►
From YouTube: IETF-ALTO-20230503-1400
Description
ALTO meeting session at IETF
2023/05/03 1400
https://datatracker.ietf.org/meeting//proceedings/
A
A
A
Okay,
we
cannot
hear
you
yeah.
D
A
E
A
C
E
A
G
Okay-
okay,
sorry
about
that!
Actually
I
switched
to
my
you
know:
PC,
my
laptop
maybe
doesn't
work
and.
G
Yeah,
so
sorry
about
that,
actually
yeah,
that's
a
you
know,
restart
actually
yeah.
This
is
Ching.
Actually,
this
third
interim
meeting.
Actually
so
thanks
everyone
to
come
to
join
this
meeting,
especially
people
who,
still
you
know,
invocation,
actually
make
a
come.
You
know
make
to
attend
this
meeting
and
my
name
is
Qing
and
my
co-chairs
Mohammed.
Actually
we
both
online
to
host
this
meeting
and
so
another
way
I
apply
yeah.
G
You
are
not
familiar
if
it
is
related
and
the
first
time
when
we
set
the
term
for
this
discussion,
you
know
actually
you
know,
for
we
really
want
to
Foster
the
auto
development.
So
we
really
want
to
know
how
many
data
Nano
data
stores
that
can
be
integrated
with
auto
server
would
readily
available
data
source.
So
we'll
you
know,
focus
on
this
available
data
source
that
we
already
have
operation
diploma
experience
and
the
second
I
want
to
emphasize.
You
know
we
not
really
identify
or
enumerate
all
the
data
source.
G
So
this
agenda
for
today's
discussion.
First,
we
will,
you
know,
invite
Luis
and
Jody
to
discuss
the
survey
on
the
list
of
the
potential
data
source
of
Interest,
thanks,
Luis
and
Judy,
to
take
a
leadership
to
do
the
homework
for
this
part
and
and
then
we'll
invite
several
problems
to
introduce
a
proposal
for
data
source
integration
and-
and
so
we
have
40
minutes
before
the
discussion
for
each
proposal.
We
assign
10
minutes.
So
please
stick
to
your
time
so
and
then
we
can
wrap
up
the
discussion.
G
Yeah
Louis:
do
you
want
to
share
by
yourself
or
okay
right?
It's
kind.
H
H
Okay,
just
one
second
I
will
look
for
the.
C
I
need
to
look
for
the
for
the
Google.
Do.
E
You
have
the
I
mean
I
mean
I'll
put
it
in
the
chat.
Yes,
please
yeah.
E
H
So
I'll
go
ahead,
yeah!
Well,
the
the
basic
idea
at
the
time
of
collecting
this
was
necessary
to
identify
a
useful
and
interesting
data
sources
that
could
be
integrated
in
Alto
right.
So
at
the
time
of
integrating
data
sources
I,
there
are
like,
let's
see
three
fundamental
questions
now:
how
to
collect
it,
how
to
process
it
and
how
to
present
it.
So,
probably
in
all
these
three
three
exes
there
is
an
impact
on
alto
here.
What
we've
concentrated
was
to
identify
the
impact
in
the
in
the
terms
of
how
to
process
it.
H
Let's
say
what
would
be
the
artifacts
in
Alto
that
could
be
impacted
somehow
for
the
different
data
sources.
So
this
is.
This
is
the
impact
that
we
reflect
here,
but
for
sure
there
will
be
some
impact
on
how
we
collect
the
the
data
source,
how
how
we
integrate
at
the
end
and
also
how
we
present
it
to
the
two
potential
Alto
clients
right
so
yeah.
H
We
essentially
the
exercise,
was
to
identify
those
sources
that
could
be
of
interest
trying
to
declare
an
objective
for
having
this
different
different
data
sources
in
the
into
Alco
than
a
rational.
Why
I
mean
a
justification,
a
motivation
for
that
right
and
then
existing
word
for
references.
This
was
essentially
the
the
exercise
also
comments
in
case
that
something
should
be
somehow
highlighted
or
reflected.
So
we
started
with
a
basic
aspects:
topology
in
in
topology.
H
We
have
a
very
good
number
of
outcomes
from
ITF
we
have
the
IDF
Network
topology,
we
have
the
idft
network,
topology
and
so
on
so
forth.
So
a
number
of
relevant
topology
models
have
already
been
developed
and
deployed
in
actual
networks.
So
yeah
is
here
essentially
is
to
use
Alto
or
I
mean
to
to
leverage
on
components
that
already
have
this
network
models
and
use
Alto
to
retrieve
from
them
the
information
and
to
integrate
with
the
rest
of
the
information
that
could
be
part
of
Alto.
