►
From YouTube: IETF102-IPPM-20180718-1330
Description
IPPM meeting session at IETF102
2018/07/18 1330
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
All
right
so
hi
I'm
Brian
Trammell,
co-chair
of
my
ppm.
This
is
AI
ppm.
I,
see
a
lot
of
familiar
faces,
so
I'm
assuming
pretty
much
everyone's
in
the
right
room.
Hialeah
welcome.
B
I'd
like
to
first
introduce
our
new
co-chair
Tommy
poly,
to
be
running
things
today,
basically
yeah,
okay,
so
everyone
welcome
Tommy
and
let's
have
a
meeting
all.
C
Right
welcome
to
IPM
everyone
I'm
looking
forward
to
joining
you
and
apologies.
If
I
don't
know,
everyone
super
well
right
now,
but
I
trust
that
everything
will
go
great.
So,
let's
just
go
through
the
ministry
via.
So
here
is
the
note.
Well,
you
have
seen
it
before
make
sure
you
do
note
it
it's
slightly
changed.
Yes,
note
it!
Well!
C
Yes,
we
are
so
sweater
is
taking
notes
for
us.
Do
we
have
anyone
who's
going
to
be
jabbar,
scribing,
okay,
Brian!
Thank
you,
alright.
So
a
quick
status
update
on
just
the
some
of
our
working
group
documents
that
we're
not
going
to
be
talking
about
this
time.
We
have
two
documents
that
are
in
the
RFC
editor
q1
is
the
ipv6
Edition
2,
2,
3,
3,
0
and
I
think
that
one
is
just
in
the
edit
state.
So
it's
nothing
particularly
new
to
update
their
the
team.
C
I'm
being
draft
does
have
its
Ayana
marking
gone
through
and
it's
just
I
think
it
has
a
dependency
on
another
draft
reference
and
yeah,
miss
Roffe,
so
I
think
that's
just
waiting
on
the
underlying
document.
Yeah
great
and
then
we
also
have
a
working
group
working
group
last
call
on
the
T
web
test
document
that
went
out
a
little
bit
ago
and
that
ends
today.
I
think
we've
only
had
a
couple
responses,
I'm
just
saying
yeah,
it
looks
fine.
If
there
are
any
other
concerns,
please
bring
those
up.
Excuse.
B
B
One
person
who
has
not
yet
said
that
14:1
test
is
okay
to
stand
up
and
say
it's
okay,
so
that
we
can
say,
we've
got
enough
people
who
paid
attention
excellent,
yeah,
okay,
perfect
good,
it's
good!
So
working
with
last
call
is
now
complete.
We
will
I
guess
it
do
we
have
a
shepherd
assigned
to
that
already
I'll
take
a
look
and
yeah
okay,
yep,
we'll
get
it
moved
along.
Thank
you.
C
Following
that,
we
have
two
other
drafts
that
are
not
working
group
items
yet
but
have
been
brought
up
in
a
hat
to
the
list
that
have
had
some
discussion
so
we'll
be
going
through
those
and
then
at
the
very
end
we
have
some
time
allocated
for
lightning
talks.
These
are
going
to
be
limited
to
five
minutes
at
most
each
and
we
will
get
through
them
as
we
have
time
so
any
agenda
bashing
on
this.
F
I'm
al
Morten
representing
a
okay,
well,
everything
else
works
perfectly
so
I'm
al
Morten,
and
this
this
is
the
initial
registry
contents
draft
that
you
see
displayed
there
that
we've
been
working
on,
obviously
for
a
long
time,
but
that's
the
draft
that
feeds
into
the
performance
metrics
registry
draft,
which
has
another
long
list
of
people
and
has
also
been
worked
on
for
a
long
time.
So
they're
really
quick
summary
because
I
want
to
get
to
the
discussion.
I
think
you're
gonna
quickly
assess
the
fact
that
I
would
like
to
see
this
done.
F
I
would
like
to
see
both
of
these
done
and
now
I
can
see
nothing.
You
can't
or
they're
there.
So
anyway,
let's
see
what
happens
perfect
great
now,
everybody's
back
online.
Even
me,
so
here's
the
quick
overview
by
the
way
who's
read
these
drafts.
In
any
version,
there
are
okay,
several
hands
coming
up.
Thank
you.
There's
a
I
mean
there's
a
lot
of
years
of
work
here,
so
if
you're
interested
in
I
ppm
at
all,
this
is
something
you
really
ought
to
be.
F
Taking
a
look
at
because
I'd
like
it
to
be
done
as
I
said.
So
we're
solving
a
problem
here,
all
the
original
rfcs
for
loss
and
delay
and
so
forth.
They
were
written
with
great
flexibility,
we're
adding
precision
here
now
so
that
if
you
make
the
measurement
according
to
the
way
the
registry
specifies
it,
and
you
do
all
the
things
the
registry
says
you're
at
least
starting
out
with
two
fairly
identical
measurement
systems.
You
can
Prost
I
still
probably
do
a
few
things
wrong,
but
you
know
with
you.
F
F
This
is
what
the
registry
looks
like
in
a
graphical
form
where
you
can
see.
We've
got
these
different
categories
of
summary
metric
definition,
methods
of
measurement,
the
outputs,
administrative
info,
all
that
sort
of
thing.
So
that's
the
quicky
quicky
summary.
So
here's
what
happened
in
versions,
7
and
15
respectively.
This
is
going
to
hit
all
of
you,
authors
and
editors.
F
Rfc
2-1-1
9
was
updated
by
RFC
8174
and,
as
Tommy
just
reminded
us,
we've
had
a
couple
address
just
moving
through
iesg
and
we
got
dinged
on
the
fact
that
neither
one
of
them
referenced
8174,
which
updates
the
terminology
in
a
nice
way.
If
it's
capitalized.
Now
it's
supposed
to
be
taken
as
this
definition,
so
everybody
needs
to
do
that.
Enough
said:
we've
revised
the
passive
tcp
metric
Yakov
Stein
provided
comments
as
as
planned.
F
G
F
We
can
we
can
definitely
improve
the
heuristics
if
we
get
the
Bessemer
review
there.
So
we
clarified
thanks
to
ignosi
Asst.
You
know
SEOs
comments,
mail
with
me
DNS,
which
is
in
Section
six.
It's
a
response
time
and
responsible
isometric
we've
clarified
that
that's
from
the
user
perspective.
There
was
some
thought
that
we
might
do
some
more
complicated
things
measuring
DNS.
What
we're?
Not
it's,
just
this
simple
thing
that
everybody
else
does
they
do
the
user
measurement
so
discussion
points
we
got
here
in
like
a
minute.
That's
pretty
cool!
F
So
now
now
is
when
you
all
sort
of
rushed
to
the
mic
and
give
me
feedback
on
these
critical
points
of
interest
for
a
DNS
response,
time
and
loss.
We
need
a
way
basically
to
track
the
fact
that
a
given
request
has
either
been
responded
to
or
not,
and
for
that
we
need
the
equivalent
of
a
kind
of
sequence
number
something
to
determine
correspondence
between
the
and
subsequent
requests
that
we
make.
F
B
B
D
B
B
B
F
F
H
F
H
F
Yes,
all
right!
So
that's
one!
