►
From YouTube: IETF114 RTGWG 20220728 1400
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
A
Yeah,
I'm
not
I'm
not
on
I'm
not
on
the
cisco
vpn,
so
it
should
be
just
direct
connect
over
google
fiber
to
to
the
fastest
link
that
they
have.
C
C
Yeah
so
welcome
to
rtcwt
session,
I'm
yunjin
chi
and
I
have
my
co-chair
jeff
tanzera
online.
So
we'll
start
the
meeting
now
and
we
have
a
very
tight
schedule.
So
we'll
try
to
keep
the
agenda
so
people
speak.
Please
try
to
keep
your
own
allocated
slot
jeff.
You
want
to
start
a
chair
slice.
B
B
B
Please
sign
in
intercession
with
media
light
client.
Please
use
mit
echo
to
join
a
microphone
key.
Please
shout
audio
and
video
off.
If
you're,
not
using
online
version.
You
see
we
have
issues
with
bandwidth,
so
save
bandwidth
and,
most
importantly,
wear
mask
unless
you're
actively
speaking
the
microphone
given
issues
with
microphones.
Please
come
as
close
to
mike
as
you
can
remote
participants
again.
Please
shut
your
audio
in
video
unless
you
are
speaking
and
we
highly
recommended
to
use
headset.
B
Idf
policies-
please
reader
fc
8179,
so
you
know
them
specifically
to
routine
working
group.
We
will
call
for
ipr
every
time
draft
progresses
from
one
stage
to
another,
and
unless
each
and
every
author
and
co-author
every
spotted
document
will
not
progress.
B
Existing
documents
we
closed
working
group
plus
call
on
reapex
standard.
It's
on
me
to
progress
document.
Further
bgp
peaks
is
pending
ad
review.
There's
some
more
going
on.
We
are
trying
to
reconnect
with
stefan
to
work
on
it,
atnbgp
some
reviews
being
done.
We'll
start
working
group
plus
calls
soon
it'll
face
still
being
reviewed
and
qs
model
will
get
updated
today.
B
The
agenda
for
today
is
we'll
get
updates
on
working
group
documents.
Young
model,
dynamic,
networking
and
ac
will
talk
about
updates
to
vrp.
We
will
have
a
revit
working
group
document,
thanks
jen
to
start
working
on
it
again.
If
the
destination
starts.
Routing
we've
got
great
research
topic
on
framework
for
evaluation
of
performance
of
routing
protocols,
really
interesting
work
and
actually
the
mod
been
doing
really
a
lot
of
research.
And
hopefully
we
will
see
him
not
only
today,
but
going
forward
and
there's
a
lot
of
new
work.
You'll
see.
B
And
we
are
planning
at
least
have
one
and
three,
maybe
more
mid-september
so
stay
tuned.
C
C
C
Okay,
maybe
we
do
the
second
one.
C
D
C
E
E
So
there
we
haven't
been
here
for
a
while,
so
we
added
a
co-author
from
azure
microsoft
and
he
brought
in
a
lot
more
problems,
and
so
we
try
to
focus
the
problems
to
the
routing
area,
the
the
problems
which
can
be
addressed
by
the
routing
area.
There
are
many
many
problems
in
in
connecting
to
the
cloud,
so
we're
not
being
be
able
to
solve
all
of
them.
Many
of
them
are
gravity
problems
just
not
being
able
to
solve
by
us.
So
we
also
revise
the
gap
analysis
accordingly,
so
here
on
this
next
slide,
please.
E
Sorry,
it's
really
hard.
I
can
take
this
off
right
when
I
talk
okay,
so
well.
The
big
problem
in
cloud
data
center
is
the
workload
in
the
cloud.
Data
center
are
connected
to
many
many
peers,
many
parties.
So
one
of
the
problems
experience
is
the
pgp
mismatch
configuration
arrows.
E
You
have
parties
in
different
places
like
configuring,
the
the
bgp
capability,
there's
a
mismatch
and
there's
also
notification
issues
and
causing
one
party
to
drop.
E
If
you
have
the
workload
in
the
cloud
and
in
the
on-prem,
you
want
to
create
a
vpn,
you
want
to
do
the
bgp
connection
like
how
do
you
create
the
bgp
consistent
configuration
on
both
sides
like
bgp
over
ipsa
configuration?
That's
the
issue.
They
cannot
be
able
to
do
it
very
consistently.
E
Second
issue
is
when
the
cloud
gate
gateway
data
center
is
connecting
to
many
sites
or
pods
pods
go
sometimes
go
aggregate
degradation
or
go
dark.
Actually,
it's
not
uncommon
that
there's
fiber
cut
or
there's
a
power
outage
or
there's
overload
and
causing
one
side
to
go
dark,
and
with
that,
when
that
happens,
large
number
of
routes
will
be
need
to
be
switched
by
the
ingress
router.
E
So,
with
today's
sending
route
withdrawal
message,
there'll
be
large
number
of
messages,
so
they
will
need
a
way
to
be
able
to
notify
to
the
ingress
router
of
a
large
number
of
routes
being
impacted
or
need
to
be
switched.
So
that's
another
problem.
We
recently
added
to
the
the
draft
next
page.
Please.
E
Another
issue
is
the
source,
the
user
device
ue
when
they
move
from
one
cell
tower
to
another
cell
tower.
They
may
maintain
the
same
ip
address,
so
you
will
encounter
of
same
source
address
actually
coming
from
different
ingress
routers.
So
that's
another
problem
added
into
the
the
cloud
network
connecting
the
cloud
next
page.
Please.
E
So
another
issue
is
when
you
have
hybrid
cloud
right,
you
have
instances
cloud
extension
by,
I
mean
hosted
by
different
cloud
providers,
and
different
providers
may
have
different
net
functions.
So
when
you
have
a
cpe
connecting
to
to
different
sites
managed
by
different
cloud
operators,
the
net
can
be
different.
Function
can
be
different.
So
when
we
establish
a
connection
like,
for
example,
establish
an
sd1
connection,
you
will
need
to
be
notified.
E
The
net
feature
like
what
kind
of
net
is
support,
so
that's
something
another
problem
which
haven't
been
defined.
Yet
another
thing
is
discovery.
Okay,
so
you
have
instances
extension
in
multiple
locations
and
how
do
you
quickly
be
able
to
identify
what
are
the
possible
locations
where
your
instances
are
located?
E
That
will
help
the
the
on-prem
router
to
be
able
to
efficiently
or
quickly
switch
the
service
from
one
place
to
another
place
I'll
be
able
to
optimally,
choose
the
the
most
optimal
location
for
the
intended
services
next
one
and
then
the
dns.
Well,
this
not
intended
to
solve
dns
problem,
but
in
the
context
of
routing
there
are
lots
of
resources.
You
extension
in
different
cloud
operators
and
your
service.
E
The
resource
are
local
to
particular
enterprise,
so
your
service,
like
your
load
in
cloud,
a
may,
need
to
resolve,
address
which
is
not
visible
or
not
located
in
that
particular
cloud
operator.
So
the
dns
you
have
to
set
up
the
policy,
so
this
dns
query
will
be
routed
back
to
your
on-prem's
dns
and
then
on-prem
dns
will
be
able
to
locate.
Where
is
the
next
cloud
operator,
where
the
actual
instance
is
so
from
the
routing
perspective?
How
do
we
make
the
instance
the
services
in
one
cloud
talk
to
another
cloud?
E
Okay,
another
issue
with
dns
is
the
client
itself
may
cache
the
dns
query.
So
if,
like
say,
for
example,
your
today's
instance
is
no
longer
like
a
single
connection,
you
connect
to
a
to
b.
The
b
you
want
to
connect
to
is
usually
located
in
multiple
places.
So
when
one
place,
one
location
on
side
part
is
having
a
degradation.
For
example.
How
do
you
tell
the
client
to
switch
to
to
the
next
one?
E
So
if
the
client
keep
the
cash
make
it
very
not
very
convenient
for
the
client
to
be
able
to
initiate
the
switch.
So
that's
why,
when
we
have
those
router
at
the
the
ingress
router,
if
they
get
notified
instead
of
client,
then
they
will
be
able
to
switch
the
service
to
the
next
site
more
optimally.
