►
From YouTube: IETF115-SAVNET-20221111-0930
Description
SAVNET meeting session at IETF115
2022/11/11 0930
https://datatracker.ietf.org/meeting/115/proceedings/
A
A
A
B
So
it's
a
Pity
that
my
culture
cannot
attend
the
meeting
and
you
know
have
your
participated,
media
remotely
foreign.
B
Setters
and
currently
there
are
already
nine
documents
that
have
been
uploaded
to
the
cell
net
reportery,
which
covers
the
problem,
segment,
use
case
and
requirement,
and
the
seventh
architecture
and
the
solution
analysis
aspect
and
for
the
first
step
for
the
first
part,
this
off
has
been
present
at
the
last
meeting,
and
the
author
of
the
lead
draft
has
clarified
the
the
concert
that
this
is
the
deal
in
the
last
practitioner
and
also
update
there
and
their
their
document
accordingly,
and
they
will
print
the
update
content
today.
B
The
second
part
is
the
cell
net
architecture,
and
this
is
the
first
time
to
practice
in
this
meeting
and
also
the
some
drops
that
for
the
some
initial
answers
for
the
essential
and
the
influence
for
the
new
solution
and
for
the
yeah.
Lastly,
there
are
also
some
detailed
solutions
that
aim
to
solve
the
similar
problem,
but
not
align
with
the
proper
sales
net
architecture
document.
We
will
find
some
manager
to
try
to
sing
up
them
later
and
now
we
are
in
according
our
Milestone.
B
B
B
And
after
that,
we
will
discuss
the
the
the
first,
the
initial
approach
to
the
solutions,
so
we
can
evaluate
the
capacities
and
and
the
as
a
requirement
for
the
that
we
have
provide
and
the
last
set.
We
went
to
Empire
that
currently
we,
our
our
initial
goal,
is
to
finish
the
discussion
of
the
problem
statement.
So
after
the
this
video
If,
there
is
no
more
issue
raised
and
we
will
take
the
adoption
code
on
the
mail
list.
B
I
have
introduced
the
and
the
first
section
is
the
problem:
seven,
there
are
three
parts
interest
and
the
incentive
incentive
policies.
After
that
we
will
have
an
open
discussion
for
the
document
and
decided.
What's
the
next
step
at
our
group,
we
will
forward.
B
And
the
second
part,
the
second
part
is
the
architecture
architecture
path,
and
we
also
will
introduce
the
info
as
solution
architecture,
interest
and
the
unsung
table
table
design
and
principle.
After
that,
we
will
introduce
a
I
want
another
solution
for
the
for
the
same
aim
and
the
the
last
one
is
the
entities
for
the
current
current
solution.
Okay,
so
I
think
this
is
the
WCR.
So
if
there
is
no
an
issue,
I
think
we
I
want
to.
B
Move
forward
to
the
to
the
first
petition.
C
C
C
Know
what
I
would
like
to
suggest
to
to
the
chairs,
of
course,
is
that
yeah,
let's
dedicate
the
time
to
that,
if
we
start
stretching
a
little
bit
at
the
time,
we
need
to
get
those
those
things
done
sooner.
There
will
be
more
time
to
talk
about
the
next
steps,
architecture
and
solutions.
E
F
So
stand
IG
and
the
Jewel
for
the
hard
work,
hello,
everyone,
I'm
Danny
from
chinhua,
University,
I'm,
sorry
again
that
I
have
to
present
the
okay
the
draft
remotely
this
time.
So
next
I
will
present
the
updated
version
of
the
introduction
rate
cap
analysis,
problem
statement
and
requirements
draft.
F
So
we
know
that
as
a
cause
of
this
draft
is
to
provide.
Is
it
a
couple
analysis
of
existing
into
domain
soft
mechanisms,
summarize
the
fundamental
problems
of
existing
instrumental
mechanisms
and
Define
the
requirements
for
the
new
info
domains
of
mechanism?
F
So
in
in
the
last
iatf
meeting,
we
presented
the
first
version
of
this
draft,
so
that
is
version
zero,
zero
based
on
the
feedback
we
got
in
the
last
meeting,
as
well
as
the
comments
and
discussions
in
the
mailing
list.
So
we
made
the
three
runs
over
revisions
for
this
draft,
and
today,
I
will
presenting
the
latest
version.
Today's
version
003.
F
F
The
first
scenario
is
for
multi-home
established
where
asymmetricology
may
need
to
improper
block
problem,
and
this
second
scenario
is
that
existing
in
into
the
massive
mechanism
that
is
English
feature:
okay
only
works
for
the
moving
traffic
Arrangement
here
from
the
sublet.
It
cannot
work
for
spoofing
traffic
from
in
One
Direction.
F
So
if
some
Edge
routers
do
not
deploy
English
featuring,
there
is
no
way
for
the
installment
Network
to
validate
the
spoofing.
Traffic
originated
from
the
subjects
that
directly
connecting
the
edges.
Authors.
Energy
fourth
scenario
is
a
misbehave
result.
The
last
time
we
presented
that
result
may
not
Faithfully
conduct
Southern
mechanism.
F
So,
since
the
last
iitf
meeting,
we
received
okay,
a
lot
of
comments
and
feedback
for
the
draft
we
presented.
We
made
the
summary.
There
are
four
major
comments.
The
first
is
that
why
couldn't
you
deploy
itself
at
all
doctors
in
the
into
the
network?
F
The
third
problem
is
that
misaligned
incentive
means
the
cost
of
deploying
soft
are
paid
by
operator
itself,
while
its
benefits
are
only
experienced
by
other
operators,
but
in
for
the
network
is
rarely
managed
by
multiple
operators
and
as
a
force
comment
is
that
are
we
talking
about
non-ip
package
as
well?
There
are
also
some
manner
comments,
so
we
just
meet
the
revisions
for
the
draft
to
address
all
the
concerns
we
received
and
the
main
updates
of
this
latest
version
compared
to
the
first
version.
Okay,
I
have
four
kinds
of
major
updates.
F
The
first
is
the
updates.
In
the
cap
analysis
part,
we
explain
the
reasons
for
partial
deployment
and
remove
the
scenario
of
a
misbehaviors
author,
so
for
partial
deployment.
I
think
we
had
okay,
not
all
of
our
discussions
in
the
business.
As
pointed
the
order,
okay
by
some
experts-
okay,
we
acknowledge
that
there
are
two
main
reasons
for
partial
deployment
of
into
domain
soft
mechanism.
The
first
is
that
the
technical
limitation
of
today's
into
domestic
mechanism
may
make
it
hard
to
deploy
itself
are
always
authors.
F
For
example,
this
Airbase
to
solve
requires
many
configuration
in
Dynamic
Networks,
stricter
your
RPF
Ingress
filtering
blocks.
Legal
traffic
in
the
scenario
of
asymmetric
technology,
I
think,
okay.
These
issues
are
pointed
out
by
some
okay
operators
in
the
mailing
list,
and
the
second
reason
is
that
some
results
cannot
support
staff
due
to
doctoral
capabilities
versions
and
vendors.
F
F
F
In
the
latest
version,
We
summarize
the
problems
of
existing
into
domain
soft
mechanisms;
okay,
as
okay,
just
a
three
major
problems.
The
first
problem
is
inaccurate
validation.
F
The
behavior
Gap
is
that
asymmetrical
routine
may
need
to
improper
block
scenario
in
okay,
multi
Home
Connection
case.
The
reason
is
that
existing
into
the
mess
of
mechanism
conducts
itself
based
on
the
local
feed,
which
may
not
match
the
real
data,
putting
forwarding
tasks
from
the
source.
The
second
problem
is
limited
protection.
The
behavior
Gap
is
that
okay,
we
may
feel
to
block
this
moving
traffic
from
outside
the
airs.
North.
F
The
third
update
is
in
the
requirements
part
we
remove
the
requirement
of
direct
incentive
at
the
requirement
of
acceptable
overhead
and
revised
the
description
of
other
requirements
in
the
latest
version.
We
summarize
the
requirements
for
allele
infra
domains
of
mechanism
as
a
okay,
three
parts.
The
first
requirement
is
that
the
mechanism
must
ensure
accurate
yourself.
The
new
soft
mechanism
should
match
must
match
the
real
data,
plane,
forwarding
parts
and
avoid
improper
block
under
asymmetric
analogy.
F
The
statement
requirement
is
that
the
mechanism
must
work
for
all
kinds
of
interdome
spoofing
traffic,
be
the
master,
be
able
to
validate
the
traffic
from
all
directions.
