►
From YouTube: IETF-IDR-20230424-1400
Description
IDR meeting session at IETF
2023/04/24 1400
https://datatracker.ietf.org/meeting//proceedings/
A
B
A
A
B
C
D
A
E
A
A
Thank
you
we're
glad,
thank
you
for
all
our
presenters
for
speaking
into
the
mic.
This
is
no
well.
Hopefully
all
of
you
have
seen
it.
If
not.
This
is
a
reminder
that
the
ITF
policies
are
in
effect,
such
as
patent
or
code
of
contact.
This
note
is
just
to
point
you
in
the
right
direction
reminder
if
you're
participating
you
agreed
to
follow
the
ITF
process
and
policies.
A
If
you're
aware
of
any
contribution
that's
covered
by
Patent
or
application,
you
must
to
close
that
fact
or
not
participate
in
the
discussion
and
as
a
participant
in
ITF
activities.
You
acknowledge
that
any
written,
audio,
video
or
photographic,
whatever
records
of
the
meeting,
will
make
public
and
personal
information
you
give
to.
Itf'd
is
handled
as
accordance
to
the
ITF
privacy
policy
and
we
agreed
to
work
with
one
another.
Please,
if
you've
not
seen
this
read
it
once
because
we
have
a
little
extra
time.
A
I've
actually
gone
through
it
and
said,
say,
see
your
note.
Well,
our
agenda
is
meant
to
cover
presentations
that
didn't
make
it
to
ITF.
Once
in
time
time
slots
you
can
get
the
materials
there
you're
on
the
meet
Echo
I'm
sure
that
G
and
your
chairs
would
love
it.
If
you
help
with
any
note-taking,
especially
if
you're
a
a
presenter,
please
make
sure
we've
got
any
comments
you
want
to
make
sure.
Are
there.
Our
agenda
is
ahead,
we'll
start
with
anchor
we're
going
to
go
with
first
adjustments
to
standard
bgp
timers.
A
A
A
G
E
G
On
the
east
coast,
in
the
US
and
good
afternoon,
for
folks
in
Europe
and
good
evening
for
folks
in
India,
China
and
the
other
Asian
countries
all
right,
so
this
the
topic
is
applying
TCP
user
time
Mark
parameters
to
pgb
sessions,
presenting
presenting
on
behalf
of
myself
and
the
course
Robert
razuk,
all
right,
just
click
on
the
next
page.
It's
not
moving
I
think
I'm
not
familiar
with
this
Sims.
G
Okay,
it's
good!
Okay
looks
like
there's
a
delay.
Okay,
let's
let
me
start
with
by
talking
about
what
it
is
the
TCP
user
timer
is.
It's.
G
It
is
it's
a
very,
very
basically
old
concept.
It's
actually
defined
in
the
pgp
in
the
TCP
based
spec,
I,
see
0793
and
also
with
the
latest,
updated
IC
9293.
It
basically
says
you
with
this
timeout
you.
Basically,
the
TCP
connection
should
be
terminated
if
it's
not
progressing
using
this
specified.
Timeout
duration.
G
So
the
interim
this,
if
you
look
at
Let's
I,
think
what
is
particularly
relevant
is
about
the
delivery
of
of
application
data
by
TCP.
They
are
basically
they
are
really
different
types
and
Concepts
one
is
basically
is
the
transmitted
data.
It's
basically,
the
the
data
has
been
transmitted
by
the
local
TCP
and
it's
waiting
for
acknowledgment
and
the
second
one
is
buffer
data.
Basically
this
if
the
data
has
not
been
transmitted,
it's
waiting
to
be
transmitted
because
the
remote
side
has
advertised
a
zero
size
window,
TCP
window
so
locally.
G
We
can
understand
anything,
although
this
is
there's
a
slight
confusion.
I
think
in
some
of
the
in
some
of
the
the
document
that
reference
this
gcp
user
time.
It's
basically,
it
didn't
really
call
out
the
second
part,
but
as
far
as
application
is
concerned,
there's
no
difference
between
these
two
cases.
Its
application
data
is
not
delivered
within
certain
period
of
time.
There's
no
difference
at
all
right
so
this
that
so
that
the
user
timeouts
would
apply
equally
in
both
cases.
G
I
have
not
asked
people
around
about
other
implementation,
but
there's
a
there's
a
Linux
implementation.
This
is
just
an
example,
and
so
it's
basically,
if
you
look
at
it,
is
the
implementation
you
introduce
a
TCP
socket
option
called
The
tcps
Go
user
at
school,
Time
March.
So
basically,
application
can
simply
call
this
safe
stock
option
to
enable
this,
the
timer
value
is
specified
in
milliseconds.
G
And
also,
if
you
want
to
Let's
retrieve
the
the
time
and
value
you
can
do
a
gay
stack
option
using
the
using
the
same
TCP
user
time
options,
what
it
does
that,
if,
if
basically
for
the
for
the
two
cases
that
show
in
the
previous
slide,
if
either
of
these
in
either
case,
if
the,
if
the
data
is
a
user
application
data,
is
not
delivered
using
the
specified
camera
value,
it
basically
will
return
easy
timeout,
the
error
to
the
application
in
the
in
the
TCP,
the
read
or
near
Circuit
read
or
write
operation.
G
It's
just
like
any
other
circuit
errors
like
if
the
remote
side,
a
closer
circuit
you,
the
application
locally,
would
receive
e-packed.
So
it's
this
is
this.
There
are.
There
are
a
number
of
errors
that
the
application
has
to
deal
with.
This
would
be.
One
of
this
would
be
a
new
addition,
obviously
stacking
option.
There
has
been.
Basically
there
has
been
a
couple
of
bug
fixes
to
the
implementation
and
which
I
I,
actually
I,
basically
fix
them
myself
and
was
committed
around
January
of
2021.
G
It
has
been
actually
more
than
two
years
ago
and
a
debate.
It's
shipped
in
Linux
distribution,
5.4,
Dodge,
I,
don't
remember
the
the
minor
version
there
and
there
are
two
of
them.
These
are
the
hashtags
for
the
other
applications
for
the
other.
Basically,
TCP
is
stack.
Basically,
if,
if
let's
say
currently,
if
it
does
not
have
a
Implement,
is
that
if
that
has
implementation,
I
would
think
the
implementation.
Actually,
it
should
be
straightforward,
because
this
is
not
just.
We
are
talking
about
the
implementation
of
timeouts,
it's
basically,
they
are
also.
G
They
are
typically
three
operations.
You
start
the
timer,
you
basically
reset
the
timer
or
you
do
the
measurements.
It's
you
know
it's.
It's
really
does
not
require
any
optimization
or
complexity.
It's
fed
forward,
there's
no
magic
there,
all
right
so
advisor
BEP
is
concerned.
So
there's
a
this.
Is
this
this
concept
I
think
fox
called?
It
is
the
stock
sessions
for
bbp.
It
basically
simply
means
that
PG
messages
are
not
delivered
for
exit
extended
period
of
time.
G
This
can
result
in
steroids
in
the
writing
system,
and
so
this
is
reasonable
and
desirable
to
have
a
marketing
to
detect
these
situations
and
also
terminalization
and
I
mean
currently
I.