H
For
instance,
there
was
a
recent
comment
in
the
mailing
list
was
triggered
by
Kai
about
the
fact
that
okay
from
the
topology,
we
can
collect,
let's
say
the
internals
of
the
connectivity
of
the
different
nodes
and
links,
and,
and
so,
but
there
is
yet
information
missing.
That
could
be
retrieved
in
some
other
way.
That
could
be,
for
instance,
the
IP
pools
that
are
declared
in
different
border
border
routers
or
the
nodes
like,
for
instance,
the
IP
addresses
of
residential
users.
H
That
is
well
simplified,
I
guess,
with
the
case
of
the
telephonica,
an
integration
of
Alto
right.
So
we
need
yet
I
mean
this
would
not
be
just
yet
the
only
source
of
information
for
building
an
overall
topology.
We
probably
need
to
rely
on
some
other
information
in
order
to
aggregate
all
of
that
and
build
the
food
topology.
H
So
this
is
a
unfortunate.
Please
interrupt
me
whenever
you
want
to
so
I
I
will
go
one
by
one,
but
it
I
should
focus
on
concentrating
someone,
please
let
me
know
so.
For
instance,
in
these
two
cases,
the
topology
and
in
the
case
of
the
raw
I.tf
Network
topology,
probably
the
impact
is
in
the
network
map.
How
We
Do
form
the
topology,
taking
into
account
this
data
source
in
the
case
of
the
IIT
in
network
topology.
H
The
impact
would
be
not
only
the
network
map
so
how
I
built
the
the
topology
view
of
in
Alto,
but
also
in
the
cosmap
in
the
Matrix,
because
we
are
incorporating
traffic
engineering
information.
On
top
of
that,
okay
third
case
will
be
the
the
performance
Matrix
this
already.
There
is
a
document
close
to
you
to
be
corner
of
C
in
Aldo,
where
we
declare
how
to
present
the
auto
performance
metrics,
but
it's
not
yet
clear
at
all
how
we
collect
the
performance
metrics
right.
So
how
can
we
integrate
those
performance?
H
Metrics,
together
with
the
topology
to
be
presented
in
line
with
this
Graphics
is,
is
is
presented
right,
so
here
is
essentially
the
point
here
would
be
to
analyze
to
understand
how
we
can
integrate
such
performance
metrics,
relying
on
bpls
or
in
IEP,
carrying
this
performance
information
and
so
on
so
far
here.
Clearly,
the
impact
will
be
on
the
cost
map,
so
how
we
build
the
the
cosmap
integrating
those
metrics
that
we
we
identify
so
far
in
the
previous
draft,
then
we
commented
on
on
the
we
identified
also
as
relevant
the
compute
resources.
H
So
the
idea
here
will
be
to
incorporate
into
Alto
information
related
to
the
available
compute
resources
connected
to
the
network.
And
here
let
me
let
me
make
a
difference
between
the
copy
resources,
the
raw
Computer
Resources
versus
the
resources
allocated
to
service
functions.
That
will
be
the
following
topic,
so
here
will
be
essentially
the
resources
that
are
available
in
certain
point
in
the
network
of
the
resources
you
you
want
the
resources
that
are
attached
to
some
point
in
the
topology
in
the
network
right
so
yeah.
H
Then,
as
mentioned
the
service
functions
in
this
case,
the
idea
would
be
to
present
information
of
service
functions.
So
this
is
Network
functions
or
applications
or
service
Services
attached
to
certain
points
in
the
network,
and
also
here
I
mean
we
can
essentially
incorporate
information
or
what
is
what
are
the
service
functions
available
where
these
service
functions
are
located
and
also
probably
even
what
are
the
resources
allocated
for
those
service
functions?
So
this
will
be
a
matter
of
analysis
right
to
understand.
The
information
is
relevant
or
not
yeah.
H
Okay,
in
this
case
as
before.
So
there
is
an
impact
on
the
network
topology,
because
we
need
to
incorporate
further
information
that
are
the
the
services
that
are
instantiated
in
the
network
in
certain
nodes
in
the
network
in
the
Border
part
of
the
nodes,
and
also
metrics
that
could
be
cooperated
related
to
these
service
functions.
Right
and
here
as
a
comment,
we
added
that
there
will
be
a
a
clear,
Synergy,
I
guess,
with
what
is
being
defined
in
cuts.
H
I
have
continuing
with
the
exercise.
We
also
consider
as
a
potential
metric
to
sorry
a
potential
data
source
to
integrate
will
be
the
available
Bank
within
a
given
path,
so
to
include
the
the
the
this
information
about
the
available
Bank
within
the
path
in
such
a
way
that
we
can
take
an
important
decision
at
the
time
of
placing
some
flows
in
given
paths.