Specifically,
it
should
be
able
to
block
the
spoofing
traffic
from
the
offset
is,
or
from
the
unemployed,
edges
altered
and
the
okay.
The
block
should
be
as
close
to
the
source
as
possible
and
in
this
third
requirement
is
that
the
mechanism
must
not
induce
Max
overhead.
F
First,
we
okay
for
the
intro
domain
software
work
scope.
We
think
that
or
IP
encapsulated
scenarios
are
in
scope,
including
both
actually
before
and
IPv6
addresses,
and
now
IP
packets
are
out
of
scope,
because
our
purpose
is
to
validate
the
source.
I
shall
addresses
not
other
source
information
for
security
consideration.
We
stated
that
staff
needs
focuses
on
launching
protocol-based
mechanisms,
so
this
so
the
security
scope
of
each
other
myself.
That
should
be
similar
to
that
of
into
the
minority
protocols.
F
B
You
yeah
so
I
know
so
if
there
is
no
question
I
I
think
this
topic
has
been
discussed
solely
in
the
last
meeting
and
also
oh
okay.
So
there
is
the
The
Cure
in
the
for
the
question.
So,
okay
with
a
moment,
okay.
B
H
Yeah,
okay,
I,
have
a
question:
I
see
the
repeated
use
of
the
term
intra
intra
are.
Are
we
sure
that
that's
actually
the
correct
term
to
use
or
should
we
be
using
enter
int
ER.
F
Oh,
you
mean
the
term
iPhone
yeah,
the
I
pre
I
presented
the
instrument
part.
Now
we
have
the
interdomen
part
and
later,
but
I
I
am
wondering
which
part
you
mean
that
we
use
the
int
to
represent
the
intro.
H
Right
yeah,
I,
believe
I
believe
the
entire
Focus
and,
if
I'm
mistaken,
please
feel
free
to
point
it
out.
But
I
think
that
we're
looking
at
inter-domain
rather
than
intra
domain.
I
H
I
understand,
but
when
I
actually
read
through
the
documents,
it
seems
like
the
entire
Focus
almost
is
on
interdomain
anyway.
It's
just
a
linguistic
thing.
I
want
to
make
sure
that
that
we
are
accurately
representing
what
we're
discussing
here
for
the
majority
of
in
the
majority
of
the
documents
and
proposals.
That's
all
I
had
thanks.
J
Could
you
go
back
a
few
slides
because
I
must
apologize
not
having
completely
read
that
the
the
changes
between
the
drafts
I'm
looking
for
the
part
of
miscellaneous
engine
there?
It
is
the
one
word
one
more
further.
So
do
I
understand
that
you
removed
the
part
of
the
misaligned
incentive
and
I
was
wondering
what
the
reason
was.
F
Yeah
last
time,
okay,
we
met
assumption
that
into
the
network
may
be
managed
by
multiple
operators.
So,
okay,
there
is
some
deployment
and
incentive
problem,
but
a
lot
of
colleges
in
the
mailing
lists
argue
that
eagerly
into
the
network
is
managed
by
only
one
operator.
So
the
incentive
problem
primarily
exists
for
interdoma
part
not
for
instrument
part.
J
Incentive
it's
a
very
commercial
incentive,
but
actually
because,
when
you're
implementing
saf,
you
will
see
that
your
traffic
on
your
network
will
go
down
for
your
network
and
for
your
customers,
which
means
that
you
have
more
capacity
on
your
network.
This
is
a
commercial
incentive
that
I.
You
know.
People
fail
to
see
because
yeah
you
have
to
do
something
for
it.
You
have
to
convince
your
your
customers
or
your
connections.
You
know
to
to
to
sign
rowers
or
to
have
proper
IR
data.
F
Okay,
I,
you
just
had
a
very
good
question:
I,
don't
know
we
made
a
lot
of
discussions
for
the
incentive
problems
and
many
names.
Okay,
the
primary
concern
is
okay:
how
to
encourage
a
operator
to
benefit
from
the
south
mechanism
they
deploy.
I
think
we
will
have
a
lot
of
discussions
is
the
two
presentations
later
so,
okay.
K
F
Yeah
yeah
actually
for
our
two
parenthesis
nature.
We
just
want
to
translate
this
requirements
into
technical
problem,
so
I
hope
that
we
can
have
more
discussions
later,
but
I'm,
not
sure,
okay,
which
part
should
be
put
into
Instagram,
part
or
Internet
part.
Maybe
that's
a
okay,
a
problem.
We
need
a
more
discussion
later
yeah.
A
I
just
wanted
to
respond
to
the
last
comment
here
at
the
mic.
If
you
are
running
a
a
network
and
a
customer
says
it
works
on
provider
a
but
not
on
provider,
B
and
provider,
B
is
the
good
actor.
There
is
a
very
strong
commercial
incentive
to
relent
on
that,
and
what
we
have
to
be
very
cognizant
on
is
that
any
sort
of
source
address
validation
may
actually
break
more
things
than
it
is
preventing,
because
it
is
the
largest
attacks
that
we
see
today
are
not
from
sourced
from
spoofed
packets.
F
L
Carefully
and
send
the
term
in
the
mailing
list
and
all
the
issues
have
been
tracked
in
the
GitHub
so
and,
of
course,
welcome
to
join
the
main
list,
and
we
can
discuss
this
draft
further
to
improve
continuously
important
this
document
and
about
the
discussion
of
incentive.
I
think
that
there
is
no
question
about
whether
is
no
miscellaneous
incentive
in
the
infra
domain.
Networks
of
the
discussion
is
about.
L
Our
colleges
will
introduce
the
internal
domain
draft
to
presenter.
How
can
we
improve
the
incentive
or
for
SLA
mechanisms
by
technical
technical
improvements?
Of
course,
we
canceled
the
incentive
problem.
B
D
D
D
D
D
D
D
Misaligned
incentive
is
one
of
the
main
reasons
why
some
asses
haven't
deployed.
Bcp
38
several
Publications
have
also
indicated
the
misaligned
incentive
problem,
bcp3h
only
prunes
the
provider
who
deploys
served
from
originating
spoofed
Source
traffic,
but
doesn't
protect
the
provider
from
receiving
spoofed
traffic
or
being
the
victim
of
an
attack.
It
means
the
cost
of
deploying
itself
are
paid
by
the
AIS
itself,
but
the
benefits
are
experienced
by
the
rest
of
the
internet.
D
D
Finally,
we
conclude
the
two
fundamental
problems
of
existing
interdomenal
mechanisms.
The
first
problem
is
inaccurate.
Stuff,
existing
interdominal
cell
mechanisms
have
improper
block
or
improper
permit
problems.
It
is
because
that
they
all
conduct
serve
based
on
local
weave,
which
means
not
match
the
real
data
plan
forwarding
path
from
the
source.
D
D
We
Define
four
requirements
for
the
new
interdominal
sound
mechanism
requirements
number
one.
The
mechanism
must
ensure
accurate
self
by
matching
the
real
data
plan
forwarding
path,
it
should
avoid
improper
block
and
reduce
improper
bleed
as
much
as
possible,
since
this
requirement
cannot
be
met
by
using
only
local
Europe
information.
Additional
mechanisms
are
needed
to
provide
the
required
information
requirement.
D
Number
two:
the
mechanism
must
provide
direct
incentive
to
the
es
who
deploys
the
it
is
required
to
validate
traffic
from
all
directions
and
help
the
deployed
as
mitigate
spoof
Source
attacks,
including
reflection
attacks,
requirement
number
three:
the
mechanism
must
support
incremental
deployment,
it
must
prove
and
Source
or
just
spoofy
when
it
is
partially
deployed
in
the
internet
requirement
number
four:
the
mechanism
mustn't
induce
much
overhead.
It
must
avoid
data
plan
packet
modification
and
the
limit.
The
number
of
control
plane
protocol
messages.
D
B
M
Hi,
so
what
you
are
saying
about
the
reflection
attacks
if
the
this
is
sriram
from
nist,
if
the
victim
of
the
reflection
attack
is
on
your
pisps
customer
cone,
for
example-
and
you
are
doing
Source
address
validation
for
your
customer
for
your
customers.
In
that
case,
the
reflection
attack
will
be
detected
and
mitigated.
So
it
is
not
so
so
it
is
not
entirely
true.
M
The
statement
you
made
that
efp
urpf
cannot
mitigate
reflection
attack
in
general
I
mean
it's
only
any
reflection
attack
that
is
originating
in
your
customer
con
and
going
to
your
peers
customer
account
where
the
victim
is.
