►
From YouTube: MPLS WG Interim Meeting, 2020-09-16
Description
MPLS WG Interim Meeting, 2020-09-16
A
So,
regarding
the
agenda
number
three
chen
chung
zhang
she's
on
leave,
so
she
asked
to
you
know
we
can
just
remove
this
agenda
from
this
intro
meeting.
B
Okay,
thank
you.
I
did
mention
that
it's
off
the
agenda
and
thank
you
for
you
know,
sharing
this
with
us
sure.
Okay,
anything
else
from
you
lower.
B
Before
we
jump
to
the
first
presentation
that
we
have
today,
so
the.
B
Oh
okay,
that's
why?
Okay!
Thank
you.
So
the
first
presenter
we
have
is
rakesh
I'll
flip
to
rakesh
slides.
D
Hi
everyone,
my
name,
is
lakis
gandhi.
I
see
the
slides
tariq
thanks
for
presenting.
D
So
I'm
presenting
this
draft
on
enhanced
performance
monitoring
in
segment
routing
networks.
On
behalf
of
the
authors
listed
and
next
like
this.
D
So
the
agenda
for
today
is
to
look
at
the
requirements
and
the
scope
of
this
draft,
the
summary
of
the
the
solution
and
the
next
steps
next,
like
this.
D
D
So
there
is
a
increased
requirement
to
run
single
protocol
in
sr
networks
to
simplify
implementation
and
development
cost,
to
simplify
the
deployment
and
the
operational
complexity
in
in
the
network
avoid
the
interrupt
issues,
so
no
reflector
dependencies
make
it
stateless
on
the
reflector.
So
it's
unaware
of
the
monitoring
protocol.
D
D
So
this
draft
was
presented
at
spring
working
group
at
the
last
itf,
and
previously
it
was
presented
at
mpls,
a
virtual
meeting,
pls
working
group
meeting
as
well
in
april,
but
we
ran
out
of
time
at
that
time.
So
hopefully
we
are
the
first
one,
so
we
have
a
good
good
chunk
of
time
today.
Next
slide,
please.
D
So
the
enhanced
performance
monitoring
uses
pm
approves
in
loopback
mode.
So
this
is
an
example
of
for
sr
policy
where
we
have
sender
and
reflector
originating
or
the
hidden
and
the
talent
of
the
policy
probes
are
sent
in
look
back
mode
using
the
segment
list
after
sr
policy
and
in
look
back
mode
probe
messages
are
not
pointed
on
the
reflector
or
not
injected
back.
They
are
just
forwarding
in
the
first
part.
D
D
D
And
the
loopback
mode
is
enhanced
for
enhanced
performance
monitoring
using
network
programming
functions,
so
this
is
to
optimize
the
operational
point
and
time
stamp
and
inject
the
packets
on
the
reflector
node,
and
they
are
fast.
They
are
forwarded
on
fast
path
to
detect
the
any
performance,
degradation
or
liveness
failure
very
quickly.
D
So
in
this
mode,
using
the
network
programming
function,
the
reflector
node
adds
the
received
timestamp
in
the
payload
of
the
probe
messages
in
the
first
part,
and
it
does
it
only
if
the
the
address
in
the
method
matches
the
local
address.
D
This
way,
it
ensures
that
the
probe
messages
come
back
from
the
intended
reflector
node
next
slide.
Please.
D
The
in
terms
of
notifications,
the
query
node
monitors
the
sr
path
and,
if
consecutive
say,
m,
number
of
probe
messages
or
delay
values
crossing
some
locally
configured
thresholds,
then
delay
matrix
are
notified
if
consecutive
n
number
of
say
messages
are
not
received
and
the
heartbeat.
Basically
loudness
failure
is
notified
as
soon
as
probes
start
to
flow,
then
liveness
is
declared
success
that
there
is
a
heartbeat
is
restored,
so
it
allows
a
pretty
fast
and
simple
notifications
for
the
various
performance
degradations
next
slide.
Please.