E
Next
one
and
then
here's
the
gap
analysis.
So
with
those
problems,
this
gap
analysis
we
made
some
changes.
One
thing
is
about.
E
Many
forwarding
is
based
on,
even
though
they
use
ip
address.
The
ip
address
is
not
quite
the
same
as
before,
because
each
ip
address
may
represent
a
cluster
of
instances
like,
for
example,
if
you
want
to
do
some
kind
of
application
based
routing
the
use
ip
address,
but
the
ipod
address
represents
many
instances
and
they
are
located
in
many
places.
E
We
need
a
simpler
or
simpler
way
to
represent
a
large
number
of
services
when
there's
a
site
failure.
Actually
that
one
has
been
covered
before
also
that
you
may
have
different
connections,
you
may
have
the
vpn
nprs
connections
and
you
may
have
the
connection
over
the
public
internet,
so
the
instance
can
be
connected
in
many
different
ways.
E
Next
place,
here's
just
another
summary
for
the
routing
area,
so
the
the
the
past
management
right.
So
when
there's
virtual
cpe
extension
in
the
cloud
can
be
reached
by
multiple
paths
from
the
routing
domain,
how
do
we
optimally
select
the
optimal
egress
router
to
reach
to
the
optimal
location?
E
E
And
also
that
we
need
the
in
the
control
plan
right,
we
need
to
advertise
advertise
the
network
condition
connecting
to
the
virtual
nodes
within
the
cloud
dc,
so
there's
also
issues
of
when
we
aggregate
traffic
over
the
private
path
and
over
the
internet
path,
and
there
will
be
when
you
have
the
port
facing
the
internet,
there
will
be
security
issues
and,
and
also
also,
how
do
you
indicate
a
different
kind
of
path
available
for
transporting
in
today
in
the
bgp
tunnel
in
cab,
doesn't
really
indicate
that
one
client
route
can
be
aggregated
over
public
internet
or
over
private
or
both
or
give
some
kind
of
policy.
E
So
that's
the
next
page,
please
so
so
this
draft
has
been
two
drafts
have
been
in
the
rtg
working
group
for
a
while
and
I'm
hoping
that
people
can
give
some
more
reviews,
and
we
want
to
call
for
working
group
last
call.
Thank
you.
Any
questions.
F
Hi
albertana
speaking
as
a
working
group
participant
thank
you
for
doing
this
and
for
documenting
the
requirements
and
the
use
cases
and
the
gaps
and
everything
else.
If
we
understood
you
at
the
beginning,
you
said
that
some
you
added
some
new
information.
F
I
understood
you
at
the
beginning.
You
said
that
you
had
added
some
new
information
to
the
drafts.
Yes
right,
so
what
I
would
like
to
suggest
to
the
working
group
is
that
you
know
documents
like
these.
I
think
I
think,
are
very
important,
because
they're
going
to
guide
work
on
actual
solutions,
I
know
there's
an
edge
discovery
draft
in
idr
that
references,
the
the
problem
statement
and
the
gap
analysis.
F
F
So
what
I
would
like
to
suggest
again
as
a
working
participant
is
that
we
maintain
these
documents
as
working
documents
in
the
working
group.
You
know
maintained
periodically
so
that
they're,
not
you,
know,
frosting
in
time
today,
right
or
whenever
we
get
to
publication,
so
they
can
be.
They
can
remain
useful
for
solution,
work.
E
Yeah,
okay,
so
the
the
key
information
we
added
in
is
in
the
first
two
pages.
Is
the
bgp
mismatch
when
the
party
is
peering
with
many
more
peers?
It's
not
like
traditional
bgp,
you,
which
you
have
pre-agreed
policy
with
your
partner.
You
may
have
negotiated
policy,
and
here
when
the
cloud
gateway
established
bgp
there
will
be
many
more
and
so
another
one
is
the
number
of
actually
that
other
one
is
partially
covered.
E
But
we
enhanced
that
to
make
it
more
formal
formally
stated
like
when
you
have
a
side,
a
part
goes
dark
or
degradated.
How
do
we
indicate
to
the
the
ingress
router
to
do
the
efficient
switching?
So
I.
G
One
question
regarding
a
slide:
probably
the
third
all.
Fourth,
one
is
showing
a
picture
going
from
microsoft,
error
going
through
public
and
the
provider
network
yeah
for
this
case
actually
there's
another
one
is
instead
of
the
provider
a
public,
there
is
a
circus
and
the
cloud
provider
will
have
its
own
backup.
H
A
Okay,
I'm
ac
lindham
and,
along
with
aditya
dogra,
we're
doing
the
vrp
version
three
viz
document
that
that's
currently
rfc
57
on
98..
A
I
just
wanted
to
say
that
a
lot
of
times
you
know
you
have
people,
you
know
you
want
the
right
people
to
do
these
biz
documents
and
aditya
is
currently
the
lead
for
the
for
vrv
versions.
Three
and
two
on
the
ios
xc,
cisco,
xc,
family
of
networking
products
so
he's
qualified,
and
as
for
myself,
I've
implemented
vrp
version
2
at
two
different
vendors
and
for
vr
version.
A
Three,
I
put
proprietary
constraints
into
it
for
sd-wan
integration,
for
instance,
whether
or
not
you
have
an
sd-wan
connection,
you
have
a
connection
to
control
or
if
you
don't,
you
don't
want
to
allow
yourself
to
remaster
state,
so
I
think
we're
uniquely
qualified
to
do
this
next
slide.
A
Yep,
it
worked
okay,
primary
motivation,
for
this
is
not
only
well
the
first
one
is.
The
ietf
has
adopted
the
nist
standards
for
inclusive
language,
which
some
and
those
of
you
familiar
with
the
vrfe
protocol
know
that
it
uses
master
and
that
just
about
changed
every
page
and
at
the
same
time,
cisco
and
other
vendors.
A
We
really
want
to
change
all
our
our
our
code
and
documentation
to
also
have
inclusive
language.
A
matter
of
fact.
If
you
go
to
a
cisco
document
now,
you'll
see
on
the
first,
you
know
an
online
documentation,
you'll
see
our
commitment
to
inclusive,
using
inclusive
language
and
all
our
documentation.
A
A
You
know
the
the
the
concept
of
master
and
backup
was
so
so
much
part
of
erp
that
it
changed
every
almost
every
other
paragraph,
so
we
went
ahead
and
did
divis
and
we
came
up
with
active
because
you
know
I.
I
know
that
in
other
aha
type
protocols
and
standards,
the
natural
corollary
of
backup
is
active.
A
Now
it
was
less
and
and
as
far
as
the
standard
is
concerned,
we
figured
well.
We
might
as
well
back
up
back
up
if
we
can
and
have
the
least
intrusive
change.
So
that's
why
we
did
that.
We
also
changed
some
instances
of
black
hole
to
talk
about
advertising
reachability,
yet
dropping
the
packets,
so
that
that
was
that
was.
That
was
the
main
impetus
and
we
probably
wouldn't
have
dubbed
is
of
rfc
5798
without
this
motivation.
Next
slide.
A
I
have
control
I'll
blame
it
on
kovac
additional
changes.
First
of
all,
I'm
I
migrated
all
to
xml
version
rfc
xml
to
rfc
version
three,
so
it
uses
all
the
new
tags
and
everything,
and
that
seemed
to
just
do.
That
sounds
like
an
internal
change,
but
it
seemed
to
result
in
better
formatting.
Just
by
making
that
switch,
there
were
a
number
of
er
errata
open
on
all
the
finite
state
machines
and
we
fixed
all
of
those
and
then
I
went
through
and
I
did
editorial
changes.
A
There
were
some.
There
were
actually
some
spelling
errors
and
some
lots
of
sentence,
fragments
and
run-on
sentences,
and
I
tried
to
get
rid
of
those,
and
I
also
tried
to
use
cons.
There
was
also
some
inconsistencies
in
the
use,
just
in
the
in
just
a
changing
master
everywhere
I
saw
there
was
inconsistencies
of
when
it
was.