If
you
are,
if
deploying
in
into
domain
Sav
using
efp
or
PF,
you
can
definitely
mitigate
those
reflection,
attacks.
D
Well,
we
can
see.
M
In
the
scenario
I
am
presenting
to,
you
is
it's
you
and
a
lateral
peer
of
yours.
The
provided
question
is
somewhat
different.
If,
if
you
have
a
lateral
peer
and
your
customer
is
from
your
customer,
reflection,
attack
is
originating
and
it
is,
and
the
victim
is
in
your
peers.
Customer
cone,
the
efp
you
are
PF,
is
able
to
catch
that
and
mitigate
it.
That's
what
I'm
saying
so.
The
statement
you
made
is
not
true
across
all
scenarios.
M
G
So
I
think
it
is
stating
too
much
that
it's
it's
always
misaligned
incentives.
There
could
be
cases
when
incentives
are
very
much
aligned.
For
example,
it
did
a
value
that
if
you
can
detect
attacks
from
your
customer
cone
and,
for
example,
inform
them
that
hey
you
have
compromise
machines.
G
G
D
Okay,
we
want
to.
We
acknowledge
that
efp
or
ipf
can
provide
some
direct
incentive
compared
to
bcp38,
but
we
want
to
a
new
into
the
missile
maximum
should
provide
more
theoretic
detail
to
help
mitigate
reflection,
attacks
and
protect
at
all
directions.
Thank
you.
C
Hi
I'm
already
future
Technologies
if
we
go
to
slide
12
I.
Just
this
is
a
individual
participant
comment,
I
think
in
slide
trials
where
you
start
talking
about
some
of
the
requirements,
and
if
we
get
to
the
point
where
we're
going
to
build
Solutions
based
on
this,
we
need
to
be
very
specific
on
what
we
are
requiring
and
many
of
these
I
don't
know
what
they
would
be,
for
example,
just
to
pick
on
one.
The
last
one
must
not
induce
much
overhead.
C
Okay,
that's
great
I
mean
I,
understand
the
general
purpose
of
that.
But
what
does
that
mean
as
much
overhead
two
packets,
10
packets,
25,
big
packets,
small
packets?
You
know
what
does
that
mean?
So
we
need
to
you
know
quantify
some
of
these
things
a
little
bit
better.
I
know.
This
is
the
first
time
that
this
text
is
in
the
drafts.
C
I
think
so,
if
the
working
group
decides
to
keep
working
on
this
document
that
adopts
it
we're
going
to
want
to
look
at,
you
know
clarifying
these
steps
right
so
that
then
later
we
can
come
back
and
say:
oh
look
at
the
solution.
It
meets
that
requirement
right,
and
so
it's
easy
to
and
we're
not
arguing
whether
overhead
is
much
or
not
much,
which
is
not
a
very
technical
term.
C
The
other
thing
that
I
want
to
sort
of
throw
out
there,
and
this
may
be
a
longer
discussion
for
the
working
group
and
something
that
that
the
chairs
should
consider
in
discussing
is
the
charter
specifically
says
that
this
working
group
is
not
going
to
extend
existing
mechanisms
right.
C
It
doesn't
say
that
the
working
group
is
not
going
to
use
existing
mechanisms
as
part
of
a
bigger
solution,
so
you
know
some
of
the
arguments
about
yes,
this
solution,
existing
solution
solves
part
of
the
problem,
but
not
all
the
problem
and
depending
on
where
you
sit
and
look
at,
which
is
some
of
the
things
that's
the
room
was
talking
about
before
you
know.
We
need
to
consider
that
and
again,
if
we
get
to
the
point
where
we're
going
to
go,
build
Solutions.
C
C
B
H
Hi
Roland
Dobbins
Scott
Arbor
a
couple
of
quick
points
on
the
misaligned
incentive
discussion.
H
First
of
all,
it
needs
to
be
acknowledged
that
there
are
in
fact
incentives
for
some
Network
operators
to
not
deploy
Source,
address
validation
and
those
are
operators
who
deliberately
knowingly
are
actually
selling
transit
to
criminal
actors
who
actually
have
control
of
the
attack
generation
infrastructure
to
for
the
reflection,
amplification
initiation
traffic,
which
is
what
is
food
and
also
for
direct
pass
move
to
tax,
and
so
it's
important
to
understand
that
there
are
different
categories
of
incentives
and
there
are
in
fact
some
incentives
for
some
operators
to
not
enforce
social
address
validation
on
their
customers.
H
The
target
we're
also
seeing
spoof
direct
path,
attacks
down
significantly
as
well,
and
so
it's
important
to
understand
that
the
incentives,
the
economic
incentives
both
internally
within
the
perceived
economic
system
of
the
operator
as
well
as
key
externalities,
can
change
over
time
and
it
has
changed
and
we're
seeing
even
more
success
with
existing
mechanisms
that
are
in
place
today.
I
just
want
to
make
sure
that
everybody
involved
in
this
discussion
is
aware
of
these
facts.
Thank
you.
B
So
video
next
necessary
for
the
incentive
consideration.
I
think
this
is
the
most
important
point.
D
D
Therefore,
to
improve
the
adoption
of
Staff,
the
new
South
mechanism
must
provide
direct
incentive
if
a
network
deploys
sub
but
finds
that
it
only
helps
other
networks.
The
network
will
not
be
motivated
to
deploy
itself
if
a
network
deploys
staff
and
finds
that
sometimes
it
can
help
itself
compared
with
not
deploying
the
network
will
be
more
motivated
to
deploy
itself.
D
Nowadays,
Source
addressed
spoofing
can
be
used
in
various
malicious
attacks
and
is
mainly
used
in
reflection.
Attacks
to
conduct
a
reflection,
attack
and
attacker
will
force
the
victim's
Source
IP
address
in
requests
sent
to
the
reflector
server.
After
you
see
me,
the
spoofed
request
the
reflector
server
will
reply
to
the
victim
with
a
lot
of
messages.
D
D
D
Here
we
first
used
an
example
to
illustrate
why
efp
uipf
feels
to
prevent
reflection
attack.
We
simplify
the
participants
in
a
reflexive
attack
into
three
rules.
They
are
attacked
as
reflector
es
and
the
victim
is
in
this
example.
As1
is
the
attacker
as
as2
is
the
reflector
as
and
as3
is
the
victim
ES
to
conduct
the
reflection
attack,
the
attacker
forces,
the
victim's
Source
IP
address
in
the
spoofed
request,
sent
to
reflector,
assume
as2
and
as3
deploy
efp
uipf
in
this
case
both
efp
uipf
algorithm
a
and
efp
uipf
algorithm
B
field.
D
D
The
attacker
in
as1
will
spoof
The
Source
IP
address
of
victim
to
the
to
send
a
spoof
request
to
the
reflector
in
as2,
assume,
es2
and
as3
deploy
EFT
uipf
in
this
case.
If
they
apply
efp
uipf
algorithm,
a
as2
can
successfully
block
the
spoofed
request,
but
if
they
apply
efp
uipf
algorithm
B
FF
to
prevent
the
reflection
attack.
D
D
Foreign,
we
have
defined
several
requirements
for
the
new
inter-domain
sound
mechanism.
In
the
previous
presentation
for
the
seek
of
description,
we
temporarily
name
a
possible
new
cell
mechanism
as
soundnet,
since
there
is
no
specific
or
mature
subnet
solution.
Yet
we
assume
cell
net
could
meet
the
following
two
requirements.
First,
it
could
validate
traffic
from
all
directions.
D
D
We
still
use
the
simplified
reflection
attack
example,
but
with
enumerate
multiple
attack
scenarios
by
changing
the
relative
positions
of
the
three
rules
in
a
reflection
attack
assume
the
victim
Network
always
deploy
sound
mechanism,
because
only
the
victim
Network
benefits
from
the
defense
against
reflection
attacks,
then,
for
any
deploying
cases
of
the
other
two
networks.
We
check
whether
the
reflection
attack
can
be
prevented.
If
so,
the
victim
can
be
motivated
to
deploy
itself.
D
A
scenario
number
three:
where
the
attack
network
is
located
on
path
between
victim
Network
and
reflector
Network
for
all
deployment
cases
we
find
none
of
them
could
prevent
the
attack.
It
is
because
South
Fields,
when
victims
Source
address,
shares
the
same
incoming
interface
with
the
attacker
Source
addressed
in
the
South
world.
