a
very
good
application
for
it.
F
Okay,
fair
enough,
I
I
I.
I
hope
that
between
my
comments
and
greg's,
maybe
some
ideas
and
how
you
could
flush
out,
at
least
like
the
introduction
of
this
document,
a
little
more
just
to
sort
of
say
what
prob
exactly
is
being
solved
and
maybe
not
get
too
deep
into
like
spring
terminology
for
people
who
aren't
familiar
with
it.
Thank
you.
A
All
right,
so
it
looks
like
we've
trained
the
queue
you
know
on
the
call
for
adoption.
We've
had
a
lot
of
good
support
from
a
number
of
people,
but
I
think
that's
also
due
to
it
being
cross-listed
in
the
work
with
spring.
The
comments
that
greg
and
martin
bring
up
are
very
good.
A
What
I
would
like
to
hear
more
of
is
comments
and
reviews
and
thoughts
on
these
particular
issues
from
you
know
some
of
the
ippm
working
group
members
who
did
a
lot
of
work
on
t,
amp
and
stamp,
and
the
design
and
the
kind
of
original
discussions
around
those
and
to
comment
on
kind
of
the
appropriateness
of
this
or
really
what
is
the
right
way
to
solve
these
problems?
A
So
I
can
repeat
that
on
the
list
as
well,
but
you
know
this
is
a
call
to
get
kind
of
more
input
in
higher
level
review
like
that
and
greg.
Did
you
want
to
say
something.
E
As
I
noted
in
my
review
comments,
this
proposal
and
a
com
and
a
similar
proposal
on
stem
does
look
like
it
defines
a
new
protocol
and
that's
actually
reflected
in
the
slides
that
this
is
a
standalone
functionality.
E
A
A
K
Ippm
connectivity
monitoring
third
version,
I
applied
for
adaption
at
the
work,
and
this
was
actually
running.
The
scope
is
monitoring
second
routed,
subpath
or
links
and
to
detect
and
locate
loss
of
connectivity
and
congestion,
and
that's
the
value,
add
detecting
the
location
of
congestion,
which
is
enabled
if
this
draft
is
supported,
it's
a
based
and
only
work
with
mphs
segment,
routing,
I'm
not
sure,
but
it
could
work
with
ipv6
segment.
Routing
too
so
properties
is
monitor,
network
sections
with
known
topology,
not
unknown
ones.
K
One
measurement
path,
monitored,
link
that
is
important.
The
design
is
so
that
you
roughly
have
the
same
functionality
as
if
you
work
with
a
ping
and
yep.
There
is
one
important
issue
change
point.
Detection
doesn't
require
well
synchronized
timestamps,
so
it's
a
differentiating
matrix
metric
and
it
applies
segment
routing
to
allow
for
an
a
prior
design
network,
tamography
evaluation
and
a
limited
number
of
monitoring
systems.
So
it's
important
you
do
not
have
to
work
on
routers.
You
can
work
on
routers,
so
it's
just
depending
on
what
you
want
to
use
it
for.
K
And
it
does
estimate
round-trip
delays
and
one-way
delays
of
monitored
links
or
path,
as
it
is
designed
right
here
and
yeah,
the
aim
is
to
standardize
a
suitable
metrics
and,
as
I
mentioned,
it's
a
tomography
approach
next
slide.
Please.
K
So
this
is
shown
here
in
the
slide.
I
do
not
want
to
go
to
details,
but
you
need
to
be
relatively
strict
in
setting
up
paths
like
shown
here,
and
I
think
that
should
be
done
in
an
automated
fashion
if
you
run
it
out
on
a
larger
scale.
But
this
is
not
part
of
the
draft.
The
draft
is
only
dealing
with
the
methodology
and
the
metric
right.
K
The
important
thing
is,
every
measurement
path
is
set
up,
as
shown
above
you
have
one
u-loop,
you
have
one
downlink
measurement,
one
uplink
and
below
you
see
that
on
each
measured
path
you
have
pretty
much
the
same
mix
of
three
different
paths
or
the
green
one
is
the
uplink
one.
The
blue
one
is
the?