You
know
how
it
was
phrased
and
whether
it
was
master
router
with
both
capitalized
or
one
capitalized
or
everything
I
made
it
consistent
and,
and
so
that
was
so
there's
a
lot
of.
A
A
Also
thanks
to
a
comment
from
pascal
touver,
we
now
recommend
for
ipv6
that
both
the
master
and
backup
accept
unsolicited
neighbor
ad
advertisements,
as
is
recommended
in
rfc
9131.
A
This
was
this
is
a
recent
change.
This
was
just
changed
recently
in
in
received
we
now
rec.
We
now
recommend
that
validation
is
done
on
the
maximum
interval
so
that
it
matches
when
you
receive
vrp
packets
for
background
compatibility.
We
not
discard
the
packets
because
in
in
rfc
5798
it
doesn't
do
that,
but
the
mismatch
is
logged.
A
We
removed.
There
was
a
lot
there
was.
There
are
dependencies
that
talk
about
fddi,
token
ring
and
tm-line
emulation
those
are
removed.
We
kind
of
cisco
has
pretty
much
put
all
those
products,
all
those
technologies
and
the
end
of
life.
It
didn't
seem
like
it
made
this
it
was.
It
was
necessary
to
carry
these
appendices
into
the
future.
A
I
beat
up
the
security
considerations
and
we'll
probably
as
we
go
through
the
a
d-
and
I
see
iesg
reviews,
I'm
sure,
we'll
beef
them
up
even
more,
and
thanks
to
some
comments
from
donald
slake
and
I
improved
the
iana
section
next
slide.
A
Okay,
what
I
want
to
do
is
get
some
additional
reviews,
like
I
said,
I've
gotten
some
good
reviews
from
drew
pascal
and
others
like
donald.
You
can
look
at
and
by
the
way,
anybody
who
acknowledge
who
did
add.
It
gave
me
a
substantive
comment.
I
put
their
name
in
the
acknowledgements
and,
and
so
I'm
looking
for
some
more
good
reviews.
A
It's
a
working
group
document
now
so
we'll
give
it
a
little
time
to
settle
and
then
I'll
probably
start
pushing
to
see
how
ready
we
are
for
a
working
work.
Group
last
call
I'm
also
gonna
aditya
and
I
are
also
gonna
do
both
of
us.
You
know
like
take.
A
You
know
a
couple
hours
and
both
of
us
go
through
it
line
by
line
and
look
for
any
inconsistencies
as
far
as
taking
brp
version
3
to
propose
standards.
I'm
not
going
to
do
this
reason.
This
I
get
at
this
point.
I
get
asked
to
do
a
lot
more
than
I
have
time
to
do,
and
this
just
isn't
a
priority
to
me.
So
I'm
not
going
to
do
this
implementation
survey
and
report
to
get
it.
A
I
mean
consider
moving
to
internet
standard,
I'm
not
going
to
do
this
if
somebody
else
is
interested
in
doing
this,
that
would
be
welcome
anyway
and
then
working
group.
So
the
next
step
would
be
working
group
class
call
and
publish
so
that
vendors
are
comfortable,
making
code
changes
and
documentation
changes.
A
Oh,
I
was
gonna,
I
was
gonna,
say
well,
I'm
not
interested
in
this
I'll
probably
start
work
once
I
get
like
a
weekend
or
something
work
on
the
yang
document
start
looking
at
revisions
to
that,
so
that
we
can
change
the
terminology
in
the
end
document
and
I'm
hoping
I
did
yeah.
I
don't
want
to
speak
for
him.
He
might
tackle
the
mib
because
we
have
an
implement
at
cisco.
We
have
an
implementation
of
the
mib
as
well.
F
Just
a
couple
of
comments
on
on
the
process
itself:
it
is
not
a
requirement
to
publish
an
implementation
report
to
move
to
internal
standard.
So
it's
nice
that
the
working
group
discusses
the
fact
that
this
is
deployed
and
used,
and
you
know
et
cetera,
but
as
we
used
to
do
years
ago,
where
we
had
to
publish
line
by
line
each
must
and
should
and
everything
else,
we
don't
need
to
do
that
anymore.
The
main
thing,
however,
is
for
the
changes
from
proposed
standard
to
internal
standard
to
be
minimum.
F
F
If
we
can
justify
that
those
are
minimal
minimum
changes,
then
it
shouldn't
be
a
problem
as
long
as
it's
documented
in
the
shepherd
report,
for
example,
that
there's
a
pervasive
implementation
and
you
know
etc,
so
just
keep
an
eye
on
that.
Not
don't
make
too
many
changes,
especially
technically
technical
changes,
so
that
if,
if
the
working
group
wants
to
move
this
winter
standard,
then
there
are
no
there's
no
opposition
to
that.
F
A
Yeah,
maybe
I'll
start
a
thread
on
that,
if
we
don't
have
to
do
that
like
I
am
like,
I
said,
every
every
other,
every
other
paragraph
is
changed
because
of
the
the
editorial
changes,
but
I
I
know
exactly
what's
changed
and
everything
I've
changed,
there's
nothing,
that's
not
backward
compatible
and
it's
mainly
just
recommendations
that
are
that
are
ad
that
are
changed
technically.
A
C
So
I
assume
you
want
to
try
again
yeah.
I
Okay,
awesome.
I
believe
I
had
some
browser
problems.
I
believe
now
it's
working
fine.
I
have
the
control
right.
I
Awesome:
okay,
hello,
hello,
everyone:
my
name
is
asim
chaudhry
and
I
am
going
to
provide
updates
on
qscan
model.
I
I
We
have
reduced
the
number
of
feature
statement
from
19
to
5.
By
doing
that,
we
have
simplified
the
model
to
a
great
extent.
Also.
We
have
kept
only
those
feature,
statements
which
are
very,
very
specific
to
particular
vendors
and
are
not
part
of
any
particular
deployments,
typically
typical
deployments
going
to
the
next
one.
So
we
have
also
worked
on
reducing
the
number
of
modules,
as
I
will
describe
in
the
later
slide,
how
we
have
achieved
that
and
what
are
the
advantage
of
doing
that?
I
Also
in
the
as
promised
in
the
last
itf,
we
have
added
the
operational
model
to
the
to
the
young
model.
We
felt
very
very
strongly
that,
as
part
of
the
configuration
model
for
the
qs,
we
should
also
add
the
stats
counters
for
it
to
be
able
to
complete,
and
so
we
have
worked
on
that
to
add
the
counters.
I
I
Okay,
so
if
we
look
into
our
current
qscan
modules,
we
have
combined
the
qs
policy
classifier
and
the
target
module
into
a
single
module
and
we
call
it
as
a
traffic
policy
module.
I
Now
the
advantage
of
having
traffic
policy
module
is
that
any
particular
traffic
related
policy
can
make
use
of
this
module
as
a
base
module
and
can
augment
or
create
their
model
of
their
own
in
a
very,
very
simple
steps,
as
I
will
describe
in
the
next
slide.
Now
all
these
blue
boxes,
like
qa,
section
the
traffic
policy.
I
These
are
base
module
and
the
orange
boxes
are
derived
modules
so,
for
example,
the
diffser
policy
we
have
created
by
sticking
together,
qs
action
with
the
traffic
policy
and
adding
some
kind
of
a
match
criterias.
Similarly,
for
queue
policy
and
scheduler
policy,
we
are
able
to
come
up
with
a
simple
modules.
I
Okay,
so
if
somebody
has
to
create
a
new
policy
module,
they
can
easily
make
use
of
the
traffic
policy
model
which
we
have
added
as
part
of
this
draft.
So
these
are
the
five
simple
steps
which
somebody
has
to
follow.
I
For
example,
you
know
somebody
has
to
create
a
new
policy
module
of
the
type
policy
forwarding,
so
first
they
have
to
define
a
new
policy
type.
Second,
they
have
to
define
the
classification
criteria
like,
for
example,
we
have
given
a
flow
tag
criteria
here
and
then
next
step
will
be
to
come
up
with
a
set
of
actions
like
a
next
hop
action
here
and
then
number
four
number.
Five
steps
will
be
augmenting
those
classification
and
the
action
criterias
to
the
traffic
policy
module,
and
that
is
pretty
much
it.