Think
the
problem
is
mostly
mostly
speculation
due
to
the
lack
of
Auto
automated
detection
mechanisms
in
the
in
the
implementation
in
pgp
implementation
today.
G
G
This
type
data
delivery
issues,
and
if
this
is
with
this
option
the
detection-
and
it's
it's
much
more
deterministic
and
precise-
it's
basically,
it
can
do
a
much
better
job
than
PDP
itself,
because
informal
implementation
for
the
implementation
of
my
health
is
a
bgp
itself
actually
does
not
have
enough
that
much
visibility
into
the
TCP
State,
whether
it's
interesting
transmission
mode
or
how
long
it
has
been.
It's
simply
does
not
know,
and
with
this
TCP
option
that
it
will
require
minimum
code
changes
in
PDP.
G
Excuse
me
so
as
a
spell
out
in
your
spec.
Basically,
we
are.
We
have
made
several
recommendations.
We,
of
course,
we
recommend
using
this
TCP
user
timeout
option
and
the
default
time
value
is
that
I
mean
in
the
graph
is
basically,
as
the
current
specifies
five
times,
the
whole
timer,
but
not
less
than
two
minutes,
and
obviously
this
should
be
configurable.
G
And
another
point
is
that,
basically,
because
when
the
PDP
comes
up
depends
on
the
the
number
of
sessions
amount
of
data
that
need
to
be
exchanged,
it
can
be
quite
heavy
computation
wise.
So
it's
so
we
basically
recommend
that
you
enabled
this
enable
this
option
only
after
UI
is
received
from
from
all
the
neighbors
also
remains
posts,
decision
establishment
and
to
basically
to
mitigate
some
of
the
possible
Force
positives.
G
Basically,
we
say
you
follow
the
gr
procedures
for
this
year,
timeout
this
error
and
the
longer
the
log,
the
events
and
because
this
is
the
most
most
likely
attribute
attribute
to
both
to
the
remote
to
the
implementation
of
the
Remote
device.
You
basically,
you
work
with
the
peer
alert,
appear,
administrator
and
work
with
them,
basically
to
for
its
resolution
and
the
next
slide.
H
Hi
Ben,
you
know
in
the
other
proposal
for
this
I'm
curious
on
how
you
came
to
30
minutes
on
your.
If
you
go
back
a
couple
of
slides.
Oh
these
were
these.
G
Are
timers
I
think
several
of
them?
We
obviously
this
should
be
configurable.
It's
just
I
think
as
a
common
practice
for
most
of
the
most
of
the
the
time
that
armor
related
values
and
most
of
the
time
we
recommend
the
default.
So
that's
30.
G
Of
fields
like
it's
long
enough
for
the
writing
to
converge
after
basically
the
pgp
comes
up,
but
if,
if
we,
if
we
your
particular
configuration,
basically
it
requires
a
longer
time.
You
definitely
should
should
configure
that
way.
So
it
is.
These
are
not
basically
hard-coded.
They
should
be
configurable.
Okay,.
E
G
I
G
Mark,
yes,
Roddy,
that's
pretty
much.
The
message
there's
already
exists.
You
know
a
very
nice.
You
know
mechanism
for
doing
that
at
the
TCP
can
do
a
much
better
job
than
PDP,
because
the
TCP
is
number
is,
is
the
transport
for
us
right
for
PDP?
It
actually
certainly
has
has
all
the
information
necessary
information
to
deal
with
it.
G
You
see
I
was
amazed
that
actually
the
TCP
was
developed
with
30
at
least
30
years
ago,
and
actually
the
the
folks
actually
had
a
vision
about
this
kind
of
problem
and
they
actually
introduced
a
magazine
for
it.
That
is
quite
amazing.
It's
a
point
of
it's
a
it's
another
addition
to
the
all
the
amazing
work
that
has
a
basically
put
into
the
TCP.
B
So
the
apologies,
if
I'm
difficult
to
understand
so
for
cases
where
TCP
is
misbehaving,
and
that
includes
like
firewalls.
This
has
good
value,
I
agree,
but
I
don't
think
it's
no.
The
core
use
case,
that's
necessarily
applicable
I.
Think
the
second
presentation
will
get
into
some
of
those
details,
but
I
think
the
this
problem
space
is
being
discussed,
has
two
partitions
one
place
where
ECP
itself
is
misbehaving,
which
this
addresses
and
other
cases
where
TCP
is
properly
behaving,
but
the
application
is
not
properly
behaving
basically.
G
Definitely
Jeff
I'm,
just
thinking
that
you
know,
although
you
understand
in
the
in
the
draft
I
thought
of
least
the
example
where
this
could
happen,
it
feels
I
feel
like
it's
more
like
a
hard
crafted
scenarios,
and
you
know
these-
it's
it's
very
hard
for
me
to
actually
imagine
this
kind
of
problem
would
occur
without
having
a
park
in
pgp
implementation,
because
if
you
look
at
it,
the
pgp
has
its
own
keep
liver.
People
keep
live
timer,
it
also
have
into
whole
time
the
whole
timer
definitely
should
have
TDM
it's
hard
to
imagine.
G
The
remote
side
actually
keeps
receiving
TCP
no
keep
not
to
keep
receiving,
keep
saving
TCP
the
PDP
people
lives,
but
another
region,
the
pgp
people
live
ad,
still
keeps
the
sessions
up
running
and
the
whole
timer
does
not
kick
in.
It's
very
is
I
would
think
this.
This
is
really
forced
into
very,
very
Corner
cases
of
some
kind
of
either
crafted
scenarios
in
the
lab
or
some
nasty
implementation
bugs.
B
Yeah
so
I'll
offer
my
commentary
that
for
my
28
years,
worth
of
working
at
bgp
I've
run
into
at
least
three
implementations
that
have
no
issues
with
TCP
and
it's
strictly
vtsp
read
applications
in
the
vgp
stack,
that's
the
problem,
so
we
have
different
experiences.
You
know
again,
this
has
value,
but
I,
don't
think
it's
the
total
solution.
Thank
you.
A
G
Works
well.
Actually,
it's
because
the
bdp
has
its
own
keep
life
as
well
the
whole
timer
they
would
kick
in.
They
should
right
for
some
for
some
Corner
cases
for
some
mysterious
reason.
So
hardcore
fast
crafty
scenarios,
somehow
the
whole
time
basically
didn't
kick
it
all
right,
yeah
I,
don't
have
any
for
the
comments.
Thank
you.
I
J
A
J
J
The
problem
we
are
trying
to
attack
is
that
in
a
while
we've
seen
scenarios
where
btp
speakers
that
are
not
under
the
control
of
the
operator,
So
speaking
from
my
own
perspective,
this
could
be
my
customers
or
my
peering
partners,
where
devices
like
that
advertise,
a
TCP
receive
window
of
xero
for
a
long
period
of
time,
and
this
can
show
up
on
certain
implementations
as
hey,
slow
peers
detected
or
an
out
cue
that
never
ever
again
decreases
and
the
Curious
Thing
is
that,
because
the
remote
system
that
has
some
kind
of
issue
is
signaling
to
us,
I
don't
have
any
room
for
new
information.