H
H
The
idea
here
would
be
to
provide
information
about
the
power
consumption
in
a
path
on
a
number
of
paths
right,
so
the
the
this
is
clearly
of
Interest
or
I
mean
there
is
interest
in
IDF
in
the
latest
atfs
for
exploring
and
working
on
on
this
area.
There
there
has
been
some
work
in
IRB
also,
there
are
some
specific
drugs
in
terms
of
green
metrics.
So
here
the
point
will
be
essentially
to
include
this
information
about
the
power
concerned
being
consumed
in
a
path
together.
H
The
eighth
data
source
of
potential
relevance
would
be
inventory
information
and
here,
by
inventory
an
example
that
we
provide
here,
because
it's
related
with
one
of
the
latest
activities
that
has
been
discussed
in
the
mailing
list.
It
could
be,
for
instance,
the
geolocation
of
the
nodes
so
that
your
location
at
the
end,
they
know
they
has
a
property
which,
where
this
node
is,
is
being
deployed
physically
geographically
right.
H
So
a
potential
source
to
be
integrated
would
be
to
to
take
information
from
the
inventory,
for
instance,
and
add
this
kind
of
information,
this
kind
of
properties
into
the
network
map.
So,
in
order
to
enrich
the
the
information
of
the
network
map
right,
new
services
that
could
be
relevant
or
impacted
for
this
would
be,
for
instance,
aspect
like
geofencing,
or
that
probably
you
I
mean
there
is
a
decision
behind
avoiding
certain
paths
or
certain
areas,
and
so
on
so
far
and
I
said
this
yeah.
H
This
could
be
probably
having
have
some
in
a
senior
USA
with
the
with
the
work
on
inventory
that
is
now
being
triggered
in
in
ITF,
and
also
the
trusted
network
activity
from
our
colleagues
from
Fujitsu
also
could
have
shown
so
relation
with
this.
H
The
nine
potential
data
source
could
be
the
one
related
to
a
planned
or
schedule
variation
of
routing,
and
here
there
is
a
clear
link
with
another
working
group
in
ITF,
which
is
the
TBR
with
this
working
group
is
explored
in
this
idea
of
skill
in
programming
changes
in
the
Denver
and
then
advertising
those
changes
so
carrying
this
in
in
in
ITP
or
or
bdp.
But
it's
not
only
that
case
also.
H
There
will
be
other
cases
that,
for
instance,
the
Declaration
of
pools,
IP
pools
in
situations
like
cgnat
or
APNs
in
the
mobile
network,
but
essentially
operators
are
the
declaring
new,
prefixes
or
prefixes
that
in
in
days.
H
G
Yeah
I
want
to
have
some
question
for
for
this
table.
Actually,
so
thanks
for
put
it
together
for
these
data
source
I
think
is
very
useful.
I
I
think
yeah.
What
it
missing
part
is,
you
know,
maybe
we
can
add
a
one
column
to
say
which
that
results
actually
already
have
an
open
source.
You
know
project
which,
which
you
know
data
source,
has
already
been
implemented,
which
is
by
either
by
the
open
source
or
by
the
vendor
or
operator.
I.
G
Think
a
bgp
and
a
nanotopology
actually
more
concrete
data
source
for
Nano
topology
has
already
been
implemented
by
the
open
daylight
and
for
bgpu
or
any
vendor,
maybe
already
implement
the
bgp.
So
for
some
other
data
source
that
you
mentioned,
inventory
I
I
think
inventory.
Not
only
they
can.
You
know
to
expose
some
of
Nano
device
properly,
but
also
you
can
expose
some
kind
of
endpoint
property
or
device
property
I
mean
the
endpoint
device.
So
this
may
relate
to
some
kind
of
you
know:
camera
Network
as
a
service
API
project.
G
H
G
F
You
yeah
yeah
I,
actually
I
was
thinking
on
the
on
the
aspects
regarding
to
the
information,
because
I
remember,
igmar,
mention
I
think
the
last
IDF.
You
know
that,
in
terms
of
you
know
how
we
we
will
collect
the
information,
because
Luis
mentioned
it
three
equations,
how
we
will
collect
how
we
will
process
how
we
will
present
and
then
you
know
we
have
a
set
of
components
running.
F
F
You
know
that
is
just
to
mention
that
we
have
got
in
our
set
of
companies
and
we
have
that
currently
are
running,
and
in
some
point
you
know
we
could
take
advantage
on
that
to
to
Define
how
is
the
process
or
how
the
data
source
is
collected
and
eventually
how
it's
processed
just
to
try
to
answer
some
of
the
questions
at
least
mentioned
at
the
beginning
of
the
presentation.