I
Also,
to
describe
a
little
bit
about
the
operational
modules,
so
the
operation
module
in
cos
consists
of
a
cos
interface
stats
container,
which
is
augmented
to
the
itf
interface
module.
I
It
has
per
direction
qs
counters
and
it
consists
of
tagged
and
untagged
counters
now.
The
tag
counter
also
called
name
counters-
are
either
of
the
aggregated
counters
or
the
non-aggregated
counters.
So
whenever
the
counters
can
be
combined
together,
these
are
aggregated
counters
and
when
they
cannot
be
counted
aggregated
together,
but
are
tagged
with
the
same
name.
For
example,
we
have
classification,
sorry
queuing
as
well
as
the
metering
counter.
They
are
tagged
together,
but
cannot
be
combined.
I
I
So,
as
a
next
step
to
the
draft,
we
will
be
working
on
the
remaining
yank
doctor
comment
and
hopefully
we
should
be
able
to
take
care
of
most
of
the
comment
by
the
next
itf.
C
So
we
don't
have
people
in
the
queue.
I
think.
I
I
J
Good
morning,
I'm
jen-
and
I
was
changing
here
back
in
2016
talking
about
enterprise
multi-homing
can
ipv6
and
how
destination
source
routing
is
supposed
to
help
us
with
that
and
recently
I
got
a
strange
feeling
that
nice
and
changed
in
the
last
six
years,
so
I
talked
to
david
and
next
slide,
please
so
yeah.
Apparently
we
did
have
a
draft
about
changes
to
the
routing
which
would
allow
the
routers
to
take
the
source
addresses
of
packets
into
account
while
making
routing
decisions,
and
the
draft
was
actually
making.
I
think,
reasonably
good
progress
and
authors.
J
Even
the
authors
even
requested
a
working
group
last
call,
but
I
did
not
find
any
evidence
that
it
happened.
So
the
draft
like
died
peacefully
like
three
years
ago
and
now
I
have
some
interest
in
making
that
stuff
finally
happen.
So
we
revived
the
draft
recently,
basically
area
submitted
essay
trees
with
just
starting
two
more
authors.
J
J
Actually,
multiple
providers
address
space
right,
so
your
host
might
pick
up
source
address
from
one
of
isps
and
obviously
that
packet
needs
to
be
routed
to
the
right
uplink.
So
we
can
adjust
route
based
on
destination.
We
need
to
take
source
into
account
as
well
similar
cases.
It's
like
traffic
engineering
or
traffic
segmentations
in
network,
like
enterprises,
when
you
have
something
like
world
garden,
and
you
want
some
particular
destinations
to
be
reached
through
different
paths
and
return.
Traffic
needs
to
be
routed
accordingly,
again,
an
ipv4.
Quite
often
it
was
sold
with
not
in
ipv6.
J
J
So
there
are
two
or
three
key
things
about
what
the
draft
proposes.
First
of
all,
the
draft
defines
the
expected
behavior.
The
specific
algorithm
of
particular
implementation
can
be
different
as
long
as
the
resulting
output
of
that
black
box
is
exactly
the
same
right,
so
we
defined
what
we
expect,
what
the
routing
decision
is
expected
to
be.
It
doesn't
matter
how
you
get
there
so
also.
J
The
idea
is
to
first
select
look
at
the
destinations
like
traditional
routers
do
and
then,
after
the
router
selects
set
of
particular
routes
which
can
be
used
to
route
that
package,
the
router
could
also
look
at
sources
and
because,
obviously
we
need
backwards
compatibility
and
we
need
to
deal
with
situation
when
we
do
not
care
about
sources.
The
idea
is
that
source
prefix
of
zero
zero
should
be
basically
should
behave
exactly
the
same
as
we
have
routes
with
no
sources
again,
the
draft
doesn't
care
how
this
is
achieved.
J
As
long
as
the
behavior
is
the
same
next
slide,
please
so
yeah.
I
do
not
want
to
go
into
the
details,
because
I
want
this
presentation
to
be
teasers,
so
you
got
curious
and
read
the
draft
but
like
in
simplified
description
of
what's
happened.
Packet
arrives
to
the
router.
The
router
is
supposed
to
find
the
most
specific
destination
route,
no,
nothing
new
here
right.
J
So
if,
for
example,
we
have
zero
zero,
slash,
32
48
and
slash
64
out,
let's
be
couple
routes
which
are
slash
64.
and
after
that,
if
it's
and
after
that,
I
will
look
at
most
the
most
specific
source
route
right.
It
might
be
that
there
is
no
route
much
in
the
given
source
like
on
this
slide
right.
The
packet
source
does
not
match
anything.
J
We
have
in
our
selected
route
set
in
this
case.
We
repeat
the
process,
selecting
less
the
next
next
most
specific
destination
route.
Until
eventually
you
get
the
route
which
you
can
use.
I
guess
in
the
worst
case
scenario,
it's
going
to
be
destination,
zero,
zero
source,
zero
0,
but
maybe
not
so
next
slide,
please
yeah!
So,
basically,
why
I'm
here
I'd
like
people
to
read
the
draft
and
I'd
like
the
working
group
to
consider
re-adoption
call
for
this,
so
we
can
move
forward.
K
This
microphone
works,
I'm
adrian
farrell,
thanks
jen,
you
teased
me.
I
read
it
it's
good
to
have
this
documented
and
I
see
no
reason
why
the
working
group
shouldn't
pick
it
up
again.
K
K
Sorry,
additional
routing
information
that
is
going
to
get
carried
into
protocols
as
a
result
of
doing
this,
and
and
what
are
the
convergence
implications?
And
things
like
that-
and
I
think
that's
just
worth
thinking
about
as
you
go
along.
D
Hi
andrew
alston,
speaking
just
as
a
working
group
participant
jen,
so
I've
taken
a
very
brief
route
of
this
document
and
I
just
wanted
to
find
out
from
the
authors
if
they'd
done
any
looking
at
what
this
type
of
routing
could
do
to
line
speed
in
the
forwarding,
plane,
etc,
and
whether
or
not
you've
you've
looked
at
if
there
are
any
implications
and
whether
this
can
be
done
at
line
rate
in
the
vendors
I'm
just
to
try
and
get
a
sense
of
where
we're
going
and
what
effects
this
could
have
on
hardware.
L
That
was
important
to
answer
that
I
did
actually
look
at
some
of
the
silicone
capabilities
and
it
varies
a
lot
for
some
silicon
can
do
it
and
they
generally
do
it
through
the
acl
engine,
which
is
more
limited
than
normal
forwarding
capabilities,
but
we
don't
have
an
actual
good
data
set
on
this.
This
is
basically
just
I
have
a
few
data
points
only
and
if
anyone
can
add
information
on
this,
I
would
greatly
appreciate
that.
J
Yeah,
obviously,
there
is
a
like
again
checking
the
net
problem
hard
to
find
how
fast
people
can
do
things
which
do
not
exist
and
you
don't
define
are
not
defined
yet
right,
so
yeah
any
feedback
from
vendors
hardware.
Vendors
is
really
welcome.
So
if
you
run
into
the
mic
telling
me
it's
not
possible
to
get
it
done
at
all,
like
yeah
I'd
like
to
hear
that
or
if
you're
gonna
tell
me
that
piece
of
cape,
we
can
do
this
I'll,
be
even
more
happy
to
hear
that.
M
So
darren
duke
cisco,
yeah,
it's
it's
varied
depending
on
the
hardware
for
the
type
of
hardware
you're.
Looking
at
it's
a
doable
thing.
A
Yeah
ac,
linda
cisco
systems,
so
I
notice
I
noticed
that
if
you're
doing
this,
you
you
you
do
it,
you
take
the
source
as
the
most
relevant
key
as
opposed
to
so
so.
If
this
is
this,
the
source,
the
source
routing
overrides
you
do
the
source
routing
first
and
then
you,
then
you
do
the
destination
route.
J
No,
we
look,
we
no,
no,
no!
No,
we
first
look
at
the
destination
like
traditional
routers
would
do,
because
the
most
specific
destinations
get
a
subset
of
routes
for
more
specific
destination
and
then
look
if
there
is
anything
much
in
the
source
as
a
most
specific
source
and
if
there
is
an
empty
set,
so
nothing
much
in
the
source
you
move
to.