D
So
some
of
the
mechanics
involved,
so
this
one
shows
the
tvamp
probe
message
formats
and
this
is
to
leverage
the
existing
wide
deployment
of
t1
implementations,
as
well
as
all
the
operational
tooling
around
this,
and
in
this
case
there
is
a
time
step
offset
16.
This
is
where
the
reflector
would
have
the
received
timestamp,
the
carrier
would
add
the
transmit
timestamp,
and
this
will
allow
to
monitor
the
end-to-end
delay
for
the
sr
path
and
next
slide.
Please.
D
So
this
is
achieved
for
sr
mpls
using
the
timestamp
label,
a
special
purpose,
timestamp
label
which
is
used
for
timestamp
and
forward
network
programming
in
srmpls,
the
in
inner
ip
header.
The
addresses
are
swap
so
it
will
come
back
to
the
source
who
sends
the
message
once
the
mpls
header
is
removed
by
the
reflected
node
and
next,
like
this.
D
Idea
is
very
similar
for
srv6
data
plane
as
well.
There
is
a
endpoint
function
are
defined,
which
does
the
similar
thing
as
timestamp
and
forward,
and
that's
the
received
timestamp
reverse
path
can
be
ip,
in
which
case
srh
is
removed
or
reverse
reverse
path
can
be
sr.
D
D
So
the
the
solution
also
allows
to
explore
the
ecmp
parts
which
are
typical
for
sr
network
for
sr
policy.
For
example,
it's
done
using
the
taking
to
take
advantage
of
the
hashing
function
in
forwarding
plane,
so
it
could
be
sweeping
the
loopback
addresses
ipv4
127
address
or
for
ipv6,
for
example,
sweeping
flow
label
in
the
header
next
slide.
Please.
D
So
this
is,
there
is
no
signaling
involved
in
bootstrapping
those
sessions,
unlike
other
monitoring
protocol,
so
this
is
achieved
using
the
provisioning
model
where
the
sender
and
reflector
nodes
in
the
network
are
provisioned,
you
would
have
look
back
mode
or
enhance
mode.
The
measuring
measurement
protocol,
the
packet
count,
the
timestamp
format,
threshold
values,
udp
ports
and.
D
Network
programming
label
can
be
also
programmed
or
it
can
be
ini
allocated
as
well.
Next
slide,
please.
D
D
There
is
a
pls
involved
here,
especially
the
special
purpose
labeled
allocation,
and
we
want
to
keep
mpls
working
group
in
the
loop
for
this
work.
So
that's
what
I
had
thanks.
B
Yeah,
I
I
have
questions
and
I
don't
I'm
gonna
check
on
the
chat.
If
there's
anybody
yeah,
I
see
multiple
people
have
requested
to
be
in
the
queue
greg
is
the
first
thank.
E
You
good
morning,
good
time
of
day
everyone
rakesh.
Thank
you
for
presentation.
I
have
a
couple
questions.
So
first
you
say
it's
a
loopback
mode.
E
Their
format
of
t-bump
and
stamp
packets,
sender,
packets
and
reflector
packets,
they're,
quite
distinct,
so
you
say
that
reflector
doesn't
have
a
state,
but
by
the
standard
the
reflector
is
supposed
to
count
texts
it
received
and
sent
to
on
a
specific
session.
E
Is
this
functionality
somehow
avoided
because
that's
a
little
confusing.
D
Yeah,
the
the
proposal
uses
the
message
format
defined
in
rfc,
5357
or
abc
162,
but
that's
basically
it
it
just
uses
the
message
format
to
complete
the
timestamps.
It's
not
the
the
reflector
does
not
follow
anything
any
procedure
for
t1,
for
example,.
E
So
if
so,
you're
not
doing
a
packet
to
us
one
way.
D
So
this
does
allow
when
you
not
not
receive
the
packets
back
at
the
querier.
You
are,
you
know
that
it's
not
being
let
his
liveness
affected
by
the
reflector
is
agnostic
to
the
protocol.
So
reflector
is
not
pm
aware
in
this
case.