K
So
this
is
a
difference
to
a
ping
where
you
do
not
know
exactly
where
the
congestion
is
located,
because
you
cannot
differentiate
that.
So
that's
the
value
add
of
this
contribution.
All
right.
I
don't
want
to
go
to
details.
If
you
have
time
read
the
slide,
it
roughly
tells
you
what
to
do.
I
received
questions
during
the
discussion
on
the
list
and
these
were
related
to
the
temporal
resolution
of
event
detection.
K
So
there
is
a
little
text
and
the
draft
the
text
on
this
slide
is
likely
a
bit
larger
and
the
illustration
is
more
better.
K
Now,
as
you
see
in
the
figure
that
all
the
packets
of
all
flows
are
having
the
same
periodical
distance
measurement
interval,
and
so
that
I
think,
is
quite
straightforward.
K
Temporal
properties-
I
just
see
that
I
was
hard
to
hear
and
but
I
didn't
think
that
through
so
this
is
a
a
quite
simple
answer,
and
that
tells
you
what
can
be
done.
The
aim
is
not
to
replace
any
hello
messages
which
are
exchanged
directly
on
a
link
with
neighbors.
The
aim
is
to
monitor
networks
on
a
time
scale-
maybe
seconds
maybe
minutes
and
just
be
aware
where
you're
forwarding
doesn't
work,
because
that
is
what
this
draft
do.
There
is
no
interaction
with
routers.
K
It's
only
measuring
on
forwarding
paths
all
right,
and
that
was
the
presentation.
If
you
are
interested,
read
the
draft
and
support
adaption.
I
don't
know
whether
this
should
be
discussed
here.
I
don't
know
how
big
the
segment
routing
mpls
segment,
routing
know.
How
of
this
work
group
is
here,
so
I
didn't
try
to
present
it
in
segment
routing.
Yet
if
there
is
a
sentiment
that
I
should
do
so,
please
let
me
know:
thanks
are
there
any
questions.
D
Greg's
in
the
queue
greg,
do
you
want
to
go.
E
I
just
wanted
to
note
that
if
you
see
the
scale
of
detection
in
terms
of
seconds
or
even
100
milliseconds,
that
might
be
too
long
period,
because,
if
we'll
take
local
protection
like
trl
tray
that
will
work
on
10
milliseconds
interval,
so
there
are
local
protection
there.
E
A
defect
that
is
locally
detected
between
two
nodes
would
not
be
noticed,
so
I
I
think
that
that's
something
important
to
really
take
into
account
that
we
might
have
a
local
switchover
and
reroute
of
the
path
and
the
system
would
not
notice.
K
I'm
not
sure
if
it
changes
the
forwarding
pattern
and
you
measure
different
times,
then
you
will
notice
that
if
you
are
looking
at
the
situation
of
a
bundle,
then
it
might
go
unnoticed.
I
agree,
but
then
that
depends
on
what
you
want
to
measure.
I
had
a
conversation
with
someone
else
who
tried
to
suggest
it
to
monitor
end-to-end
paths
and
then
that's
something
completely
different.
K
D
Okay,
thanks
jay
you're
next.
A
A
All
right
yeah,
so
we're
not
hearing
from
you
jose,
ignacio.
If
you
want
to
unmute,
please
do
otherwise.
I
think
you
don't
have
any
other
people
in
queue
yeah,
so
we
haven't
gotten
a
lot
of
feedback
on
the
list
on
this
one.
Thank
you
al
for
sending
your
comments
and
greg.
If
you'd
like
to
kind
of
reply
on
the
list,
that
would
be
great
as
well.
A
I
would
ask
that
we
get
a
few
more
people
reviewing
this
document
and
commenting
whether
or
not
they
think
it
should
be
adopted
before
we
move
on,
but
otherwise
that's
good.
Thank
you.
So
much
berger.
K
L
Okay,
hello,
it's
xiaomi
speaking.
This
presentation
is
echo
request,
reply
for
enabled
in-situ
oem
capabilities.