The
less
specific
destination
destination
is
first.
N
Okay,
gene
in
fact,
I'm
a
a
little
confused
about
this
work
because
you
know
we
have
this:
the
flows
back
route
or
the
acr
when
they
do
the
forwarding,
we
will
match
the
destination
and
the
source
and
not
only
destination
source,
but
also
designation,
port
and
the
source
port.
So
I
think
the
esl
or
the
flows
back
may
also
satisfy
the
requirement,
I'm
not
sure.
What's
the
difference
between
the
two.
L
H
J
J
List,
probably
here,
if
you
can,
then
if
you
can
send
this
to
the
list,
we'll
think
about
it.
Thank
you.
Thank
you.
Dean.
C
C
Yeah
I
granted
it.
P
This
is
a
joint
work
between
a
grammatical
university
in
italy
and
the
university
of
obligate
in
so
what
is
the
problem
that
we
tried
to
solve
the
world?
We
know
that
the
networks
are
complex
and
characterized
by
high
density
networks
and
with
a
wide
variety
of
routing
protocols
with
heterogeneous
implementations
and
devices.
P
P
P
P
P
P
The
methodology
defines
two
main
test:
categories:
failure
and
recovery
failure,
tests
and
use
many
types
of
common
faults
in
data
centers
and
measure
how
implementation
react
to
those
faults.
While
the
recovery
tests
measure
how
implementations
react
when
the
network
is
restored
from
a
vote
to
cover
all
the
failure
and
recovers
these
cases,
we
define
a
failure
and
recovery
test
for
each
type
of
node
and
for
each
type
of
link
for
each
test.
We
analyze
four
main
aspects:
the
convergence,
the
messaging
load,
the
locality
of
the
event
and
the
rounds.
P
P
The
second
aspect
is
the
messenging
load.
The
purpose
is
to
determine
if
the
protocol
hoverhead
grows
smoothly,
with
respect
to
the
topology
parameters
that
are
the
values
of
k
and
r,
to
evaluate
the
ability
of
an
implementation
to
limit
the
blast
radius
of
a
failure
already
cover
event.
The
methodology
defines
a
locality
metric
that
is
computed
with
the
following
algorithm.
P
P
However,
considering
this
this
aspect
of
being
oblivious
as
much
as
possible
with
respect
to
diaspora
timing
is
a
challenging
task
and
to
do
that,
the
methodology
introduces
another
representation
that
describes
how
the
states
of
node
are
synchronized
by
routing
protocol
packets.
We
call
it
not
state
graph
here.
We
have
an
example
of
a
k
equals
two
r
equals
two
factorial
and
another
state
graph
relative
to
an
event
on
this
topology.
P
In
particular,
this
node
states
graph
is
the
one
generated
by
a
leaf
node
failure
on
a
bgp
implementation
to
compute
the
node
state
graph
from
the
outside.
We
use
an
intermediate
representation,
called
the
node
state
timeline.
So
let's
consider
our
usual
example
of
a
leaf
node
failure
on
a
bgpk2,
r2
topology.
P
P
P
P
P
Okay,
we
now
know
how
to
build
a
non-state
graph,
but
what
can
we
do
with
it?
Well,
this
graph
allows
to
perceive
the
number
of
state
changes
that
the
network
directs
us
before
the
convergence,
because
this
state
changes
rounds
and
using
the
graph
for
each
vertex.
We
can
compute
the
number
of
rounds
as
the
length
of
the
longest
path
from
swords
to
that
vertex,
and
we
call
this
measure
state
run
value
of
a
vertex.
P
P
The
simple
architecture
presents
levels
of
modifiability
and
assemble
existing
tools
with
new
ones
specifically
device
for
the
purpose.
For
the
network
emulation
we
we
choose.
Katakana
category
is
a
container-based
network,
emulator
developed
that
commater
and
we
choose
it
since
it
can
leverage
on
kubernetes
for
the
virtualization
and
dance
it
is
able
to
scale
up
in
a
cluster
to
faithfully
emulate
large-scale
networks.
P
Okay,
but
how
does
it
work
civil?
What
is
what
did
it
do
during
an
experiment.
P
It
generates
and
they
produce
a
topology.
It
starts
to
capture
all
the
pdu's
and
then
it
starts
the
routing
protocol
after
he
performs
the
required
action
for
a
specific
type
of
test
like
another
failure,
and
then
it
performs
the
convergency
check
to
ensure
that
the
protocol
has
converted
after
each
shutdown
the
containers,
and
then
it
performs
the
analysis
on
the
damped
traffic
of
putting
the
results
without
the
metric
and
some
novel
and
some
graphics,
interactive
graphics
to
monitoring
the
network.
P
Okay.
But
what
can
we
do
with
cbi?
Well,
there
are
many
possible
use
cases
and
in
the
following,
we
will
see
some
of
them.
A
first
possible
use
case
is
the
complete
analysis
of
a
routing
protocol
implementation,
as
example.
We
decided
to
analyze
bgp,
since
it
is
the
de
facto
standard
for
plus-based
data
centers,
and
we
choose
the
africa
routing,
bgp
implementation.
P
We
executed
all
the
tests
of
the
methodologies
of
technologies
up
to
10
chapters
and
more
than
33
000
interfaces
in
the
next
I'll
present.
You
only
a
summary
of
the
results,
but
you
can
find
all
in
the
repository
and
in
the
paper
so
from
a
messenging
load.
The
perspective
we
found
that
implementation
does
not
manifest
any
performance,
degradation,
varying
k
and
our
motherboard.
We
found
that
the
number
of
packets
exchanged
by
node
is
not
affected
by
the
number
of
nodes
in
the
network,
but
it
only
depends
on
the.
P
I
want
to
show
you
only
an
example
of
the
graphs
automatically
produced
by
siebel.
These
are
representations
are
very
intuitive
and
they
can
be
exploited
to
reason
about
locality
and
rounds
on
the
left.
You
can
find
a
topology
label
with
the
number
of
packets
that
traverse
each
link
during
an
event,
and
this
could
be
very
helpful
to
think
about
locality.
For
example,
we
can
easily
notice
that
exactly
two
packets
are
exchanged
on
each
link
on
the
right.
P
P
P
P
P
We
found,
however,
that
risk
python,
even
a
physical
prototype,
prototype
exhibits,
results
showing
the
credible
potential,
especially
from
a
locality
point
of
view.
As
you
can
see
from
the
pictures
where,
in
case
of
a
leaf
failure
leaf
only
fluids,
the
lift
only
floats
the
nodes
inside
the
part
where
the
voltage
happens.
P
P
Ferragaya
stylus
and
the
flag,
open
fabric
anaya,
says
modification
that
aims
to
reduce
the
flooding
in
a
leaf
spine
topologies.
You
can
find
all
the
results
in
the
repository
year.
For
time
reason,
I
only
showed
the
node
state
graph
for
a
lip
knot.
Failure
on
a
k2r2
topology
and,
as
you
can
see
here,
the
effects
of
the
flooding
reduction
are
easily
noticeable
with
the
knot
stage.
Graph
of
open
fabric
that
becomes
a
tree
other
possible
use
cases
are
the
bug.
P
Finished.
Sorry
are
the
bug
and
inefficiency
inspection
using
the,
not
state
graph.
You
can
find
an
example
of
it
in
the
paper
where
we
spot
some
inefficiencies
in
the
juniper
rift
implementation.
Also,
we
have
the
verification
of
category
configuration
and
the
test
of
interoperability
between
implementation
of
different
vendors.
P
I
am
at
the
end,
so
thank
you
for
your
time.
Here
you
can
find
some
useful
links
and
some
contacts
you
can
find
also
the
mail
of
mariano
that
is,
the
colleague
guy.
We
I
work
with
on
the
development
of
sible.
If
you
have
a
question,
I
am
happy
to
answer,
but
if
we
are
out
of
time,
feel
free
to
contact
me
in
private
or
send
anime
also
to
mariano.
Thank
you
for
the
opportunity
to
be
here.