So
then
we
have
this
other
point
in
section,
8
we've
got
this
one
specific
metric
for
UDP,
one-way
delay
and
loss,
the
probability
variation
and
things
like
that
there
too,
and
we
left
this
open
so
that
people
could
provide,
have
have
some
flexibility
about
the
the
packet
stream
that
they're,
generating
and
so
forth,
but
I'm
I'm.
My
tendency
is
that
we
should
make
this
fixed
time
parameters
or
fixed
parameters
that
we
choose
for
the
for
the
packet
periodicity,
for
example,
in
the
packet
size.
F
F
No
opinions
all
right
so
we'll
try
something
and
we're
gonna
make
it
fixed
and
and
if
we
don't
review
it,
then
we're
all
going
to
live
with
it.
How
about
time
I
there
are
tons
of
measurement
systems
that
do
something
that's
very
VoIP
like
so
I.
Don't
think
we
can
go
wrong
if
we
go
that
way,
but
good
thanks.
F
F
Jonathan
was
kind
of
flipped
out
by
the
size
of
the
document
and
how
big
the
reference,
though
each
entry
was,
but
but
it's
an
RFC
to
you
know,
so
it
has
the
function
of
of
being
an
RFC
for
these
things
as
well.
So
yeah
I
think
you
guys
agreed
to
work
on
some
new
metrics,
just
nod
or
say
no
she's,
coming
to
the
mic.
I
Yeah
yeah
so
yeah
in
the
last
meeting
we
discussed
this
and
I
think
yeah.
Some
people
may
think
that
you
know
have
a
registry
but
be
useful
yeah.
Some
people
may
still
think
it
doesn't
have
value
to
do
that.
So.
F
I
I
C
F
You
and
innocently
the
blue
sheets
up
here
for
folks
coming
in
late.
That
might
want
to
sign
it
all
right.
So
really
the
TCC
TCP
section
URIs
--tx,
is
where
we're
looking
for
more
review
and
that's
it
for
this
one.
So
I.
Thank
you
for
your
feedback
today
and
the
status
reports
and
stuff
like
that
cool.
Thank.
C
F
C
It
also
just
one
comment
on
that,
because
I'm
fairly
new
to
these
since
I
was
reading
through
these
documents,
they
are
quite
long
but
I
think
they're,
very
well-written,
they're,
very
good
introduction
to
the
space,
and
so
not
everyone
did
raise
their
hand
on
this
for
having
read
it.
So
I
encourage
you
to
go
through
it
and
do
get
feedback,
because
these
are
important.
J
F
F
So
here's
our
here's
our
quick
background.
We
got
this
going
really
in
draft
form.
Far
before
I
see
our
ATF
1999
and
we
worked
it
through
to
adoption
in
101
got
lots
of
good
comments
along
the
way
so
that
it's
jumping
quickly
to
the
good
verbal
feedback.
We
got
from
the
last
session
that
we
needed
to
qualify
what
can
be
discovered
here
for
the
kind
of
the
active
and
traceroute
methods,
and
we've
actually
done
that
three
times
in
the
text
now
so
there
can't
be
missed.
F
We've
got
the
new
methods
where
we
even
move
them
now
into
the
methodology
section
on
temporal
composition
and
this
routing
class
see
that
we've
been
talking
about
and
waving
our
hands
about.
Those
are
in
the
new
methodology
section
we're
kind
of
referring
to
addition,
existing
tools
there
for
sure
in
terms
of
parish,
trace
draft.
F
In
fact,
when
we
started
this
draft,
Bryan's
first
advice
to
us
was
start
with
parish,
trace,
wrap,
and
that
was
a
good
place
to
start
and
then
the
I/o
and
am
stuff
appeared
as
we
were
working
on
it,
and
that
was
another
place
to
start
in
the
hybrid
space,
so
that
we
also
had
comments
about-
let's
incorporate
MPLS
in
here
somehow.
So
we
generalized
our
definitions
to
do
that.
F
But
now
we've
actually
got
a
proposal
for
an
appendix
which
we
included
in
the
draft
text
from
rüdiger
guide
that
talks
about
MPLS
and
carlos
has
reviewed
it.
We've
got
some
few
things
to
fix
there
anything
else
on
this,
oh
yeah,
and
on
this
on
the
reporting
section,
we
basically
said:
look
there's
a
place.
This
RFC,
where
we
talked
about
storage
of
trace
routes.
That's
in
in
retrospect,
that's
not
going
to
cover
everything
we
talked
about
here
and
I.
Think
I'll
get
to
that
in
a
moment.
F
Actually
I'm
gonna,
because
I'm
in
control,
I'm
just
gonna,
go!
Do
that
so
you
know
trace
routes
are
composed
of
hops.
Each
route
is
represented
as
an
ordered
graph.
So
this
is
the
ordered
Oriya
graph
for
a
single
member
route
and
with
with
by
designating
the
host
with
this
hij
terminology,
we
can.
We
can
actually
make
a
kind
of
a
multi-dimensional
tupple
here.
F
F
Yes,
so
in
version
2,
as
I
mentioned,
we've
worked
hard
on
temporal
composition
for
route
metrics.
Why?
Because,
if
you've
ever
done
trace
routes
and
then
done
it
again
and
done
it
again,
a
lot
of
them,
the
results
are
highly
continuous
repeatable.
You
get
the
same
answer
many
times.
That's
the
way.
To
put
it
the
spot.
F
What
you
would
basically
do
is
is
kind
of
reduce
your
repetitive
tests,
so
the
cases
that
are
interesting
and
we're
kind
of
saying
that
that
really
the
load
balancing
places
are
the
ones
that
you
really
want
to
look
more
and
more
often,
but
we've
got
your
wrist
except
they
look.
If
you
see
anything
else,
change
in
a
spot
check
of
the
other
things
that
you
think
are
constant.
Now
you
got
to
rerun
the
whole
thing.
So
so
that's
what
we've
got
in
there
for
temporal
composition.
It's
not
a
predictor
of
the
future.
F
So
going
back
to
our
terminology,
each
member
root
of
a
root
ensemble
has
a
routing
Class,
C
and
I
should
say
the
I
should
say
the
thing
about
why
I'm,
using
this
word
routing
this
adjective
routing,
there's
there
used
to
be
Class
C
addresses,
and
it
came
up
in
discussion
that
people
flipped
out
that
we
thought
we
were
talking
about
Class
C
here
for
the
old
address
space.
You
know
the
class
classical
address
space,
we're
not
we're
not
it's
a
different.
It's
a
fully
different
class
seat
here.
F
So
so,
as
I
said,
each
member
of
the
route
has
a
Class
C,
so
there's
a
synergy
with
the
temporal
composition.
In
that
now
you
know
the
kinds
of
things
that
we're
interested
in
determining
about
the
load
balanced
portions
of
the
path,
temporal
composition
actually
helps
us
determine
those
and
and
avoid
testing
around
them.
So
we
think
it's
very
useful
to
know.
In
fact,
it
says
that
both
in
twenty
three
thirty
and
seventy
seven
ninety-nine-
and
then
your
line
is
how
useful
is
it
come
on
pay
attention?
Everybody,
let's
get
with
the
program
here.
F
F
It
becomes
important
when
you
make
a
midpoint
passive
offers
observation
of
a
stream
that
you
can
only
sort
of
observe
and
because
you
observe
that
stream
having
a
problem.