L
L
L
L
L
L
L
L
Since
still
one
version
of
this
draft,
the
defined
tv
sub
tvs
are
reclassified
and
we
adjust
the
relevant
tov
sub
tv
definition.
Accurate
test
and
echo
replied.
Tuvs
are
divided
into
two
separate
sections
list
of
sub-tuvs
is
removed
from
the
echo
request.
Tov
tracing
capabilities
sub-tuv
is
divided
into
pre-allocated
tracing
capabilities,
subtle,
v
and
incremental
tracing
capabilities
sub
to
v.
L
L
L
L
We've
also
received
a
good
comments
from
drove
study
before
this
meeting,
and
the
updates
on
addressing
these
good
comments
will
be
included
in
the
next
revision
of
this
chapter
next
slide.
Please.
L
Next
steps,
this
job
is
in
call
for
adoption.
Currently,
we
are
looking
forward
to
the
conclusion.
F
There
we
go
okay,
sorry
yeah!
This
is
there's
a
little
latency
in
this.
Thank
you
xiao
for
the
presentation.
So
am
I
understand
that
then
every
one
of
these
echo
requests
generates
and
echo
replies
forever
and
advertising.
F
L
F
Okay,
so
it's
there's
not
an
amplification
problem.
Wonderful
have
you
considered,
possibly
if,
if
the
is
it
is,
is
the
desire
here
to
figure
out
what
is
like
the
actual
common
set
of
capabilities,
that
the
entire
path
is
able
to
support?
And
if
so,
could
you
not
like
have
a
essentially
like
a
list
of
capabilities
that
is
travels
through
the
through
the
path
and
filters
out
the
the
capabilities
that
are
not
common?
It's
like,
for
instance,
hop
one
would
say.
F
Oh,
I
can
do
ten
different
things
and
the
hop
two
says
I
do
eight
of
those
and
so
then
eliminates
two
of
the
fields
and
so
on.
Until
you
end
up
with
you
know
the
five
common
things
that
the
entire
path
uses,
so
you
could
do
it
all
on
one
message
instead
of
a
bunch
or
is
that
not
what
you're
trying
to
achieve.
F
E
E
Yes,
sorry
to
jump
well,
that
is
one
of
the
scenarios
that
martin
described
so
finding
their
lowest
common
denominator,
but
at
the
same
time
the
goal
is
just
to
learn
the
capabilities
of
each
node
and
that
can
be
used
in
some
other
scenarios.
B
C
B
L
L
So
I
think
it
it's
more
like
a
tool,
a
complementary
iom
tour,
so
I
I
don't
think
it's
management
specific.
L
Secondly,
for
the
ippm
working
group,
whether
or
not
ibpm
is
the
appropriate
working
group
for
this
work.
I
think
I
have
already
sent
a
sentence
this
question
before
and
I
think
the
the
comments
to
my
question
is
that
this
is
iom
specific.
So
I
need
to.
L
Present
this
chapter
and
get
adopted
in
ippm
before
I
can
do
any
next
actions.
Next
steps
to
actually.
I
have
already
said
in
my
presentation
that
I
asked
for
guidance
from
the
working
group
chairs
on
whether
to
request
ir
requests
in
this
job
or
in
separate
jobs
in
other
working
groups.
C
M
Since
the
draft
is
try
to
extension
to
the
icmp
and
the
mtsp
particle,
so
I'm
not
sure
this.
This
is
the
right
working
group
to
program
this
work.
L
Yes,
I
have
already
explained
to
frank,
so
I
think
it's.
I
asked
for
guidance
from
working
group
chairs
on
this.
N
Hi
guys,
okay,
one
comment
like:
can
this
mechanism
to
be
used
in
like
multi
domains
or
basically
something
like
this
because
like
by
using
the
icmp,
I
think
it
can
like
get
information
of
the
nodes
from
other
domains,
but
I
think
it
will
introduce
some
security
considerations
issues
like
what
do
you
think
about
this.
L
A
All
right,
thank
you
and
just
to
kind
of
reiterate
that
point
any
of
these
comments,
particularly
the
ones
that
are
raising
some
concerns.
Please
do
reply
to
the
awkward
option
for
that.