J
We
cannot
send
them
a
keeper
live
message
now
under
normal
circumstances.
We
would
expect
the
remote
system
to
disconnect
us
because
they
haven't
received
our
keeper
lives
for
some
time,
but
we're
dealing
with
systems
that
are
misbehaving
for
one
reason
or
another,
and
many
implementations
do
not
disconnect
from
such
a
broken
situation.
J
J
So
the
consequences
of
this
are
can
be
quite
nasty,
as
I
mentioned
this
or
scenarios
that
that
happen
for
for
hours
on
end
and
the
internet
churns
significantly
over
the
span
of
a
few
hours.
So
there
I
think
there's
three
aspects
to
this.
J
The
inability
to
live
here,
HP
messages
to
the
remote
site
means
that
the
remote
site
is
not
aware
of
our
latest
updates
and
with
trusts,
which
might
mean
that
the
remote
site
is
sending
us
packets,
for
which
we
have
no
routes,
and
we
did
tell
we
did
queue
up
a
message
for
the
remote
period
that
we
withdrawled
routes.
J
It
also
means
that
if,
if
the
session
is
not
killed
at
some
points,
the
likeliness
of
us
having
routes
pointing
towards
the
broken
tier
are
outdated
because
I
don't
think
it's
reasonable
to
assume
that
a
remote
care
that
is
no
longer
a
progressing
btp
messages
is
only
has
issues
with
that.
One
specific
session,
but
more
likely
has
issues
in
in
all
directions
and
feel
you
have
to
disconnect
from
such
a
stuck
here
also
hinders
our
ability
to
inform
the
rest
of
the
network
that
routes
no
longer
exist.
J
So
we've
had
it's
a
two
large
periods,
large
discussions
over
the
last
two
years
and
my
takeaway,
but
this
is
just
my
summary
and
I'm
happy
to
to
stand
corrected.
If
people
feel
that
I
did
not
characterize
it
correctly.
J
J
J
Actionable
suggestions
that
followed
from
these
discussions
I
think
the
general
consensus
was
that
Central
timer
should
be
generous
in
the
sense
that
we
don't
unnecessarily
love
connections
where
systems
are
just
churning
through
the
data
and
as
a
transient
State
signals
please
hold
on
sending
me
more
and
from
the
get-go.
We
have
I
agreed
with
that
principle,
because
the
this
whole
timer
really
is
a
backstop
of
Last
Resort.
This
it's
it's
like
the
last
thing
you
should
do
in
in
problematic
situations.
J
It
should
not
be
something
that
fires
off,
because
reset
things
too
tight.
J
Other
suggestions
were
to
change
the
tone
of
voice
of
the
internet
draft
to
be
a
little
bit
less
alarmist,
because
in
our
when
we
originally
discovered
this
issue,
I
personally
considered
it
more
of
a
vulnerability
than
as
a
simple
shortcoming
and
some
implementations,
okay,
and
it
would
also
maybe
be
helpful
to
document
what
central
timer
doesn't
solve.
J
J
J
So
this
means
that
the
remote
peer
is
not
accepting
our
messages
but
depending
on
how
large
the
buffer
is
a
a
small
or
a
large
number
of
HP
messages
has
to
be
generated
and
sent
to
the
remote
peer,
and
if
the
local
systems
outbound
buffer
has
filled
up,
then
the
count
the
timer
starts
in
open,
bhpd
caused
by
the
soccer
being
Unbreakable
and
once
a
timer
expires.
So
that's
eight
minutes
from
the
top
of
my
head.
J
Then
it
proceeds
with
the
this
connection
of
the
remote,
pier
and
basically
does
the
same
thing
you
would
do
if
whole
timer
expired
so
ends
to
ends.
It
takes
more
than
eight
minutes,
even
though
Central
timer
is
set
to
be
eight
minutes
now.
I,
don't
consider
this
an
issue
because,
as
I
mentioned
before
these,
this
is
a
mechanism
that
should
kick
in
and
really
pathological
situations
that
don't
seem
to
resolve
themselves
in
some
other
way.
J
These
are
some
links,
I
really
recommend
to
working
group
to
take
a
look
at
the
there
is
the
BHP
zero
window
test.
That
is
a
simple
go
program
that
you
can
connect
to
your
own
BHP
implementation
to
inspect
how
your
bgp
implementation
behaves
if
receive.
Windows
0
is
advertised
for
a
prolonged
time,
and
this
is
the
program
we've
used
to
test
whether
the
open
HPD
changes
were
having
an
effect
or
not,
and
the
BHP
is
a
window
test.
Emulates
Behavior
I
have
seen
in
the
wild.
J
So
it's
it's
not
a
intended
as
a
contrived
test
case.
This
this
I
think,
is
a
very
accurate
representation
of
what
what
BHP
implementations
might
interface
with
involvement.
J
G
Okay,
are
you
able
to
hear
me
yep,
hello,
yeah?
Okay,
thank
you,
yeah.
My
concern
with
this
approach.
I
think
it's
a
it's
a
it's
also
a
concern
shared
by
I
think
several
people
on
the
mailing
list.
It's
basically
to
solve
the
subset
of
the
problem
as
the
job
listed,
the
conditions
there's
several
other
conditions.
G
One
of
the
big,
if
is
basically
the
local
TCP
buffer,
is,
is
full
and
the
only
thing
this
Market
would
kick
in,
and
that's
that's
a
very
if,
if
we
consider
this
problem
as
as
a
series,
that's
a
very
big
assumption,
because
the
the
local
bar
TCP
buffer
may
not
be
full,
but
it
is
a
big
important
bdp
update
message
use
the
TCP
buffer
that
will
be
stuck
there
for
a
long
time.
At
least
immediately
will
not
be
able
to
detach
that's
it.
G
The
basically
damage
will
continue
to
work
on.
So
this
I
consider
this
solution
as
a
young
complete,
and
it's
a
subset
of
the
problem
that
addressed
by
the
TCP
this
user
time
option.
I,
don't
think
this.
This
draft
is
this
proposal
is
actually
necessary
or
even
useful,
because
it's
incomplete
it
does
not
solve
a
very
important.
You
know
btp
update
message
stuck
in
the
TCP
buffer.
A
J
The
cases
we
observed
in
a
while
were
of
machines
that
somehow
were
able
to
keep
generating
keeper
lives
and
send
those
each
keeper
life,
accompanied
with
the
signal
that
the
TCP
received
window
is
zero.
So
there
could
not
be
replies,
but
of
course,
keeper
lives
are
not
a
conversation
where
you
reply
to
each
other.
J
It's
it's
two
unidirectional
streams
of
information,
and
my
my
feeling
is
that
the
central
timer
operates
at
an
application
Level
and
offers
some
value
in
in
the
case
where
the
remote
peer
has
a
fully
functional
space
stack,
but
something
else
is
broken
and
and
I
think.
Your
proposal
is
advantations
in
in
the
case
that
there's
an
issue
on
the
transport
layer,
so
I
don't
perceive,
though,
actually
exclusive.