What
happens
next,
you
want
to
figure
out
where
the
problem
is
right.
Maybe
it's
got
a
round-trip
time
larger
than
you
might
expect.
Well,
where's,
the
congestion
half
or
the
delay
variation.
That's
causing
that
round-trip
time
to
expand.
F
I
think
this
is
a
great
idea,
but
what
can
we
do
this
well
at
the
hackfest
I
hacked
around
with
spoofing
the
source,
IP
address
and
I
figured
this
would
work
and
and
and-
and
it
actually
has
to
work
if
you
can
observe
both
paths.
Through
your
observation
point
and
and
that's
one
of
the
things
that
we
at
least
in
this
path,
that
you
should
be
able
to
do,
the
bad
news
about
spoofing
your
IP
address
or
somebody
else's
IP
address-
is
that
now
they're
going
to
get
unexpected?
F
You
know
a
destination,
not
reachable
from
certain
points
along
the
path,
probably
not
good,
not
good,
but
and
it,
but
it's
an
interesting
way
to
start
this.
You
know
to
sort
of
see:
okay.
Is
there
really
stuff
out
there
that
we
can
look
at
so
it
turns
out.
The
end-to-end
flow
conforms
to
a
specific
routing,
Class
C.
F
B
I
do
want
to
help
hack
on
some
tools
and
I'm
going
to
I.
Don't
think
he's
here
so
I
can
volunteer
on
his
behalf
on
a
volunteer.
Yari
yard
also
do
some
hacking
on
these
tools
cuz.
This
is
so
it's
interesting
to
me
that,
like
traceroute
is
kind
of
old
and
boring
and
dead
not
dead,
but
I
mean
so
if
you
look
at,
if
you
look
at
like
almost
all
different,
like
you
know,
most
of
the
trace
routes,
sort
of
in
open
measurement
now
are
done
on
right.
B
That
was
nodes
because
it's
a
huge
network
and
if
you
look
at
though
at
the
code
in
the
right
that
lets
note,
it's
actually
got
comments
from
Van
Jacobs
and
in
it.
So
it's
like
the
original
trace
right
yeah,
but
it's
old,
there's
other
stuff
that
you
can
do
in
this
space.
That's
really
interesting!
So
I'm
gonna
put
my
chair
hat
back
on
now.
A
lot
of
this
stuff
is
actually
new.
We're
developing
new
algorithms
for
traceroute
like
stuff
so
Rory's
interested
in
this,
which
is
why
I'm
volunteering
them
yeah.
B
The
document
and
and
look
at
his
sort
of
he's
doing
a
little
bit
of
hacking
around
on
stuff
like
this,
and
it
makes
me
wonder
about
the
status
this
document.
So
what
is
it
now?
B
F
Okay,
but
there's
I
mean
when
we're
talking
about
methodologies,
yeah
and
we're
talking
about
tools
to
back
up
the
methodology.
Great
and
and
I've,
like
I,
said:
I've
started
hacking
around
with
parish
traceroute
the
MDA.
You
version.
K
F
And
give
their
comments,
so
we've
got
some
to
do's,
some
of
which
we've
done.
Actually,
we
did
the
first
one
about
investigating
parish,
trace
routes,
v6
capability
it
we
have
to
have
some
other
stuff
to
do
there,
especially
adding
cautions
about
the
methods
we've
already
added
questions
about
the
metrics,
but
you
can,
but
you
can
probably
say
some
things
too,
that
are
useful
about
what
happens
when
you
use
good
measurements,
badly.
I
think
and
we
did
an
MPLS
appendix
so,
but
we've
got
more
to
do
with
that.
F
F
All
the
material
on
that
topic
is
in
Section
four
and
we're
still
looking
for
review
for
the
later
sections,
which
Ignacio
is
the
primary
primary
author
of
and
where
he's
used
the
round
trip
delay
to
do
analysis
to
find
things
like
congestion
points
and
stuff
like
that
determining
link
types
and
so
forth.
So
that's
all
to
do
I
think!
That's
it
yeah
yeah!
That's
it.
A
K
So
we'll
talk
about
the
the
main
data
graph
that
we
have
adopted
for
awhile
now
and
we're
talking
about
a
couple
of
individual
contributions
for
now
that
we're
hoping
to
get
adopted
very
soon
on
encapsulating
iom
data
into
the
expand,
GP
age
and
eve
GRE,
and
now
so
ipv6.
There's
a
new
draft
on
that
and
well.
The
obvious
question
is:
how
do
you
get
the
data
off
box?
K
We
have
an
export
draft
for
that
now,
so
that
the
D
capsulated
note
can
go
on
to
export
the
data,
we're
not
munching
around
with
the
data
we're
just
exporting
it,
which
is
why
it's
called
well
export
and
then
mostly
as
an
FYI.
The
SFC
working
group
between
last
meeting
and
today
has
adopted
a
draft
on
Io
I'm,
Ana's,
HN,
encapsulations
or
making
progress
there,
and
they
also
adopted
the
prove
of
transit
draft
so
to
actually
prove
that
a
packet
made
it
through
a
sequence
of
hops.
K
So
you
can
go
and
put
that
data,
because
well,
the
one
link
might
be
operated
by
the
guys
who
runs
OpenStack
and
harass
the
jean-yves
overlay
and
the
underlay
might
be
by
the
v6
underlay
operator.
So
we
try
to
go
and
clarify
that.
Hopefully,
that's
good
enough
for
everybody
by
now,
because,
while
after
proposing
that
language
also
publicly
we've
not
really
seen
any
further
comments.
Well,
we
also
clarified
and
that's
based
on
a
comment
of
Tom
Herbert.
K
Is
that
once
you
do
the
encapsulation,
you
should
closely
follow
the
recommendations
for
reuse
of
port
numbers
or
use
of
port
numbers
and
7605,
which
basically
says
well.
Udp
port
numbers
are
not
globally
unique.
If
you
haven't
well,
have
an
uncontrolled
environment
like
the
global
internet,
so
don't
touch
content
just
because
it
carries
a
genève
UDP
number.
The
content
might
not
be
genève.
So
don't
blindly
update
information
in
transit
nodes
because
you
believe
it
might
be
iom
data
carried
within
genève.
Don't
do
that
it's
there.
K
So
there
is
a
must,
not
statement
for
kind
of
global
Internet
deployment
in
the
various
encapsulation
drafts
we're
going
to
discuss
in
a
second,
then
we
also
started
to
do
the
laundry
that
we
have,
probably
that
we
promised
in
the
last
meeting
largely
around
the
security
section
and
many
thanks
to
tell
to
go
and
do
that
work.
So
there
is
now
a
variety
of
potential
attacks
that
you
can
go
do
if
you
have
IOM
present.
K
So
obviously,
you
can
fake
the
IOM
data
to
make
the
consumer
of
iam
data
believe
that
the
information
well
leads
you
to
failures
that
don't
exist
or
you
can
kind
of
capture
failures
or
hide
failures
in
a
positive
and
a
negative
way.
You
can
well
expose
Network
information
to
people
who
shouldn't
know
that.
Well,
the
network
information
is
there,
so
you
cannot
get
an
understanding
of
the
network
topology,
despite
the
fact
that
you
don't
want
to
go
and
expose
that
you
can
use
it
to
consume
resources
because
we
add
data
to
the
package.