I
don't
think
I've
seen
a
lot
of
those
on
that
particular
thread.
So
please
do
chime
in
all
right.
D
O
O
Thanks
all
for
your
comments,
we
have
evolved
several
versions
to
align
with
the
latest
iom
data
draft
and
the
iom
directory
support
draft,
and
the
latest
version
corrected
the
admin
type
scenes,
as
suggested
by
tom,
including
the
reference
copyright
tree
diagram
and
the
correct
reference
of
the
proof
of
transit
and
so
on
and
after
we
submit
submitted
this
slide.
We
received
two
more
detailed
comments
from
the
roof
and
the
russian
not
updated
this
slide,
but
I
will
discuss
this
later
next
slide.
O
The
model
structure
is
like
this:
the
model
is
organized
as
a
list
of
profiles,
as
suggested
by
the
working
group.
Each
profile
container
is
enabled
by
by
the
corresponding
features,
so
the
device
do
not
need
to
support
all
the
features
each
profile
associates
with
one
flow
and
the
corresponding
iom
information.
O
The
flow
is
typically
identified
by
pcl
and
can
be
identifi
defined
flexibly.
We
support
five
profiles
here:
the
incremental
tracing
the
pre-allocated
tracing
the
director
export
proof
of
transit
and
the
end
to
end
multiple
iom
data
types
can
be
encapsulated
into
the
same
ion
header.
O
O
O
O
O
The
i1
proof
of
transit
data
is
to
support
the
pass
or
service
function
chain
verification
use
cases
it's
imported
from
the
draft
in
sfc
working
group.
Richard
made.
This
latest
comments
made
the
latest
comments
on
this,
so
one
discussion
is
whether
we
should
incorporate
the
parts
directly
in
this
i1
model
or
still
like
this.
We
import
it
from
the
sfc
and
in
comments.
O
A
Thank
you.
It
looks
at
this
point
that
there
isn't
anyone
in
the
queue
someone
speaks
up
now
and
we
did
get
a
lot
of
good
feedback
on
the
list.
So
thank
you
for
everyone
who
did
respond
there,
all
right
so
yeah.
Thank
you
for
your
presentation
and
we'll
get
back
on
the
list
about
the
status
for
adoption.
O
A
All
right
so
now
we're
going
to
go
on
to
the
documents
that
are
not
in
current
call
for
adoption,
and
actually
I
was
expecting
that
we
would
have
the
hybrid
two
step
next.
A
Apologies
or
actually,
I
don't
know
if
it
matters
too
much
if
tomorrow's
already
ready.
Oh
okay,
let's
keep
the
agenda
as
it
was
and
take
ten
minutes
to
talk
about
the
hyper2
step.
That's
okay,
ready.
N
E
Please,
okay,
so
we
had
a
good
discussion
on
the
mailing
list
and
edit
how
hybrid
two-step
can
use
iom
or
and
alternate
marking
them
domains
and
clarify
their
scenario
of
the
late
full
packet.
So
when
their
htc
timer
expires,
and
we
welcome
how
you
as
a
call
for
netflix.
E
So
the
telemetry
can
be
viewed
as
a
union
of
three
tasks,
generating
telemetry
information,
collecting
transporting
telemetry
information
and
then
correlating
telemetry
information.
E
We
are
working
on
a
number
of
solutions
to
their
telemetry,
and
that
includes
iem
trace
options
and
when
their
information
is
collected
and
transported
in
the
data
packet
itself
and
the
trace
option
can
have
two
sub
variances
pre-allocated
and
incremental.
E
Then
we
have
a
direct
export
which
combines
the
proposal
that
postcard
based
elementary.
I
there
is
a
postcard
based
telemetry
m,
which
uses
as
a
trigger
method
of
packet
marking,
for
example,
as
the
described
in
rfc
8321,
so
in
direct
export
or
marking.
So
the
difference
only
in
what
triggers
the
information.
E
Generation,
their
method
of
each
node
collects
the
information
and
then
transported
to
their
collector,
using
a
out
of
band
packet,
so
out
of
band
in
terms
of
relative
to
the
data
flow
which
generated
this
telemetry
information
next
slide.