D
In
summary,
for
any
attack
scenarios
or
deployment
Keys,
we
find
that
soundnet
could
work
better
or
not
worse
than
efp
uipf.
Therefore,
a
network
could
have
more
essential
to
deploy
soundnet
as
the
sound
mechanism,
because
it
would
have
higher
probability
of
Defending
against
reflection
attacks.
Thank
you.
D
H
Guess
hi
a
couple
of
comments
here.
First
of
all,
there
is
a
an
overemphasis
on
reflection
amplification
and
an
under
emphasis
on
direct
pass
boost,
attacks
and
I.
Think
that
that's
something
that
we
need
to
address
throughout
all
the
documentation
and
and
discussion
here
about
the
problem
of
smooth
traffic
in
general.
Secondly,
there
is
insufficient
differentiation
between
the
spoofed
attack,
initiator
traffic
and
the
Amplified
attack
traffic.
You
know
the
Amplified
attack
traffic
is
actually
not
spoofed
on
the
leg
from
the
reflectors
amplifiers
to
the
targets
and
I
know.
H
Most
of
the
folks
involved
in
this
discussion
understand
that,
but
I
think
that
it
needs
to
be
clearly
documented.
Thirdly,
I
I
wanted
folks
to
understand
that
we're
increasingly
seeing
attacks
where
attackers
who
can
in
fact
launch
reflection,
amplification
attacks
are
deliberately
choosing
reflectors
amplifiers
that
are
topologically
cited
near
the
Target,
because
the
fewer
administrative
boundaries
that
the
Amplified
attack
traffic
has
to
be
traversed
provides
fewer
opportunities
for
detection
and
classification,
Trace
back
and
mitigation.
H
Another
I
guess
this
is
almost
a
question
to
my
knowledge,
efp
or
PF
hasn't
actually
been
implemented
in
any
any
routing
Hardware,
at
least
in
my
knowledge
and
I
personally,
don't
know
of
any
operators.
Who's
actually
deployed
it,
and
so
I'm
a
little
bit
puzzled.
By
seeing
these
statistical
comparisons
of
efp
RPF,
which
at
least
has
been
specified
but
to
my
knowledge,
has
never
been
deployed
and
then
some
assertions
about
savnet,
which
hasn't
even
been
fully
defined
so
I'm
kind
of
confused
by
this,
and
perhaps
the
previous
speaker
can
help
clarify
that.
D
Yes,
as
far
as
I
know,
efp
uipf
has
not
been
implemented
in
existing
routers,
maybe
eager
can
could
clarify
the
deployment
for
efpu
ipf.
So
we
are
talking
about
the
technique,
technical
analysis
between
EF
View,
ipf
and
the
7net.
So
there
is
no
mature
solution
yet
and
we
just
assume
a
possible
sound
mechanism
could
meet
the
requirements
we
proposed.
H
Yeah
I,
don't
think
this
is
actually
a
technical
analysis.
I
think
that
this
is
a
theoretical,
hypothetical
sort
of
positing
of
of
some
opinions.
It's
not
really
based,
in
fact,
thank
you.
B
So
I
think
open
discussion
for
the
for
the
schedule,
so
I
think
we
can't
continue.
We
can
also
begin
the
open
discussion
about
the
periods
would
be
cut
to
five
minutes.
Okay,.
M
You
can
call
me
sriram
I,
have
a
quick
comment
for
you
or
a
request
for
you
before
I.
Go
to
my
question:
I
put
a
request
in
the
chat
box
for
you
to
share
pointers
to
the
work
that
you
were
mentioning
earlier
about
measurements
that
show
that
the
Adidas
attacks
have
reduced
because
of
the
deployment
of
urpf
Etc.
I
would
appreciate
that.
M
D
M
Yeah,
so
one
thing
to
note
here
is
that
if
as1
as1
is
a
provider
of
as2
and
the
attack
is
assuming,
that
attack
is
not
originating
from
as1
itself,
because
it's
a
provider,
it
is
originating
from
a
customer
of
as1,
and
if
you
deploy
the
search
technique,
your
EF,
erpf
or
or
whichever
one
is
best
on
the
customer
side
of
as1.
M
That
is
where
that
attack
is
most
successfully
prevented,
because
you
cannot
move
up
the
chain
of
customer
to
provider
and
try
to
implement
these
techniques
as
a
very
effectively
at
the
higher
levels
in
the
higher
hierarchy.
So
what
you
have
to
do
is
you
look
back
down
towards
the
customer
cone
again
from
as1
down
to
into
its
customer
cone,
and
if
the
sav
mechanisms
are
implemented
they
there,
then
the
attack
is
prevented
in
the
cust,
even
like
it's
not
successfully
emerging
out
of
as1
to
proceed
towards
as2.
M
So
so
that's
something
that
to
keep
in
mind
and
as
you
move
higher
in
in
the
hierarchy
in
in
the
customer
provider
chain.
What
happens
is
that
it
it
becomes
I
mean
it
becomes
less
effective.
You
have
to
I
mean
unless
you
do
a
lot
of
things,
additional
a
lot
of
additional
overhead
unless
you
do
that,
what
happens
is
that
at
the
higher
levels
in
the
hierarchy,
it
is
harder
to
implement
it.
It
is
harder
to
achieve
the
objective
of
directionality
that
you
have
been
talking
about.
M
You
would
have
to
sort
of
resort
to
a
feasible
yeah,
a
loose
urpf.
M
For
example,
it
becomes
harder
to
implement
a
achieve
a
good
sense
of
directionality
at
that
level,
because
yeah,
basically,
because
the
the
the
complexity
is
higher,
the
customer
cone
is
much
larger
and
the,
and
there
are
difficulties
with
the
implementation,
of
course,
at
that
level,
but
at
the
lower
levels,
if
you
implement
it
on
both
sides,
not
on
the
not
only
on
the
on
the
on
the
customer
side
of
of
as2
as3,
but
also
on
the
other
side
of
as1,
where
the
traffic
is
then,
then
it
then
these
things
work
much
better.
M
So
that's
one
comment:
if
you
have
I.
D
Yes,
to
achieve
accurate
style
at
provider,
interface
is
much
more
difficult
than
a
TV,
accurate
style
at
customer
interface.
Yes,
but
I
see
8704,
it
recommends
to
deploy
loose
uipf
at
provider
and
the
pr
interfaces
and
deploy
efp,
uipf
air
algorithm,
B,
add
customer
interfaces,
but
to
simplify
the
menu
configuration
overhead
for
deploying
sound
is
required
to.
We
need
one
mechanism
that
can
validate
traffic
from
all
directions
instead
of
applying
different
stuff
at
different
directions.
D
M
Yeah
and
my
other
question
is
like
Roland
pointed
out-
what
you
are
describing
here
is
savnet.
You
haven't
described
what
savnet
is.
Perhaps
you
are
referring
to
the
another
one
I
think
that
is
the
architecture
document,
so
so
at
this
point,
I
think
it
is
still
like
a
continuation
of
the
requirements,
how
efp
or
PF
or
other
urpf
don't
meet
the
requirements.
How
and
how
the
the
new
requirements
for
savnet
will
overcome
the
limitations
of
the
previous
methods.
M
So
I
think
that's
what
you
are
doing
here,
but
but
you
don't
really
have
a
solution
that
you
can
call
savnet
at
this
point
unless
you
are
referring
to
the
another
document.
That
is
the
architecture
document,
so
I
would
I
will
be
careful
to
say
that
there
is
this
solution
and
it
performs
forms
better
than
efp
urpf.
L
From
our
technology,
so
this
document
some
something
about
this
document.
This
is
a
supporting
director
and
is
temporary,
so
it
should
incentive
is
a
whole
topic
on
how
many
discussions
in
the
mailing
list.
So
we
have
a
desire
to
make
a
individual
draft.
Actually,
it
should
be
included
in
the
inter
domain
problem
statement.
Crafter
and
the
traveler
is
actually
a
technical
analysis
result
presents
some
technical
analysis
results
way
to
an
auto,
give
some
implementations
and
something
may
be
misleading
in
the
presentation,
something
that
is
not
a
solution.
L
Maybe
two
more
Solutions,
like
information
is
provided
in
the
present
Edition.
Actually,
there
is
a
normal
pool.
Some
United
Center
can
perform
a
candidate
already
requirement
in
the
previous
problem
statement
after
okay.
That's
all.
B
So,
let's
begin
the
second
part
of
for
the
serenade
architecture
so
before
we
introduce
the
into
our
internal
domain
architecture.