S
Yeah,
let
me
talk
clearly
20p
juniper,
so
full
disclosure-
I
mean,
I
I
work
with
them
pretty
closely
since
quite
a
long
time.
I
highly
encourage
people
who
are
heavy
into
routing
protocol
special
implementation
to
have
a
deep
look
at
this
stuff,
because
this
goes
far
beyond
a
framework
that
allows
you
to
compare.
You
know,
protocols
and
find
interesting
bugs.
S
You
know
they
kicked
a
couple
of
things
out
of
my
implementation,
which
stuff
was
working
just
fine
and
we
didn't
look
for
like
correctness,
we
were
looking
for
correctness
in
terms
of
ultimate,
I
think
converge
timing
and
so
on
that
they
found
stuff
that
were
due
to
protocol
inherent
self-stabilization
was
working
fine,
but
it
was
not
efficient
right.
We
were
like
retransmitting
too
often
something
like
this.
S
So
that's
very
applicable
immediate
interest,
but
the
framework
they're
building
is
actually
allowing
to
observe
this
distributed
system
in
a
very
formal
way,
like
leslie
lamport
kind
of
vector
clock
graph.
If
you
do
it
and
that
could
lead
us
to
a
point
where
we
actually
can.
S
We
can
start
to
think
with
a
formal
proof
of
correctness
of
routing
protocols
is
viable
because
until
today
we
just
don't
know
ospf
converges
yes,
but
who
said
we
don't
have
any
formal
proof
bgp,
we
all
know
I
mean
stable,
oscillations
and
so
on,
and
this
kind
of
framework
gives
us
an
access
to
an
algebra
which
is
a
complete
game.
Changer
right,
so
I
highly
encourage
people
to
like
at
least
read
the
paper
or
otherwise
just
kick
off.
It's
a
phenomenal
firm
framework.
S
O
Hello,
this
looks
really
neat
and
you
said
that
you
implemented
the
negative
disaggregation
and
we
I
don't
didn't
get
much
feedback
on
on
it
and
whether
it's
really
useful
and
whether
it
performs-
and
since
you
mentioned
deltamin,
I
would
have
liked
to
see
yeah
how
it
improved
the
situation
or
made
it
worse.
O
P
Inverted
the
negative
disaggregation
using
the
rift
implementation
so
using
the
juniper
implementation
of
tony.
So
I
don't
know
because
in
the.
P
Works
as
expected
when
he
implemented
it,
but
this
implementation
is
written
in
python,
so
it
has
it
converges
only
on
very
small
topologies,
and
I
think
that
we
cannot
state
that
it's
working
well.
There.
G
The
probably
the
first
it's
called
the
fed
factory,
not
notation
the
things
for
the
cloud
data
center
are
in
the
price
data
center.
Actually,
there
are
multiple
plans,
the
first
it's
just.
It's
not
restricted
to
only
one
plan.
The
second
part
is
you're
not
showing
is
like
a
virtual
port
channel
like
a
cisco
brocade.
They
have
a
virtual
fabric,
arista
amlag,
c-lag,
extreme,
all
kinds
of
things.
G
So
when
you
consider
the
thing
together,
a
stand
alone
like
the
leaf
span,
super
span,
node
will
not
be
considered
by
itself
only
actually
going
to
be
aggregate.
In
that
case,
see
I
have
to
acknowledge.
I
haven't
read
through
your
work,
but
from
your
presentation.
It
seems
you
put
everything
by
itself
without
considering
the
thought,
tolerance
mechanism
that
have
been
put
into
the
cloud
or
enterprise
data
centers.
C
P
B
Yeah
I
just
wanted
to
encourage
tom
to
thomas
to
continue
this
work.
I'm
really
happy
to
see
this
here
and
I
hope
to
see
the
team
on
regular
basis,
updating
us
that
has
been
tested,
and
so
obviously
you
will
see
more
comments
coming
from
more
enterprise
people.
So
the
comment
about
of
use
mc
lock
is
proper
one
and
if
you
have
more
time,
probably
could
look
in
less
hyperscale
topologies
into
more
enterprise
development.
But
practically
it's
really
good
work
and
thank
you
for
doing.
C
A
Hey
just
a
quick
one.
You
want
to
post
the
links
on
the
on
on
the
on
the
routing
working
group
list,
because
you
know
in
the
pdf
version
of
the
draft,
there's
no
link
for
the
paper
and
what
everything.
T
T
Okay,
this
this
job
to
provide
a
simplified
protection,
magnesium
of
fast
failure
for
multi-home
egress
scenarios.
T
T
Okay,
the
this
slide
shows
the
extension
of
sih
flex,
which
defines
the
b
flag
in
the
s
h
flex.
T
N
C
Comment
right
wayne:
please
continue.
Oh
okay,.
T
This
is
the
procedure
of
the
multi-home
egress
protection
protection
on
srv60
path.
In
this,
as
shown
in
the
picture,
one
p
p1
is
the
ingress
note,
and
the
p3
p3
is
the
primary
egress.
Note
and
p4
is
the
backup
egress
note
when,
when
the
ingress
node
p1
encapsulates
the
srh
srh
header
judge
whether
the
primary
vpn
seat
of
the
egress
node
has
a
backup
sheet.
T
Okay,
here
is
an
example
of
srv60
path
scenario:
when
pe1
received
the
packet
from
c1
to
c2
p1
in
capsule
in
encapsulates
the
package
with
ipv6
header
and
inserts
the
backup,
egress
seat,
a4
column,
column
100
into
the
production
of
srh
last
entry.
The
second
list
in
sih
is
designed
as
a11821
a3100
a4
100..
T
The
sl
is
set
to
2.
The
last
entry
is
set
to
3
and
b
flag
is
set
to
1.
when
p2
receives
a
packet,
that's
destined
to
end
and
dot
exit,
a2
column
column,
one
in
normal
operations.
It
forwards
the
packet
with
resource,
a
zero
color
color
one
and
the
destination
ps3
vpn
seat,
a3,
column,
column,
100
from
the
link
between
b
p2
and
the
p3
according
to
and
dot
exceed.
T
The
multi-home
egress
note
protection
processing
on
the
sr
v6b
path
is
consistent
with
that
on
the
srv60
path.
The
only
difference
is
that
when
p1
encapsulates
the
package
with
ipv6
header,
only
the
active
seat,
the
active
seat,
a3
column,
color
100
and
the
backup
seat,
a4
column,
column,
100
are
filled
in
the
segment
list
of
the
sih
header.
T
T
R
Hello,
I'm
also
called-
and
I
would
like
to
respond
to
you,
so
I
I
think
it
we
have
already
worked
with
some
of
our
vendors
to
have
a
protocol
prototype
and
it
works
very
good.
R
X
X
X
X
X
X
Next,
I
will
introduce
a
srv6
midpoint
protection
mechanism
secretary,
pass
scenario
when
node
t
is
failure.
Nab
node
c
performed
for
plugged
forwarding
offset
the
asset
up
to
the
next
segment
and
the
send
packet
to
the
next
segment.
Node
yi
lose
pass
in
yellow
when
node
d
is
failure,
because
net
node
c
isn't
an
s
as
a
v6
endpoint
node.
In
this
version
we
modify
only
and
point
nodes
can
perform
process
for
audio
so
notice.
It
can
can't
perform
press
forwarding
after
igp
conscience,
node
b
deletes
the
fib
to
d.
X
Pursuit
we
collect
feedback
from
chess
print,
springboard
group
chance,
node
platform
product
forwarding
will
modify
as
h,
which
is
against
eight
to
zero.
In
this
version,
we
added
a
restriction
only
as
a
v6
endpoint
can
run
the
middle
point,
propagation
and
change
chances.
Node
can
do
this.
X
X
C
Do
I
have
questions
otherwise
we'll
bring
it
to
the
list,
because
we
are
running
short
on
time
noah.
You
want
to
say
something
quick.
C
Okay,
can
you
bring
the
question
to
the
list?
Please
we'll
start
with
the
next
presentation.
Thank
you.
R
U
U
There
are
central
cloud
edge
cloud
and
also
some
private
cloud
in
custom
site.