G
If
you
go
back
to
the
slides
where
you
list
there
are
this
conditions,
it
would
work
all
these
conditions
in
that
situation,
as
actually
the
TCP
user
time
option
would
solve
the
problem.
So
so
I'm
not
questioning
the
problem
statement.
Actually
that
that
part
is
fine.
It's
just
a
solution.
I
think
here.
What
it
is
proposal
does
is
the
partial
solution,
and
if,
if
we
do
the
TCP
user
time
or
whatever
problem
are
you
trying
to
solve
here-
is
all
resolved
by
the
TCP
right?
You
you
much
more
deterministic
manner.
A
Okay,
anchor
working
and
and
job
we're
gonna
go
on
to
John
and
Claudio,
so
they
have
a
chance
to
make
their
comments.
Folks,
we're
going
to
continue
this
for
about
four
to
five
more
minutes,
so
John
go
ahead.
Thank
you.
Enki.
E
A
K
So
I
wanted
to
say:
I,
don't
agree
with
NK
that
the
TCP
user
timer
is
covering
more
than
the
sandhold
timer.
It's
actually
the
other
way
around,
because
the
TCP
zero
window
is
not
the
only
way
of
triggering
this
issue,
because
the
the
zero
window
only
happens
when
the
receive
site
buffer
on
the
TCP
level
fills
up
and
and
and
cannot
progress
anymore.
But
if
the
the
remote
system
is
actually
emptying
the
tcp's
buffer
all
the
time
but
is
actually
not
making
any
progress
on
that
level,
then
we
would
still
have.
A
Okay,
we'll
pick
that
up
in
a
few
minutes,
I
think
I'll
if
we
can
we'll
take
a
break
later
on,
but
we're
gonna
go
on
with
the
next
presentation
and
I'll
try
to
cycle
back
for
comments
after
pradash
pradash
we're
gonna
pick
up
you
job.
Thank
you.
I'm
gonna
try
to
give
us
another
time
slot
for
comments
later
on.
A
F
A
F
Yep,
thank
you.
Sue
and
I
want
to
start
by
thanking
everyone
who
has
already
looked
at
this
draft
and
Who
provided
feedback
and
questions
and
comments.
They
have
been
very
useful
in
us.
The
authors
we
get
together
every
other
week
to
have
discussions
about
this,
to
make
the
text
better
and
also
I
think
to
show
interest
in
in
this
topic.
So
thank
you.
Thank
you
for
that.
F
We
had
in
case
you
missed
the
emails
on
the
list.
We
had
posted
the
version
zero
of
this
right
before
the
ATF
meeting
in
Japan.
There
were
some
comments
and
a
lot
of
good
feedback
there
as
I
said,
and
we
posted
zero
one
on
Friday
I'm
going
to
say
up
front.
We
didn't
address
all
the
comments
because
we
ran
out
of
time,
but
we
wanted
to
post
something
at
least
to
show
some
progress
for
this
talk.
F
I
have
at
the
end
of
the
slides.
We
put
a
list
of
a
to-do
list
of
the
outstanding
items.
So
please,
when
we
get
there,
take
a
look
at
that.
Make
sure
that
if
you
run
up
an
issue,
it's
there
and
and
if
not,
maybe
we
already
addressed
it
and
the
slides
are
going
to
talk
about
what
the
current
state
of
the
draft
is.
So
some
of
the
comments
that
you
already
made
may
not
be
addressed
here
already.
F
So
what
is
the
problem
that
we're
looking
at
I
I
wouldn't
really
care
for
us
it
as
a
problem
per
se?
But
you
know
the
situation
that
we
have.
There
are
many
networks,
operator
networks
or
big
Enterprise
networks,
where
the
administrative
domain
The
Entity,
who
administers
the
network,
has
multiple
lists.
This
can
be
for
many
many
reasons.
It
can
be
the
network
group.
This
way
it
can
be.
There
were
mergers,
Acquisitions
Partnerships,
other
relationships
where
the
span
of
control
of
an
administrator
really
goes
across
multiple
autonomous
systems.
F
We're
calling
that
not
a
single
initial
domain,
we're
calling
it
one
administrative
domain
OED.
So
that's
where
the
OED
name
of
the
draft
comes
by
and
of
course,
the
point
of
all
this
is
that
there
is
a
desire
to
do
better
policy
control
and
influence
route.
Selection
across
is
boundaries
using
attributes
that
would
normally
not
go
across
as
boundaries.
So
that's
why
that
disclaimer
goes
into
a
whole
bunch
of
drafts.
We
want
this
to
be
across
a
single
or
OED
the
one
other
domain
that
we're
talking
about
here.
F
F
The
naming
is,
we
did
it
this
intentionally,
because
the
idea
is
for
the
session
to
behave
as
an
ebgb
session,
but
because
it
is
part
of
this
OED
across
multiple
answers,
with
the
ability
with
the
optional
ability
to
have
to
be
able
to
do
exchange
of
attributes
that
would
normally
not
go
across
apgp
sessions
or
there
wouldn't
be
use
in
the
receiving
side
of
the
evgb
session.
So
a
simple
example
is
local
preference.
F
The
ROC
says
you
shouldn't
send
it.
If
you
send
it,
if
you
receive
it,
you
should
ignore
it.
So
in
this
case
for
the
ebgp
OED
session,
we're
saying
it
is
optional
for
you
to
send
it
and
if
you're
the
receiver
on
the
other
side,
you
should
receive
it
and
use
it
because
again
we're
extending
that
domain
across
where
it
used
to
be
ebgp
sessions
again
default.
F
Behavior
is
ebgp,
plus
the
option
controlled
by
policy
to
send
other
things
across
currently
The
Way,
It
Is
defined
in
the
draft
is,
is
this
is
controlled
by
configuration?
I
know
Jeff.
You
had
made
the
comment
about
a
capability
that
is
on
our
to-do
list
to
think
a
little
bit.
More
about
this
and
consider
that
so,
as
I
said
before,
this
is
what
draft01
still
says
we
want
to
follow.
F
Ebgp
rules,
so
attributes
like
the
aspath
or
example
would
go,
would
would
be
forwarded
or
would
be
transmitted
the
same
way
they
are
in
ebgp,
so
the
aspath
we
would
add,
we
would
prepend
the
as
right,
so
just
like
normal
epcp
and
again,
in
addition,
optionally,
some
ibgp
stuff
or
non-translative
attributes
the
draft
talks
about.
We
have
added
more
of
the
attributes
to
the
list.
That
was
a
big
comment
that
we
got
from
zero
zero.
F
We
had
started
what
we
thought
were
some
of
the
exceptions,
but
now
we
have
added
a
almost
all
the
attributes
are
discussed
there.
There
are
some,
for
example,
that
we
still
don't
want
to
go
across
the
epgp
OED
session,
like
originally
your
ID
and
cluster
list,
and-
and
the
reason
here
is
simple-
we
don't
think
you
know.
Route
reflection
should
span
across
these
types
of
sessions
as
well.
F
So
again,
this
is
the
the
Crux
of
the
of
what
the
draft
says.