For
the
author
will
like
to
introduce
the
cell
type
of
service
table
architecture
for
the
discussion
of
the
next
tools
like
okay,
so
I
can
change
the
slide,
really
quiet.
Okay,.
B
L
Hello,
everyone
glad
to
give
our
presentation
here.
This
chapter
is
another
solution
draft
we
will
describe
abstractly
the
social
surprise,
validation
and
some
typical
validation
modes,
so
that
we
can
easily
to
understand
which
kind
of
offers
ISO
weight
table
can
be
generated
or
used
in
different
scenarios
size
please.
L
So
why
there
is
such
a
document.
Isolate
tables
are
used
for
filtering
specific
social
prefixes
and
they
can
be
generated
manually
or
automatically
in
the
data
plane.
They
are
implemented
by
SL
office
or
some
other
resources.
It's
important
for
engineers
and
operators
to
understand
which
kind
of
associatables
should
be
used
in
particular
scenarios,
but
existing
so
mechanisms
have
their
own
data
structures.
These
data
structures
are
coupled
with
corresponding
underlying
implementations
and
there
is
no
unified
data
structure
in
for
the
SOA
table
in
different
scenarios.
The
levels
for
validation,
accuracy
and
the
validation
strictness
are
different.
L
Iso
weight
table
and
any
existing
tables
can
be
expressed
by
this
abstraction
based
on
the
abstraction
we
describe
the
four
typical
validation
modes
and
as
well
as
their
application
conditions.
Finally,
we'll
start
using
multiple
actions
for
different
operations
requirements.
L
L
Any
existing
is
waitable.
Generating
mechanisms
is
to
fill
this
table
in
the
table.
Each
cell
indicates
the
validity
state
of
a
pair
of
source
prefix
and
the
interface
the
state
can
be
valid
invalid
or
unknown
so
more
complicated.
The
table
is
the
more
strict
the
validation
is
and
the
more
accurate
the
table
is,
the
less
improper
block
and
improper
permit.
There
will
be-
and
there
is
only
one
isolated
table
abstraction
in
this
document
on
one
device.
L
Next,
please
or
we
can
configure
Excel
rules
to
do
Ingress
filtering
at
different
interfaces,
and
we
we
can
easily
express
this
kind
of
isolate
table
by
using
the
describe
the
expert
table.
Abstraction
under
another
example
of
full,
strictly
RPF
can
be
found
in
the
backup
slides
as
this
okay.
What
can
we
do
you?
By
using
the
abstract,
I
swear
table?
We
Define
some
typical
validation
modes
that
describes
the
typical
value
station
procedures.
L
Why
do
we
need
this
most?
Why
do
we
Define
these
modes
in
different
application
scenarios?
L
The
Suva
tables
may
have
different
levels
of
accuracy
and
validation
strictness,
so
we
Define
some
validation
modes
that
are
suitable
to
different
scenarios
and
decimals
can
help
us
to
easily
understand
which
kind
of
vegetable
can
be
generated
under
input
and
enabled.
In
the
data
plan
there
are
four
validation
modes
module.
One
is
interface-based
prefix
on
our
list.
We
are
very
familiar
with
this
mode.
Most
of
the
existing
activities
are
try
to
generating
accurate
terms
or
isolate
table
under
mode
one.
L
This
model
is
used
a
full
circle
in
great
filtering.
If
we
want
to
use
Mode
web,
we
need
to
learn
the
complete
set
of
source
prefix
that
may
allow
at
the
the
kindergarten
interface
and
the
mode
2
is
interface-based
prefix
block
list.
If,
if
we
are
sure
some
prefix
are
not
invalid
for
this
interface,
we
can
put
them
into
the
block
list
and
the
business
model
and
more
tool
will
have
mode
3
and
mode
4..
They
are
prefix
based
interface
on
our
list
and
the
prefix
is
the
block
list,
actually
the
that
are
similar.
L
So,
let's
talk
about
the
model
3,
only
most
trade
is
less
considered
in
the
existing
Sue
mechanisms.
If
we
want
to
protect
the
specific
Source
prefix,
we
can
use
modestly
or
model
4.,
and
if
we
cannot
obtain
the
complete
set
of
those
prefix,
we
can
also
use
this
mode
so
effect.
Yes,
by
choosing
different
modes
in
the
suitable
application
scenarios,
we
can
provide
as
much
protection
as
possible.
L
L
Yeah,
there
are
many
available
actions
that
can
be
picked
for
the
validator
package
that
there
are
permission
block
already,
limiting
or
sampling
UFO
invalid.
Even
for
the
package
with
invalid
State,
we
can
use
permeate
block,
really
limiting
and
sampling
actions
to
do.
We
can
use
sampling
actions
to
do
further
analysis,
for
example.
Yes,
there
are
for
meeting
interest
value
operation
requirements.
L
For
one
more
word,
let's
drop,
it
doesn't
focus
on
how
to
generate
and
implement
the
sway
table.
Yeah.
It's
a
supporting
craft.
A
Shared
much
so
if
you
go
back,
I
I
think
it
was
one
slide
where
you're
talking
about
you
know
the
potential
to
You
Know
sample
the
packet,
you
know
or
even
rate
limit
them
or
something
no
move
forward.
A
You
know,
but
it
you
know.
This
is
something
that
many
of
The
Operators
are
already
doing,
is
they're.
Actually,
you
know
they're
already
collecting
data
through
either
netflow,
ipfix
or
S
flow
based
collectors
to
go
and
Trace
back
when
there
are
spoof
packets
on
their
Network,
but
I'm
also
struggling
a
little
bit
in
in
a
multi-tenant
environment.
So
if
I
have
a
you
know
an
L3
VPN,
you
know
type
of
environment.
You
know,
I
may
not
know
all
of
the
source
IP
addresses
that
are
valid
in
that
entire
environment.
L
Thanks
Jared
yeah,
it's
an
interesting
question.
The
answer
is
no
wait.
Wait
wait!
Don't
assume
that
the
other
routine
protocols
will
work
for
some
net
yeah
or
we
just
said
that
we
can
take
a
sampling
action
into
consideration
because
it
will
help
us
to
analysis
where
the
texts
come
from,
so
that
we
can
do
better
Ingress,
filtering
or
yeah,
but.
A
Out
of
all
of
you
know
all
of
the
routing
protocols
involved
and
if
we're
going
to
try
and
describe
how
to
you
know,
hear
where
you're
talking
about
some
of
the
prefixes
are
known.
All
of
the
prefixes
are
known
and
some
are
known
and
some
are
unknown.
A
We
need
to
have
a
way
to
take
all
of
those
connected
interfaces
that
may
be
assigned
from
a
provider
an
internet
exchange
point
you
know
for
peering
or
something
like
that.
That
is
a
valid
source
address
for
a
packet,
especially
like
you
know,
an
exception
packet
like
a
icmp
error
or
TTL
expire
or
or
something
like
that.
We
need
to
ensure
that
all
of
those
are
still
flow
through
these
Networks
and
are
not
dropped,
because
that
is
important.
A
L
Yeah
thanks
yeah
we're
just
provide
some
options
for
the
validation
modes.
We
must
classify
which
model
can
be
used
in
the
Target
networks.
So
we
give
some
application
suggestions
in
the
draft
yeah
thanks.
B
K
B
I
think
a
little
little
closer
to
the
back.
Okay.
K
Oh
okay,
my
controller
slide
yeah.
B
Okay
controller
slide
now.
B
K
K
K
Oh,
is
it
gonna
work
for
me,
I
think
I.
Think
okay,
I
I
work
hard.
Okay,
let's
start
now.
This
is
the
phone
call
from
Jungle
International
in
Beijing.
The
topic
of
my
presentation
today
is
about
architecture
of
internment.
Satellite
I
will
introduce
this
in
the
following
three
aspects:
goals
and
the
requirements
analyzing,
some
like
flow
and
the
effect
of
Supply
implementation.
K
I
did
the
first
beginning:
let's
have
a
review
about
the
gaps
or
problems
of
existing
into
them
itself,
negatively
for
the
urpf.
It
have
the
limitation
on
celerio,
which
is
produced
at
the
improper
block
and
the
asymmetrical
routine
scenario,
and
it
also
have
the
limitation
on
the
directions.
It
cannot
cover
the
smooth
traffic
confirmed
outside
the
As
and
come
from
the
uncapable
or
undeployed
age
routers.
On
the
other
hand,
ACL
based
increase
as
filtering
seems,
walkable
and
more
cellulose
or
technical,
but
it
need
continues
and
complex
menu
configuration
to
Cable
accuracy.