K
So
we
increase
the
packet
lengths,
that's
potentially
bad.
If
it's
uncontrolled,
we
consume
resources,
that's
potentially
bad
if
it's
uncontrolled,
so
all
of
these
things
are
potentially
bad
if
they're,
uncontrolled,
the
good
thing
is,
I
is
defined
for
a
constrained
domain
for
an
operational
domain
as
opposed
to
it
runs
on
the
global
Internet.
So
all
of
these
things
need
to
be
taken
into
consideration,
so
we've
basically
listed
all
the
bad
things
that
can
happen.
K
One
thing
that
came
up
also
with
the
layering
discussion
and
I
touched
on.
That,
briefly,
is
if
you
have
a
genève
layer
and
the
other
v6
layer,
underneath
we
always
refer
back
in
the
document.
Like
all
the
information
that
we
have
an
old
ID,
a
interface
ID,
that's
all
its
own
namespace.
It
all
relates
to
a
domain,
an
operational
environment.
K
K
So
we
refer
to
the
namespace,
but
we
don't
carry
the
namespace
identifier
anywhere
because
yeah,
if
you
have
that
namespace
identifier,
if
you
just
export
data,
then
well,
the
exporting
information
would
not
only
have
the
individual
node
ID
ie
13
or
interface
ID
25,
but
you
also
have
the
contacts
that
tells
you.
Ok.
This
is
the
namespace
that
this
information
is
relevant
in.
K
So
what
we
want
to
go
do
and
that's
a
proposal
and
I
want
to
go
and
hear
or
feedback
from
that
is
add
a
namespace
identifier
to
well
top-level
iom
data
so
that
you
can
say
well.
This
is
the
namespace
that
I
refer
to
and
all
the
additional
data
that
I
have
is
to
be
understood
for
within
this
particular
namespace
and
well.
If
we
had
this,
it
also
helps
us
with
some
ambiguity
and
some
additional
requirements
that
we've
seen
floating
around.
K
So
there
is
a
couple
of
graphs
that
we've
even
seen
being
read
out
here
about
limitations
that
you
see
with
the
current
IO,
am
data
field
definition,
so
people
dream
off
additional
fields
for
specific
use
cases
you
can
say
well,
there
is
a
couple
of
fields
in
IO
am
today
that
are
almost
free
format.
We
had
app
data,
for
instance,
app
data
is
32
or
64
bits
free
format.
K
Well,
so
it
depends
on
the
namespace
how
you
interpret
that
now.
If
we
have
a
clear
linkage
to
the
namespace
well,
then
somebody
can
say
well,
this
app
data
means
GPS
geolocation
for
my
Pacific
use
case,
or
it
refers
to
a
template
ID
like
what
the
me
IFA
guys
want,
so
that
an
individual
note
can
say
well,
I
updated
just
this
portion
of
the
overall
data
in
the
IOM
feel
as
opposed
to
all
the
things
or
I
can
say.
Well,
this
app
data
is
to
be
interpreted
as
well
into
packet
gap.
K
So
if
we
have
a
clarification
of
this
namespace
I
think
we
make
it
far
more
clear
that
certain
things
are
really
contained
within
a
context,
and
so
I'd
love
to
hear
more
comments
on
that
later
on,
as
we
go
through
the
discussion,
so
that
I
think
is
the
updates
or
the
considered
updates
for
the
data
graph.
Then
we
have
a
bunch
of
encapsulation
graphs
and
we
briefly
discussed
them
in
the
last
meeting
and
they
haven't
really
changed
much
above
and
beyond
the
discussions
on
layering.
So
we
added
a
couple
of
additional
notes
there.
K
So
we
have
an
a
capsulation
graph,
4vx
LAN,
GP
e,
and
the
format
is
almost
always
the
same.
So
we
add
the
IOM
data
fields,
just
after
the
typical
shim
header
that
you
would
use
for
one
or
the
other
encapsulations
4vx
LAN
GP.
We
use
the
next
protocol
header
and
then
you
typically
have
the
typical
structure
and
I
I.
K
So
in
genève
we're
just
using
the
available
POV
format
capability
so
that
you're
using
in
geni,
if
you're,
using
an
option
class
for
iom
and
then
the
type
is
either
one
of
the
four
types
that
we
have
in
IOM:
ie
pre-allocate
a
trace,
incremental
choice
and
two
and
and
to
end
or
or
approve
transit.
And
so
then
again,
you
just
slot
in
all
the
data
that
you
would
naturally
have
in
an
IOM
context
that
we've
done
with
the
reformatting
earlier
on.
In
the
last
revision,
the
the
IOM
space
is
almost
self-contained
by
now.
K
The
same
goes
for
GRE
and
with
GRE
I.
Think
that's
the
de-facto
and
the
easiest
way
to
go
and
do
I
am
over
r-e,
for
because
all
you
would
need
is
spend
four
bytes
to
go
and
slot
in
the
GRE
header
there
and
then
well.
You
can
go
on
again
dump
in
the
before
and
from
the
IOM
information.
And
what
is
new
and
just
recently
published.
Is
we
have
a
draft
for
ipv6?
By
now?
K
So
that
means
for
a
e2e
you're
using
destination
options
because
they're
running
into
end
and
for
everything
else
like
incremental
and
pre-allocated
trace,
as
well
as
P
ot
you're,
using
hop-by-hop
options
and
well
that's
basically
it
so.
We
just
need
coat
points
for
those
things
from
six-man
and
some
same
just
it's
not
gonna,
be
a
just,
probably
because.
K
A
K
L
So
everything
we
just
talked
about
was
pretty
much
in
the
data
plane.
How
do
we
carry
all
this
fun?
Io
am
data
from
one
node
to
another
node,
you
know.
Eventually,
we
want
to
get
it
off
the
network,
so
this
figure
is
just
explaining
what
we
think
raw
export
is
and
why
it's
a
problem
that
we're
trying
to
address
so
the
green
on
the
bottom
packets
are
going
through
a
bunch
of
nodes.
Some
node
decides
okay,
I'm
I'm
done
with
this
stuff,
maybe
up
the
at
the
edge
of
the
domain.
L
Take
out
all
that
iom
data
that
we
have
now.
What
do
I
do
with
it,
and
we're
saying
that
in
order
to
really
optimize
the
data
plane,
nodes
and
those
IOM
nodes
at
the
bottom,
we
want
to
make
the
export
as
simple
as
possible.
So
raw
export
means
we're
basically
taking
chunks
of
packets
and
I
OEM
headers,
stuffing
them
in
a
simple
format
and
getting
them
off
to
some
data
processing
system.
L
From
that
point
on
whether
the
data
processing
system
does
analytics
itself
or
it
then
aggregates
in
or
interprets
or
processes
info
and
then
punts
that
up
to
an
analytic
system,
whatever
you
want
right,
there's
there's
many
options
for
how
to
put
the
pieces
together.
So
the
goal
here
was
as
simple
as
possible
for
the
data
plane
nodes,
the
exporting
node
does
not
interpret
aggregate
or
reformat
IOM
data
before
it's
exported
we're
offloading
all
that
to
the
data
processing
system.
L
So
when
we
looked
at
what
can
we
do
for
the
format
of
getting
all
this
stuff
off
of
the
system?