E
But
so
you
said
that
you
derive
the
notion
of
take
it
to
us
on
your
sr
policy
by
the
fact
that
you
didn't
receive
the
round
trip
packet.
E
But
how
you
can
differentiate,
how
you
detect
the
packet
was
on
the
reverse
path
because
it
traverses
how
over
ip
network
over
srm
pls
network
over
a
srv6
network.
D
So
the
the
the
mechanism
is
similar
to
other
monitoring
protocol
for
liveness
detection,
where
you,
the
querier,
would
send
the
packet
and
it
will
wait
for
the
response
to
come
back
at
the
querier
and
if
x,
number
of
packets
are
not
received
back,
then
it
declares
the
liveness
down.
So
this
it's
no
different
than
other
protocol.
D
From
that
point
of
view,
having
said
that,
it
ensures
that,
because
of
reflector
checks
for
the
address
in
the
packet
matching
the
local
one,
you
know
that
it's
coming
back
from
the
right
reflector
note
that
it's
intended,
but
but.
E
So
if
reflector
doesn't
have
a
notion-
and
the
only
arm
is
by
your
data
plane
programming,
how
do
you
know
that
it's
not
some
other
node
that
responds
to
you
and
for
the
performance
measurement?
E
So
if
your
reverse
path
is
over
ip
network,
so
the
time
travel
over
srm
pls,
for
example,
tunnel
and
ip
network-
might
be
quite
different.
So
how
can
you
achieve
the
accurate
delay
measurement.
D
So
it's
it's
up
to
the
query
on
how
to
send
the
probe
messages
to
control
the
reverse
path
as
well.
So
if,
for
a
srm
pls
example,
it
can
put
reverse
sr
part
in
the
mpls
header,
as
well
in
case
of
srv6,
also
in
the
srh
to
receive
a
reply
back
on
a
specific
reverse
path
and
the
first
question
that
you
had
it.
It
would
return
back
from
the
intended
responder
node,
because
responder
only
responds
if
the
address
matches
its
local
address.
E
On
your
first
assumption
that
if
you
program
the
loopback
path
so
the
round,
then
this
packet
will
just
touch
the
reflector
or
because,
if
it
processes
in
all
paths
in
in
all
nodes,
so
you're
just
traversing
the
node.
There
is
no
time
stamping
on
the
far
end,
that's
been
already
defined
in
spring.
E
It's
hard
to
see
how
why
why
do
you
need
to
t1
or
stamp?
It
could
be
any
probe
because
you're
effectively
a
measuring
round
trip
delay,
assuming
that
you
are
targeting
the
right
probe.
D
Yeah
so
reason
we
are
used
using
t1
in
probe.
Diff
messages
is
the
wider
implementation
and
deployment
of
the
message.
Formats
is
widely
understood
by
a
large
number
of
hardware
and
silicones
and
lookback
messages
are
not.
D
Having
said
that,
it
extends
the
loopback
concept
with
network
programming
to
achieve
the
one-way
delay
measurement,
so
you
can
do
loopback
and
you
can
get
the
round-trip
delay,
but
the
use
case
for
sr
being
addressed
is
the
one-way
forward
direction,
delay
by
doing
the
t1
and
t2,
and
this
message
formats
allow
to
put
t1
and
t2
using
the
existing
mechanisms
existing
format
that
all
the
hardware
understand,
and
this
way
we
achieved
the
the
one-way
performance
monitoring,
as
well
as
the
liveness
detection,
using
the
simple
and
single.
E
Another
question
I
have
is
about
their
allocation
of
news
special
purpose
label
or
enhanced
personal
purpose
label,
so
this
looks
like
oem
functionality.
So
why
not
to
use
a
gal
label.
D
So
gal
is
usually
at
the
bottom
of
the
stack
and
it
has
a
purpose
of
punting.
The
pro
asset
pro
message,
the
use
case
here
is
not
it's
just
to
do
opposite
of.
That
means
not
to
point
the
packet
but
forward
it
at
line
rate
with
network
programming
function
to
add,
received
timestamp.
So
the
gal
gal
is
the
complete
opposite
of
what
you
want
to
achieve.