E
So
each
of
these
methods
that
we're
pretty
much
well
familiar
with
has
some
advantages,
but
at
the
same
time,
challenges
so
the
hybrid
two
step
is
a
method
that
is
complementary
to
all
the
method
that
I
mentioned,
and
it
uses
their
specially
constructed
message:
a
follow-up
message
to
collect
telemetry
and
transport
it
along
the
same
path,
as
was
followed
by
their
packet,
that
triggered
their
telemetry
information.
E
So,
since
the
trigger
packet
is
a
network
layer,
specific
the
follow-up
packet
uses
the
same
encapsulation
so
to
ensure
that
it
follows
the
same
path,
but
the
method
can
be
flexible
so
that.
E
E
So
another
quality
of
hybrid
two-step
proposal
is
that
there
is
no
limitation
of
how
of
amount
of
information
collected
along
the
path,
because
any
transit
node
would
be
capable
or
required
to
be
capable
to
generate
additional
follow-up
packet
using
the
same
encapsulation
as
their
original
data
pack.
E
Please
so
this
is
outlines
their
proposed
format,
a
follow-up
packet,
the
transport
network
encapsulation,
which
is
should
be
identical
to
the
one
that
used
for
a
layer,
specific
trigger
packet
and
then
information
that
is
required
for
identification
of
the
flow.
The
sequence
number,
it's
a
sequence
number
in
a
queue
of
follow-up
objects
that
are
triggered
by
the
same
data
packet
and
then
telemetry
information
organized
as
a
tovs.
E
So
here
it's
in
the
illustration
how
their
method
of
collecting
transporting
their
telemetry
information
works.
We
have
a
flow
of
data
packets.
Then
there
is
a
trigger
packet
which
produces
the
follow-up
packet
and
follow-up
packet.
The
original
for
packet
may
be
filled
with
their
telemetry
information.
E
Then
their
transit
note
in
this
diagram.
It's
a
node
c
can
generate
another
additional
follow-up
packet
and
then
they
will
arrive
at
the
egress
node,
which
will
remove
them
from
the
network,
because
it's
a
destination
of
the
data
packet
and
then
the
information
can
be
analyzed
or
sent
for
the
processing.
The
advantage
of
it
is
that
it's
easy
to
correlate,
because
all
the
information
like.
E
Iom
trace
options
collected
together
next
slide.
E
So
the
previous
diagram
shows
how
hybrid
to
step
method
works
in
a
point-to-point
case.
It
works
in
a
multicast
distribution
tree
as
well,
and
that's
works
efficiently,
because
there
is
no
replication
of
telemetry
information.
E
So
I
would
not
go
through
this.
You
can
read
the
draft
or
bring
the
question
on
the
list,
but
the
diagram
shows
that
only
their
information
of
the
sub
multicast
sub
3
is
collected
in
a
follow-up
packet.
So
the
information
that
was
collected
upstream
is
not
replicated
on
all
their
branches.
E
So
iom
can
be
used
with
a
hybrid
two-step
method
of
collecting
information,
and
we
added
a
description
of
how
what
information
from
their
I
header
can
be
used
to
trigger
and
how
this
information
can
be
encapsulated
into
their.
E
Payload
of
follow-up
packet,
so
that's
an
intention
of
using
their
information
defined
in
iem
data,
but
encapsulating
in
tovs
for
flexibility
next
slide.
Please.
E
So
we
are
looking
for
a
for
comments,
discussion
on
the
list
and
very
encouraged
by
their
thought
of
adopting
this
work
at
some
point,
thank
you.
J
Hi,
I'm
wondering
if
that's
actually,
this
work
is
actually
in
scope
for
ippm,
because
it
somehow
rather
defines
an
entirely
new
protocol
than
like
a
metric
or
model
or
whatever,
so
that
should
definitely
be
discussed
before
starting
working
group
adoption.
I
think.
E
E
So
but
that's
something
for
working
group
shares
and
ad
to
decide
but
well
appreciate
their
consideration
and
advice
where
this
work
can
be
anchored.