Ip
fix
is
commonly
used
for
flow
related
things
today,
it's
more
statistics,
but
you
can
also
do
packet,
sampling
and
various
things
like
that,
and
so
we
took
a
course
look
at
IP
fix
and
figure
that
it
has
a
lot
of
the
things
that
we
might
need
and
there's
just
a
few
extra
information
elements
that
we
could
add
and
then
we
would
have
a
full
format
for
doing
all
this
raw
export.
L
One
thing
I
note
just
the
bullet
on
the
bottom:
that's
not
actually
in
text
it's
in
the
draft,
but
depending
on
your
implementation,
if
you're
highly
optimizing
the
data
plane,
often
in
Hardware
you're,
probably
not
going
to
support
the
full
flexibility
of
hey,
you
know
tell
me
what
template
you
want
and
so
on.
You
may
find
that
implementation
does
a
small
number
of
fixed
formats
that
it
can
highly
optimized
and
that
it
will
adjust
alignment
constraints
and
such
so
that
it
can
process
that
at
billions
of
packets
per
second.
L
So
among
the
IP
fix
information
elements
that
already
exists
that
we
can
just
use.
We
have
an
IP
header
packet
section
and
note
that
an
IP
header
packet
section
it
starts
with
the
IP
header,
but
depending
on
the
length
of
that
packet
section,
it
can
keep
going
into
various
headers
after
that,
including
possibly
IO
a.m.
that
is
embedded
within.
You
can
do
the
same
thing
with
Ethernet
using
datalink
frame
section
and
there's
other
things.
There's
a
section
exported
octets
and
forwarding
status.
If
you're
reporting
drop
packets,
then
you'll
find
forwarding
status
of
great
interests.
L
Basically,
the
first
three
bits
are
telling
you
did
I
forward
the
packet
that
I
drop.
It
did
something
else
happen
and
then
the
next
five
bits
like
if
you're
dropping
you
can
give
you
can
specify
why
the
packet
was
dropped.
So
what
we
wanted
to
add,
we
wanted
to
add
a
few
information
elements.
So
one
is
report
Flags.
Maybe
that
should
be
renamed,
but
it
was
just
trying
to
give
a
kind
of
the
broad
categories
on
why
I
want
to
report
this
particular
info,
so
did
I
drop.
A
packet
did
I
see
congested
queues.
L
What's
in
the
data
draft,
if
it's
one
of
the
others-
and
you
start
with
the
shim
header
that
Frank
just
explained
in
the
various
end
caps,
you
go
continue
on
into
the
I
data
and
possibly
continue
after
that,
and
when
you
have
four
data
chunks
corresponding
to
the
four
type
shion's
of
IOM
that
we
have
today
the
pre-allocated
incremental
and
an
edge
edge
and
proof
of
transit.
L
So
once
we
have
all
those
it's
up
to
you,
how
you
want
to
put
them
together,
but
we
thought
it
was
useful
to
list
some
examples
of
how
you
might
layout
these
information
elements
in
a
template.
So
of
course
it's
not
mandated
to
use
exactly
this
template.
The
simplest
way
to
go
would
be
to
only
have
we
have
most
of
the
info,
just
in
an
IP
header
packet
section,
and
then
it's
long
enough
that
it
continues
on
into
the
IOM
data.
L
L
What
the
flow
is
that
you
care
about
and
then
pull
out
some
of
the
IOM
and
phone
put
it
on
top
and
you
can
do
various
combinations
of
which
things
are
fixed
length
which
are
variable
length
and
so
on,
so
that
pretty
much
covers.
What's
in
the
draft,
I
did
have
some
questions
and
I
sent
a
more
detailed
message
to
the
mailing
list,
but
these
were
more
around.
There
are
certain
fields.
L
If
you
look
at
the
definitions
of
information,
almost
like
IP
header
packet
section,
they
had
a
certain
restriction,
so
you
have
to
put
things
like
section
exported
architects.
In
certain
cases
you
may
or
may
not
be
allowed
to
put
padding
and
it
just
seemed
like
you
could
cut
out
a
few
bytes
if
we
updated
those,
but
that's
kind
of
an
open
question.
Can
we
change
the
definitions
of
fields?
L
B
So
I'm
just
gonna
go
ahead
and
get
in
line
because
me
so
I,
Brian
Trammell,
not
as
chair.
Actually
as
one
of
the
designated
experts
for
the
I
Anna
IP
fix
information
elements.
The
short
answer
to
your
question
is:
read:
RFC
7013
and
dry
again,
the
most
of
the
changes
that
you're
proposing
here
are
non
interoperable
with
existing
implementations.
As
far
as
I
see,
right,
like
you,
actually
likes
relaxing
restrictions
that
might
break
collectors
that
are
relying
on
those
restrictions.
However,
this
is
not
a
very
scarce
number
space.
B
A
win
it
might
make
sense
to
do
IOM
specific
packet
section
by
ease
right
so
design.
What
you
want.
The
downside
is
that
you
know
existing
collectors
will
have
to
then
get
the
update
but
you're
talking
about
changing
a
definition
that
might
break
those
existing
collectors
anyway.
So
it's
the
same
amount
of
time
for
the
existing
collectors.
Update.
Okay
through
is
data
line
frame
section
not
deprecated.
I
thought
we
deprecated
that
one
I,
let
me
go
actually
have
a
look
at
the
at
the
I
was.
B
Yeah,
it
would
have
been,
it
would
have
been
deprecated
along.
Oh
no!
No!
No!
No!
No!
Sorry!
No,
that
one's
good!
No
sorry
that
I
thought
it
was
a
section
offset.
No,
that's
fine,
yeah
yeah!
So
short
answer,
read:
1713!
Okay,
slightly
longer
answer:
reads
already:
thirteen,
and
and
really
consider
whether
you
want
to
reuse
these
II's,
because
I
don't
think
you
can
tweak
them
to
be
exactly
what
you
want.
I
think
you
can
basically
create
new
ones
by
reference.
B
B
B
So
it
was
kind
of
optimized
for
that,
but
not
really
so.
Yeah
I
definitely
consider
whether
or
not
these
are
really
what
you
want,
because
the
what
what
you're
trying
to
do
to
take
the
IOM
information
out
is
semantically
different
enough
from
just
random
parts,
the
IP
header
that
it
might
make
sense
to
define
to
do
ie
anyway.
It's
not
that
big
a
deal
so
I
guess.
L
B
H
Million
elkins
ever
two
different
kinds
of
comments,
one
functionally
and
I'm
from
a
higher
level.
This
is
extremely
cool,
very
good
data
in
a
you
know
that
if
we
can
kind
of
like
I,
if
I'm
understanding
it
correctly,
you're
gonna
grab
all
this
stuff
passively
and
then
search
and
just
take
the
IOM
data,
which
is
very
cool
yeah.
L
I
mean
the
thought
here
was
that
you're
getting
iom
data
at
your
edge
nodes,
whether
they
do
any
stateful
processing
to
figure
out
which
things
they
want
to
export,
which
things
they
just
want
to
don't
want
to
export
because
they
think
it's
redundant,
that's
always
an
option.
Usually
you
wouldn't
do
the
entire
packet.
H
Yeah
we
can
talk
offline,
I,
guess
what
I'm
saying
is
on
a
lot
of
enterprise
networks.