It
would
be
great
when
you
go
and
read
it
to
make
sure
that
what
I
just
said
in
the
slides
is
what
is
actually
written
in
the
text
there.
We
think
it
is,
but
you
know
just
just
to
make
sure
the
other
thing
that
the
draft
talks
about-
and
this
goes
back
to
the
point
about
the
capability
that
I
made
before
and
Jeff's
comment
is
that
you
know
currently
the
draft
talks
about
the
ability
to.
F
Configure
this
behavior
on
not
just
session
by
session
basis,
but
on
a
peer-by-peer
basis.
You
know
the
rationally,
meaning
that
we
believe
right
now
and
that's
why
we
still
have
that
to
do
item
to
go.
Consider
the
capability
that
the
way
7606
is
defined,
that
we
could
enable
the
bgp
OED
session
on
one
side.
If
attributes
that
are
not
needed
would
be
sent,
they
would
be
ignored.
F
In
other
words,
standard
dbgp
would
apply
for
the
receiving
end,
and
the
session
would
remain
operational,
so
I
need
I
would
just
want
to
say
this
again
and
and
that's
why
I'm
going
to
show
the
next
slide.
One
of
the
action
items
or
the
to-do
list
that
we
have
here
is
to
go.
Consider
the
advantages,
the
potential
advantages
of
having
that
capability
to
negotiate
the
current
time
and
among
one
of
them
may
be.
You
know
avoiding
some
extra
errors
in
there.
So
this
is,
as
I
promised
the
to-do
list.
F
We
almost
are
almost
done
with
the
enumeration
of
the
attributes
of
all
the
attributes
were
missing
some
like
origin,
and
we
probably
want
to
look
go
look
at
the
container
that
is
used
for
white
communities
just
because
that
is
already
outside
of
the
working
group
right,
it's
with
the
ASG
somewhere.
So
we're
going
to
look
at
those,
and
then
you
know
finish
enumerating
those.
Please
go.
Take
a
look
at
the
list.
F
We
have
right
now
and
and
provide
your
comments
on
those
on
what
the
behavior
is
or
what
we're
proposing
is
going
to
be
and
again
most
of
the
behaviors
there
are
going
to
be
optionally.
You
may
send
non-transitive
attributes
across
the
epgp
OED
session.
We
need
to
the
other
item
on
the
to-do
list
is
considered
new
role
definitions.
F
Currently,
the
draft
says
that
the
use
of
the
roles
defined
in
1934
are
not
recommended
for
the
apboad
type
again.
You
know
something
we
need
to
go
address
and
think
more
in
our
like
Wiki
meetings
that
we
have.
We
need
to
go
talk
about
configuration
and
what
may
be
configuration
requirements
that
we
have
someone
mentioned
on
the
list
that
maybe
we
need
to.
We
want
to
have
a
list
of
other
OED
Asus
in
the
vicinity.
For
example,
now
the
draft
talks
about
the
definition
of
one
session.
F
There
are
some
use
cases
that
we're
looking
at
where
this
would
be.
You
say
with
a
stub
as
because
there's
a
special
business
relationship
there
and
we
would
want
to
just
have
it
there,
for
example,
but
it
is
obvious
that
we
could
an
operator
could
use
ebb
OED
sessions
to
connect
all
their
Asus,
for
example.
F
So
one
thing
that
we
are
considering
also
is
you
know
what
is
the
effect
of
that
to
the
place
where
we
have
only
ebhb
session
from
that
place
to
the
place
where
we
have
all
ABC
LED
sessions,
for
example,
if
the
default
is
of
ebgpoed
is
ebhp?
Well,
what
happens
is
we?
You
know
through
policies,
start
allowing
things
through
we're
thinking
right
now
that
this
would
be
a
separate
draft,
meaning
this
draft?
The
current
one
would
talk
about
the
session
type
itself
and
then
another
one
we'll
talk
about.
F
Well,
what
happens
when
you
use
it
in
specific
ways,
not
just
everywhere
to
connect
all
your
Oasis,
for
example.
You
know
migration
considerations
and
you
know
things
like
that,
but
also
other
things
that
came
up
around,
for
example,
exchange
points
or
router
operation.
So
that
may
be
an
operational
consideration,
type
thing
that
we
put
it
into
a
separate
draft.
F
There
is
a
draft
that
was
pointed
out
on
the
list
that
we
need
to
go
think
a
little
bit
more
and
add
some
text
around
and
preparation
between
OED
and
the
RPG
pediatrician
announcement
draft,
and
we
need
to
add
text
on
what
we
expect
to
happen
with
new
attributes
going
forward.
In
other
words,
we
are
listing
the
current
existing
attributes
and
the
behavior
that
is
expected
from
there
in
this
draft.
F
But
the
intent
is
for
this
draft
to
say
for
future
attributes
non-transitive
future
attributes
they
would
still
optionally
apply
to
the
ebgp
OED
session
type.
In
other
words,
my
policy
and
operator
can
allow
those
non-transitive
attributes
to
be
announced
to
evgpo
the
peers,
and
so
that
way
we
won't
have
to
come
back
and
update
this
draft
all
the
time,
and
if
there
is
any
exception,
of
course,
any
future
attributes
would
should
or
not
should
must
indicate.
F
And
there
was
the
other
point
about
disjoint
the
essence.
There
were
a
couple
of
people
who
were
talking
about
what
happens
in
my
oeda
SSR,
separated
by
someone
else's
saying.
Yes,
well,
that's
out
of
scope
for
for
this
right,
because
the
continuity
wouldn't
be
there.
F
So
was
that
clearly
that
at
a
scope,
so
what
we're
looking
at
is.
We
are,
as
I
just
said,
in
the
process
of
addressing
all
the
comments
that
came
from
draft
zero,
zero.
We
would
love
for
you
to
go
and
read
the
current
version
of
the
draft.
Look
at
the
this.
If
you
already
looked
at
zero
zero
and
you
know,
provide
some
more
comments
and
more
feedback.
F
F
So
that's
it
for
this
presentation.
If
anyone
has
comments,
I've
seen
some
things
go
back
by
the
chat,
I'm,
not
sure.
If
she
is
this
or
not,
but
any
case
comments,
questions
please.
F
So
I
see
Claudio
on
the
Queue.
If
you
want
to
go
ahead.
A
I
want
to
warn
you
Claudia
that
we're
going
to
only
take
bgpoa
D
comments
right
now.
I'll
try
to
fit
the
send
Hole
timer
comments
at
the
end
of
the
session.
So
are
you
working
with
the
oam
comments?
Claudio?
Yes,.
K
F
Great
thanks,
yes,
we're
looking
into
that.
You
know
both
from
the
existing
roles
point
of
view
to
maybe
potential
neurals.
So
thanks.
A
Any
other
comments,
Alvaro,
you
might
look
at
the
original
ISO
version
of
bgp
7
10757
for
that
Confederation
concept.
I
think
you're
sharing
some
of
the
same
things
I'll,
send
specifics,
ready.
A
I
know
but
Randy
your
turn.