A
Thanks
for
bringing
that
up,
I,
I
think
also
there's
a
you
know,
general
question
that
this
raises
to
me
of
making
sure
that,
as
we're
doing
the
work
such
as
direct
export,
that
we
have
a
kind
of
good
and
consistent
story
and
we're
not
doing
too
much
overlapping
or
conflicting
work
and
to
see
how
much
can
we
unify
these
approaches
all
right.
A
A
Yes,
it's
a
little
quiet,
so
just
speak
up
other
than
that.
We're
good
okay,.
G
G
G
Okay,
there
are
some
implementations
that
are
already
made
and
demonstrated
in
the
acaton,
for
example,
during
the
quick
measurement
project
and
the
implementation
are
actually
our
free
telecom
italia.
Implementation
is
on
android
mobile
phones.
The
probe
is
inside
the
mobile
phones,
so
the
idea
is
to
put
the
pro,
not
in
the
network,
but
also
in
the
devices
and
end
user
devices.
There
is
some
implementation
on
core
network
probes
and
the
orange
jakama
implementation,
the
first
one
already
in
field
by
two
years,
the
akamai
production,
cdns
and
core
network
props.
That's
right.
G
I
recap
shortly
the
round:
try
trip
time
spin
bit
that
is
used
to
mark
the
packet
in
a
client
server
protocol
to
measure
the
round
trip
time.
So
the
idea
is
to
create
a
square
wave
signal
that
bunches
between
client
and
server,
and
the
length
of
the
signal
is
equal
to
rtt
time
round
trip
time
next
slide.
G
G
So
an
observer
put
on
the
path
can
measure
the
distance
between
two
delay
sample
two
double
market
packets,
and
so
the
measurement
is
more
precise
because
there
isn't
ambiguity
caused
from
packet
loss,
because
if
the
packet
is
not
detected
in
size,
the
spin
bit
period
that
the
measurement
is
not
done
and
the
delay
beat
detects
a
specific
packet.
So
there
isn't
a
possibility
that
you
can
measure
different
packets
on
the
edges
of
the
periods
next
slide.
G
G
G
Okay,
this
is
about
the
one
it's
different
before
the
previous
one,
because
this
is
one
direction.
One
way:
packet
loss,
the
traditional
packet
loss,
because,
because
the
round
three
packet
loss
is
a
new
kind
of
measurement
in
which
we
can
detect
where
the
packet
loss
occurred,
but
in
this
case
measuring
the
one-way
packet
loss,
we
can
measure
the
direction
where
the
packet
loss
occurs.
G
G
In
this
case
the
measurements
happens
between
the
end
point
and
the
measurement
point
in
is
this
quite
different,
because
the
previous
techniques
about
country,
plus
or
round
trip
time
in
this
case
the
measurement
is
not
a
end-to-end
measurement
but
is
between
the
client
and
the
server
in
one
direction
or
the
servant
or
serving
the
other
direction.
So
we
can
measure
the
loss
between
the
endpoint
and
the
observer
next
slide.
G
If
we
have
two
measurement
point,
we
can
do
some
inter
domain
packet
loss
because
we
can
do
the
difference
between
the
packet
arrived
in
a
server
that
and
the
packet
arriving
the
second
observer.
So
we
can
measure
the
difference:
the
packet
loss
in
between
the
two
server.
G
G
The
7
bit
idea
is
to
mark
a
packet
with
that
is,
albeit
put
to
one
when
a
loss
is
detected
by
the
protocol.
So
the
idea
is
to
have
a
connection
between
the
protocol
measurement
inside
application
and
the
network
measurement
marking
the
a
bit
in
a
packet.
G
Every
packet
loss
event
detect
inside
the
protocol,
so
is
only
a
one,
is
a
end-to-end
measurement.
You
know,
server
measure
in
every
point
is,
is
put.
It
always
observe
the
end-to-end
measurement
on
this
direction
on
this
path
so
complement.
The
next
slide.
Complement
is
this
kind
of
measurement
with
the
previous
one,
the
cubit
we
can
have.