So
hopefully,
some
days
we
can
see
the
cloud
services
will
be
as
the
same
as
electricity
and
water
provided
to
the
customers.
U
So
computer
resources
are
connected
by
the
networks
and
served
to
different
workloads,
and
you
end
users,
cell
phone
customers,
home
users
and
enterprise
users,
so
it
needs
to
be
to
be
to
to
provide
some
differentiated
and
deterministic
sre
guaranteed
capabilities.
U
U
So
we
what
we
want
to
to
do
is
to
unify
the
protocol
of
the
ip
network,
so
makes
a
collaboration
between
at
least
between
the
network
and
the
cloud.
U
Easier,
that's
that's
that
that's
to
simplify
that
networking
technology
and
also
to
enhance
the
the
the
cross
domain,
scheduling
ability.
U
So
the
picture
shows
very
simple
architecture
that
is,
we
have
unified
network
running
the
we
have
network
run
unified
as
our
base6
particles,
and
that
connects
the
compute
computing
resources
and
also
the
computer
resources
is
kind
of
central
cloud
or
the
ad
cloud.
Even
it
has.
It
can
be
the
private
cloud
in
the
customer
side
now
those
those
computing
things
will
serve
to
our
customer,
so
so,
based
on
this,
we
we
have.
We,
we
are
have
some
use
cases.
U
U
The
potential
solution
is
we.
We
can
have
some
kind
of
slice.
The
slice
can
be
just
like
some
kind
of
topology
serves
to
different
kind
of
class
of
service
and
also
can
be
some
kind
of
hard
slice
that
is
have
some
of
the
resources.
Q,
resources
or
memory
resources
reserved
for
different
different
slides.
U
Then
we
need
to
the
important
thing
is
we
need
to
the
cross-domain
ability
which,
in
internal
mobile,
we
have
men?
We
we
have
provinces
network
and
also
has
backbone
networks.
We
need
to.
U
U
Then
we
have
to
need
to
consider
to
put
the
connection
and
the
computing
services
together,
and
so
we
can
here
we
can
use
srv6
again
to
to
to
to
do
both
of
the
connection
and
also,
if
the
some
of
the
the
the
computer
services,
like
some
firewalls,
like
that
they
can
be
srv6,
enable
this
can
be
put
into
the
the
ssid
list.
U
U
U
E
This
is
linda
dembra
from
future
way,
so
I
do
have
questions
that
you
mentioned
about
the
computing
aware
of
the
computing
aware
of
the
user,
and
so
is
this
related
to
china.
Mobile's.
E
Y
L
Y
U
U
The
because
of
the
different,
the
the
you
know,
you
know
the
different
kind
of
the
customer
have
different
of
the
the
delay
requirements.
U
U
Doesn't
have
at
all
because
you
know
most
of
the
access
network
and
also
some
of
the
some
of
the
networks
they
are
in
the
party
of
the
access
network
and
aggregation
network.
They
are
some
form
of
the
tree,
so
in
case
of
the
that
way,
then
there
must
be
some
kind
of
congestion
happened
in
the
root
of
the
tree.
C
Okay,
can
you
please
continue
the
discussion
on
the
list?
Link
you'll,
have
a
question.
Please
keep
it
short.
Q
Lingham
from
future
week,
a
quick
question,
so
how
do
you
define
the
computer
resource?
Is
it
just
by
name
or
by
five
tuples
of
the
address
and
ports?
Thank
you.
U
V
I
already
get
the
answer
from
from
the
questions
linda,
I'm
not,
but
I
still
like
to
strongly
recommend
the
authors
to
consider
take
consider
about
the
computer
world
network.
The
scenario
thank
you.
Okay,.
H
G
Good
morning,
everyone
kenji
here
on
behalf
of
all
the
authors,
I'm
talking
about
problem
and
requirements
of
satellite
constellation
for
internet.
The
work
has
been,
I
think,
has
been
talked.
G
Yeah
the
world
has
been
talked
twice.
This
is
the
the
version
three
from
some
changes,
the
next
one
for
the
this
agenda.
Well,
just
you
know
for
the
first
two,
a
big
bullet
bullets
so
just
to
try
to
give
some
a
little
background
about
how
significance
of
the
the
work,
basically,
the
satellite
the
requirement
and
some
problem
statement
there
and
for
the
for
the
third
and
the
fourth,
the
bullets
is
about
what
has
been
changed
and
then
what
will
be
the
next
next
slide?
Please.
G
Thank
you.
Okay,
here,
just
give
the
background
about
all
the
three
gp
related
tech,
spec
tech
documents.
Now
the
importance
of
critical
for
the
success
of
us,
the
the
foundation
for
the
work
in
the
draft,
so
the
last
two
bullets
is
mentioned
as
essay
2,
that
is
the
3
gbp
working
group.
G
Those
are
two
ongoing.
The
study
ids
id
study
projects
have
been
explored
and
studied
in
3gb
side.
One
is
the
5g
you
satellite,
as
the
backhoe
background,
from
the
genome
b
toward
the
cornet,
the
core
network
is
a
mobile
core
network
and
the
other
one
is
satellite
as
the
access
network.
So
basically,
it's
from
on
the
rent
side,
while
the
geno
b.
G
Okay,
here
this
one,
this
is
actually
the
new
the
change
added
to
the
version.
Three-
and
this
is
it's
very
high
level
background
from
thirty
thousand
foot
bird
wheel.
It's
about
the
the
satellite
insertion
point
on
the
three
gbp
wireless
access
part,
the
first,
the
the
top.
The
top
diagram
is
showing
the
stateline
network
is
after
the
mobile
access
that
is
before
the
internet.
G
Well,
the
bottom
part
is
showing
the
satellite
is
on
the
access
side,
so
basically
from
your
own
device
and
go
through
the
satellite
and
then
go
through
the
mobile
access.
So
the
next
slide,
I'm
going
to
yes,
yeah,
I'm
going
to
explain
what
this
means
for
the
top
left.
It's
just
like
very
simple
things,
and
then
you
can
consider
the
satellite
as
a
reflector,
is
hanging
around
on
the
sky
and
give
you
a
simple
reflection
and
the
the
top
right
one
is
about
regenerative.
It's
a
normally
for
large
areas.
G
Okay,
but
in
the
top
right
it
starts
to
in
include
the
multi
asl,
inter
in
the
satellite
link.
Well,
for
the
bottom
left
is
actually
it's
like
the
access
part,
so
you
have
the
the
the
ginobi.
Sorry,
you
have
the
the
du
geno
du
du
cu.
Those
are
the
the
the
wireless
the
base
station
part
and
then
one
on
board
like
a
du
on
board
on
satellite
and
the
cu
is
on
the
ground.
G
But
here
there's
some
big
challenge
regarding
some
timing:
timing,
things
like
for
the
wireless
you
have
to
get
some
slot
within
some
timing
budget
like
one
millisecond
and
you
use
some
super
frame,
those
type
of
things
to
split
into
a
different
sub.
You
know
a
slot,
so
this
one
really
pose
some
challenges
so
for
the
box
trial,
the
bottom
right,
the
thing
is
well
for
all
the
three
bullets
here,
the
middle
one
is
important,
just
to
say:
okay,
you
know
for
all
the
work
here,
it's
critical
for
the
three
gpp
it
has.
G
It
has
the
setups
the
projects
and
then
the
5g
commercial
deployment
is
there.
I
think
it's
close
two
million
five
g
base
stations
already
lots
of
applications,
and
then
satellite
is
important.
So
in
order
to
do
that,
the
cellular
network
is
the
you
know
some
standardize.
The
work
need
to
be
done
next,
one
yeah
here,
it's
on
the
on
the
layer,
three
just
to
show
okay.
Well,
you
know,
if
you
consider
a
satellite
as
a
node
hiding
around
on
the
sky
and
then
how
the
igp
potential
eb
gp,
ibtp.
G
Those
type
of
things
will
be
pairing
together,
so
for
the
work
that
has
been
has
been
attacked
in
previous
ietf
one.
I
think
one
two,
I
forget
so
previous
to
talk
about
the
how
extreme
dynamics
in
the
red
font
throughout
the
bottom
left
about
all
the
link,
quality
cost,
all
kinds
of
things
and
the
igp
btp
mutual
advertisement.