I
I
I
I
don't
want
in
Myas
I,
don't
want
you
an
ebgp
peer
to
be
able
to
signal
to
get
my
internal
data
like
multiple
pref
was
your
example.
I
believe
I've
actually
cheated
and
read
the
draft
I
believe
that
won't
happen,
but
I
think
some
security
text
is
useful.
F
Sure,
yes,
what
you
mean?
Yes,
we're
not
changing
the
behavior
of
ebgp
session,
but
yes,
I
see
what
you
mean
that
you
could
end
up
with
the
wrong
negotiation
or
something
and
prompt
you
to
send
something
that
you
didn't
want
to
send.
Yeah
thanks.
A
Yep
and
yes,
ISO
10
747
is
the
right
number.
Okay,
any
more
comments
for
Alvaro.
L
Good
morning
and
good
evening,
everyone,
my
name-
is
Anjin
from
Huawei
technology.
I'm
gonna
give
a
presentation
about
a
flows
pack
redirect
load
balancing
group
Community
next
slide.
Please.
A
L
Okay,
so
this
is
the
second
time
this
draft
was
presented
on
intf,
so
the
last
time
was
itf1
sorting.
So
it's
been
a
year.
I'm
gonna
do
a
recapulation
about
the
motivation
and
problem
segment.
Also,
we
made
a
few
updates
and
added
some
chapters
to
the
draft.
After
the
last
last
presentation,.
E
A
A
L
May
have
some
delay.
Okay,
so,
as
we
all
know,
bgp
flow
specification
allows
flow
road
that
carries
traffic
policies
to
zero
traffic.
It's
always
being
used
everywhere.
So
in
some
scenarios
we
may
want
to
steer
traffic
to
a
load,
balancing
group
either
ecmp
or
ucmp.
So,
as
shown
in
the
graph,
we
might
want
to
do
a
traffic
steering
at
R1
to
R2
R3
R4
in
a
ratio
like
a
one,
two
two
to
two:
the
ratio
could
be
changing
according
to
network
running
status.
L
So
for
now
we
have
a
redirect
IP
extend
Community,
IPv6
external
community
and
we
can
reach.
We
can
steer
traffic
to
srv6
turnover
with
a
Cutler
extent
community.
So,
but
in
some
in
some
implementations
we
can
do
ecmp
of
redirect
IP
actions
by
encoding,
multiple
redirect
extended
communities,
but
the
current
set
of
communities
can
hardly
support
either
ecmp
off
srv6
externals,
nor
a
ucmp
of
either
types.
L
So
we
introduced
the
redirect
load,
balancing
group
Community,
it's
an
extension
to
PHP
Community
container
attribute,
so
it's
a
new
type
of
a
wide
community,
so
the
the
format
of
The
Wider
Community
complies
with
the
definition
in
a
wide
Community
draft.
Besides
that
it
must
contest
only
one
parameter:
TRV
within
the
parameter
TRV
we
have
a
list
of
sub-series
each
represents
redirection
actions.
L
L
Since
last
time,
we
made
several
updates
based
on
the
comments.
Firstly,
we
reference
on
the
rephrasing
on
the
atoms,
so
now
we
call
them
sub
trvs
for
reject
actions,
because
column
atoms
could
be
confusing
with
the
atom's
definition
of
the
wide
Community
itself.
L
Also,
we
add
some
chapters
about
interactions
with
other
redress
reactions
so
away.
What
we're
suggesting
is
to
do
a
configuration
control,
so
the
original
extended
community
actions
may
take
a
precedence
by
default
after
you,
based
on
configuration
on
the
device,
the
what
the
redirect
group
Community
take
precedence.
L
Also,
we
add
a
validation
procedures
to
the
draft.
So
each
past,
CRVs
of
the
within
the
wide
Community
within
the
radio
group,
Community
need
to
be
validated
separately
and
the
reject
group
is
considered
verified
only
after
all.
Past
theories
are
verified.
L
The
verification
procedure
for
each
one
redirects
to
IP
follows
the
normal
religion
procedure
of
that
also
we're
thinking
about
the
compatibility
with
flows
backward
and
two,
but
this
Tech.
This
mechanisms
was
mainly
designed
for
flows
back
where
one
scenarios
it
didn't
conflict
with
it
does
not
conflict
with
a
flows.
Fact
we're
in
two
action
wide
community.
L
Oh
sorry,
I
forget
to
move
the
slides
and
that's
it.
Thank
you
and
more
comments
and
discussions
are
welcomed.
B
So
I
think
for
your
flow
spec,
B2
considerations.
I,
don't
think!
That's
really
well
discussed
at
this
point
in
time.
Oh
much
like
what
we
have
for
flow,
spec,
V1
and
all
the
other
action
communities
that
we
have
defined.
Currently,
the
main
considerations
are:
how
do
they
interact
with
each
other,
especially
if
you
stack
more
than
one.
This
is
just
another
addition
to
it.
So
I
I,
don't
think
this
specific
proposal
changes
too
much
from
flow
spec
V1
to
V2.
At
this
point
in
time,.
L
Yes,
within
the
interactions,
without
a
redirection
actions,
we
all
we
talk
about
is
the
interactions
with
the
mechanisms
that
already
exist
in
philosophy
one.
We
didn't
talk
much
about
interactions
with
philosophy
too,
in
this
draft,
so
doing
that
we
may
want
to
add
some
new
sections
later
or
proposing
a
another
drop
to
talk
about
use
of
this
influence
back
way
too.
A
I
would
recommend
putting
the
interactions
with
the
flow
spec
V2
in
this
draft.
If
that's
all
that
you're
missing,
please
look
at
the
actions
that
are
in
flow,
spec,
V2
and
let
us
know
I
think
that
would
be
better
and
I.
I
should
say
that
as
a
working
crew
participant
and
not
the
chair.
A
That's
fine
I'm
going
to
then
go
on
unless
there's
any
other
comments
on
this
flow.
Spec
we're
going
to
go
on
to
a
second
flow
spec
presentation.
A
E
C
Be
on
the
bottom
yeah:
okay,
hello,
everyone,
my
name
is
Hunter
I'm
from
Bali
and
I
will
talk
about
a
few
drops.
Actually
I
won't
related
to
the
cats
communion
where
traffic
theory
I'm
going
to
talk
about
the
background
and
the
intention
of
bgprs,
and
he
will
talk
about
the
attention
to
PVP
of
those
back
foreign.
C
I
will
introduce
the
background
on
the
cats,
so
Cass
is
a
new
working
group
on
the
routing
area,
the
chartered
to
consider
the
problem
of
how
the
network
ads
can
steal
traffic
between
canines
and
and
the
service.
The
scenario
is
that
one
service
can
deploy
multiple
service
instance
in
different
sites,
and
the
Ingress
router
can
wrote
the
connected
request
to
the
suitable
site
based
on
their
Computing
centers,
for
example
their
Computing
node
or
Computing
latency.
C
In
order
to
do
that,
the
Ingress
need
to
know
the
Computing
metric
of
each
site
due
to
the
traffic
theory,
and
this
working
group
is
limited
to
to
the
overlay
model.
This
is
the
Ingress
router
and
the
egress
router
will
run
on
top
of
Eternal
so
that
the
Computing
metric
will
not
affect
the
underlying.