E
Well,
the
the
the
situation
with
their
time,
stamping
their
the
format
for
the
timestamp
is
different.
So,
basically
you
need
to
reformat
the
packet.
You
cannot
really
put
the
timestamp
in
the
t1
packet
or
stump
stem
packet
as
it
was
received.
D
So
the
hardware
on
the
reflector
node
adds
the
timestamp
at
the
right
location,
so
on
reflector
today,
packets
is
punted,
then
the
control
plane
has
time
stamp
at
the
right
location
and
injects
it
back.
This
whole
process
is
a
very
heavy
processing,
work
and
complexity
and
interop
issue
and
many
other
things,
and
this
is
the
simplification
that
comes
from
the
network
programming
function
in
segment,
routing
networks.
D
It's
a
simple
mechanisms,
widely
used
network
programming
concepts
where
using
the
ability
of
the
hardware
to
do
this
time,
stamping
deadline
rate
and
achieve
the
the
you
know,
the
the
requirements,
the
performance
goals
in
in
the
new
5g
networks,
for
example,.
B
Okay,
okay,
I
I
have
I
I
tried
to
capture
most
of
your
questions
on
the
ether
pad,
but
you
know
I
encourage
you
to
review
the
questions
and
you
know
to
be
fair
with
the
people
in
the
queue.
Okay.
F
E
B
Understand,
thank
you.
Okay.
Let's
take
it
to
the
list,
yeah.
Okay,
sure
thanks
lol
you're!
Next,
in
in
the
queue
do
you
want
to
ask
your
question.
C
C
C
D
Yeah,
so
there
are
so,
the
requirement
is
to
have
network
programming
label
that
reflector
node
understands
and
this
when
it
receives
a
packet
with
that
label.
It
knows
what
to
do
put
the
timestamp
and
forward
the
packet.
There
are
two
ways
of
achieving,
maybe
could
be
more
than
two
ways,
but
at
least
two
ways.
One
is
that
we
we
have
ayanna
allocated
the
special
purpose
label
and
this
way
all
nodes
know
this.
D
What
this
label
is
for,
and
the
second
is
that
controller
allocates
a
global
label
and
it
programs
the
nodes
in
the
network,
and
then
this
way
the
reflector
also
knows
that
this
label
is.
This
is
the
purpose
of
this
label.
So
there
are
multiple
ways
of
allocating
this
label
and
I
think
the
this
is
similar
to
some
of
the
other
concepts.
For
example,
iom
schemes
also
talk
about
similar
models
and,
but
I
think
the
more
standard
way
or
is
to
allocate
a
label-
and
this
way
it's
it's
all.
D
Vendors
have
implemented
the
special
work
purpose
label
and
this
way
it's
is
standardized
and
it's
well
understood,
behavior.
But
having
said
that,
controller
can
also
program
and
using
apis,
and
this
can
also
be
achieved.
D
So
I
I
will
look
for
a
feedback
from
the
working
group
on
which,
which
way
is
you
know
which
direction
it
can
we
go?
I
think
both
methods
would
exist.
Even
when
we
have
the
special
purpose
allocate
label
allocated.
D
It
doesn't
mean
that
now
you
know
the
controller
cannot
pro
use
api
to
program,
a
global
label,
for
example.
So,
but
having
said
that,
I
I
look
for
opinion
and
feedback
from
the
working
group
on
them.
C
D
Right
one
right:
why
two.
C
Can
you
back
up
the
slide?
I
thought
there
was
one
one
more
they've
been
listed.
D
C
So,
okay,
I
need
to
read
and
I
will
be
back,
but
I
have
one
other
you
and
that
is
that
when
you
get
close
to
working
with
glasgow
in
the
spring
work
group,
I
would
like
you
to
at
least
before
that
request
a
earlier
location
of
the
special
purpose.
Labels.
C
B
Okay,
here,
let's
go
back
to
the
queue
now
and.
D
Yeah,
so
there
is
mpls
data
plane,
srm,
pls
data
plane,
as
well
as
srv6
data
plane
support
in
this
draft.
