►
From YouTube: IETF110-DETNET-20210309-1430
Description
DETNET meeting session at IETF110
2021/03/09 1430
https://datatracker.ietf.org/meeting/110/proceedings/
A
Hello
and
welcome
to
net
at
ietf
110
online,
I'm
lou
berger.
We
have
with
us
yamashifarkas
and
our
secretary
ethan
grossman.
This
is
our
second
session
and
the
information
is
posted.
A
Yesterday,
this
is
an
itf
meeting.
I
think
you
all
know
that,
and
I
most
are
familiar
with
the
rules
that
govern
our
participation
and
our
process.
If
you're
not,
please
go
look
up
the
note
well
everything
we
say
here:
it
becomes
part
of
our
public
record
and
it
is
being
recorded
we're
using
meat,
echo
you're
here
with
me.
So
of
course
you
know
that
and
meet
echo
automatically
handles
our
blue
sheets
and
it
does
have
a
chat
channel.
The
jabber
is
still
there.
A
If
you
want
to
go
directly
into
that,
note-taking
you've
seen
in
chat
we're
using
code
emd,
please
jump
in
and
help
us
take
notes
and
capture.
What
is
said.
That's
particularly
important
if
you
make
a
comment,
whether
it's
in
jabber
or
in
at
the
mic,
I'd
like
to
have
that
captured
comments
that
are
set
in
jabber
will
be
repeated
in
general.
A
If,
for
some
reason
you
don't
want
it
repeated,
that's
fine,
but
then
don't
add
that
to
the
minutes,
as
I
mentioned
yesterday,
the
tools
page
seems
to
be
having
a
little
bit
of
trouble.
So
you
might
want
to
go
to
data
tracker
I'll
point
out
that
the
data
tracker
now
has
the
htm
the
tools
format
html,
which
personally
works
much
better
for
me,
so
may
wish
to
go
there
we're
in
our
second
session.
A
We
have
a
joint
session
on
friday,
with
that's
being
hosted
by
pals,
it's
joint
with
mpls
in
spring,
and
it's
really
focused
on
the
mpls
label,
stack
and
pseudo
wire,
related
control,
word
and
oem,
and
how
those
things
are
going
to
progress
and
there's
lots
of
lots
of
work
going
on
in
that
area.
So,
if
you're
interested
in
that
topic,
I
suggest
you
join.
A
A
As
I
mentioned
yesterday,
we've
added
a
new
step
to
our
itr
disclosure
process.
Specifically,
if
you
join
a
as
a
contributor
or
as
a
co-author
to
a
working
group
document,
we
ask
that
you
make
an
ipr
statement
at
that
time.
We
reviewed
the
status
of
our
drafts
yesterday,
so
I'm
not
going
to
redo
that
today.
A
Oh,
I
should
mention
this
being
asked
for
working
group
adoption.
This
oem
framework
draft
is
the
first
one
that's
going
to
be
presented,
and
so
please
pay
attention
and
we
still
know
news
on
ietf
111.
So
with
that
we're
going
to
jump
right
in
to
oem
greg.
Just
please
make
sure
to
say.
C
Next:
okay,
ready:
okay,
thank
you
and
let's
go
to
the
next
slide.
C
Okay,
so
we've
been
working
on
on
a
document
addressing
the
comments
and
discussions
that
we
had
since
previous
109
meeting
and
the
structure
is
the
same.
What
was
updated
is
that
we
added
the
requirements
section
that
previously
was
a
part
of
a
working
group
adopted
document
on
oem
over
mpos
data
plane,
and
it.
C
Next,
so
originally
we
used
their
document
that
was
spawned
from
raw
oem
support
and
we
discussed
it
in
109
we
received
updates,
and
that
reflects
what
we've
done.
C
C
Net
constantification
there
are
different
methods
that
has
been
classified
as
passive.
One
of
them
is
a
in
a
group
of
onpath
telemetry
collection,
iom
and
alternate
marking
method.
C
So
the
telemetry
information
can
be
the
process
can
be
separated
into
two
sub
processes.
One
is
a
trigger
and
originated,
and
the
second
is
a
collect
and
transport
for
the
network.
Analytics
iem
tracing
uses,
two
methods
that
characterizes
end-to-end
and
hot
by
hop
in
addition
for
iem
tracing
is
defined,
direct
export,
which
every
node
that
originates,
telemetry
information
exported
to
the
collector
separately
and
another
mode
of
collecting
transporting
was,
is
proposed
and
discussed
in
npp
and
working
group.
It's
a
hybrid
two-step.
C
It
is
using
the
follow-up
packet
that
follows
the
same
path
as
a
trigger
packet
using
the
same
encapsulation
and
collects
telemetry.
Why?
We
think
it's
important
to
discuss
it
because,
especially
for
the
net
net,
where
the
network
resources
dedicated
to
detmet
service
are
scarce
and
precious
their
method
of
collecting
and
transporting
on
path.
Telemetry
is
important
because,
as
you
see
for
iem
tracing
it,
it
embeds
the
telemetry
information
into
the
data
packet
itself
into
trigger
packet,
whereas
direct
export
and
hybrid
two-step
can
be
transported
out
of
that.
C
C
And
another
important
part
that
of
the
update
to
this
draft
is
the
list
of
the
requirements.
I
I
didn't
call
we
didn't
copy
the
list
itself,
so
there
are
more
requirements.
We
will
try
to
summarize
that
will
fit
in
one
slide.
C
So
I
I
don't
want
to
read
them.
You
can
look
it
up
and
as
part
of
their
meeting
materials,
the
slides
are
there.