There
are
huge
packet
capture
devices
already
that
are
capturing
the
entire
packet,
including
the
I/o
I,
mean,
of
course,
I
mean
the
IOM
data
comes
with
it.
It's
just.
What's
that
huge
packet
brokers?
You
know
having
said
that,
having
said
that,
this
is
very
cool,
because
you
know
you're
not
having
to
filter
and
stuff,
and
it's
just
the
data
you
want.
Is
that
kind
of
your
point.
Yeah.
B
Know
we
are
running
up
on
our
time
here,
so
let's
cut
the
line
and
make
this
quick
yeah.
So
this
is
I'm
relaying
jakob
Stein
who,
as
you
know,
job
or
who
is
basically
raising
the
point
that
any
time
you're
taking
bits
of
packages
they're
flowing
by
and
exporting
them
that
may
raise
serious
security
issues.
I'll
also
note
that
the
on
that
point
when
piece
and
went
through
the
whole
process,
there
were
a
lot
of
2804
concerns.
K
L
Okay,
so
so
next
step,
so
the
I
data
draft
has
been
fairly
stable
and
there's
a
question
whether
we're
ready
for
a
working
group
last
call.
B
M
B
Yeah
so
yeah
crank
crank
the
revision
and
ping
bitshares,
maybe
hard
I'm
going
on
vacation
for
this
for
a
couple
weeks.
But
Tommy
knows
how
to
yeah
push
the
buttons
so.
N
Just
not
on
the
data,
but
on
the
encapsulations,
of
course
one
is
genève.
Has
the
ethertype
field
for
next
protocol
just
like
GRE
does,
and
it
would
be
nice
if
there's
a
rationale
for
them
being
different
to
have
that
articulated,
because
otherwise
you're
just
adding
complexity,
I,
don't
understand
yet
why
I'm
sure
there's
a
reason
if,
but
please
take
a
look
at
that,
and
also
the
packet
format
for
the
GRE
version
doesn't
reflect
2890,
which
updated
the
base
G
re
SPECT.
So
it
should
really
do
that.
Particularly
people
are
doing.
You
know
implementation.
N
So
minor
points
on
the
namespace
question:
how
I
I
like
the
idea
of
namespace
it
also
lets
you
do
different
administrative
domains.
You
can
have
the
ability
to
say
strip
out
information
when
it's
crossing
a
point,
so
I
think
it
has
a
lot
of
power.
It
also
lets
you
handle,
say
overly
under
play.
Situations
I
think
it's
got
flexibility
beyond
I'm,
not
sure
you
thought
of
it
beyond.
N
L
B
Yeah,
so
the
idea
was
to
bring
them
here,
because
there
was
not
a
whole
lot
of
interest,
but
to
have
all
of
the
discussion
echoed
over
to
those
lists
and
to
have
the
working
group
last
call
also
split
out
into
the
to
the
appropriate
working
groups
of
encapsulations.
It's
basically
is
the
encapsulation
here,
but
the
point
of
the
review
here
is
to
make
sure
that
it
all
works
well
with
IOM
and
the
point.
The
review
there
is
to
make
sure
it
isn't
break
the
end
caps,
where.
K
So
in
Geneva
we
add
that
discussion
GRE
not
so
far
as
far
as
I
know
our
v6-
definitely
not
because
that's
brand
new,
but
again
so
that
the
problem
that
we
face
and
we
we
started
the
other
way
around
right.
Nsh
was
okay,
the
genève
people
in
nvo
3
I'm,
not
saying
that
they
couldn't
care
less,
but
there
was
a
load
of
not
a
load
of
energy.
There.
We
talked
to
the
list
guys
on
the
Excel
and
GP
and
the
lights.
K
The
problem
is
that
you
typically
start
from
scratch
and
explain
all
the
contacts
and
then
you
a
load
of
questions
on
IOM
as
opposed
to
we
need
a
coke
point
guys.
All
we
need
is
a
coke
point
from
you
guys,
and
so
we
thought
that
that's
mature
it
here
where
people
have
the
context
and
if
people
in
I
ppm
think
this
is
okay,
then
we
go
and
well
discuss
the
protocol
related
aspects,
but
not
the
I
related
aspects
with
the
respect
of
Winkle
I,
think
that
was
the
rationale
because
sodding
the
other
way
around.
H
K
B
C
N
As
frank,
it
said
point
a
point
of
and
they're
moving
us
here,
so
you
get
the
focus.
I
don't
think
that,
due
to
scheduling
etcetera,
that
there's
been
a
lot
of
looking
at
the
impact
of
some
of
the
Appalachians
but
they're
like
the
impact
of
it,
it
sort
of
be
nice
to
at
least
get
routing
directorates,
or
you
know
RT
gwg
or
this
kind
of
thing,
I,
don't
think.
There's
any
concern,
I,
don't
think.
There's
a
territory
argument
about!
N
The
excellent
GP
is
interesting
because,
as
far
as
I
know,
the
excellent
GP
isn't
standardized
or
being
standardized,
there's,
not
an
RFC
for
it.
It's
explicitly
their
will.
At
some
point,
I
mean
assuming
that
nvo
3
does
what
we
agreed
on
a
while
ago.
Eventually
there
will
eventually
be
an
informational
document
saying
the
excellent
GP.
It
was
a
choice
that
was
not
selected
to
be
standardized,
but
I
feel
like
it's
different.
N
N
M
So
I
think
what
we
need
to
do
is
I.
Have
a
look.
I
need
to
have
a
look
at
all
these
documents,
because
I
didn't,
but
for
for
the
I
think
it's
actually
just
the
decision
of
the
other
working
group
to
redirect
this
work
to
this
working
group.
It's
not
that
we
can
decide
it
over
here,
so
it
probably
makes
sense
to
coordinate
between
the
chairs
and
the
ATS.
What
to
do
but
I
mean
both
is
possible
right.
M
You
can
like
the
question
is
where
to
actually
adopt
it
and
right
around
the
formal
process
but
like
having
working
a
joint
working
group.
Last
call
and
discussion
can
also
have
them
in
those
groups
where,
wherever
there's
more
interest
like,
even
if
you
get
it
adopting
the
other
working
group,
you
can
still
presented
here
and
discuss
it
here.
K
B
M
M
B
We
already
had
that
with
with
video,
so.
M
B
B
Hello,
hello,
yeah,
so
just
yeah
we
didn't
get
the
request
in
for
remote
presenting.
So
if
you
can
just
tell
us
next
slide,
we
can
we
can
do
that
for
you,
okay,.
O
This
is
an
update
of
the
multi-point
alternate
marking
draft,
so
we
changed
to
works
on
things
after
the
long
term
meeting.
So
the
disease's
light
gives
you
a
recap
about
the
document
change.
We
moved
from
0
to
2
0
to
version
and
then
to
0
4.
We
got
uncomment
after
Londo
meeting
I
want
to
thank
algae,
Intel
Rachel
for
the
feedbacks,
and
there
is
a
list
of
the
main
points
that
have
been
modified
in
during
the
last
batch.
O
So
in
summary,
we
detail
better
the
description
of
multi
point
approach,
the
example
of
the
cluster,
the
algorithm
for
cluster
partition.
We
have
more
details
from
delay.