D
So
I'm
not
sure
I
look
forward
to
feedbacks
from
the
chairs
of
spring
and
mpls
or
may
even
be
six
man,
I'm
not
sure,
to
see
where
this
drop
should
be
hosted.
I
don't
have
any
preference
but
good
to
have
a
drop
that
covers
this.
This
solution.
D
Okay,
yeah
we'll
keep
that
in
mind,
and
I
will
let
spring
chairs
know.
B
Okay,
going
back
to
the
queue-
and
you
know
I'm
trying
to
manage
the
time
you
keep
track
of
the
time.
I
was
in
the
queue
next
I'll
ask
my
question
very
quick:
there
is
in
the
liveless
detection.
This
is
stark
from
juniper,
so
I'm
on
slide
six.
I
think
why
there
we
have
something
called
sbft
that
does
live:
liveness
detection,
a
seamless
bfd.
D
So
there
is
increased
push
for
simplification.
Sr
brings
a
lot
of
simplification
in
the
network
and
having
multiple
protocols
the
development
cost,
as
well
as
deployment
operational
cost
in
networks.
They
are
quite
heavy
with
multiple.
You
know.
Monitoring
protocols,
tvamp
is,
is
heavily
implemented.
Hardware,
a
lot
of
vendors,
actually
most
vendors,
support
them
and
is
deployed
and
tooling
and
operational
is
well
understood
and
there
is
increased
push
for
latency
based.
D
You
know,
monitoring
in
5g
networks.
So
with
the
you
know,
innovation
and
the
push
of
new
requirements,
especially
coming
from
5g
using
it's
using
t1
message
formats.
I
would
say
that
it
can
use
different
message
formats
too,
but
the
the
solution
allows
to
do
enhance
it's
a
leap
forward,
leapfrog
for
them
monitoring
protocol,
for
that
does
multiple
things
in
one
shot.
B
B
There
was
a
concern
posed
on
on
on
on
a
different
draft,
which
I
think
somehow
applies
here
I'll
share
with
you
that
concern,
and
maybe
you
can
clarify
how
you
solve
that
on
your
side.
Yeah,
okay,
thanks
thanks,
satanic
I'll
go
back
to
the
cube,
see
if
we
have.
We
have
adrian.
F
Adrian,
I
think
adrian
is
here:
yeah
thanks
thanks
rakesh
for
this,
I'm
not
sure
whether
your
slides
have
got
ahead
of
your
draft
or
something.
But
anyway,
if
in
your
draft,
you
were
able
to
fix
figure
six,
I
think
there
might
be
two
fixes
here.
The
first
is
to
show
that
this
is
an
extended
special
purpose
label.
F
In
other
words,
you
need
two
label
stack
entries
for
it,
and
the
second
is
is
some
confusion.
You've
just
been
talking
about
the
network
programming
label
and
the
draft
is
rest
referring
to
a
timestamp
label.
I
think
it
might
be
helpful
to
add
some
clarity
about
that.
D
Yes,
so
yeah,
it
is
a
genetically
networked.
Programming
function
and,
in
case
of
mpls
is
network
programming
label.
That
does
the
time
stamping
in
this
use
case
and
for
srv6,
is
network
function,
network
programming
and
function.
D
So
we
can
clarify
in
the
draft,
and
you
are
right
that
the
extension
label
just
put
it
just
before
posting
the
slide.
We
added
it
in
the
slides,
but
draft
needs
to
update
sure
sure
these
things
happen.
F
D
So
if
there
are
different
use
cases,
then
we
either
we
have
a
different
label
or
we
change
the
network
function
for
it.
At
this
point,
we
are
not
aware
of
it,
but
working
with
the
working
group
and
if
the
innovation
we
can
do
to
solve
an
you
know,
a
bigger
problem
or
bigger.
You
know
use
case.
We
can
definitely
work
together
and
develop
the
solution.
D
Yes,
yeah:
that's
a
excellent
feedback!
Adrian
well,
we'll
think
more
about
this
and
and
come
back
with
our
thoughts
on
this
okay,
lovely
thanks.