F
G
G
G
So
Luis
this.
H
Yeah
yeah,
this
is
but
a
very
let's
say,
straightforward,
rational
I
guess.
So
the
point
is
essentially
the
the
fact
that
we
operators
are
very
used
to
work
with
BEP
communities.
So
it's
a
way
of
handling
and
identifying
set
of
prefixes
in
the
network
for
different
operational
purposes.
For
instance,
in
the
slide
you
can
see
what
are
the
purposes
which
is
for
the
building
connections
and
so
to
determine
what
is
the
local
preference
at
the
time
of
receiving
the
traffic
right.
H
This
is
just
an
example,
so
the
point
is
that
somehow
BJP
communities
and
aggregation
of
information
of
different
prefixes
so
somehow
is
similar
to
the
concept
of
the
PID
in
Alto
right.
However,
Community
could
not
be
exactly
the
same,
so
so
how
we
could
even
consider
defining
bgb
communities
aggregating
more
than
one
PID
or
or
alternatively,
one
PID,
having
or
aggregating
more
than
one
bgp
Community,
for
instance,
fixed
users,
mobile
users,
so
that
could
be
different
company
Nation.
H
So
we
could
have
one
to
one
one
to
n
or
n21,
depending
on
on
the
purpose
of
the
bgp
community.
So
here
the
very
basic
proposal
would
be
to
evolve.
Alto
in
such
a
way
that
we
can
work
with
the
concept
of
bgp
community,
so
so
how
expanding
the
the
fight
of
having
the
pids,
okay,
but
also
work
with
the
BB
communities
in
such
a
way
that
we
can
integrate
existing
processes
in
the
operator
side
for
different
kind
of
purposes.
H
B
Yeah
I
think
Luis.
This
is
a
very
interesting
idea.
Do
you
have
any
I
should
read
it
the
the
proposal
very
carefully?
Do
we
have
any
particular
suggestion
in
terms
of
the
protocol,
in
terms
of
both
in
terms
of
modification
of
output,
Auto
particle
or
also
in
terms
of
what
bgp
can,
because
you
talk
about?
Essentially,
this
one
can
have
impact
on
both
sides.
Right
one
is
impact
on
auto
protocol.
One
is
have
impact
on
the
potential
bgp
carrier
of
communities.
B
Can
you
give
a
very
quick
comment
about
what
it
might
look
like,
for
example,
what's
the
extension
on
the
bgp
auto
protocol?
So
are
you
seeing,
for
example,
we
Define
cost
map
to
say,
hey
cost
map
one
is
associated
with
is
associated
with,
for
example,
given
a
pgp
community?
Would
that
be
a
possibility?
So
therefore
we
announced
all
this
stuff
in
ird.
Is
that
one
way
to
do
it.
H
A
different
process,
you
know
the
information.
This
is
from
the
process
in
part
for
the
the
way
of
retrieving
the
information.
Also
either
even
bgp
communities
are
advertised
through
bgp
I'm,
not
sure
it
will
be
sufficient
to
listen
to
the
I
mean
with
the
VIP
speaker
to
collect
this
information
or,
if
or
if
we
will
require
something
else
at
the
time
of
defining
the
policies
to
have
some
consistency.
H
Sorry,
not
the
policies,
the
VIP
communities,
to
have
some
consistency
in
the
in
the
definition
of
Bev
communities
in
the
in
the
routers
and
also
in
Alto
I'm,
not
sure
to
be
honest,
if
just
speaker
for
collecting
the
information
from
bgp
would
be
sufficient
or
not.
And
finally,
at
the
time
of
presenting
the
information,
probably
will
be
required
to
modify
the
alto
protocol
for
yeah
for
for
asking
for
VIP
communities
and
and
yeah.
H
Probably
the
clues,
because
bgp
Community
can
involve
more
than
one
PID,
as
we
find
today
is
how
we
could
handle
and
manage
that
information.
I
mean
if
we
will
require
something
else,
because
a
a
multimodal
will
have
bid
could,
for
instance,
address
a
region
rather
than
a
single
router.
So
so
how
we
were
questioning
throughout
the
protocol
impact
for
a
region
so
and
let's
put
an
example,
Performance
metrics
or
whatever
so?
Can
we
Define
a
particular
performance
metric
for
a
region
or
not
so
probably
something
we
need
to
think
about.
H
So
what
kind
of
information
could
be
applicable
in
a
straightforward
manner
for
the
for
the
for
the
VIP,
Community
and
yeah
and
and
work
on
that
to
understand
if
everything
can
be
done?