We
can
split
them
the
packet
loss
between
the
server
and
the
server
and
between
the
client
and
the
server.
So
we
have
all
the
possibility
to
split
the
measurement
to
locate
the
losses
next
slide.
G
The
alternative
to
l
beat
is
to
use
the
cubit
coupled
with
the
airbeat.
Their
beat
is
a
measurement
that
simply
reflects
the
losses
from
one
side
to
the
other
side,
so
they
could
be
the
mark,
the
packets,
the
64
packets,
for
example,
if
you
find
less
than
64
packet
market
from
the
other
side
of
the
network.
To
the
other
end
point,
this
endpoint
reflects
a
period
that
starts
with
this
number
63,
for
example,
62.
G
If
there
are
this
one
or
two
losses
in
one
side,
so
we
can
measure
on
the
other
side
the
losses
on
the
first
way
and
part
of
the
lawsuit
the
second
way,
so
we
have
in
the
next
slide
next
dice.
Please
the
possibility
to
complement
it
is
to
measurement
to
have
to
split
the
losses
between
the
two
direction
on
one
direction.
For
example,
we
can
subtract
the
qubit
measurement,
the
loss
velocities
direction
to
the
three-quarter
measurement,
seeing
the
reflection
bit.
So
we
can
measure
in
one
direction
the
end-to-end
losses
in
the
opposite
direction.
G
This
using
only
one
direction
server.
This
next
slide.
If
we
add
two
of
two
direction
server,
we
can
split
the
measurement
in
two
parts:
around
three
pluses
right
around
three
plus
is
left
or
the
the
single
losses
between
the
server
and
the
client
and
between
the
client,
the
server
and
the
server.
Okay.
Next
slide.
G
G
Using
three
bit,
we
have
three
possibilities
to
measure
the
precise
rtt
using
two
beat
and
round
three
packet
loss
using
one
bit
or
using
only
one
bit
for
a
spin
beat
for
delay
at
two
beats
for
the
packet
loss.
One
way
using
the
two
alternative
possibility
could
beat
and
that'll
beat
ocubit
and
their
beat
okay
next
slide.
G
So
these
metrics
are
deeply
discussed
on
the
working
group
inside
the
mailing
list.
We
demos
this
technique
inside
the
and
we
joined
the
two
previous
drafts
that
are
about
the
packet
loss
and
we
joined
in
one
draft.
That
is
this
one
comparing
lost
bit
and
the
reflection
bit.
There
are
the
two
alternative
using
jointly
with
the
equipment.
G
A
On
the
system,
all
right,
yeah,
I
think,
thank
you
and
I
don't
think
we
have
time
to
go
over
those
slides
now,
but
those
are
available.
Does
anyone
have
comments
on
this.
D
Thank
you
for
merging
these
two
drafts
by
the
way,
and
I
I
think
this
is
pretty
interesting
work
overall,
so
I
would
definitely
appreciate
more
feedback
on
the
list.
A
All
right
looks
like
we
have
a
empty
queue,
so
yeah
yeah
again.
Thank
you
for
doing
this.
Thank
you
for
doing
the
work
to
emerge,
the
documents
and
proposals,
and
I
think
this
is
one
that
we
would
like
to
consider
for
adoption,
great
and
so
for
people
watching
the
clock.
We
are
pretty
much
at
the
end
of
our
time
and
we've
gone
through
all
of
the
documents
that
we
wanted
to
kind
of
have
full
slots
available
to.
A
Unfortunately,
it
looks
like
we
don't
have
time
for
any
of
our
lightning
talks,
but
I
would
like
to
remind
people
that
those
slides
are
available
and
to
please
take
a
look
at
those
slides
and
if
you
have
any
questions
about
those
documents,
shoot
them
to
the
list
and
what
we'd
like
to
hear
is
kind
of
more
good
feedback
on
those
kind
of
before
we
progress
or
have
larger
slots
for
those
documents.
A
And
thank
you
to
andrew
for
doing.
The
notes
looks
like
those
are
nicely
filled
out,
and
we
very
much
appreciate
that
and
your
help.