B
Okay,
thank
you.
I
see.
Greg
is
back
in
the
queue
we
have.
Rakesh
is
up
next
for
another
presentation,
I'll
consider
greg
to
ask
the
question.
Maybe
after
rakesh
finishes
his
next
presentation
just
to
make
sure
we
have
enough
time,
probably
we're
running
out
of
time
so
rakesh
I'll
flip
to
the
next
set
of
slides
and
and
and
sorry
if
I
cut
off
anyone
in
the
queue
so
go
ahead,.
D
Thanks
tariq
so
again,
rakes
gandhi,
from
cisco,
presenting
a
draft
on
pm
using
rfc
6374
for
sr
mpls
networks.
D
On
behalf
of
the
authors
listed
on
the
slide
and
next
slide,
please
so
the
agenda
requirement
and
scope
of
this
draft,
the
history
of
the
draft
it's
been
around
for
some
time,
the
updates
that
we
made
since
we
presented
a
couple
of
ideas
ago
and
the
summary
and
next
steps
next
slide.
Please.
D
So
this
draft
addresses
the
pm
delay
and
loss
measurement
requirement
for
srm
pls
links,
as
well
as
end-to-end
paths
for
as
well
as
advertising
the
metrics
in
the
network.
D
D
So
draft
has
been
around
for
over
two
years
now
and
it
was
it
was
in
spring
and
we
presented
an
mpls
working
group
and
now
it
has
been
adopted
by
the
mpls
working
group
as
a
working
group
document.
D
So
before
adoption,
mplsrt
expert
review
was
done
and
we
have
updated
draft
to
address
the
comments
and
the
comments
were
to
combine
support
the
combined
dm
plus
lm
message
format.
There
were
some
questions
about
p2mp,
so
we
have
added
details
for
p2mp
srmpls
parts
added
a
section
on
backwards.
Compatibility
iona
was
missing
a
registry
for
the
return
sub
tlv
returned
parts
of
tlv,
so
this
was
added
in
the
draft
and
various
editorial
changes
suggested
by
the
experts.
D
D
Hello,
I'm
at
the
end.
Okay,
it's
not
refreshing
for
me,
but
okay
yeah.
So
at
this
point
we
are
looking
for
feedbacks
and
suggestions
from
the
working
group.
It
is
a
working
group
document
and
welcome
your
comments.
Thanks.
B
Okay,
are
you
asking
for
you're
asking
for
feedback
from
the
group?
Is
this
in
a
good
shape
to
progress?
The
working
group
last
call:
do
you
think.
D
Yeah,
it's
it's
in
very
good
shape
to
progress
as
working
group
last
call.
Definitely
we
haven't
formally
asked
for
it,
but
yeah.
It's
quite
ready.
It's
been
around
for
two
and
a
half
years
and
I
would
think
so.
B
Okay,
I
just
wanted
to
get
a
sense
from
you
all
right,
thanks
a
lot
okay,
I
I
did
a
promise
greg
for
another
question.
So
if
it
is
one
question,
please
go
ahead.
Greg.
E
No
actually,
I
already
put
my
thank
you.
I
already
put
my
comment.
It
was
in
regard
to
sbfd
in
in
case
of
point
to
multipoint,
to
the
best
of
my
understanding.
Svfd
is
not
applicable
to
point
to
multipoint,
okay,.
B
All
right
I'll
switch
to
the
next
presentation
we
have
plan
and
the
floor
is
yours
fan.
Are
you
present.
G
Yes,
yes,
can
you
hear
me?
Yes,
ma'am?
Okay,
thank
you.
Actually,
as
one
of
the
one
of
the
authors
of
this
draft,
I
I
need
to
do
the
presentation
instead
of
billion,
because
here
she's
joining
another
itot
meeting
right
now
so-
and
this
is
a
this-
is
a
new
draft.
Zero
version
draft
of
about
the
signal
degree
degree
date,
indication
used
in
the
npr's
network
actually
not
last
time.