So
what
we
emphasize
is
that
each
and
every
node
in
that
net
must
be
capable
of
acting
as
a
source
or
addressee
of
oem
session,
so
act
as
a
map.
C
C
Controller
obviously,
there's
important
there
network
failure,
detection
and
itf
mechanisms
that
could
be
used.
The
protocols
among
them
are
bfd
and
being
in
trace
route.
C
C
By
sharing
the
same
encapsulation
of
their
flow,
that
is
being
monitored,
active
oem
is
in
band,
and
so
thus
we're
able
to
do
interpretation
of
their
failure,
detection
or
performance
metric
collected
and
apply
it
to
their
network
analytics.
Another
requirement
is
that
oem
must
support
one-way
measurement,
whether
it's
a
one-sided
or
two-sided.
C
And
in
case
of
two-sided,
so
their
results
can
be
collected
out
of
band
path.
Mtu
discovery
is
important
so
as
a
packet,
replication,
duplicate,
elimination
and
water
preservation
function,
and
because
this
is
a
very
dead
net
specific.
C
It's
something
that
we
are
planning
to
take
charge
and
have
a
proposal,
probably
by
the
next
meeting.
A
That
this
this
slide
and
this
section
of
the
document
we
didn't
get
a
lot
of
attention,
as
we
do
adoption
and
I'll
just
remind
people
that
they
can
make
their
comment
on
the
list
and
say
they
want
it
addressed
as
part
of
the
working
group
process.
It
doesn't
have
to
block
adoption
just
because
you
disagree
with
one
of
these
points,
but
it
is
good.
You
express
your
view
during
adoption.
C
Now
working
group
documents
such
as
defnet
oem
in
mpls,
with
mpls
data
plane
and
another
document
with
ip
data
plane,
they
referenced
the
requirements
and
now
they
reference
this
document.
So
we.
C
By
moving
this
list
of
document
to
its
more
logical
place
in
the
oem
framework
in
the
devnet
created
the
dependency,
so
it's
a
really
much
appreciated
consideration
of
adopting
this
working.
C
D
Thank
you
greg,
and
I
I
like
the
approach
actually
and,
as
you
said
now,
the
the
this
would
be
an
oem
framework
at
that
point
framework
and
to
which
the
mpls
om
and
the
ipym
could
refer
to
this
one
would
specify
the
common
aspects
for
that
notorious.
A
Comments
anyone
want
to
come
into
line,
we
have
a
minute
or
two.
I
think
it's
really
useful
to
get
any
objections
to
adopting
this
document.
Normally,
if
we
were
in
person,
we
would
do
our
usual
poll,
but
it's
a
little
hard
on
meat
echo,
but
getting
objections
is
not
so
if,
if
I
may.
A
I'd
like
to
just
ask
people
if
they
have
objections,
because
if
we,
if
we
do
the
tool
and
ask
for
objections,
we
don't
know
what
the
objection
is.
So
I
I
really,
I
think,
we'd
prefer
to
hear
what
the
objection
is,
whether
that
someone
types
it
in
jabber
or
says
it
at
a
mic.
I
think
those
are
good.
Now,
of
course,
we
are
going
to
do
this
on
the
list,
so
there's
always
an
opportunity
there.
A
Personally,
I
actually
have
some
technical
comments
on
the
document,
but
I
don't
think
it
blocks
adoption
so
I'll
make
those
comments
during
the
adoption
call
and
we'll
address
them
during
work,
normal
working
group
processing.
But
if
someone
has
an
objection,
probably
a
strong
one
it'd
be
great
to
hear
hear
from
them.
A
All
right
well,
thank
you
very
much
for
this
work
and
we
should
expect
to
see
an
adoption
call
on
the
list
and
hopefully
get
some
good
comments
moving
to
next,
I
think
we
have
control
planes,
bring
that.
A
E
Time
I
was
just
waiting
for
the
slides
to
show
yeah
if
you're
good
to
the
next
one.
Please.
E
Yeah
thanks:
that's
the
an
update
to
the
we
presented,
the
first
version
of
this
in
the
in
a
109
and
in
the
interviews
meeting
in
december.
E
We
had
discussions,
particularly
in
the
rings
meeting
on
what
we
intend
to
you,
know
the
focus
of
the
draft
to
be,
and-
and
I
highlighted
the
keywords
here
that
are
copied
from
the
from
from
from
the
abstract,
which
means
we
are
proposing
changes
to
rcp
and
surf
for
a
better
alignment
with
ours
with
tsn.
We
called
it
rsvp
dash
tsn
in
the
document,
and
that
means
we're
limiting
the
scope
of
the
draft
to
to
those
changes.
E
We
had
a
bit
of
discussion
in
the
interview
on
whether
or
not
this
this
would
have
a
broader
scope
to
something
like
rsvp.net,
which
we
initially
planned,
but
we
kind
of
focused
this
down
to
the
tsn
part
initially,
but
that
doesn't
limit
us
to
maybe
go
for
an
accompanying
exercise,
for
you
know
other
changes
to
rsvp,
but
we
wanted
to
start
with
that.
E
E
Yeah
thanks
yeah,
I
think
you
you,
you
jumped
you
jumped
the
the
structure
slide.
I
think
there
was
a
slide
in
between,
though.
A
A
A
A
One
question:
do
you
want
comments
as
we
go
or
do
you
want
comments
at
the
end.
E
Oh
yeah,
that
was
the
question.
I
didn't
really
quite
get
I'm
fine.
I'm
fine
either
way.