Similarly
to
the
PID
a
case
or
we
need
to
to
those
something
different.
G
H
H
Probably
we
could
Define
different
communities
for
fix
and
mobile
because,
for
instance,
the
different
resolution
of
the
content
or
whatever,
whatever
property
that
you
would
like
so
for
sure.
I
guess
that
with
this
granularity
either
because
we
can
be
more
granular
or
even
we
can
I
mean
we
can
zoom
in
and
zoom
out,
probably
we
need
to
adapt
the
artifacts
in
Alto
for
reflecting
this.
This
fact
and
and
I'm
sure
that
there
will
be
some
modifications
in
that
respect.
Yeah.
G
Okay,
I
think
we
we
can,
you
know,
take
it
to
the
list
that
you
have
further
discussion
for
for
this
open
question.
Thank
you.
Let's
move
to
the
second
proposal
and
bgprs,
oh
setting,
do
you
want
to
make
a
comment
for
this.
G
I
I
know
we
are
running
out
of
time,
but
the
I-
maybe
it
would
be
good
so
since
Luis
provided
some
examples,
it
would
be
good
to
like
formalize
them
in
some
further
discussion
or
document
to
justif
to
explain
why,
for
example,
it
is
okay.
I
Sometimes
you
don't
have
a
one-to-one
map
that
justifies
this
world
and
why
it
is
not
possible
in
in
the
in
case,
you
define
PID
with
respect
to
some
geographical
criteria
and
within
this
geographical
region,
you
would
like
to
Define
several
pgp
communities,
so
why
why
it
is
that
not
possible
just
to
Define,
maybe
a
second
Network
map
that
would
subdivide
these
pids,
why
it
is
not
possible?
Is
it
because
of
scalability
or
maybe
I
missed
something
or
so
yeah?
That
was
just
a
comment
so.
G
D
D
So
it's
supposed
to
summarize
some
experience
and
practices
the
fact
that
they
try
to
decide
using
the
standard
PPI
in
our
implementation.
So
you
can
go
to
the
excise.
D
The
the
auto
match
generation
using
the
API
they
found
they
open
the
last
PCP
packing,
and
then
we
report
the
from
this
technique,
issues
in
the
information
and
then
replace
the
fantastic
team
of
Luis
and
actually
referred
to
to
the
department
and
to
have
mercury
I'm
already
invention
to
the
activityp
and
reported
in
the
item
number
five.
D
D
Okay,
so
let's
get
the
progress
updates
on
the
go
to
the
the
details
about.
D
So
as
soon
as
you
can,
you
can
see
the
slides
but
yeah
that
is
still
loaded.
So,
but
it's
done
in
a
single
as
some
requirements
to
leverage
the
VPN
visualize
to
generate
the
automatic.
So
first,
the
Audi
yes
for
the
router
I
must
logically
connect
to
at
least
one
certificate,
roster
factor
and
in
the
each
ITP
area.
D
The
pgpris
must
be
configured
at
least
one
node
and
in
the
in
practice,
because
we
want
to
minimize
the
number
of
the
nodes
to
configure
the
vpri
typically
because
it
only
can
be
around
the
error
but
the
router.
So
this
is
a
figure
for
the
test
that
we
try
to
try
the
problem
in
the
types
of
the
website
and
the
all
the
nodes
configure
the
video,
as
must
locally
connect
to
the
other
one
rotary
vector
and
then
the
auto
server.
D
D
D
Then
the
text
model
will
summarize
for
the
PCP
advantages.
We
just
need
to
generate
the
network
map,
so
we
need
to
use
the
path
attribute
down
to
the
perfect
information,
and
so
this
project
is
very
straightforward.
We
just
need
to
what
we
did
it
to
use
the
gas
number
on
the
laptop
to
prove
the
cell.
The
prefix
attack
to
about
the
router.
D
D
So
there's
a
pass
Factor
on
the
network
map,
so
we
can
compute
the
apid
to
the
other
cosmet.
So
that's
the
basic
workflow
so
to
leverage
the
address
in
a
single
AF
and
the
learn.
If
something
say,
If
try
to
move
this
to
the
multiple
as
actually
it's,
not
some.
No
issues
need
to
figure
out,
so
one
major
relationship
is
that
so
yeah
dot
screen
may
add
so
for
the
end-to-end
flow
Ronaldo,
the
what's
the
Ingress
point
for
this
flow,
so
there's
some
potential
solution
to
to
practice.
So
we
can.
D
We
can
monitor
this
traffic
to
know
what
the
the
acting
response
for
the
for
giving
flow
and
can
also
compute
the
effector
for
the
outer
potential
in
response
and
waiting
for
the
information
from
the
apps
we
may
ask.
