►
From YouTube: IETF-DETNET-20230426-1200
Description
DETNET meeting session at IETF
2023/04/26 1200
https://datatracker.ietf.org/meeting//proceedings/
A
A
C
E
C
A
A
C
C
C
C
C
C
F
C
C
It's
similar
to
the
last
meeting,
quick
intro,
which
we're
in
the
middle
of
now,
including
the
note,
well
slide
and
a
general
sort
of
discussion
of
process
oriented
topics.
How
do
we
make
progress
from
here
and
we
have
a
presentation
coming
up
from
ping?
Sorry,
let's
see
xiaofu
I
think
on
a
deadline
based
forwarding
and
I
see
Chef,
who
is
on
good?
Okay.
C
C
C
Sure,
if
you've
sure,
if,
if
you
want
to
talk
about
revisions
to
it,
because
one
of
the
things
I
would
like
to
get
started
after
this
meeting
is
to
to
take
the
start.
Looking
at
how
the
TSN
queuing
scheduling
mechanisms
Fair
against
the
requirements,
it's
not
so
much
that
we
expect
the
meat
requirements.
We
expect
them
not
to
meet
most
of
the
requirements,
but
we
ought
to
get
started
with
something
that
is
known
and
neutral.
C
G
C
G
C
G
Will
come
save
those
eyes
I'm
asking
yes,
okay,
so
here's
the
proposed
change
of
the
requirement,
meaningful
section
three
I
think
from
jindong.
He
proposed
to
add
three
Apartments.
The
first
one
is
basically
about
to
a
large
number
of
hops,
whereas
the
complex
technology
and
my
second
one
tolerates
the
birth
and
this
accumulation
and
the
third
one
we
tolerate
High
utilization
and
he
proposed
to
add
a
three.
So
if
we
really
add
it,
we
will
have
a
10
requirements
and
it's
a
proposed
text
from
him.
G
C
B
No
don't
need
to
change
the
title
right
so
just
just
as
as
an
add-on
to
to
the
proposed
Tech
TBD.
Look
into
that,
and
maybe
the
the
folks
who
presented
that
last
week
would
would
know
exactly
what
to
add
to
highlight
that
problem,
because
that's
exactly
the
combination
of
number
three
five
we
do
have
rings
with,
with
with
traffic
being
injected
in
multiple
places
and
then
number
three
seven
of
the
microbirth
problem.
G
G
G
G
G
C
G
G
E
E
D
E
C
That's
a
really
good
question:
I've,
probably
given
a
bad
answer
earlier,
so
let
me
try.
Let
me
try
to
explain
my
way
out
of
this
I
think
we
need
to
put
all
the
requirements
in
the
requirements
document.
This
set
of
meetings
is
going
to
be
focused
initially
on
data
plane,
but
I
think
we
should
capture
the
control
and
management,
so
I
wouldn't
want
to
lose
it
I.
Just
I
just
like
to
separate
it.
E
E
F
Thank
you
so
I
might
duplicate
your
comment,
but
I
still
want
to
mention
this.
The
last
sentence
for
the
3.8.
Basically
it
it
says
to
provide
a
scalable,
Network
planning
and
queuing
solutions.
I
want
to
I
want
to
point
out
the
the
queuing
solutions
to
tolerate
whether
they
tolerance
support
the
high
utilization.
I
think
this
is
a
valid
one,
because,
for
example,
in
some
of
the
query
mechanisms
with
like
cqf
original,
you
try
to
use
the
that
time,
but
the
more
the
longer
that
time
we
try
to
put
then
the
lower
utilization.
F
D
H
C
E
D
Document
that
nubuckers
was
just
me
to
add
some
tests
for
the
the
enhance
dialect
to
to
add
this
test
to
the
working
group
working
job.
So
we
also
can
add
new
function
about
the
control
plane,
so
I
think
we,
we
should
suggest
to
add
this
function.
I
agree
to
add
this
function
and,
like
the
lateral
planning
I
think
it
is
very
reasonable
to
ask
you
that
drop.
D
G
E
E
C
D
E
B
E
B
G
G
B
H
C
F
B
F
Yeah,
okay
I
got
it
so
so
the
controller
that
one
the
the
working
group
document.
Actually
it's
talk
about
the
controller
plane,
so
it
talks
about
the
past
computation,
pce
part
and
the
past
establishment.
So
I
think
what
tallest
suggested
is
more
like
from
the
forwarding
plane.