G
So
here
is
really
the
challenging
part
for
the
layer
three,
especially
for
the
leo
the
low
earth.
Obviously
like
this
network.
G
Okay,
now
back
to
the
the
draft
here
or
this
draft
is
the
the
third
version
of
the
second,
the
bullet
in
the
top
box.
That's
actually
the
first
bullet
in
the
top
box
is
the
red
font,
and
then
this
is
one
of
a
series
of
ids
on
the
seller
networks.
The
the
things
have
been
discussed
in
previous
ideas.
M
This
is
dino,
so
it
kind
of
makes
sense
that
from
ground
to
space
you're
using
ebgp,
but
it
looks
like
you're
using
one
single
igp
for
the
entire
satellite
network.
Is
that
suppose.
G
Q
G
Okay,
by
the
way
link
is
the
chief
editor
of
the
draft.
So
yeah
here
is
the
the
bunch
of
the
ids
okay,
and
here
these
are
two
cases
for
from
the
3gbp.
That
means
it's
the
real
thing
that
being
started.
It's
so
far
is
in
the
stage
two
after
the
sit
down
after
stage
two
going
to
trick
down
to
stage
three
and
then
about
the
you
know
all
the
kind
of
communication
protocols.
All
kind
of
thing
will
be
involved
the
next
one
here
yeah
well.
G
This
is
a
case
to
show
you
know
the
leo
with
marty
sl
as
a
5g
backhoe,
and
then
you
know
it's
well
basically
just
say
suppose:
you
have
a
two
ues
who
anchor
on
different
ups
and
then
when
they
want
to
talk
to
each
other
over
a
session
not
spend
more
than
five
minutes
and
because
of
the
extreme
dynamics
of
the
of
the
satellite,
it
will
cause
some.
G
G
So
if
you
consider
things
together,
like
suppose,
from
ue1
to
ue2,
it's
going
through
the
layer,
3
so
you're
going
to
involve
the
routing
the
extremely
dynamic
routing
protocols,
locking
algorithms,
with
the
qs
inside
and
also
the
3gpp
case,
are
talking
about
monitoring.
You
have
to
get
some
feedback
from
the
network
in
order
to
adjust
or
improve
it.
G
So
when
you
put
everything
together,
so
this
is
a
very
important
critical
to
justify
why
a
cellular
network
is
important
and
it
it
has
its
some
position
in
ietf
to
to
do
some
sdo
work
and
then
the
the
second
bullet
on
the
bottom
box.
Just
to
show
you
know
dynamic
does
not
mean
anything
about
the
render.
So,
actually,
you
know
the
the
track
of
a
satellite
actually
has
some
predefined
tracks,
so
you
may,
or
we
may
be,
able
to
utilize
those
information
in
order
to
do
the
mechanical
work.
G
J
G
It
yeah,
so
this
is
just
a
conclusion
about
the
other
things
and
then
you
know
we
have
a
you
know
this
is
our
time
and
then
we
hope
you
know
by
the
way
there
will
be
asa
meeting
in
next
in
uk
london
for
itv105,
and
we
do
welcome
out
comments
contributions,
and
you
know
all
kind
of
feedback
we
believe
is
important
and
the
real
case
is
there
through
gpp
are
working
on,
it
are
pushing
it.
Okay,
thank
you.
M
Yes,
this
is
dino
with
respect
to
the
instructive
routing
draft.
Do
you
guys
from
all
the
research
you've
done
so
far?
Have
you
do
you
believe
or
have
an
opinion
if
gps
based
routing
will
be
done
on
the
satellite
network,.
G
Q
Actually,
that's
just
one
solution,
we've
given,
but
we
we
know
that
there
are
many
research
topics
for
this
regard,
so
I
I
think
we
welcome.
Other
people
also
contribute
different
solutions,
because
that's
another
easy
routing
problem.
Q
W
M
Q
Right
now
we
don't,
we
don't
expect
there
will
be
a
interworking
between
different
satellite
network,
but
one
satellite
network
has
to
exchange
internet
laws
with
internet.
Yes,
it's
different
now,
yeah,
okay,
so.
M
Q
C
So
we
only
have
one
minute
for
the
second
to
the
last
presenter.
We
apologize
to
the
last
presentation
we
won't
be
able
to
get
to
it.
We'll
have
an
interim
to
cover
that.
C
C
Z
Yes,
thanks:
I
will
give
a
quick
break
and
we,
I
will
present
the
two
use
cases
and
in
kerala
networks
the
operators
may
change
the
page.
Yes,
operators
may
deploy
multiple
data,
centers
or
computing
resource
pools,
distributed.
Lift
dataset
centers
can
has
a
different
types
of
value
added
service.
Z
From
the
perspective
of
an
operator,
we
need
to
think
how
to
provide
about
better
services,
in
addition
to
the
cost
of
land
tendency
of
network
process
between
the
data
centers,
whether
computing
resource
should
be
considered
and
how
to
measure
a
computing
resource,
the
number
of
virtual
cpu
or
cpu
utilization
memory,
utilization
or
other
parameters
and
what
the
technology
is
used
to
achieve
computing
of
our
function,
and
I
will
talk
about
the
two
use
cases.
Why
is
the
computing
of
our
sd1
and
the
other
is
computing
overall
sfc.
Z
Yes,
if
sd1
can
be
aware
of
the
computing
resource
of
a
application
deployed
in
the
cloud
by
virtual
cpes
and
selected
application
instance
for
the
client
to
visit
according
to
the
computing
power
and
the
network
state
of
one,
as
shown
in
the
page,
on
cloud
1
and
cloud
2
visiting
application
1
in
cloud
1.
If
the
load
of
application
in
cloud
1
is
50
and
the
load
of
application
in
cloud,
2
is
20
and
the
same
network
stage
how
about
calendar
3
and
we
think
to
consider
let's
say
loaded.
Z
The
application
sd1
selects
a
application
one
in
cloud,
two
for
the
calendar,
three
in
branch
office.
So
the
path
is
shown
in
the
in
the
page
and
the
other
case
use
cases
is
about
computing,
aware
sfc.
Z
Yes,
we
can
see.
Each
data
center
may
have
different
types
of
service
functions.
High
usage
service
functions
are
deployed
in
edge
or
regional
data,
centers.
Well,
other
low
usage
service
functions
are
deployed
in
global
also
to
central
data
centers,
so
sfcs
with
different
types
of
service
functions
may
span
multiple
data
centers
and
when
we
consider
the
process
computation
of
srp
according
to
the
cost
of
all
lanterns
of
a
network
process
between
the
data
centers
house
hosting
the
service
functions
nowadays.
Z
A
user
orders
out
other
sfc
with
a
chain
of
two
valued
added
services
for
its
access
to
internet
service.
The
the
services
secretaries
of
the
fcc
as
a
to
assess
farewell
and
the
vo,
which
dc
is
better
to
provide
as
a
function.
Z
Fair,
walk
to
consider
the
lately
loaded
service
functions
computing
as
slp
as
a
repair
center,
as
in
the
page,
and
we
have
mentioned
the
procedure
for
sfp
creation
in
the
draft,
and
we
also
to
give
the
potential
solutions
could
be
considered,
maybe
to
collect
the
service
functions,
location
and
computing
power
by
bgp,
ls
or
net
config
from
the
router
connecting
the
data.
Centers
and
dynamical
get
get
the
association
relationship
and
the
independent
collector's
function,
location
and
the
computing
power
by
other
means
and
statically
configure
the
association
with
the
network
on
the
control
plane.
Z
Okay,
okay,
we
will
just
add
you
to
two
to
to
solicit
comments
and
refine
the
draft,
and
we
wanted
to
can
to
s
to
see
it
a
little.
Any
other
pair
means
meters
to
be
considered
and
any
better
solutions.
Thanks.
C
Okay,
so
thank
you
for
the
presentation
and
we
don't
have
time
for
any
questions,
sorry
for
the
delay
and
all
the
technical
hiccups,
and
we
will
for
the
unfinished
presentation
we
will
have
an
interim
in
london
also
for
the
request
that
you
make
on
the
slot.