K
So
if
some
men
want
to
resolve
all
these
issues,
the
following
functions
need
to
be
considering
or
provided
for
the
perspective
of
accuracy.
Supply
needs
to
obtain
completely
social
prefix
in
the
mighty
homie
solution
and
identify
accurate
interfaces
in
the
asymmetric
routine
scenario,
just
like
the
PBR
or
even
the
different
bi-directional
HP
codes
to
design
I.
Think
all
of
those
are
some
popular
and
ordinary
deployment
in
most
of
commercial
networks
and
for
the
aspect
of
Direction
Sublime
need
to
try
to
support
validated
traffic
from
all
the
directions.
K
If
it
could
work
and
more
routers,
you
don't
have,
it
will
have
the
chance
to
protect
the
subwoofing
traffic
as
close
to
the
source
as
possible
and
could
have
the
chance
to
block
the
suburbing
traffic
from
outside
of
the,
as,
on
the
other
hand,
providing
a
approach
to
generate
some
rules.
Dynamically
is
also
important
goals
under
requirement,
as
it
could
help
to
make
the
maintenance
course
to
be
affordable
and
the
correct
seven
rules
according
to
changes
in
the
real
time.
K
So
we
want
to
make
all
those
goals.
The
key
idea
of
sublight
is
to
try
to
generate
an
independent
and
accurate
subtitle
on
each
router,
which
follows
the
related
plane
exactly
its
Discovery,
the
related
platform
in
parts.
We
are
hoped
by
hope,
notification
and
then
generate
some
rules,
hope
I
hope
along
the
path.
After
some
rules
generated
on
the
deadline,
we
have
a
diagram
figures
showing
how
it
process
when
a
package
comes
in
before
it
look
up
the
field
table
button
pattern,
destination
IP
address
the
source
address
will
be
validated
in
the
subtable.
K
There
are
the
seven
rules
and
through
this
new
procedure
it
can
support
validated
package
received
from
all
directions
and
the
protect
the
source
profits
of
the
deployed
error
from
be
forged.
K
Okay,
I
will
show
you
the
overall
architecture
through
the
diagram
on
the
right
side.
There
are
two
components
on
the
control:
plane,
Sapa
and
SPD.
Both
of
them
are
targeted
to
generate
self
growth
on
a
data
plane
and
each
one
have
accuracy
bounding
protocol
message.
The
first
component
on
control
plane
is
the
spa
Source
prefix
advertisement.
It
has
three
tasks
and
first
they
need
to
obtain
completely
so
local
Source
prefix.
Then
it
will
support
to
generate
seven
rules
on
the
Ingress
interface
of
each
each
routers.
K
Then
support
SPD
to
generate
some
rules
by
provided
resource
prefix,
their
Spa
messages.
They
mentioned
the
SPD.
Just
now,
sapd
is
the
second
component
on
control
plane.
We
call
it
as
an
Source
pass
Discovery.
It
will
identify
the
accurate
incoming
interface
for
Source
prefixes
by
discovering
the
real
data
plane.
40
parts
then
generate
some
rules
on
network
to
network
interface
of
each
intermediate
routers.
K
K
Okay
in
the
following
tonight,
I
will
expand
on
the
co-op
flow
of
the
satellite
and
show
how
it
works
in
each
step.
By
some
examples,.
K
Record
Target
of
the
Sapa
is
to
obtain
the
completely
solved
prefix.
Let's
take
a
look
at
the
figure
on
right.
Each
router
in
the
domain
will
conduct
as
an
orange
return
and
lungs
is
also
prefix
to
all
other
routers
in
the
domain
by
the
spa
messaging
under
the
sapn,
receiving
router
will
generate
a
table
to
store
the
relationship
between
the
orange
root
ID
and
the
source.
Prefix
belongs
to
each
it's
just
as
shown
in
the
bottom
left.
K
If
something
sounds
not
so
complex,
I
want
to
highlight
here
in
the
my
homies
may,
take
a
look
at
the
root
value
and
the
rooted
2.
For
example,
a
tag
mechanism
will
be
adopted
here
for
the
source.
Prefix
synchronization
between
these
two
routers,
a
tech
value
either
identified
of
the
same
consumer,
and
it
is
taken
in
the
Sapa
message.
It
will
identify
the
source,
prefix
and
the
interface
belongs
to
the
same
consumer
and
have
to
go
to
the
computer
source
prefix
for
the
two
routers
and
the
interface
pair
next
after
Spa
procedure.
K
Let's
take
a
look
at
what
what
happened
in
the
subtable
of
the
age
routers
about
the
serverless
generation
of
the
Ingress
interface,
basically
take
them
some
table
of
root,
one
and
Rule
2,
for
example,
and
it's
autonomy
bottom
left.
The
record
in
the
black
color
is
the
source.
Prefix
comes
from
the
local
information.
So
it's
the
signal
since
with
the
current
submachry
zip,
but
the
record
with
green
color
is
the
source
prefix
achieved
by
Spa
synchronization
and
just
this
new
green
record.
K
Well,
that
is
true,
avoid
improper
Block
in
the
Dual
homine
cellular,
and
that
is
definitely
one
of
the
reason
why
adopt
a
mechanism
of
a
corporate
between
routers
and
counter
plane?
Okay,
then,
let's
go
to
the
next
step.
It's
a
source
pass
Discovery,
SPD,
plus
CU
SPD
procedure
is
Task.
2
generate
some
rules
on
the
intermediate
routers
and
the
network
to
network
interfaces.
K
The
main
idea
of
spt
is
neglect.
Each
routers
send
the
sapd
message
on
a
preferred
pass.
According
to
the
real
data
plane
information,
then
the
receiving
return
notice,
the
incoming
interface
and
funding
this
with
the
source
profits
and
generate
some
rules
on
the
date
plane.
Actually
SPD
is
operated
Group,
which
consists
with
three
sub
steps.
We
look
at
the
figure
on
the
right
this
time
we
take
the
root
3.
For
example,
unit
3
will
generate
a
orange
spt
message
for
the
prefix
of
P1
and
P2
on
the
left
to
the
next
hope
group.
Six.
K
After
the
receiving
sap
team
message
from
root
3,
it
will
generate
some
rules.
According
to
the
incoming
interface
at
first
then
execute
the
remaining
operator
to
operation
to
the
received
SPD
message.
Then
it
will
going
to
the
next
hope
of
root.
5.
out
of
the
five
operation
is
the
similar
with
Root
6.
So
we
are,
we
will
repeat
here
and
go
ahead
at
last.
The
SPD
message
arrived
at
reach,
the
last
node
of
the
destination
P1
and
P2.
The
root
1
and
root
2
will
generate
sub
rows
at
first.
K
According
to
the
incoming
interface,
then
execute
determination
to
the
received
SPD
messages,
and
here
now
the
whole
spt
procedure
is
done
and
some
rules
generated
on
each
hop
of
the
whole
spt
pumps
I
want
to
highlight
here
is
that
in
order
accuracy
of
that
incoming
interface
during
the
four
SPD
procedure,
all
factors
that
influence
the
volume
Parts
just
negatively
performing
or
PPR.
We
are
determine
the
directions
and
outgoing
interface
of
the
sap
message
on
each
hub.
K
Okay,
that's
all
of
the
co-op
flow
and
they
have
a
quick
summary
again.
It
is
the
benefit
achieved
by
sublet
and
comparing
with
the
existing
star
magnitude.
We
look
at
the
diagram
again
by
this
new
mechanism.
Some
light
can
apply
to
more
Solarium,
just
connect
to
the
1.1
money,
homie
and
1.2,
or
symmetrical
routine,
and
the
supply
can
protect,
deployed
errors
from
spoofing
attack
on
all
directions.
K
K
We
are
focused
on
the
workflow,
considering
most
of
the
people
were
interested
with
the
convergency
and
the
partial
deployment
scenes,
so
I
simply
missed
some
design
principle
of
them
here
for
the
convergency,
the
principle
is
balancing
performance
and
overhead,
so
it
will
be
first
new
sub
rules
installing
and
snow
some
rules.
Removing
the
best
mechanism
for
convergency
is
a
mixture
of
periodical
updating
and
aging
time
experiencing.
K
If
we
want
to
fast
the
convergency,
a
trigger
updating
mechanism
or
backup
sub
rules,
preparation
in
the
if
I
asked
celerio
is
also
could
be
adopt
and
for
incremental
deployment.