What's
there
what's
there,
what's
the
at
the
forwarding
plane
on
like
the
more
like
Source
routing
kind
of
thing,
I
guess.
B
The
audio
on
and
off
is
is
very
slow
today,
so
I
think
I
need
to
stay
unmuted
yeah,
exactly
that's
that's
what
I
was
fearing,
because
I
think
some
of
the
hop
by
hop
steering
mechanisms
that
we
have
wouldn't
allow
another
independent
hop
by
hop.
You
know
header
for
for
the
Qs,
so
the
combination
might
be
an
interesting
work
item
later
on.
C
I
was
thinking
that
that
I
think
that's
a
different
point.
I,
it
might
be
worth
having
a
list
of
the
steering
and
routing
mechanisms
that
we
care
about,
obviously,
is
a
great
deal
of
interest
in
Source
routing,
but
list
listing
all
of
them
or
a
representative
subset
of
them
and
a
requirement
that
could
replace
requirement.
Four
in
this
slide
would
be.
The
queuing
mechanism
ought
to
be
compatible
with,
with
with
all
of
them
not
specific
to
one
of
them.
B
I
Yes
to
to
David's
point,
there
was
this
Insider
of
using
the
currently
spent
latency
of
the
packet
in
the
queuing,
in
which
case
it
has
to
be
on
the
Hub
on
the
other
packet
headers
or
possibly
on
the
Hub.
You
want
to
Signal
how
much
latency
we
have
at
every
hub,
and
so
that
would
be
a
case
where
David's
point
is
obliged
that
it
cannot
be
a
state
in
the
network
but
has
to
be
a
state
in
the
back
end.
I
C
And
Pascal,
one
of
the
things
we're
trying
to
dance
around
here
and
I'll
admit
it's
it's.
An
interesting
dance
is
to
figure
out
the
queuing
mechanisms
without
committing
to
the
design
of
the
addition
of
the
headers
needed
to
carry
additional
per
packet
information.
The
queuing
mechanisms
require
so
it's
sort
of
an
interesting
dance.
Where
we
know
some
of
the
mechanisms
are
going
to
require
the
information
to
be
carried,
but
not
try
not
to
commit
to
try
to
figure
out
the
queuing
mechanisms
before
embarking
on
detailed
design
of
the
additional
headers
or
header
extensions.
G
B
Well,
I
I,
guess
that,
given
how
we
understand
the
problem,
maybe
it's
also
something
we
can
independently
start
to
to
figure
out
what
we
could
even
do
right
in
terms
of
saying
we
don't
exactly
know
what
the
fields
are
going
to
be,
but
go
to
six
men.
If
that's,
for
example,
one
of
the
working
groups
that
would
need
to
take
responsibility
later
on
anyhow-
and
you
know,
make
suggestions
or
try
to
get
guidance
that
way
we
wouldn't
need
to
fully
serialize
it,
but
we
could,
you
know,
probably
faster,
get
some.
C
If
we,
if
we
focus
on
what
has
to
be
carried
for
the
queuing
mechanisms,
I'm
not
sure
the
right
word
to
use
to
use
to
use
is
to
use
for
it.
It's
it's
it's
a
model
of
some
form
because
we're
not
specing
the
actual
data
field
that
we
carried
and
then,
as
you
say,
for
example,
for
for
for
a
V6
extension
header
we
go,
we
would
go
to
six
men
and
we'd
have
something
to
point
to
it
says
this
is
the
data
we
need
to
be
carried.
Please
help.
C
Okay,
so
ping
ping
and
zhangdong
and
everybody's
Community.
Thank
you.
It
sounds
like
you
need.
There
needs
to
be
a
little
more
discussion
among
the
authors
and
then
please
don't
wait
for
the
next
meeting.
Please
go
ahead
and
discuss
on
the
mailing
list
and
see
about
revive
revising
requirements.
Draft
based
on
that
discussion.
C
C
So
I
would
hope
we
can
get
the
requirements
draft
revised
and
maybe
for
the
next
meeting
the
requirements.
Draft
authors
could
pick
one
of
the
TSN
queuing,
queuing,
queuing,
scheduling
mechanisms
and
just
come
up
come
up
with
an
initial
draft
made
Draft
line
of
the
Matrix.
That
shows
just
that
shows
how
how
it
looks
against
the
requirements
in
the
requirements
draft.
So
the
expectation
is
is
that
by
and
large
you
won't
meet
them,
but
let's,
let's
at
least
get
started
with
something
that
is
known
and
hopefully
understood.