What
is
kind
of
like
a
VPN
login.
C
Actually,
we
have
a
existing
method
to
pass
this
Computing
metric
using
the
IDR
working
group.
It's
called
the
service
metadata.
This
service
method
defines
defines
a
set
of
computing
metric
capacity
using
the
pgp.
It's
a
new
pass
attribute
and
if
we
are
running
VDP
between
the
site
and
the
egress
router
and
the
pgp
between
the
Ingress
and
the
Ingress,
for
example,
this
R1
and
R3-
and
in
this
case
the
Computing
metric
of
the
site
can
be
tested
to
the
R1,
the
Ingress
router
and
combined
with
the
terminal
information.
C
This
R1
can
generate
the
final
Computing.
We
are
road
to
steal
the
traffic
to
the
appropriate
site
ready,
but
we
are
proposing
a
new
solution
that
we
do
not
want
to
touch
the
PCP
between
the
Ingress
router
and
the
egress
router.
So
we
can
use
the
controller
to
do
some
kind
of
Renee
renew
the
Computing
metric
on
the
for
the
control
that
you
connected
the
connected
the
Computing
metric
from
egress
router.
We
we
use
pdprs
and
for
it
to
distribute
the
Computing
metric.
We
extend
the
bgp
flows
back.
C
C
So
first
intention
into
the
bdprs
to
collect
the
Computing
metric
from
the
U-Verse
router
to
the
controller.
Basically,
you
are
reusing.
The
the
ad
service
method
data
chart,
we
reuse
the
computer
Computing
metric
defined
in
this
route.
For
example,
we
have
site
preference
and
the
capacity
index
and
the
node
measurement
and
pass
it
using
the
pgprs
and
then
we
also
add
a
corner
attribute
TRV.
This
corner
is
just
like
the
the
road
color,
the
natural
color,
to
indicate
the
service
level.
C
G
D
Okay,
hello,
everyone
I,
am
an
ECC
from
technical
for
the
next
new
friends.
I
will
explain
the
distribution
of
computing
metric.
D
We
proposed
an
extension
of
BHP
flows
back
to
distribute
Computing
metric
in
the
single
World
shows
the
distribution
degree
of
service
data
using
mid
B
flossberg
extension,
add
a
service
metadata
to
the
first
bag
or
routine.
All
its
data
sent
by
the
b2p
prospect
controller
to
the
entry
node
to
describe
the
calculation
Matrix.
D
The
Computing
metric
can
be
described
in
two
ways.
The
first
method
is
to
refer
to
the
original
metadata
path
attribute
in
in
the
drafting
is
the
pgp
extension
for
February
Edge
service
metadata,
the
sector
method,
aggregate
aggregate,
some
calculation
attributes
and
the
describers,
then
by
numerical
values.
D
A
A
A
D
I
I,
just
it's
been.
A
A
A
Okay,
then
we
will
go
on
with
rants
presentation
has
ran,
appeared
yet
today.
M
M
Yeah,
this
is
the
introductions,
segmental
routing
policies,
all
this
list
of
segments
that
represent
represent
South
rotate
policies
and
ITF
Network
slices
introduce
the
concept
concept,
Network
NRP,
which
is
a
step
a
subset
of
the
resources
and
Associated
policies
in
the
underlying
Network
and
the
job
idea.
I
start
policy
and
RP
defines
the
extensions
to
ptps
policy
to
specify
the
ownership
which
the
SR
policy
currently
pass
is
associated
with.
M
In
this
stop
Movement,
we
defined
a
new
tra
which
enables
1008
108
to
report
the
configuration
and
the
state
of
the
RP
with
the
SR
policy
candidate
pass
is
associated
with
this
is
the
bdprs
policycontinence
for
NRP.
In
this
document
we
defined
a
new
ISR
policy,
State
Theory.
We
called
it
an
abstra
way
to
carry
it
in
RP
with
as
a
policy
a
country
pass,
it
Associates
is
a
that
is.
This
is
the
format
of
the
rptioa
rpid
is
a.
M
This
is
the
procedures,
as
a
policy
currently
pass.
Maybe
in
this
is,
is
tested
with
a
specific
NRP
on
the
108
note
by
a
local
configuration,
PCP
or
pcpsr
policy.
Similarly,
and
then
the
state
and
the
attribute
of
the
RP
is
the
current
part
of
as
a
policy
is
associated
with
is
encoded
in
the
bgpis
a
tribute
field
as
a
policies,
data
tra,
the
iso
policies,
digital
wave,
it
is
defined
in
draft
idea,
pdpr
assets
policy
and
not
change.
It
can
be
used
to
report
as
a
policy.
L
Hi
good
morning
and
good
evening
again,
it's
me
again
tangent
from
Huawei
Technologies.
This
presentation
is
about
Road
Target
constraints
for
pgp
flow
and
a
bgp
ISO
policies.
This
draft
is
an
informational
draft,
so
it
does
not
propose
to
any
new
a
new
new
piece
of
Technologies.
It's
just
a
combination
of
existing
mechanisms.
L
So
during
a
distribution
of
Sr
policies
or.
L
There's
already
an
controller
used
to
distribute
this,
so
if
the
controller
Santa
flows
back
roads
and
or
as
a
policy
to
an
RR,
they
are
need
to
reflect
the
rows
to
all
peers
and
the
client
will
apply
an
Ingress
Road
policy,
either
static
or
dynamic,
to
filter
those
those
roads
and
to
give
the
road
a
targeted
Zone.
This
procedure
may
cause
a
waste
of
benefits
and
network
congestion.
L
Let's
see
this
example
in
the
graph.
So
if
the
P1
and
P2
Center
this
row
Target
membership
NRI
with
a
with
this
as
number
and
a
router
ID
to
RR
through
the
row,
Target
rows
row
Target
Ari
after
receiving
the
update
message,
the
r
will
trigger
the
rebounds
of
the
flows
back
roads
to
match
the
egress
policies
with
the
exist
with
a
road
Target
extent
community
on
the
roads
and
updates
those
roads
to
the
PES.
L
L
If
multiple
layers
of
R
are
deployed,
there
are
need
to
advertise
all
received:
Road
Target
membership
in
our
eyes
to
the
upper
layer
r,
so
the
proposed
format
of
the
roads,
membership
NRI
is
defined
like
this.
So
we
we're
using
a
ipv4
address,
specify
a
specific,
extended
Community
as
a
road
Target.
The
subtype
is
a
gonna,
be
two
row
Target
and
the
Global
administrator
is
set
to
zero
and
the
local
administrator.
L
Oh
I'm
sorry
I
forgot
to
show
the
slides
again.
This
is
my
first
time
to
control
the
slides
online
since
I
just
have
time.
I'm
gonna
re-reading
this,
so
so
we,
this
is
a
format.
The
proposed
format
of
the
road
Target
membership
in
our
eye.
So
we
we're
gonna
use
a
ipv4
address
specific
extended
communities
as
a
road
Target,
so
the
the
subtype
is
zero.
Two
photo
Target
and
the
global
administrator
is
a
road
identifier
of
the
node.