But
for
this
we
need
some
extension
for
the
multitudinal
settings.
They
see
me
some
information
exchange
between
the
different
areas.
D
And
for
the
data
encoding,
they
will
be
we
discussed
yesterday,
so
the
status
of
listener
actually
controls
the
different
formats
to
store
the
information
Chris
from
the
pgbis
and
the
pgps.
It
depends
on
the
indentation
so
in
our
current
practice.
So
in
the
open
day
lab
it
can't
be
the
young
data
model
and
the
interactive
GPS
can
be
the
same
thing.
So
it's
it's
a
magnetic
because
they
help
the
equipment
to
the
effect
model.
We
just
scrap
the
industry,
work
slide.
D
And
we
also
got
some
the
useful,
some
other
useful
PHP
and
the
details,
extensions
and
it
can
improve
the
performances.
G
Yeah
thanks
GSM
for
our
presentation
and
so
any
any
comments
from
audience
for
this
proposal
and
I.
Think
I
I
want
to
make
a
few
comments.
You
know.
G
Okay,
Richard,
you
want
to
comment
first,
yeah.
B
Very
quick
question,
so
of
course
pdpr
is
even
one
of
the
main
use
case
of
hpis1
is
Auto.
One
is
a
pce
I'm,
always
wondering
because,
essentially
you
collect
using
bdp.
We
collect
the
the
essentially
that
policy
right.
B
They
know
the
links
and
prefixes
and
then
on
the
essential,
Auto
server
would
repeat
the
algorithm,
essentially
to
compute
layout,
to
compute
the
it
out,
but,
of
course,
in
real
life,
sometimes
they,
they
might
add
a
traffic
engineer
right
so
I'm
kind
of
curious
from
maybe
to
Luis
as
well
from
for
other
people
who
are
operators
how
common
there's
traffic
engineering
pathways
are
very
typical,
like
in
real
life.
B
Essentially
it's
a
purely
ospf
in
the
sense
you
know:
oh
Isis,
incredible
igp.
Therefore,
this
information
actually
most
times
we
mostly
accurate.
You
are
not
going
to
have
large
number
of
exception,
meaning
overlays
trap
engine
impacts
on
top
of
this
shortage
path.
Idp
path
is.
G
H
B
Okay,
so
my
understanding
is
the
volume
typical
because
the
in
most
routers,
you
have
a
relatively
complex
decision
process
right
and
you
have
a
bgp
to
basically
for
every
Central
i5b
right,
the
forwarding
engine
and
manager
or
whatever
you
call
it.
You
have
all
kinds
of
result:
decisions
like,
for
example,
you
know
igp,
which
I
think
using
pdpls
and
with
also
the
standard
algorithm.
You
can
compute
the
next
hub
and
then
you
have
pgp
imported
path.
B
You
also
have
you
know
you
basically
can
learn
the
password
from
multiple
sources
and
then
there's
essential,
ranking
system
right,
a
decision
system
to
say
okay
I'm,
going
to
pick
X.
So
for
my
quick
question
is
bgps
essentially
gives
you
one
of
the
multiple
sources
right.
B
It
gave
you
a
source
of
essential
liquid
topology.
Therefore
you
emulate
the
decision
process
of
the
igp,
but
then
there
are
other
traffic
engineering
overwriting
this
one.
Maybe
it
will
be
configured
with
higher
priority
for
given
destination
and
then
you
might
have
bgp,
you
know
imported
would
give
a
higher
decision
process.
You
know,
prefer
igp
approve
IDP
over
ibgp
prefer
eye,
you
know
ebgp.
So
my
question
typical
in
logical
systems
and
how
reliable
this
kind
of
emulation
of
igp
process
using
pdpos
LS
is
accurate
or
maybe
that's
not
essential
in
real
life.
C
G
Okay,
I
think,
thanks
for
clarification,
I
I
think
for
this
proposal.
You
know
bgbrs,
we
really
want
to.
You
know,
focus
on
you
know
the
topology
topology
that
can
you
know
Connect
using
bgbrse,
so
Jason
you
mentioned
you
use
a
bgp
update,
I
think
maybe
not
already
the
the
relevant
pattern.
Maybe
you
know
bgps
will
connect
to
you
know,
know
the
information
linking
information
we
should
focus
on.
You
know
how
these
kind
of
information
can
be
mapped
to
the
translated
into
the
Auto
Network
map
custom
map,
how
to
compute
the
you
know.
G
Network
map
customers
from
this
data
source-
using
you
know,
vbrs
and
maybe
bgb
I
was
a
young
beta
model.