C
F
A
F
H
C
B
C
E
C
Xiaofu
go
ahead,
we've
got
at
least
an
hour
for
you,
so
what
I
guess
I
would
ask
is
for
folks,
let's
take.
Let's
take
clarifying
questions
only
during
the
presentation,
the
hopes
that
we
can
get
through
the
slides
and
then
feel
free
to
ask
any
questions
when
we
get
done
so
xiaofu
go
ahead
and
share,
slides
and
get
started.
J
C
C
C
J
C
C
J
J
Here
the
meaning
of
valencies,
including
the
following
Public
Health,
we
use
five
foot
to
replace
a
missed
for
each
of
a
hardware
implementation
and
without
requiring
code
package
to
be
sorted
as
frequently.
Instead,
multiple
keys
are
sorted
by
polarity
and
the
dynamically
located
the
polarity
on
the
introduce,
and
now
we
could
concern
behavior
for
keto
sensitive
service.
J
So
this
is
the
overview
of
2017.
The
idea
of
dominant
Queen
rule
is
so
simple,
which
is
to
maintain
tools
with
counted
on
type
attribute,
to
store
parties
with
the
related
residence
type
plus.
We
know
that
each
level,
as
counted
on
time,
that
is
decreased
by
rotation
interval
under
descending
duration,
is
18..
J
J
J
J
J
J
J
J
The
packet
that
made
in
including
including
summer
packages
with
the
same
delay
level
along
with
the
simultaneously
as
The
observed
packet
and
the
Summer
Park,
is,
with
higher
levels,
arrive
with
the
simultaneously
early
in
the
future
under
summer
practice
with
the
lower
delay
level.
But
the
arrived
more
earlier
than
the
Observer,
the
package
political,
the
next.
J
Therefore,
it's
important
that
the
the
actual
arrival
traffic
of
each
level
I,
don't
know
the
master
node
that
you
can
see
that
it
is
function
so
that
the
the
security
ability
condition
is
always
better.
Otherwise,
it
should
receive
a
traffic
to
obey
the
contribute
function,
all
reward
or
policy.
The
probability
of
the
traffic
by
latency
conversation
to
identify
the
legitimate
emergency
traffic.
J
J
Is
below
especially
when
the
bandwiders
of
each
level
is
negligible
due
to
large
parts
of
the
interval
and
the
small
parts
of
the
site,
then
the
maximum
positive
source
of
each
level
May
reach
the
full
value
I
hit
the
full
one
is
same:
multiple
80,
minus
f
in
general.
The
bandwiders
of
Summoners
cannot
be
negligible.
That,
of
course,
the
ambassador
of
the
source
of
the
no
the
low
polarity
level
cannot
reach
the
full
running.
J
J
So
this
is
an
example
for
the
extremely
case
that
the
world
killer
with
the
bandwiders
resource
are
made
any
zero
due
to
large,
but
the
interval
and
the
each
level
may
have
positive
resource
with
the
full
one.
So
if
concluding
the
past
of
water
laws
with
the
full
value,
arrived
the
first
one
first,
two
versus
three
and
so
on,
who
will
be
sending
one
by
one
in
seconds.
B
J
Yep,
for
example,
because
world
class,
the
battle
positive
resource
and
the
bundle
where
this
resource
of
world
class.
Where
you
meet
at
the
schedule,
we
need
a
condition
so
there's
an
extreme
cases.
Okay,
this
is
just
an
example.
There
is
an
extreme
case
that
the
bundle
by
the
same
resource
of
ET
deterministic
lower
is
a
proximate
zero,
for
example.
C
B
What
exactly
are
the
parameters?
I
thought
that,
for
example,
in
this
example,
you
said,
the
the
interval
between
the
births
is
one
year,
but
that
is
not
a
flow
parameter.
The
flow
parameter
is
still
explicit
rate.
That
would
only
allow
you
to
send
these
bursts
very
rarely
like
once
a
year.
So
you
still
need
that
bandwidth
right
to
express
the
distance.
True.
J
J
C
J
Okay,
we
we
see
the
example
two.
This
is
another
example
for
the
case,
the
the
some
delay
levels
abundant
by
the
resource
cannot
be
negligible.
For
example,
possible
resource
of
D1
is
a
four
value,
and
the
bundle
Y
is
the
source
of
D1
is
the
half
of
C.
We
were
the
seasonal
Source
rate
of
the
scheduler.
J
J
J
So
this
page
is
for
buffer
says
the
content.
Type
of
each
code
is
dynamically
changed,
so
it
is
not
bound
to
a
fixed
delay
level.