I
I
I
heard
something
about
end,
but
that
must
have
been
when
you
had
audio
problems.
E
Thanks
so
we
we,
we
clarified
the
use
cases
as
a
change
from
the
last
version.
We
now
present
two
use
cases
what
I
would
call
a
hybrid.net
over
to
your
center,
where
customer
network.
The
difference
is
that
we
have
in
in
one
case,
a
a
core
network
with
rsvp
support
and
the
other
ones.
We
don't
that's
a
difference
in
the
use
cases.
E
Section
three
is
now
focused
on
the
design
rationale,
which
are
the
former
sections
two
two
two
three
two
five,
so
we
merged
them
into
a
new
section
and
then
a
focal
section,
four
on
the
rsvp
tsn
proposal,
which
took
the
layer
into
actions
revised
from
the
previous
version.
So
it's
it's
a
revised
part,
but
it's
largely
taken
from
from
the
previous
draft
and
we
also
changed
the
api
descriptions
to
align
with
the
terminology,
which
was
another
comment
we
received
so
in
the
structure.
If
you
go
to
the
next
slide,
please.
E
Yeah
thanks
so
section
two
is:
are
the
use
cases
also
there
we
revise
and
learn
the
terminology
as
a
request.
E
Design
rationale,
section
three,
as
I
mentioned
before,
and
then
section
four
is:
is
it's
the
proposed
design
which
which
has
the
layer
interaction,
the
the
interactions,
the
the
api
descriptions
and
the
message
formats
will
come
in
the
next
version?
We
didn't
finish
them
in
time
for
the
deadline.
So
that's
the
last
subsection
of
section
four.
I
don't
know
why
the
numbers
went
wrong.
That's
obviously
not
section,
seven
there
but
never
mind.
That's
at
the
end.
E
Please
so
the
we've
received
feedback
both
in
the
in
the
in
the
109,
but
also
the
interviews
meeting
here
that
we
tried
to
address
in
the
new
version.
So
the
the
scenario
was
the
one.
You
know
what
type
of
scenario
we
had.
That's
why
we
revised
the
use
case
section
to
provide
their
hopefully
a
clearer
example.
E
We
clarified
the
relation
to
the
to
the
tsn
specific
data
plane
drafts
by
providing
revised
api
descriptions
in
section
four
also
clarified
the
use
of
the
flow
information
model
by
aligning
the
terminology
in
those
api
descriptions
again
and
also
did
a
scrub
hope
that
we
caught
most
of
the
things
where
we
had
misalignments
across
the
sections
where
we
were
diverging
from
the
terminology.
E
Thanks
to
the
next
one,
please
so
the
next
steps
I'd
like
to
do
is
to
fill
in
the
message
formats
and
the
protocol
information,
as
I
didn't
get
that
done
in
time
and
obviously
would
like
to
seek
more
input
from
the
list
potential
which
brings
me
to
the
last
slide.
If
you
can
just
jump
to
that
one
already,
which
means
we
would
like
to
hear
feedback
comments,
potentially
anybody
who's
happy
to
help
us
co-authors
and
contributions
are
very
welcome.
E
F
F
If
you
saw
the
presentation
on
the
yang
on
monday,
does
this
brand
intend
to
cover
all
of
the
deer
playing
scenarios
that
were
outlined
in
that
yang
presentation?.
E
No,
we
haven't,
we
haven't
reviewed
that
yet
I
I
noticed
that
yesterday
I
sent
a
note
to
my
co-authors
as
well
to
to
have
a
good
look
at
that
they
they
they
couldn't
make
it
to
the
meeting.
So
I
can't
tell
you:
okay,.
F
Let
me
suggest
that
the
the
the
presentation
material
did
a
really
nice
job
on
covering
all
the
various
data
planes
that
would
work,
it'd
be
great,
rsvp
covered
all
of
those,
and
as
part
of
that,
could
I
suggest
you
find
an
acronym
other
than
tsm,
which
is
going
to
going
to
create
some
bad
implications
about
which
data
planes.
This
applies
to.
A
So,
unlike
david,
I
think
your
your
the
title
is
appropriate
because
you
really
are
limited
there,
but
I'm
going
to
go
back
to
a
comment
that
I
believe
we
made
at
the
last
couple
times
that
you've
talked
about
this
draft.
Is
we
don't
have
a
definition
of
what
rsvp
looks
like
for
debt
net
in
general
and
that's
for
all
the
use
cases
that
david
just
talked
about
where
for
ip
for
mpls
for
aggregation?
A
We
don't
know
what
that
looks
like
and
there's
a
a
large
body
of
work
we
can
build
on,
in
particular
the
traffic
engineering
extensions
to
rsvp,
and
I
know
that
you're
not
building
on
those
you're
building
on
inserve.
So
you
know,
what's
your
thought
about
addressing
these
base
capabilities
from
the
comment
before
from
david's
comment,
as
well
as
from
previous
meetings.
E
Yeah,
so
when
we
discussed
about
how
to
move
forward
because
the
camera
as
you
as
you
said,
was
made
during
the
last
meeting
to
go
for
an
rsvp
deadnet,
we
found
the
the
access,
at
least
for
us,
for
the
for
the
authors
that
we
had
too
large.
So
we
honed
it
down
to
tsn,
but
I
do
agree,
and
but
that
was
the
discussion
we
had
in
the
last
meeting,
but
there's
a
whole
body
of
work
to
be
done
on
that
net
view
and
rsvp.
E
If
you,
if
you
will,
but
it's
probably
that
we
didn't
want
to
take
that
on
so.