Particular
there
are
a
two
way
of
my
measurement
one
that
is,
there
is
multi-point
cut
very
this
approach
and
the
other
one
that
is
based
on
single
path
bases,
and
then
we
add
the
new
section
on
timing
aspect
that
extend
the
corresponding
section
of
air
FCAT.
321
I
want
to
thank
thanks,
Alfred
these
quit
bullyin
put
so
next.
O
So
next
yeah.
This
is
the
basic
base
idea,
that
is
the
cluster
pack
across
general.
If
we
have
a
multi-point
to
multi-point
network,
we
can
find
the
partition
of
these
networks
with
sub
networks,
and
the
clusters
are
the
smallest
sub
networks,
where
the
packet
rows
property
is
valid,
so
pocket
rows.
Property
means
that
the
number
of
packet
incoming
are
equal
to
the
number
of
out
to
in
path.
O
So,
in
general,
the
network
clustering
approach
is
very
useful
because
without
a
natural
clustering
we
can
monitor
only
the
entire
network
or
single
probe.
Otherwise,
with
network
clustering,
we
can
use
the
natural
caustic
or
Ethan
to
perform
the
needed
degree
of
detail.
So
we
can
make
clusters
zooming
on
the
network,
like
a
figure.
Next.
H
O
Some
comments
received
so
in
general
there
is
a
two
step
powdery.
The
first
step
is
to
group
all
the
links
where
there
is
the
same
starting
node
and
the
second
step
enjoying
the
cupid
links
at
least
one
ending
node
in
comma.
So
in
the
figure
we
can
see
that
all
the
nodes
we
put
the
first
step,
we
can
have
five
groups
and
with
the
second
step,
the
group
number
two
and
group
number
three
join
you
to
have
the
cluster,
so
in
general
that
in
that
net
multi-point
partner,
fine.
O
This
is
just
to
give
you
an
example
on
how
to
apply
these
to
step
out
during
next
life.
This
is
a
very
new
section
that
has
been
used
after
being
from
owl
in
general.
Of
course,
if
we
have
a
lot
of
source
nodes
that
mark
our
topic,
we
can
have
an
alignment
in
marking
intervals
of
each
marking
the
nodes,
because
in
general,
if
we
have
an
endpoint
network,
we
can
have
multiple
marking
nodes.
So
we
have
to
define
the
measurement
intervals,
and
that
case
we
have
to
consider
the
misalignment
in
the
timing
that
consideration.
O
The
first
way
to
perform
the
delay
measurement
is
a
multi-point
at
basis
measurement.
So
in
that
case,
the
delay
value
is
representative
of
the
entire
multi-point
path.
This
means
that
example,
that
we
want
to
apply
the
min
delay.
The
min
delay
is
difference
between
the
mean
time
stamps
of
the
set
of
output
nodes
and
the
set
of
input
nodes.
So
the
mean
clay
is
representative
of
boot
appointment.
Otherwise
we
could.
We
can
apply
the
same
measurement
on
single
packet
basis
measurement,
so
it
became
a
point-to-point
measure,
but
it
is
based
on
the
multi-point
path.
O
That
means
that
multiple
paths
just
use
Erdem
that
aid
to
couple
the
the
samples
between
input,
nodes
and
output
nodes,
single
marking
based
on
Priscilla's
packet,
cannot
be
used
because
the
batch
boundary
not
valid
in
multi-point
approach.
Double
marking
were
neglected,
marking
words
the
little
limitation
at
the
time,
detailer
laughs,
but
there
is
no
time
to
explain
this.
O
The
combination
is
to
RFC
give
an
easy
way
to
perform
the
delay,
because
we
can
use
the
basic
cache
so
methodologies,
where
they
alternate,
marking
split
the
continuous
flow
into
marking
batches
and
these
anchors,
the
hashing
samples
for
for
each
video.
This
is
simple
way
to
apply
the
aschen
methodology,
but
to
utilize
a
cache
in
a
number
of
simple
depends
on
packet
pre.
O
Otherwise,
we
should
use
a
sort
of
dynamic
cache
that
can
be
applied
with
an
iterative
value
it
where
it
is
possible
to
use
a
number
of
sample
is
almost
constant
or
each
marking
period
and
detail
of
diesel.
Karate
vulgar
and
statistically
converge
for
each
mark
Imperial
to
a
value
to
a
constant
value
of
over
a
simple,
also,
in
that
case
the
iterative
algorithm
in
detail
within
the
draw.
There
is
no
time
to
describe
so
next.
H
H
O
Multi-Point
or
ultimate
marking
allows
a
flexible
dynamic.
That's
the
end
of
the
performance
management,
because,
if
you
have
our
the
controller
can
configure
a
row
row
performance
management
so
without
examining,
and
only
if
we
we
experience
some
pocket
release
like
you
can
go
in
deep
and
we
can
detail
about
filtering
criteria
and
cross
the
partition
in
order
to
perform
a
detailed
analysis.
So
this
is
the
what
we
means
for
dynamic
performance
measurement,
and
there
are
two
ways:
one
way
is
the
filtering
and
the
other
way
is
in
the
faster
cause.
O
Cluster
can
be
also
combined
between
each
other.
So
next,
okay,
just
the
last
slide
that
so
this
methodology
are
the
new
points
of
you
to
do
ultimate,
marking
method.
It
adds
flexibility,
dynamic
performance
measurement,
Nichols
or
use
other
techniques
that
are
actual
technique,
for
example,
that
is
not
so
used,
but
the
coupling
between
ashing
technique
and
multi-point
ultimate
marking
is
varied,
and
that,
since
that
we
had
a
lot
of
comments,
we
want
to
ask
for
working
group
adoption.
B
So
hi
thanks,
Giuseppe
I,
don't
think
we're
gonna
one
calls
for
adoption
this
time
around
we're
still
trying
to
drain
the
queue,
hopefully
between
now
and
Bangkok,
at
the
latest,
at
Bangkok
cuz.
She
a
show
of
hands
here.
Who's
read
these
records
draft
okay,
so
we
have
a
few
I
think
it
yeah.
It
seems
it
seems
perfectly
reasonable.
It's
doing
adoption
call,
but
we
do
try
to
keep
the
sort
of
the
docket
of
the
working
group
a
little
bit.
B
Slimmer,
we're
gonna
get
so
there's
one
working
group
last
call
that's
done
we'll
see
if
we
can
get
another
one
out,
I
think
stamp
is
getting
pretty
close
to
done
as
well,
and
once
we
get
those
out,
this
will
be
next
up.
Okay,
okay,
cool
we'll
keep
working
on
it
right,
like
I,
mean
there's
the
there's.
You
know
if
running
it
through
and
then
adopting,
and
then
you
know
we
can
also
adopt
and
do
a
relatively
quick
jump
to
wgl
see
as
well
right.
B
B
H
Okay,
this
one
hour,
this
draft
combines
the
marking,
method
and
also
PDM,
and
the
reason
we
did
it
is
because
PDM
is
in
two
end
and
marketing
has
airing
marking
needed
to
be
v6,
v6
stuff
and
we
wanted
pass
stuff.
So
these
are
the
PDM
fields.
That's
that's
RFC
8250.
I
believe
we
already
have
it
and
that's
the
end-to-end
information
and
you
know
you
can
get
the
round-trip
delay.