The
principal
leader
to
con
concerned
about
the
device
capability
and
the
igp
arrow
design,
it
isn't
recommend
to
deploy
from
the
higher
ability,
device
and
aggregation
or
core
layer
and
the
backbone
area
of
the
ATP,
and
then
it
expands
to
the
lower
capability
device
access
nail
and
no
backbone
error
of
the
ICP
going
to
be
step
by
step.
K
It's
just
some
simple
principle
here
we
can
have
further
discussing
offline
and
Marine
list.
Okay
is
the
final
summary.
We
mentioned
the
goals
of
introdomen
subnet
and
an
architecture
of
into
domain
Sublime,
which
is
consisted
by
spa
and
sapd
procedure
on
control,
plane
and
the
subtable
in
the
display
Advanced.
The
architecture
is
protocol.
Independent
extension
of
the
routing
protocol
are
not
focused
and
mentions
in
today's
document.
Okay,
that's
all
about
the
topic.
B
Okay,
London
I
think
because
the
time
is
critical,
so
we
must
make
a
comment
every
conveniently.
Okay,.
H
Yeah
very
quickly,
I
just
want
to
point
out
that
in
the
intro
domain
case,
it's
important
to
understand
an
engineer
domain
case
for
that
matter
that
the
majority
of
spoofed
attack
traffic,
whether
it's
direct
path
to
dance
attack,
whether
it's
a
spoofed
attack
initiation
traffic,
that's
used
to
stimulate
reflection,
amplification
attacks,
the
majority
of
this
traffic
is
actually
generated
by
server-class
machines
in
data
centers,
internet
data,
centers,
idcs
and
there's
a
well-known
mechanism.
It
was
actually
described
in
an
informational
Roc
which
I
put
in
the
chat.
H
That's
called
draft
Baker
Saba,
Cisco,
IP,
sourcegard,
zero,
zero,
it's
called
IP
sourceguard
and
it's
a
mechanism
that
actually
operates
at
the
switch
level
and
I'm
kind
of
it's
surprising.
That
I
haven't
seen
any
discussion
of
that
in
in
this
working
group,
especially
in
the
intro
domain
use
case,
and
then.
Secondly,
when
we
come
to
Cable
types,
cmts
types
of
access
systems,
there's
another
mechanism
called
cable,
Source
verify
which
is
very
similar.
H
There
is
no
draft
that
actually
describes
us
to
my
knowledge,
but
it's
not
patented
technology,
and
so
I
think
that
we
need
to
take
a
look
at
this,
especially
since
it's
this
is
a
mechanism
that
is
an
intra-domain
deployed
mechanism
that
can
work
quite
well.
I
just
want
to
make
sure
everybody
in
the
working
group
was
aware
of
these
two
mechanisms:
that's
all
I,
oh
and
finally,
there's
a
typo.
H
The
convergence
ends
with
an
e
and
not
with
a
Y.
Just
a
minor
thing,
thanks
very
much.
That's
all
I
had.
K
C
Future
with
Technologies
comment
as
an
individual
participant.
So
when
I
look
at
this
architecture,
you
know
clearly
that
table
that
we're
trying
to
build
seems
to
be
the
focus
of
the
architecture.
However,
this
document
defines
or
is
based
on
the
Assumption
of
a
specific
solution.
Right
there
are
many
mentions
in
the
presentation
around.
You
know:
periodic
updates
and
even
a
discussion
at
the
end
around
incremental
deployment,
and
how
do
we
converge
faster
and
things
like
that
which
are
related
to
a
solution
right,
not
related
to
what
the
actors
might
look
like.
C
The
point
here
is
that
you
know
the
architecture
can
still
be
met
with
different
solutions,
and
what
I
would
like
to
see
is
this
document
to
be
a
general
document
of
what
we
would
like
to
again
the
high
level
architecture
to
be
not
the
solution.
Solution
is
going
to
come
later,
and
it
is
something
that
we
will
need
to
to
discuss.
Of
course,
so
I
think
that
again,
this
document
can
definitely
be
generalized
without
losing
the
the
advantages
they're
trying
to
be
to
be
presented
here.
Thanks.
N
L
We
will
share
some
preliminary
ideas,
that
is,
to
narrow
the
gaps
of
existing
inter
domain
as
a
way
mechanisms
here
is
the
three
main
gaps
of
they
are.
They
have
been
described
in
the
problem
statement.
Draft
first
is
in
power
permit
in
part,
and
the
second
is
in
power
block
and
the
third
one
is
running
in
San
Diego
next
time,
please
so
how
to
narrow
the
gaps
we
think
about.
L
Oh
okay,
so
hot
narrow,
the
gaps
we
think
about
generating
accurate
as
we
rules
at
all
directions
of
the
validation
as
the
validation
s
is
a
lesson
conductor,
social
class
validation.
There
are
two
key
points.
Firstly,
is
accurate
as
well
rules
that
means
I
call
The,
Source
prefix
and
the
accurate
incoming
interface
and
second,
is
all
at
all
directions
generate.
That
is
generator
rules
at
our
interface
of
a
interface.
B
L
Are
two
expected
results
if
those
one
is,
if
we
have
accurate,
more
accurate
as
well
rules,
we
can
reduce
improper
permit
as
a
provider
interface
and
the
incentive
can
be
improved,
can
also
be
improved
a
little,
and
the
second
result
is
if
how
accurate,
as
we
rules
as
a
customer
interface,
where
you
can
avoid
the
improper
block
and
the
improper
parameter
in
some
particular
scenarios.
Yes,
please
tour
trip.
This
goal
is
not
easy.
We
need
to
overcome
some
challenges
for
validation,
as
if
the
validation
S12
generator.
L
As
we
rules
for
the
source,
prefix
Source
prefixes
of
orange
AIS,
there
are
usually
Three
Steps
step.
One
is
to
get
the
source
prefixes
of
the
100,
as
Step
2
is
to
obtain
the
real
incoming
direction
of
the
package
coming
from
the
100
as
and
the
substrate
is
to
generate
as
well
Rules
by
bending
the
source
prefixes
to
the
incoming
interface.
The
channel
one
comes
from
Step
One.
How
can
we
get
the
accurate
and
complete
set
of
photos?
L
Prefixes
of
of
the
orange
as
and
the
challenge
tool
is
how
to
get
accurate
and
complete
incoming
directions
of
the
only
AIS
by
using
the
local
rape.
Information
is
not
enough,
so
the
main
idea
is
to
allow
axis
to
cooperate
with
each
other.
They
can
exchange
some
extra
information
to
address
challenge
one.
We
we
use
Source
prefix,
either
attachment
process
that
is
Spa
process.
The
main
idea
is
simple:
so
orange
s
can
or
knows
the
heightened
prefixes
to
the
validation,
as
even
the
prefixes
are
not
or
non-stop
by
pgp
the
validation.
L
S
can
learn
the
prefixes
owned
by
the
owner
or
original
AIS
through
the
spa
message.
This
process
can
be
used
in
the
DSR
scenario.
Yes,
please,
and
the
second
is
SPD
process.
Only
as
can
advertise
is
preferred
as
passes
to
the
validation
as
soon
so
as
passes,
connects
to
the
orange
s
and
the
validation.
As
then,
the
validation
s
will
learn
the
real
incoming
Direction
and
any
other
directions
will
be
considered
as
invalid.
You
and
the
their
relationship
between
the
neighbors
and
the
validation
is
is
set
to
be.
L
If
you
see
our
V2
here
next,
please
yeah
here
are
some
Financial
improvements.
Improvement,
we
can
deal
with
the
improper
block
problems
in
the
DSR
scenario
and
second
is
eliminated
in
proper
Block.
In
the
no
export
scenario
next
and
the
improved
3
is
about
the
incentive,
we
can
have
a
relatively
more
strict
validation
as
a
provider
interface
and
the
reflective
attacks
can
be
blocked
here,
so
the
social
prefix
can
be
protected
by
the
validation
s
and
finally,
is
no
improper
parameter
in
yeah.
L
Okay,
of
course,
tour
child
is
go
by
using
spa
and
the
SPD
is
not
enough.
If
the
as
fast
in
the
control
player
and
the
data
plan
are
not
inconsistent,
the
results
may
be
not
good.
I
mean
there
may
exists
in
proper
block
problems.
L
Advertise
the
redirection
as
access
to
the
Andreas
so
that
the
validation
s
or
can
also
learn
the
redefraction
pass
from
the
Orange
is
yes,
please
here
are
some
considerations.