Is
a
you
know,
another
interesting
Direction.
You
know
I
see
you
know
you
this,
the
you
know
open
daylight,
already
Implement
these
bgps.
As
you
know,
Standalone
beyond
the
model,
you
know
similar
to
the
BMP
young
data
model.
I
think
this
could
be.
You
know
one
of
the
direction
we
we
can.
You
know
Define
some
bgb
areas
younger
model
and
we
can
have
this
kind
of
data
source
and
second
part.
G
G
So
Ronan,
do
you
have
any
comments.
J
Just
a
short
comment
from
for
my
eyesight,
so
traffic
Engineering
in
general
is
of
interest
for
us
as
operators.
J
If
you
trust
the
poor,
let's
say
a
shortest
path,
then
you
get
a
little
bit
of
asymmetric
regarding
the
traffic
in
in
your
network
and
therefore
we
are
today
using
tools
like
equal
cost
multipath,
where
we
try
to
have
some
kind
of
distribution
or
load
shifting
balancing
of
our
networks,
and
in
that
sense,
if
there
is
let's
say
some
kind
of
information
that
that
can
use-
or
you
have
an
AP
I
that
gets
additional
information
and
in
a
sense
compared
to
what
it
does
Network
exposure
information.
J
I
think
this
might
be,
or
is
an
interesting
use
case.
It
for
sure
depends
and
must
be
evaluated
with
all
the
traffic
engineering
mechanisms
that
are
available
or
and
use,
because
here
it's
focusing,
let's
say
I
would
say
on
the
general
traffic
and
not
only
let's
say
that
an
application
server
can
distribute
traffic
by
or
is
handled.
J
So
we
in
in
general,
when
we
use
traffic
engineering,
it
is,
let's
say,
traffic
distribution
or
balancing
over
the
network
and
with
Alto
we
have
some
special
additional
requirements
which
might
be
yeah
interesting
in
the
sense
that
an
application
or
certain
features
could
could
have
a
dedicated
routing.
In
that
sense,
I,
don't
know
whether
this
draft
is
covering
this
or
just
focusing
only
let's
say
on
the
lower
layer,
but
I
think
this
is
something
where,
where
I
see
an
invalue,
let's
say
what
that's
my
personal
opinion.
G
Yes,
thanks
Ronan
to
Syria
opening.
Maybe
we
can
have
more
discussion
on
this.
For
this
part,
yeah
I
think
my
impression
you
know.
First,
they
should
just
focus
on.
You
know
the
lower
layer,
maybe
on
the
layer
but
overlay
part.
Actually
maybe
it's
a
complete
conformed,
Computer,
end-to-end,
Solution,
yeah
I
think
we
can
have
more
discounts
on
the
list.
For
this
part,
yeah
I
I
think,
let's
move
to
the
next
component.
E
So
this
is
something
that
Liz
and
I
have
prepared,
also
coming
from
the
ITF
one
on
the
last
idea
of
preparation
of
the
edge
Services
draft.
Also,
so
is
that
here.
G
But
president.
E
Thank
you
so
so
yeah,
but-
and
this
conversation
is
about
extending
out
to
to
support
compute
information
and,
as
you
know,
this
has
been
a
topic
that
has
been
discussed
for
a
while,
even
the
apps
actually
and
one
reason
that
we
have
revived
also
the
service
edge
draft,
but
dates
all
the
way
back
to
2019..
E
And,
of
course,
if
you
look
at
the
history
of
of
Alto,
starting
with
peer-to-peer
networks,
then
evolving
into
cdns
and
and
now
communication
networks
moving
into
the
edge,
basically
the
importance
of
the
edge
and
enabling
all
these
applications
XR
distributed.
E
Ai
training
inference
iot,
which
require
you
know,
exclusion
requirements,
QA
requirements,
right,
The,
Meta
versions,
one
so
in
the
way
we
think
of
Alto
is
that
it's
it's
very
well,
positioned
as
an
API
to
help
us
expose
this
information
to
the
application,
and
it's
not
no
longer
enough
to
have
communication
information.
We
also
need
to
have
compute
information
because
the
problems
we
need
to
about
you
know
placing
placing
workloads
and
placement
problems.
So,
given
our
work,
let's
say
an
aids
with
workload.
C
E
Nodes
running
in
the
edge
and
all
the
way
to
the
cloud
potentially
so
we
need
to
know
not
just
communication
information
again
with
latency,
but
also
compute,
rather
the
CPUs,
the
gpus,
whereas
the
memory
and
the
storage
so
that
we
can
make
optimize
placement
decisions,
that's
the
motivation
for
this
work
and
and
and
coming
also
from
Qualcomm.