Hence
we
hung
ourselves
down
into
this
end,
but
it's
a
little
bit
kind
of
avoiding
the
work,
but
I
think
it's
maybe
also
starting
the
work.
So
we
wanted
to
be
concrete
from
the
from
the
tsn
angle
and
can
then
maybe
reopen.
You
know
that
body
of
work
or
that
discussion
once
you
know
we
move
forward
if
you
will
right
so
that
was
that
was
a
little
bit.
I
would
compromise
in
a
way.
A
Okay,
well
with
sort
of
chair
hat
on
my
perspective,
is
follow
david
black's
guidance.
G
Yes,
this
is
also
what
I
wanted
to
highlight,
so
I
I
think,
thank
you
very
much
for
the
update
based
on
that.
It
is
much
more
clear
what
was
your
focus
with
this
draft
and
I
I
would
like
to
join
the
previous
comments
that
the
title
might
be
misleading,
because
what
you
are
focusing
here
is
if
rsvp
is
used
for
that
net
signaling
and
you
are
using
graph
for
tsn
signaling
how
these
control
plane
signaling
can
interact
in
a
sub
network
scenario.
G
So
I
think
that
is
your
use
case,
and
this
is
pretty
well
described
in
the
updated
draft.
So
thanks
for
that,
but
I
think
also
that
this
is
something
that
must
be
highlighted
in
the
title
as
well,
that
you
have
a
special
focus
on
a
special
scenario
and
not
solving
the
control
plane
signaling
for
that
net.
E
The
best
thing
to
do,
instead
of
reflecting
the
actual
focus
in
the
draft,
so
that
one
I
do
agree
with,
I
said
I
mean
there
is
an
interest
to
come
back.
I
mean
I
discussed
this
with
the
with
my
courses
to
to
look
at.
You
know
the
larger
problem
still
of
control,
plane
signaling,
you
know,
but
it
would
be.
E
I
see
this
as
a
body
work
that
probably
goes
beyond
the
competences
that
we,
the
three
of
us,
have
and
that's
why,
in
the
meantime,
we
would
probably
like
to
just
stick
to
it.
We
can
just
reflect
that
by
narrowing,
also
the
scope
clearly
in
the
title
and
renaming
the
title,
but
certainly
being
interested
in
addressing
the
conor
open,
signaling
problem
at
some
point,
as
contributors.
D
F
The
obvious
question:
if
the
draft
scope
is
limited
to,
if
rsvp
is
used
for
detnit,
then
this
is
how
it
works
in
tsn.
Do
do
the
folks
involved
have
running
code
for
rsvp
insert
for
net.
F
E
Assumption,
yes,
sorry,
I
I
again,
I
had
a
hiccup
a
dropout
somewhere,
but
what
I
said
from
the
question
is:
yes,
that's
that's
what
we
do.
Okay,.
F
Yeah,
I
advise
checking
it
because
I
think
that
the
question
is
lurking
in
here
is:
it
is
a
very
convenient
assumption
and
it'll
be
nice
if
it
were
true
that
rsvp
int
serve
can
work
for
debt
net.
That
really
ought
to
be
ought
to
be.
Sanity,
checked
somehow.
A
D
I
have
have
similar
concern,
like
the
previous
speakers
said,
that
sort
of
the
basis
would
be
to
figure
out
how
rsvp
works
for
that
on
its
own,
and
then
we
can
think
of
integration
with
tsa
and
without
having
the
basic
stuff.
D
It's
it's
hard
to
see
or
how
to
do
the
work.
I
think
I
understand
you
want
to
narrow
your
focus,
but.
F
Yeah,
I
think
youngest,
I
are
about
the
same
page,
not
not
clear
that
rsvp
in
surf
for
detonate
is
the
proverbial
right
thing
to
do.
Although
I
do
understand
the
the
concern
about
massive
massive
expansion
of
scope
of
your
work.
A
A
And
with
that
we're
going
to
move
on
to
microburst
young
peng,
if
you
would
come
to
the
mic,
find
your
slides.
A
Yes,
just
make
sure
to
say
next,
when
you're
ready
for
the
next
slide.
Please
go
ahead.
I
This
is
a
talk
about
microburst
and
I'm
tong
hongduo
from
china,
mobile
and
next.
Please.
I
A
One
comment
about
I'm
sorry
for
the
delay.
One
comment
on
on
the
slide
you
say:
dettnet
has
scoped
a
single
administrative
domain.
It
also
can
support
a
closed
group
of
administrative
control,
so
it
could
include
cooperating
administrative
domains.
So
I
think
you
yours
your
comment
there.
That
sub-bullet
is
not
aligned
with
our
charter,
so
just
be
aware
of
that.
I
I
The
conclusion
of
conclusion
is
that
nowadays,
we
are
short
of
a
common
and
a
simple
method
to
provide
this
loneliness
service
in
large-scale
sp
networks,
and
our
method
mentioned
in
the
draft
show
a
potential
solution
that
is
both
simple
and
scalable.
Meanwhile,
it
does
not
need
the
type
factorization.
I
I
I
So
we
think
that
they
should
know
both
in
the
network,
and
in
this
case
the
main
problem
is
to
decrease
the
microburst.
In
our
mechanism.
We,
it
uses
flow
shipping
on
the
edge
and
on
the
intermediate
nodes,
we
do
the
interface
graduated
shipping,
we
change
the
regular
behavior
or
on
the
router.
It
is
to
say
that
the
router,
following
all
the
traffic
as
soon
as
possible
in
our
maximum,
we
use
the
shipping
method
in
the
tsn
so
that
the
traffic
will
be
forwarded
in
out
of
the
way
next
week.