In
a
way,
you
actually
store
water
levels
of
traffic.
During
editor
CT
development
process,
the
following
Give
an
example:
we
will
see
that
at
the
t0
t0
plus
one
10,
microseconds
t
0
Plus
90
seconds.
J
J
J
So
this
is
an
example
that
the
several
parts
from
the
same
flow
arrived
on
the
Ingress
P.
According
to
the
admission
control,
some
of
us,
for
example,
positive
2
4
5
than
others,
we
should
be
no,
no,
we
don't
know
the
the
latency.
Conversation
must
be
based
on
the
shape
of
the
traffic
instead
of
the
original
traffic.
C
J
J
C
So
careful
what's
commenting
on
is,
let's
take
a
look
at
burst
number
three,
just
because
of
the
way
you've
laid
the
slide
out.
The
the
number
three
on
the
received
line
is
to
the
right
of
the
number
three
on
the
sending
line.
Now,
clearly
that
that's
not
what
happened,
the
packet
has
to
have
been
received
before
it
was
sent,
but
it
does
make
the
slide
a
little
difficult
to
understand.
J
J
C
J
J
J
J
J
By
Leading
the
conversation
with
a
smaller
currently,
however,
the
CT
running
of
the
coup
is
not
declaration
in
real
time,
so
that
will
impact
to
one
or
two
will
set
it
to
the
Clue.
The
city
one
is
seen
by
Target.
2
is
the
same
as
dedicine
by
power
one,
so
the
product
2
may
put
it
to
the
clue
before
the
Target
one,
so
that
is
the
disorder
happens,
so
how
to
adjust
this
way.
We
will
give
you
two
options.
J
J
So
this
page
is
the
partially
upgraded
simulation
in
partially
upgrade
the
case.
Several
border
nodes
are
upgraded
to
support
the
deadline
mechanism
other
than
that
path
consists
of
several
application.
Border
nodes,
for
example
r1234
and
traditional
SP
domain
SP
passes,
should
contain
few
hubs
with
the
bounded
latency
behold.
J
J
B
J
Yeah
there's
a
steady
in
the
Parkade
in
current
fees
we
introduce
a
local
officer
at
the
time
that
is
carved
in
the
packet
to
indicate
each
node
for
this
gardening.
For
the
dominance
good
evening,
the
the
local
offset
timer
is
a
warranty
type
that
is
applied
to
each
node
as
I
described
before
it
is
computed
by
the
under
to
under
delay.
Minus
the
the
accumulated
link,
propagation
delay,
10
divided
by
The
Hope
accounts
to
get
the
ability
delay.
A
C
So
I
think
the
crucial
word
there
was
relative
shafu.
Let
me
check
my
understanding.
You
don't
need
to
do
time,
synchronous
to
do
wall,
clock
time
synchronization,
but
you
are,
depending
on
on
on
on
limited
clock,
skew
so
that
the
routers
May
one
router,
may
think
it's
12
a.m,
the
other
another
Rider
may
think
it's
5
p.m,
but
as
long
as
they
both
advance
so
that
they
both
Advance
five
minutes
in
the
same
five
minutes
you
need
that
is
that
correct.
C
K
K
C
K
J
K
Okay,
I,
don't
think
I
have
fully
understood
your
answer,
but
okay,
we
can
discuss
it
offline.
My
another
question
is
that
yeah.
As
far
as
I
understand,
there
is
a
very
well
known
scheme
called
RC
EDF,
that's
a
rate
controlled,
always
deadline.
First,
we
just
first
introduced
by
the
Ferrari
yeah
and
in
that
scheduling
scheme
they
combined
two
things:
the
the
terrific
Schaefer
for
each
flow
and
then
all
this
deadline
for
a
scheduler.
K
H
B
K
B
That
that
you
don't
have
the
per
Flow
State
that
you
need
to
read,
calculate
and
update
on
every
hop
right,
the
shape
even
The,
interleaved
Regulators.
They
they
have
per
Flow
State
with
two
variables
that
they
need
to
update
on
on
every
packet
arriving,
read
calculate
right.
Then
you
have
a
deadline
effectively
from
from
which
you
send.
H
B
J
J
K
Yeah,
just
just
one
more
question
is
that
the
the
shaper
and
the
scheduler
can
be
combined.
Those
are
those
are
the
known
facts
that
I
understand,
but
in
in
this
implementation,
the
the
because
of
those
the
shaping
aspect
of
the
scheduler,
the
normal
conserving
behavior,
is
natural.