We
have
some
analysis
in
the
draft
next,
please
yeah,
really
a
quick
presentation,
because,
due
to
time
limitation,
this
architecture
is
protocol
independent,
welcome
to
Lee.
We
are
comments
or
questions
in
the
chat
box
and
the
mailing
list.
We
will
respond
to
term
one
member
thanks.
B
This
is
yeah,
I,
don't
know
who
do
you
like
to
also
I
think
we
have
one
minute
to
foreign.
M
G
You'll
drive
the
slides.
Okay,
thank
you.
Hey
everybody!
I'm
Igor!
Last
time,
last
ATF
we
talked
a
little
bit
about
barthev,
our
one
of
our
Solutions.
This
is
a
work
we're
doing
with
cider
apps,
but
it's
super
relevant.
So
we
wanted
to
make
sure
we
keep
you
appraised
of
what
we're
doing
next.
G
Thank
you.
So
this
is
just
a
list
of
requirements
for
Sav
as
we're
looking
at
them
again
we're
not
doing
we're
not
boiling
the
ocean,
we're
just
protecting
a
network
looking
at
only
the
customers
and
peers.
We
are
not
protecting
from
traffic
from
coming
from
provider
just
to
before,
but
anyway,
the
requirements
for
service
seniors
are
the
most
important
thing
by
far
is
improper
block?
G
That's
basically,
if
you
have
improper
blocks,
that's
what
gets
you
not
implemented,
because
that's
that's
a
network
optimized
for
the
most
don't
break
their
customers
and
obviously
improper
permit
or
otherwise.
What's
the
point
so
just
can't
emphasize
it
more
the
rate
of
improper
block,
you
should
really
be
competing
against
loser
payoff.
G
So
that's
that's
the
goal
and
comes
hand
in
hand.
A
high
quality
implementation
is
required,
no
matter
if
your
algorithm
is
vague
or
too
complicated,
and
it's
not
implemented
right
like
it's
almost
as
good
as
you
don't
have
a
good
algorithm.
G
Next
is
incremental
deployability
super
important
business,
economic
incentives
as
a
key.
If
you
are
the
first
adapter
and
you
don't
get
benefits,
you
don't
implement
it,
and
therefore
we
don't
have
adoption,
especially
there
should
be
benefits
for
early
adopters.
G
So
there
are
some
benefits
to
protecting
just
traffic
going
up
from
your
customers
or
from
your
peers.
I
mean
your
reputation
of
uis
is
go,
is
important.
Stopping
your
traffic
telling
your
customers
that
they
have
a
problem
is,
it
could
be
useful,
useful
service
and,
like
it
was
said
before,
if
you're
spending
a
lot
of
bits,
a
lot
of
capacity
and
garbage.
Well,
that's
your
network
capacity
too,
and
depending
on
your
cost
structure,
it
may
be
actually
a
good
idea
not
to
pay
for
transit
for
traffic.
G
That's
garbage
all
right
and
ease
of
adoption.
It's
usually
an
overlooked
Factor,
but
it's
very
important.
Human
factors
are
huge,
so
it's
very
important
thing
and
network
effects
that's
kind
of
later.
If
everybody
is
adapting
in
this,
have
maybe
some
government
will
require
it,
but.
G
G
Not
gonna
the
next
slide,
please
thank
you,
I'm
not
going
to
describe
it
in
detail.
Basically,
the
power
save
is
just
taking
BJP
data.
That's
usually
that
other
algorithms
are
using
and
augmenting.
Is
this
rpki
data
for
cases
where
there
is
a
symmetric
routes
and
or
any
other
sort
of
traffic
engineering?
We
can
get
information
about
prefixes,
as
well
as
customer
relationships
from
rpki
through
the
detailed
presentation,
so
I'm
not
going
to
go
into
much
more
detail
here.
Next.
G
Again,
this
is
a
general
operation
of
it.
You
can
read
the
slides,
basically
compute
the
customer
code
by
following
asps
relationship
in
bgp
as
well
as
asba
when
available.
Once
you
have
the
customer
code,
which
is
set
of
os
numbers
in
that
your
customer
for
any
particular
interface,
you
can
build
a
separate
interface.
You
look
at
again
DGP
matching
originating
AES
and
prefix,
as
well
as
raw
data.
G
That's
registered
and
compute
the
set
of
cider
blocks
and
that's
your
as
filter
for
the
save
next,
and
this
is
basically
my
last
slide
how
we
are
measuring
against
improp
what
we're
doing
for
our
requirements.
We
are
trying
to
reduce
person
for
most
improper
block
by
discovering
more
prefixes
and
more
as
numbers
using
information
not
available
in
BJP.
G
G
It
doesn't
require
signal
for
from
anybody
else,
so
it's
Deployable
and
it
doesn't
require
people
to
learn
any
new
technologies,
Beyond
bgp
and
rpki.
So
it's
like
easier
for
people
to
understand
right.
It
doesn't
require
new
hardware
really.
As
long
as
you
have
a
little
bit
of
memory
for
the
new
sub
list,
you
can
use
your
an
existing
Hardware.
You
just
need
to
make
sure
you
can
load
it
and
that's
about
it.
B
H
To
make
a
journal
comment
that
this
direction
seems
to
be
to
me
to
be
more
in
alignment
with
the
actual
statement
in
the
working
group
Charter
about
making
use
of
existing
mechanisms
and
not
extending
mechanisms,
but
rather
we're
actually
looking
at
potentially
weaving
them
together
and
so
of
all
of
the
approach
potential
approaches
that
have
been
discussed
so
far.
This
General
approach
to
me
seems
to
be
the
one
that
at
least
a
vector
that
has
a
higher
potential
degree
of
success
than
some
of
the
others.
H
N
Chair,
can
you
hear
me
yeah?
Okay,
okay,
thank
you.
I'll
be
quick,
hello,
everyone!
This
is
mungshao
from
new
h3c
technologies.
This
presentation
is
about
analysis
of
some
data
playing
Performance
Based
on
an
implementation
of
independent
step
table
foreign.
N
N
N
In
order
to
improve
the
accuracy
in
new
format,
mechanisms
are
expected
to
generate
step
rules
based
on
the
real
forwarding
path,
so
independent
self
table
may
be
required.
Actually,
the
existing
efp
uipf
already
requires
independent
self
table
or
referred
to
as
the
RPF
list,
but
efp
record
has
never
been
coded
in
the
draft.
We
describe
a
possible
method
to
realize
independent
cell
table
on
data
plane,
as
shown
in
the
figures.
Each
line
count
has
a
local
table
which
stores
part
of
same
rules
that
are
related
with
its
own
interfaces.
N
The
keys
of
local
table
are
Source
prefix
and
valid
incoming
interface,
and
no
result
zone
is
required.
A
global
table
is
also
used
to
store
all
the
known
source
prefixes,
which
is
a
union
of
source
prefixes
in
all
local
tables,
and
the
figure
shows
the
packet
processing
procedure
and
a
you.
You
can
read
the
draft
for
more
details
or
just
a
scoop
here.
N
Yeah,
this
slide
shows
the
performance
testing,
including
some
existing
mechanisms
and
the
independent
subtable.
The
tested
equipment
is
an
interesting
CCR
device.
The
test
of
Graphics
are
IPv6
small
package
Flows
at
line
rate.
My
performance
is
evaluated
by
PPS
compared
with
playing
forwarding
the
performance
loss
of
different
theft.
Mechanisms
are
all
around
five
percent.
The
performance
of
independent
self
table
is
roughly
equivalent
to
uipf.
N
Yeah
and
the
next
step
is
to
revise
the
draft
according
to
feedbacks,
and
any
questions
or
comments
are
welcome.
Thank
you.
E
Yeah
I
was
I
was
curious
based
on
the
requirements.
There
is
there's
only
a
few
requirements
that
I
saw
in
the
analysis,
drafts
and
I
was
wondering
how,
like
we've
seen
this
SPD
table
stuff
and
I'm
just
new
to
this,
so
excuse
my
unfamiliarity
with
it,
but
I've
seen
the
SPD
table
stuff
as
kind
of
one
solution,
maybe
not
an
architecture,
but
it
looks
like
a
specific
solution
and
then
the
the
bars
have
stuff
is
has
the?
Is
there
any
analysis
that
the
group
has
done
on?
E
N
B
H
H
Very
helpful
if
those
could
be
allowed
to
examine
the
source
code.
Thank
you.
B
Okay,
thank
you
for
overcoming,
so
we
must
close
the
media
and
for
the
foreign,
okay.