Please.
I
We
also
give
some
analysis
about
the
purpose
method
when
we
compare
comparing
and
with
the
traditional
ip
forwarding
with
the
priority
in
a
relatively
light
loading
network,
our
mechanism
can
offer
a
better
folding
for
the
rotated,
low,
latency
traffic
with
lights
microburst
and,
of
course,
we
have
some
cause:
the
cost
of
shipping
on
the
on
the
edge
and
keeping
the
shipping
result
on
the
intermediate
nodes.
I
Comparing
the
cancer
mechanism,
our
mechanism
is
simpler
and
more
scalable.
Of
course,
we
have.
Some
disadvantage
is
about
the
uncertainty,
everything
that,
in
our
maximum
uncertainty,
perhaps
is
higher,
but
okay,
when
the
uncertainty
is
higher,
perhaps
some
of
the
final
use
case
cannot
be
exported
in
our
mechanism
next
page,
please.
I
This
page
is
about
the
work
at
the
a
step
we
will
try
and
find
whether
some
of
the
that
net
use
case
can
be
supported
in
certain
conditions.
I
We
think
that,
for
example,
in
a
network
that
the
critical
traffic
is
not
too
much
or
low
density
traffic
is
not
that
critical
and
the
perhaps
the
the
first
step,
the
loaded
traffic.
The
note
that
critical
will
be
would
appear
in
the
large
scale
as
united
workforce,
and
we
also
think
about
to
do
this
approximate
evaluation
of
the
society.
I
We
need
to
check
whether
the
mechanism
we
can
work
can
work
for
those
low
latency
traffic.
They
are
not
was
that
critical
and
we
think
that
the
replication
and
elimination
method
in
testing
can
have,
because
in
our
mechanism
perhaps
we
cannot
ensure
that
a
hundred
percent
losses,
this
packed
replication
of
the
elimination,
private
hype
to
grow
some
packets,
that
have
lost
the
network,
and
this
is
the
last
page.
I
think
if,
if,
if
you
have
any
problem
in
any
question,
you
can
talk
now.
F
Yeah,
so
looking
at
the
description
of
this,
the
approach
looks
a
lot
like
diffserv,
where
the
flows
are
individually
policed
at
the
edge
and
then
there's
a
flow
aggregate
traffic
traffic
conditioning
conditioning
in
the
core.
The
scalability
comments
are
definitely
on
the
mark
is
that
that
was
one
of
the
reasons
that
diffserv
was
designed.
The
way
it
is
I'd
encourage
of
the
authors
to
take
a
look
at
a
diffserv
framework,
that'd
be
rfc
2475
place
to
start.
A
I
K
A
E
K
I
understand
that
that's
what
has
to
be
done,
maybe
one
day
meet,
echo
will
have
a
button
for
us
to
push
to
move
someone
else's
slides,
hi
everyone
jakob
stein,
it's
nice
to
be
back
in
the
itf
after
a
long
time.
Next,
next
slide,
please
the
reason
I
haven't
been
coming
to
the
itf
for
a
long
time
is
I've
been
wasting
my
time
with
5g
exhale
transport,
and
I
really
could
use
this
time
to
plug
a
book
which
just
came
out,
but
I
won't
do
so.
There
are
problems
having
to
do
with
delays.
K
I'm
sure
you
all
know
when
doing
backhaul
of
mobile
and
it's
gotten
much
worse
in
5g
and
people
have
thought
that
perhaps
tsn
or
detnit
mechanisms
are
can
be
adapted
to
solve
this
problem
and
it
turns
out
it
doesn't
work
very
well.
In
particular,
I'm
talking
about
qbv,
which
I
realize
is
not
a
death
net,
it's
more
of
a
tsn
approach,
but
I
have
to
define
my
first
line
here.
K
The
mechanism
I'm
talking
about
is
for
what
I
call
a
time.
Sensitive
packet
flow,
where
I
mean
by
time
sensitive
simply
that
the
packets
have
to
be
delivered
within
only
a
little
bit
more
than
the
physically
possible
minimal
time
to
get
through
the
network,
and
when
you
try
to
use
mechanisms
that
do
time
gating,
it
turns
out
that
they
don't
scale
and
if
you
actually
try
to
give
an
upper
bound,
the
upper
bound
is
much
higher
than
what
is
physical,
physically
possible.
K
And
if
you
don't
plan
it
correctly
and
it's
almost
impossible
to
play
it
correctly.
The
efficiency
on
the
line
goes
down
and
we're
wondering
if
there
is
a
mechanism
that
doesn't
suffer
from
all
of
these
problems
and
it's
interesting
by
the
way
that
the
802.1
people
themselves,
the
tsn
people
in
their
cm
document
for
exalt,
recommend
frame
preemption,
but
don't
recommend
their
own
mechanism
for
for
this
this
case
and
which
I
think
is
basically
because
it
doesn't
scale
very
well.
Can
I
have
the
next
slide.
Please.
K
Just
before
I
introduce
what
I
want
to
talk
about,
I
just
want
to
give
a
model
for
the
forwarder
which,
from
now
on
I'll
call
router,
although
it
could
be
a
switch,
an
sdn
switch,
an
mpls
lsr
or
whatever
you
want,
but
basically
I'm
talking
about
a
forwarding
device
which
has
two
parts,
forwarding
logic
which
takes
the
packet
and
decides
which
output
port.