L
B
Yes,
I
appear
to
be
in
the
queue.
Would
you
return
to
slide
two?
Please
okay,
so
my
question
here
is
pretty
straightforward:
I
I
understand
you
want
to
apply.
You
know
to
Sr
policy
routes
in
flow
spec.
You
know
RT
constraint,
it's
very
straightforward,
I,
don't
understand
what
full
spec
or
ref
is
here.
L
B
Okay,
I
guess:
I
need
to
read
the
specification
more
detail.
I
also
placed
into
chat
a
link
to
ji
Dong's
node
to
Target
Community
I.
Think
you
may
find
this
might
be
a
better
fit
than
a
standard.
Drought,
Target.
L
Oh
yes,
I
talked
I
talked
with
him,
we're
colleagues,
so
this
this
mechanisms
could
also
use
with
no
target
Community
yeah.
A
Okay,
thank
you
very
much,
then
I'm
going
to
allow
in
case
we
want
to
get
back
to
comments
that
we
had
on
the
send
timer.
Would
anyone
like
to
provide
additional
comments?
Claudio
you
had
questions
and
I
believe
Anchor's
still
here
or
so.
Are
there
any
further
questions
or
comments
on
the
sent
whole
timer
or
Anchor's
draft?
We
will
add
it
now
go
ahead.
Claudio.
K
Say
before
was
that,
in
a
way
the
the
two
drafts
are
kind
of
doing
the
same
thing,
and
what
we
could
do
is
like.
K
The
important
thing
is
we
want
to
actually
detect
this
and
how
it's
implemented
is
is
kind
of
open
for
people
for
the
individual
implementation
and
I
think
there
should
be
an
agreement
that,
like
both
drafts,
are
working
like
it's,
the
technical
implementation
and
I
think
either
either
solution
would
actually
work
and
for
open
bgpd.
The
version
that
I
did
I'm
actually
quite
happy
with
the
implementation
we
have
and
it
seems
to
work
very
reliable
at
detecting
this
situation.
G
Comments
Claudia
I
think
earlier
you
mentioned
that.
Basically,
you
have
some
specific
cases
that
basically
the
TCP
user
time
would
not
work,
but
the
pgp
timer
would
work.
I
would
like
to
basically
see
that
look
at
more
detail
yeah
if
you
can
share
that
that
either
privately
on
the
mailing
list.
That
would
be
great
second
comment
is
about
the
open,
bgp
implementation.
G
K
Yes,
it's
more
or
less
how
he,
how
it.
G
Works?
Yes,
it's
so
it's
a
bit
high
level!
Okay!
That's
it
yeah!
That's
what
what
I
like
to
point
out?
I
think
I
pointed
out
earlier.
Basically
with
that,
if
statement,
basically,
if
the
you
know
it's
it's
a
subset
of
the
problems,
I
think
for
you
guys
to
point
it
out,
because
if
there's
a
importing
the
pgp
update
the
message,
that's
stuck
in
the
local
TCP
buffer
at
buffer
is
not
full
it's
and
it's
not
being
delivered
by
extended
period
of
time.
It's
the
same
damage.
K
Yeah
for
us,
the
big
thing
is
that
the
BSD
Network
Stacks
don't
implement
the
TCP
user.
Get
sack,
opt
so
for
us.
K
G
Coding
of
timers
is
pretty
straightforward
at
that
point
out.
It's
always
probably
it
does
not
require
that
much
code.
To
just
add
the
TCP
us
into
the
stack.
G
Yeah
so
as
as
I
said,
you
know
it's
like.
Basically
it's
a
it's,
you
know,
TCP
user
timer
would
be
more
complete
for
for
that
issue,
because
the
way
that
the
local
buffer
CCP
buffer
is
full
or
not
as
long
as
the
data
is
stuck,
it's
not
deliberate
for
extreme
pure
time.
The
TCP
do
the
timer
would
kick
in.
So
it's
a
more
complete
solution.
That's
that's
what
I
like
to
point.
G
No
I
keep
I,
keep
repeating
myself.
It's
it's
easy
so
that
the
PDP
Solutions
solves
some
cases,
but
it's
incomplete
the
I.
Don't
I
would
not
characterize
them
as
equivalent
yeah.
K
What
what
does
your
solution
cover
that
the
the
other
one
is
not.
G
Yeah,
basically,
what
I'm
saying
that
the
TCP
user
timer
does
not
care
whether
the
local
TCP
buffer
is
full
or
not
as
long
as
the
package
is
not
delivered
on
time
or
not
deliver
for
extended
period
of
time
that
time
that
the
TCP
user
time
will
time
out,
but
in
the
PDP
based
solution,
because
yeah
there's
not
much
you
can
do
you
don't
have
visibility
into
TCP
buffer.
So
that's
you!
Basically
what
you
do
you
say.
Basically,
if
the
local
TCP
buffer
is
full?
Yes,
you
can.
K
I
I
think
in
the
end,
we're
just
splitting
hairs
over
something
that
doesn't
really
change
the
aspect
of
the
overall
solution
of
detecting
stock
routes
and
detecting
stock
sessions.
B
A
Offline
and
and
come
back
and
report
to
us
to
the
working
group
via
email
does:
is
there
anyone
else
that
would
like
to
chat
about
the
bgp
timers.
B
No
come
inside
for
me,
although
I'll
repeat
this
on
the
mailing
list,
the
missing
details
that
people
seem
to
be
talking
past
each
other
and
the
send
hold
timer
and
the
TCP
user
timeout
values
send
hold.
Timer
is
fundamentally
about
black
sockets,
which
can
happen
in
some
implementations,
even
for
non-zero
windowed
environments.
You
know,
basically,
if
you
have
a
big
enough
update
and
you're
trying
to
send
a
entire
update,
then
you
can't
fit
in
the
window
that
can
manifest
in
some
of
the
implementations
that
way
and
for
the
user
timeout
piece.
B
The
requirement
is
that
you
don't
get
an
acknowledgment.
The
things
the
said
hold
tight
as
the
tsp
timeout
authors
don't
see,
the
beginning
is
zero.
Winding
can
happen
and
you're
still
getting
acknowledgments
back
you're,
just
getting
acknowledgments
back
that
have
nothing
in
them.
You
know
it's
telling
you.
The
zero
window
is
still
there,
but
you
first
came
to
acknowledgment
within
the
time
frame.
G
B
If
you
look
under
c793
under
reliability,
it
says
if
the
ACT
is
not
received
within
a
timeout
interval,
the
data
is
retransmitted,
the
ACT
is
coming
back,
you're
sending
you
know
a
packet,
no
you're,
still
sending
your
keep
lives,
you're
getting
X
back.
You
know
that
says
that
you
know
the
receivers
may
have
no,
not
Advanced
your
window.
You
may
get
it.
You
may
continue
getting
asked
back
that
just
keeps
telling
you
the
zero
Windows
there
and
eventually
you
know
you're
still
acknowledging
their
frames,
but
they
won't
eat.
Yours,
no
X
are
still
flowing.
G
B
A
Here
in
in
the
chat
window,
someplace.