To
me:
can
can
you
make
it
or
conserving,
even
with
the
the
latency
compensation
algorithm.
J
K
D
J
J
J
K
B
J
B
Well,
I
was
Prime,
yes,
I
mean
tcqf
is
a
very
simple
calculus
right.
So
in
this
case
here
I'm
I'm
still
not
sure
that
I
fully
understood
the
calculus.
So
I
was
wondering
where
the
proof
for
the
calculus
was
done.
Was
that
done
by
you
for
this
work,
or
was
that
done
earlier?
In
some
other
research
paper.
J
B
B
Also
right
so,
the
deadline
that
is
in
the
packet
will
need
to
have
some
resolution.
It
could
be
a
micro
second
or
whatever.
The
unit
of
resolution
is
and
then
what
is
the
maximum
size
that
that
it
could
have
right.
So
it
would
be
good
to
to
get
some
example,
because
when
we
want
to
design
a
header,
it
would
be
good
to
understand
how
what
size
of
data
we
would
need
for
a
long
path
with
menu
hops.
To
do
this
algorithm.
J
J
B
So
wait
a
second,
so
in
one
mode,
The
Hop
gets
a
deadline,
which
is
the
minimum
a
relative
time,
and
the
packet
must
not
be
sent
before
that
and
in
the
other
mode
there
is
a
maximum
deadline
and
you
for
a
particular
hop.
You
only
need
to
signal
one
of
the
two
values.
Do
you
understand
that
correctly?
No.
B
J
J
B
J
C
Is
that
there
is
a
deadline,
that's
the
D
in
this
figure
and
then,
if
the
packet
arrives
well
before
that
deadline,
the
the
the
mode
choice
at
the
node
is
whether
to
opportunistically,
send
it
onward
well
before
or
try
to
delay
it
and
send
it
just
before
the
deadline.
That's
the
difference
between
in
time
and
on
time,
but.
B
K
C
C
Relative
to
the
fact
that
the
clocks,
Don't
Drift
I,
think,
is
what's
going
on
here,
you
can
have
the
routers
independent
this
independently
believe
it's
one
o'clock,
two
o'clock
and
three
o'clock,
but
as
long
as
they
all
advance
to
to
one
minute
after
the
hour,
two
minutes
after
the
hour
three
minutes
after
the
hour,
because
they
because
they
don't
drift
this.
This
should
all.
B
B
F
J
C
A
quick
comment
for
the
requirements:
draft
I
think
the
material
requirements
draft
on
synchronization
ought
to
capture
not
just
clock,
synchronization
or
lack
there,
or
the
requirement
or
ABS
requirement
for
clock
synchronization,
but
also
we've
now
seen
two
of
these
mechanisms
that
don't
require
wall
clock
time
synchronization,
but
do
assume
that
there's
effectively
no
skew
between
in
the
advanced
rates
of
the
clocks
that
that
that
would
be
good
to
capture
in
the
requirements
to
wrap.
If
it's
not
already,
there.
C
Okay,
so
when
we
talk
about
clock,
synchronization,
we're
usually
talking
about
wall,
clock
time
synchronization
and
we
have
a
really
important
requirement
in
Aquarius
draft.
That
says
we
don't
require
wall,
clock
time
synchronization
and
what
we've
seen
now
is
that
clock,
skew
or
rate
of
advance
needs
to
be
aligned
for
the
two
mechanisms
we've
seen
so
far,
that
don't
do
the
wall
that
don't
need
the
wall
clock
time
synchronization
they
do
need.
They
do
need
the
rates
to
match.
The
rates
are
Advanced
to
match.
C
I
K
C
C
Yeah
and
and
I
think
this
algorithm
and
and
the
one
that
we
saw
from
engineer,
are
both
depending
on
that
and
and
you
you
do
this
for
the
you
do
this
with
high
quality
crystal
oscillators,
for
example,
if
you
can
get
them
to
independently
Advance
the
same
rate
by
by
carefully
controlling.
What's
going
on
there.
B
So
the
way
I
understand
clock
synchronization
is
that
you
know
your
wall
can
be
arbitrary,
but
clock
synchronization
actually
means
that
over
arbitrary
periods
of
time,
the
difference
between
these
two
clocks
that
you're,
comparing
that
you're
synchronizing
against
each
other,
will
not
differ
more
than
a
maximum.
That's
the
so-called
maximum
time
interval
error.
Mtie!
That's
that's
clock
synchronization.