K
It
has
to
go
to
and
scheduling,
logic
which
decides
which
packet
of
all
those
packets
that
have
to
be
transmitted
over
the
output
port
should
be
first,
and
I
just
want
to
mention
that
when
you
think
about
it
in
terms
of
segment,
routing
and
tsn
or
deathnet,
rather
than
saying
which
port
and
which
packet
to
send
over
that
port,
you
can
think
of
it
as
where
to
send
the
packet
and
when
to
send
the
packet.
In
other
words,
forwarding
is
basically
saying
given
this
packet,
it
just
arrived.
K
K
So
what
am
I
proposing?
I'm
proposing
an
alternative
to
using
qbv
and
similar
mechanisms
that
relies
on
using
a
stack
data
structure
added
to
the
packet
in
the
packet
headers
I'll
mention
in
a
moment.
What
has
to
be
in
that
stack?
There
are
two
options.
One
is
only
for
the
scheduling
problem
once
one
for
the
both
for
the
forwarding
and
the
scheduling
problem,
and
the
nice
thing
about
this
alternative
is
that
it
can
be
optimized
in
a
scalable
way
it
can.
Its
configuration
can
be
adapted.
K
In
other
words,
you
can
add
a
new
flow
without
having
to
start
touching
a
lot
of
network
elements,
a
lot
of
routers
in
the
network,
during
which
time
you'll
have
inconsistencies
and-
and
you
might
have
packed
quote-
either-
be
lost
or
fail
to
meet
their
deadlines,
and
it
can
get
down
to
very,
very
close
to
the
physical
delay
and
the
mechanism
using
a
stack.
It
has
sort
of
an
analogy
with
segment
routing.
K
K
K
But
it
hasn't
been
that
long
in
the
system,
so
you
might
decide,
let's
use,
what's
called
earliest
deadline
first
or
edf,
and
that
simply
puts
the
packet's
deadline,
the
absolute
time
that
it's
supposed
to
reach
the
other
end
of
the
network
or
the
final
host,
and
that
sounds
better.
It
is
better,
however,
that
obviously
prioritizes
packets
with
an
earlier
final
dead
time
deadline
in
each
router
along
the
way,
without
knowing
how
far
it
is
from
this
router
to
the
final
destination.
K
So
you
might
have
a
case
where
you
are
prioritizing
a
packet
which
is
just
has
one
more
hop
to
go
over
another
packet,
which
has
which
looks
like
it
has
more
time,
but
actually
has
a
long,
a
large
number
of
routers
to
get
to
traverse
so
both
lis
and
edf,
and
many
other
variations
where
you
put
one
time
into
the
pocket,
aren't
good
enough
next
slide.
Please.
K
So
if
for
each
router,
you
traverse
you
pop
the
top
of
the
stack,
you
look
at
that
time
and
you
do
some
kind
of
scheduling
which
might
be
edf
or
just
in
time
or
something
I
call
pedf
look
at
the
draft
or
any
other
mechanism
and
of
course
you
need
some
mechanism
for
figuring
out
these
deadlines.
I'm
going
to
show
you
one
in
the
next
slide,
which
I
mentioned
the
draft,
but
certainly
not
the
only
mechanism,
and
I
refer
to
other
ones
in
the
draft
as
well.
K
Three
minutes,
okay,
of
course,
this
requires
something
more
complex
in
the
router
than
a
simple
fifo
q
per
qos,
like
is
normally
done
in
a
router
or
a
switch.
However,
I
think
it's
less
complex
than
using
time
sensitive,
gates,
time
scheduling,
gates
next
slide.
Please.
K
And
why
do
I
say
it
is
analogous
to
segment
routing,
because
if
I'm
already
putting
a
stack,
which
has
an
entry
for
every
router
to
be
traversed
into
the
packet,
I
might
as
well
combine
it
with
segment
routing
and
have
both
the
forwarding
decisions
and
the
scheduling
decisions
in
the
same
stack.
In
other
words,
I
put,
I
can
put
a
stack
in
where
every
entry
in
the
stack
has
two
sub
entries,
one:
a
forwarding
sub
entry.
Basically,
where
do
you
send
the
packet
and
one
is
scheduling,
sub
entry
which
says
when
to
send
the
packet?
K
K
This
can
be
combined
with
best
effort
traffic
with
strict
priority,
and
you
can
even
have
several
priorities
of
time.
Sensitive
packets
next
slide.
Please.
K
Here's
a
simple
example
to
give
you
the
idea
why
these
numbers
are
what
they
are.
You
can
see
in
the
draft.
However,
the
idea
is
the
host
in
this
case,
I'm
do
using
it
like
in
segment
routing
rather
like
in
sort
routing
source
routing.
So
you
actually
see
the
header
which
is
being
inserted
only
after
the
first
router,
and
it
has
the
number
of
entries
as
the
number
of
routers
minus
one,
but
basically
you
will
see
the
stock
there.
K
The
srtsn
stack,
if
you're
doing
both
where
you
have
both
the
forwarding
sub
entry
and
the
scheduling
sub
entry
and
each
time
one
is
popped
until
you
get
to
the
final
one
and
you
can
even
save
a
bit
by
having
a
special
code
for
a
for
bottom
of
stock
rather
than
wasting
a
bit.
But
that's
we
don't
really
have
to
go
into
that
right
now.
Next
slide,
please
I
realize
I
only
have
one
minute.
K
I
was
asked
to
talk
a
little
bit
about
how
much
room
this
takes
if
you're
thinking
of
128-bit
ipv6
addresses
and
64-bit
timestamps.
This
would
sound
like
it
would
take
a
tremendous
amount
of
room,
but
not
so
you
can
actually
make
do
with
much
many
fewer
bits.