It's
all
in
the
destination
options
header,
but
what
we
don't
have
is
the
one-way
path,
delay
and
or
the
middle
box
stuff.
H
So
marking
has
some
real
interesting
things
about.
It
is
that
you
can
batch
packets
together
and
look
at
the
timestamps
and
stuff,
and
you
and
you
can
do
average
packet
delay
so
with
giuseppe.
We
were
talking
about
it
and
it
looked
like
this
might
be
a
way
to
meet
both
our
ends.
So
now,
I'm
gonna
stop
talking
a
little
bit
slower
here.
This
is
actually
the
only
slide,
that's
worth
looking
at
sorry
Janell,
but
this
is
the
only
slide.
That's
that's
of
interest.
What
this
has.
H
It
say
this
is
a
quite
a
deal
of
interest
to
them
as
well
is
like
as
it's
going
along
the
hop.
Where
is
the
packet
getting
messed
up?
Where
is
it
stopping
who's
who's
doing
what
with
it
who's
the
last
guy
that
touched
it?
That
was
of
interest,
and
so
that's.
What's
in
here,
we've
been
through
a
bunch
about
what
exactly
middle
box
identify
your
means,
we're
very
open
to
yeah
very
open
to
comments
on
that.
H
H
Address
but
I
don't
know
how
you
get
that
in
a
16-bit
field,
but
anyway,
we're
will
open
real
open
to
that
Brian
yeah
keep
going
yeah,
we
got
it,
you
know
they
say
this
is
the
only
one
of
interest
we
room
for
five
middle
boxes.
It'll
wrap
around
the
last
EM
field
is
the
last
person,
the
last
middle
box
who
identified
so
you
can
quickly
hop
down
to
updating
the
one
that
you've
got
that
we're
putting
this.
This
is
not
gonna.
The
packet
length
is
not
going
to
change.
The
source
is
going
to
put
out.
H
The
whole
thing
is
a
no
field
and
then
every
metal
box
along
the
way.
It's
going
to
say
this
is
who
I
am.
This
is
when
I
got
this
packet.
This
is
real
data.
This
is
the
real
packets.
This
is
not
a
an
active
test.
This
is
hybrid
measurement.
This
is
not
a
test
packet
I
mean
you
could
devise
a
test
with
it,
but
this
is
real,
live
actual
data,
and
so
he'll
just
say
this
is
when
I
got
it
and
and
of
course
you
would
not
want
to
do
this
all
the
time.
H
B
H
Is
not
single
administrative
domain
bTW
PDM
is
the
is,
is
internet
wide?
This
could
be
single
administrative
domain
I.
Would
it
be,
as
I
said,
I've
already
had
people
from
ISPs
say
this
would
be
very
cool
and
they
would
like
to
have
it
so
I
would
I
would
hope.
This
is
not
single
administrative
domain.
So.
B
B
Iii
almost
kinda
want
to
just
say,
read
the
plus
drafts
and
do
that
instead,
but
you're
gonna
need
yeah,
you
need
to
be
bigger.
You're
gonna
need
some
way
to
handle
integrity,
protection
and
you're.
Gonna
need
some
way
to
like
if
it
costs
the
administrative
domain,
where
you're
not
really
sure
who's
gonna
be
messing
with
the
stuff,
then
you
need
to
go
a
little
bit
more
complicated
than
what's
here.
I
have
another
question:
just
really
clarifying
I'm
gonna
try
and
make
this
real
quick.
So
we
can
get
to
Ignacio.
B
H
H
B
H
K
H
Yeah
yeah
yeah,
no,
as
I
said
we're
putting
this
out,
and
you
know
it's
getting
feelers
getting
feelers
from
everybody.
It
cuz
I.
Think
our
first
thing
is:
do
people
think
this
is
useful
and
then
and
then
we'll
continue
work
on
it?
That's
that's
kind
of
where
we
are
right
now.
If
people
think
it's
useful,
we
think
it
is,
but
we
want
to
do
long.
You
know
validity
testing
yeah.
H
C
B
Please
say
comments
the
list
so
Ignacio
we
got
you
five
minutes,
go
ahead
and
pull
it
up.
So
here
he
comes
good
yeah.
Q
If
we,
if
you
could
put
because
this
is
okay,
thank
you
well
just
this
is
a
proposition
to
get
synchronizing
internet
clocks
in
secure
way
and
who
is
needing
which
application
measurements.
Our
our
working
group
localization
cryptic
coins
games,
and
there
are
surely
another
applications
next
please
well.
This
is
this
chart
represent
more
or
less
the
road
that
you
can
get
in
different
protocols,
and
the
idea
is
our
protocol
is
the
green
one
SiC
you
can
get
frequency
synchronization
in
some
kind
of
secure
way.
Next
slide,
please!
H
Q
Other
one
and
you
can
clone
you,
you
can
get
the
code
of
the
the
idea
in
that
link
next
slide.
Please,
and
this
is
the
updates
that
we
we
got
from
the
first
to
the
second
version.
New
router
is
added
through
the
gap
we
stress
at
the
frequency
objetive
and
we
compare
with
the
IQ
related
standards
and
we
provide
detailed
information
about
security
issues
and
also
an
example
of
some
kind
of
limitation
in
with
ntp
in
some
rationals
of
the
internet.
Well,
this
is
all
I,
don't
know
if
I
can
get
the
question
or
not.
B
D
This
is
the
young
leader
model
for
Iran.
Last
time
we
presented
this
in
London
and
we
have
several
comments
from
Greg
and
the
Rashaad,
and
we
make
the
for
Iowa
option
profiles
as
features
for
flexibility.
We
modified
the
ACL
live
reef
to
align,
waster
is
their
latest
ACL
model
and
we
add
a
demean
configure
container
for
a
clear
hierarchy.
D
D
B
All
right,
thank
you
very
much,
IOM
folks
that
sound
like
a
plan.
Okay,
so.
K
B
Me
ask
you
another
question:
would
you
be
scared
if
the
chair
is
imposed
an
experiment
on
your
yang
model
by
Fiat
we're
looking
into
using
github
for
Frank
model
development,
as
opposed
to
the
document
stream
for
inside
the
working
group,
I've
been
asked
by
some
of
the
net
rod
people
if
we
can
try
that
out,
I
have
no
idea
what
it
is.
I'm
gonna
find
out
what
the
actual
details
are
on
Friday.
B
So,
if
that's,
if
that's
not
scary
to
you,
then
we
would
try
that
with
this,
when
we
yeah
right,
yeah,
okay,
so
well,
yeah
well
we'll
try
and
bring
this
in
with
the
other
gigantic
pile
of
IOM
documents
sometime
between
now
and
we're
gonna
count,
but
we're
gonna
count
all
the
end.
Caps
is
one
document
against
the
document
cap,
but
because
otherwise
we'd
never
be
done
right.
We
have
the
phrase
condition
on
the
number
of
documents
that
work
so
yeah.
So
we're
not
gonna.
B
Do
the
call
for
adoption
now
we'll
do
it
with
the
rest
of
those
we're
gonna
look
into
how
we
would
manage
the
TN
model
with
this
new
process
that
I
don't
know
anything
about.
Other
than
github
and
from
there
cool
all
right,
thank
you
very
much
that
was
IP
p.m.
we
will
well
I.
Won't
Tommy,
we'll
see
you
all
in
Bangkok.