This
is
also
where
our
interest
is
in,
as
we.
E
E
That
we
go
through
the
set
of
you
know:
potential
apis
that
are
available
in
blue
are
SEO
standard
apis
in
green
are
the
factor
standards.
So
this
goes
all
the
way
through
scanning,
through
Auto
accounts,
also
on
LFC
766,
for
virtual
machines,
dmtf
cntt
and
then
going
to
the
de
facto
standards
like
kubernetes
and
openstack.
E
This
table
you
can
take
it
offline,
but
the
way
to
think
of
this
is
that
this
is
a
way
to
understand
what
are
the
apis
available.
What
are
what
is
the
what's
the
running
code
that's
available
and
to
understand:
where
do
we
from?
Where
do
we
capture
this
information?
How
do
we
expose
it?
E
Then?
It
goes
into
cad's
an
outcome
discussion.
So
this
is
the
draft
the
service
edge
draft,
that
I
mentioned
that
we
have
provoked,
and
so
you
can
take
it
offline,
also
and
check
the
latest
updates
where
we
actually
show
some
some
use
cases
a
table
comparing
the
autocad's
Alto
versus
cards,
also
some
of
the
things
that
we
have
that
have
been
discussed
in
this
group
and
also
in
the
cats
working
group
and
then
the
the
slide
that
just
goes
through
these
different
apis.
E
So
some
examples
of
how
does
how
do
we
exact
compute
information
from
kubernetes,
so
examples
of
apis,
whether
it's
through
their
through
the
rest,
API
or
through
the
command
line
and
interface?
But
you
have
information
like
memory
CPU.
These
are
things
that
are
very
important
for
the
problems
you
want
to
address,
and
so
exposing
these
applications,
information
to
the
application
is,
is
relevant
for
us.
E
You
know
to
solve
displacement
problems,
so
this
kubernetes
same
thing
for
openstack
examples
of
how
this
information
can
be
queried
Etc
so
again,
just
as
coming
through,
what's
available,
whether
it's
a
standard,
an
sdo
standard
or
or
the
factor
standard,
and
then
this
idea
that
I
out
there
could
be
used
as
a
as
a
single
point
of
voluntary
as
a
neutral
API.
E
Some
of
these
Solutions,
like
kubernetes
openstack,
are
domain
specific
exclusions
or
solution,
specific
Technologies
or
apis
and
Outdoors
is
in
this
sweet
spot.
Where
you
know
it's
a
neutral
interface,
it's
an
open
standard
and,
and
so
it
could
serve
as
as
a
catalyst
to
help
expose
this
information
from
a
variety
of
sources
in
a
neutral
way.
Basically,
so
that's
the
proposal
here
to
to
help
before
this
idea
of
extending
the
protocol
to
to
incorporate
complete
information
for
the
New
Balance.
E
We
need
to
address
it
and
also
again,
of
course,
working
with
cats
to
avoid
Reinventing
the
wheel
and
some
of
the
work
that
auto
has
already
done
so
I
can
probably
stop
here.
G
Yeah
Jody
thanks
for
presentation,
I
think
we
need
to
I
think
this
worker
also
related
to
the
cat.
So
maybe
we
should
you
know
another
cast
people
to
know
what
is
going
on
in
Auto
and
how
this
is
related
to
the
cat
worker.
Also
I'm,
very
appreciative.
You,
you
know,
do
a
lot
of
homework
to
investigate
the
open
source
tool
to
use
these
open
source
to
to
overtake
or
gather
the
compute.
G
Update
yeah
yeah:
we
can
do
that
yeah,
okay,
I
think
almost
you
know
don't
have
a
time
to
for
for
the
children's
presentation,
maybe
yeah
we
can
skip
this,
and
so
anyone
who
have
some
you
know
last
words
for
this
presentation
method.
You
have
any
comments
on
this.
G
G
G
Probably
yeah,
we
can
conclude
this
session
and
I
think
we
hear
a
lot
of
interesting
proposal.
We
will
you
know
everybody.
Can
you
know
post
your
your
proposal
to
the
Lister
have
a
further
discussion
with
additional.
You
know
the
question
we
raised
on
the
list
that
we
say
how
how
this
question
can
be
addressed.
Oh.
E
G
E
Just
a
question:
do
you
have
a
suggestion
for
geofang's
talk
whether
he
should
bring
it
offline
to
the
mail
list
or.
G
Yeah
sure
how
maybe
you
you
can
post
to
the
manager
for
your
ideas
proposal?
Well,
what
is
what
is
the
current
status
operator
say
this
proposal
and
we
can
see.