K
Actually
I
didn't
give
you
one
extra
bit
here,
because
if
you
do
that
trick,
I
mentioned
before,
but
the
bottom
of
stock.
You
don't
need
that
plus
one
at
the
end,
but
you
actually
in
many
cases
for
the
kinds
of
networks
where
this
is
important,
where
you
have
several
hundred
routers
and
the
maximum
amount
of
time
that
a
packet
can
go
through.
The
network
is
on
the
order
of
milliseconds.
K
This
can
end
up
being
16
bits
or
maybe
32
bits
if
you're
really
going
to
squander
them,
which
means
that
between
four
and
eight
hops
only
require
as
much
room
as
a
single
ipv6
address.
Next
slide.
Please,
and
to
sum
up
what
I'm
asking
from
this
working
group-
and
I
asked
something
similar
to
this
spring
working
group-
and
I
was
going
to
ask
pce
as
well,
but
they
didn't
get
me
into
their
schedule,
is
where
should
this
work
be
done?
It
is
similar.
It
has
elements
of
segment
routing.
K
It
has
elements
of
time
sensitivity
of
the
ditnet
variety
and,
of
course,
requires
an
optimization
algorithm
and
by
the
way
it
was
interesting
to
hear
just
before
about
rsvp.
I
also
have
a
non-centralized
mechanism
where
you
don't
have
to
do
measurement
of
the
link
delays
ahead
of
time,
but
rather
send
an
rsvp
like
packet
ahead
of
actually
several
of
them
and
measure
the
the
acquire,
timestamps
and
sort
of
measure
the
delays
as
you
go,
but
simply
I'm
asking
for
the
chairs
to
coordinate
where
and
if
they
want
this
work
to
progress.
K
And,
of
course,
if
there's
interest
in
this
work
at
all,
that's
it.
A
Okay,
we
didn't
leave
time
for
questions
because
we
do
have
one
more
slot
and
we're
rapidly
running
out
of
time,
we'll
take
the
action
to
coordinate
so
absolutely
and
it
seems
like
there's
some
interest,
because
we've
got
good
discussion
on
the
list,
which
is
always
nice
to
see
on
a
new
draft.
With
that
we're
going
to
move
to
the
last
presenter.
H
H
H
H
The
packets
send
by
upstream
node
and
circle
x
must
be
received
by
node
b
under
the
same
cycle
in
a
cyclex
node
a
send
all
pancakes
in
the
buffer
to
node
b
in
the
same
cycle
x,
node
we
use
above
or
to
accumulate
all
pancakes
from
e.
H
H
Therefore,
if
the
pancreate
transfers
h
port
h
hopes,
the
maximum
literacy
is
h,
plus
one
times
tc.
What
is
what
is
tctc
is
the
buffer
of
a
cycle,
and
the
the
minimum,
let's
see
is
h
is
a
h,
minus
1
times,
t
c.
The
g
abundant
is
2
times,
t
c
as
a
great
square.
H
H
H
H
H
A
Do
you
have
a
question
or
okay
well,
thank
you
very
much
for
the
presentation.
I'm
not
sure.
If
we
see
any
questions
I
actually
see
david
is
unmuted
david.
Do
you
want
to
make
a
comment.
F
Yeah,
let
me
see
if
I
can
choose
my
words
carefully
here:
there's
been
some
extensive
behind
the
scenes
discussion
between
myself,
the
chairs
and
with
the
ads
watching
about
this
draft.
F
The
advice
I
would
give
to
the
authors
is
to
split
the
mechanism
away
from
the
location
of
the
bits,
and
then
I
think
the
crucial
question
of
the
debt
net
working
group
is
whether
the
problem
that
the
authors
propose
to
solve
is
something
that
the
debt
network
group
wants
to
take
on.
In
which
case
a
plausible
path
forward
would
be
to
work
on
the
mechanism
here
and
then
go
look
for
the
bits
elsewhere
and
I
can
think
of
at
least
four
places
where
one
might
be
able
to
find
bits.
L
Thank
you.
How
is
this
so
so
david?
How
is
this
different
from
yakov's
draft
with
respect
to
cueing
mechanism
versus.
F
L
Right
so
from
my
perspective,
certainly
the
normative
thing
that
that
that
we
require
would
be
an
encapsulation
for
the
interoperability
right,
and
I
think
that
applies
both
to
yakov
and
our
draft,
and
you
know
I
I
can
just
observe
that
would
be
the
similar
type
of
you
know.
Encapsulation
that
was
done
in
support
of
prior
in
that
matter.
Allow.
F
L
Right
and
and
the
the
mechanisms
are
defined
like
tsn,
they
may
not
have
been
defined
in
the
that
networking
group,
but
the
encapsulations.
You
know
in
the
same
way
as
pre-op
right
where
was
pre-op
defined,
and
why
did
that
net?
Take
on
the
encapsulation
mpls,
for
it.
A
All
right
we
have
yaakov
and
q,
stur
and
q
and
I'll
point
out
where
over
time
they
usually
let
us
run
a
few
minutes
over,
but
we
will
have
a
hard
stop
when
meat
echo
says
we're
done
so
we'll
keep
going
as
long
as
we
can,
but
just
be
aware,
we
are
over
yakov.
K
I'll
be
really
really
quick,
just
as
a
response
to
what
therapist
said.
I
am
not
at
this
point
recommending
any
particular
encapsulation,
I'm
bringing
up
a
an
issue.
I
think
this
issue
is
a
valid
one.
K
It's
one
of
the
problems
that
I
mentioned
also
that
all
kinds
of
gating
have
this
problem,
but
they
have
to
absorb
the
amount
of
time
between
the
switches,
and
I
think
that
the
right
thing
to
do
is
if
the
mechanism
actually
can
be
proven
to
be
efficient,
is
you'll
have
to
define
either
a
universal
encapsulation,
which
is,
in
addition
to
everything
or
find
a
way
to
stick
the
bits
in
and
as
was
mentioned
before,
there
are
several
different
places
you
can
put
it.
K
J
Well,
it
seems
to
me
that
yeah
yeah
sorry
is.
It
seems
to
me
that
this
is
definitely
a
problem.
That
net
has
the
largest
interest
in
it
being
the
the
working
group,
that's
working
on
the
most
critical
form
of
of
time-related
communications,
and
so
I
think
it.
This
is
really
the
place
to
at
least
start
this
piece
of
work.
J
If
we
decide
that
we
need
to
export
it
to
a
specialist
encapsulation
group,
then
that's
fine,
but
I
think
this
working
group
is
the
working
group
that
needs
to
to
own
this
set
of
problems
associated
with
getting
packets
to
the
right
place
in
the
right
time.
A
Okay,
thank
you
I'll
point
out
I'll
point
out
that
to
date,
we
as
a
working
group
have
only
taken
on
informational
documents
related
to
queuing,
not
standards
track
because
we're
that's
outside
our
current
charter
and
more
than
that,
the
transport
area
typically
owns
queuing
mechanisms.
I
don't
know
of
any
case
where
a
queuing
mechanism
has
been
defined
in
the
routing
area.
A
The
definition
of
that,
though,
is,
if
is
not
outside
the
scope
of
the
ietf,
and
if
the
ads
and
the
transport
area
agree
that
you
know
the
right
place
for
it
to
be
done
is
that
net?
You
know
we'll
follow,
but
we
are
being
very
careful
as
chairs
to
stay
within
our
charter
and
to
coordinate
with
the
transport
area
on
the
definition
of
new
queuing
mechanisms,
referencing
existing
queuing
mechanisms
and
how
they
operate
with
debt
net
and
whether
they
need
special
bits
on
the
wire.
A
I
believe
that
is
in
scope,
that
is
within
the
discussion.
It's
the
queuing
part.
That's
not!
I'm
gonna
go
to
david
out
of
cue
just
because
we
might
get
cut
off
and
if
we're
still
going
I'll,
we'll
come
back
to
toilets
and
stewart.
So
david.
F
Wearing
my
transfer
working
group
hat
as
we're
likely
to
be
one
of
the
places
where
the
big
hunting
goes
on,
the
expertise
on
cyclic
cueing
is
very
clearly
in
debt
net,
not
in
the
transport
area
working
with
uncomfortable
going
looking
for
bits.
If
the
mechanism
is
defined
here,
I'm
not
comfortable
being
the
primary
venue
for
defining.
A
J
M
Uma
hi,
so
thanks
thanks
states,
I
think
my
my
opinion
is
basically
the
problems.net
has
been
talking
over
a
period
of
time
with
respect
to
bounded,
latency
and
also
the
jitter
case.
This
draft
need
to
be
solved
in
that
net
and
also
it's
not.
This
is
like
you
know
that
net
is
more
focused
on
p
e
to
p
e
k
s
not
to
end
to
end
so
to
speak
from
what
psn
tsv
working
group
is
doing.
M
So
I
think
this
has
to
be
done
in
depthnet
to
the
associated
problems
and
use
cases
have
been
discussed
for
many
years
here.
A
Okay,
stuart:
are
you
still
in
queue?
I
assume
you're
out
and
all
right?
Well.
Thank
you
all
for
the
discussion,
as
I
mentioned
before,
a
big
issue
with
queuing
is
our
charter
and
that
queuing
isn't
done
in
the
routing
area
traditionally
in
the
ietf,
but
we'll
we'll
work
to
find
a
home,
and
so
we'll
work
with
the
transport
area
chairs,
which
includes
david
and
also
with
the
with
the
area
directors,
who
you
know
the
isg
says
where
work
goes
so
we'll
work
with
them
on
that
and.
L
Can
I
just
jump
in
with
a
question
there
not
about
the
queuing,
but
you
know
if
I
observe
yarkov's
draft,
which
I
find
very
interesting
from
the
perspective
of.
Actually,
you
know
in
contradiction.
What
some
people
in
spring
thought,
I
think,
is
an
ideal
case
of
exploiting
segment
routing,
whether
it's
mpls
or
v6.
We
we
haven't,
you
know,
done
sr
as
a
forwarding
plane
in
that
net.
A
That's
a
question
we've,
so
sr
has
been
raised
a
few
times
in
the
working
group.
We've
said
we're
gonna
coordinate
with
the
sr
chairs
and
we've
been
talking
with
them
over
the
last
two
or
three
ietfs
privately,
we've
said
to
them.
A
It's
time
we
have
to
make
a
call
as
to
which
working
group
owns
this,
so
we've
said
to
them
that
we're
happy
to
do
it
in
debt
net,
but
it's
also,
you
know,
that's
their
forwarding
technology,
so
we
want
to
coordinate
with
them
and
we'll
let
you
know
which
working
group
that
work
should
go
to.
L
A
All
right
appreciate
everyone
staying
over
we're
gonna
end
the
meeting
now
and
really
appreciate
the
contribution
and
look
forward
to
further
discussion
on
the
list.
Thank
you
all
so
much
for.