G
Last
itf
meeting,
we
present
another
problem
statement
about
this
same
topic
and
and
we
I
know
that
we
states
that
it
is
more
special
for
the
sr
or
sro
mpls
network.
But
here
maybe
we
we
can
have
some
discussion
about
the
scope,
because
maybe
we
can
maybe
in
this
draft
that
we
propose
a
general
proposal
for
general
solution
for
the
mps
network,
but
we
can
discuss
it
later.
G
And
in
this
draft
that
we
solved
some
problem
from
the
last
last
draft
that
we
think
that
there
is
some
basic
understandings
because-
and
this
signal
degrade
is
is
brings
significant
impact
to
the
networks
and
especially
to
the
services,
and
it
should
be
noticed
and
detected
and
maybe
reported
to
to
to
the
controller
or
to
the
other
protocols
to
to
trigger
the
other
particles
and
yes,
and
in
this
new
drafted,
we
proposed
two
protocol.
Extensions.
G
One
is
for
the
networks
using
the
bfd
or
sbfd.
Another
is
the
networks
using
the
npr's
tpom
or
the
npr's
tp
bfd,
and
next
please.
G
Yeah
and
in
this
page
that
we
I
listed
the
the
encapsulation
of
the
bfd
protocol
that
we
would
like
to
use
one
of
the
reserved
we
would
like
to
use
the
reserved
beat
for
the
of
of
the
bfd
diagnostic
in
in
the
bfd's
encapsulation.
G
To
indicate
that
there
there
is
a
signal
degrade
yeah
that
I
think
that
would
be
the
that
would
be
a
very
general
general
modification
for
the
for
the
bfd
or
spfd
prototypes.
And
next.
G
Please
yeah
in
this
page.
Actually
I
I
made
some
modification
about
the
encapsulation
here
on.
Not
not
this
one
yeah.
G
I
think
the
last
one
I
think
the
the
encapsulation
of
the
bfd
is
is
a
little
bit
different
from
the
drop
from
the
the
one
in
in
my
draft,
but
that
may
be
a
different
way,
but
it
doesn't
matter-
and
for
this
for
this-
for
the
for
the
networks
using
the
mps,
tp
oem
or
the
nps
tp
bfd
protocols
that
we
would
like
to
have
the
different
modification
or
different
extension
here,
that
for
the
mps
tp
oem,
that
we
would
like
to
use
the
and
the
pdf,
the
pdu
definitions
inside
of
the
ampere's
label
and
the
gel
label
to
use
one
of
the
reserved
beads
in
the
in
the
flag,
set
the
flat
place
in
the
flag
and
that
the
yellow
part
is
the
is
the
extension
and
for
the
networks
using
the
mps
tpbfd
that
some
extension
could
be
done.
G
And
I
have
some
actually
I
have
some
questions
here
and
that
I
I
know
that
there
are
in
two
different
ways
or
two
different
proposals
in
my
draft
and
I
I
I
need
to
and
figure
out
that
for
the
bfd
extension
that
should
be.
Should
this
work
be
done
in
the
mps
working
group
or
the
bfd
working
group.
Should
I
split
the
the
split
to
different
parts
to
two
parts
of
this
draft,
and
that
is
one
question
another
is
to.
G
I
would
like
to
collect
the
interest,
especially
from
the
operators
and
the
the
vendors
and
the
equipment
vendors,
if
they,
if
they
are
interested
in
the
signal,
degrading
the
great
indication
the
internet
works,
because
I
I
I
feel
that
during
the
offline
discussion
I
feel
there
there
are
different
operators
that
do
need.
This
signal
degrading
indication
that,
in
the
amperes
or
in
the
layer,
3
network
in
the
internet,
in
the
layer,
3
particles-
and
there
are
also
some
minor
issues-
maybe
related
to
the
signal
degrade.
G
But
they
are
they.
They
come
out
from
the
the
other
people,
the
others
of
the
dislike
of
the
offline
discussion
that
like,
for
example,
how
to
measure
this
signal
degrade
and
should
be
there.
G
Any
new
parameters
be
defined
here
because
I
see
there
is
maybe
a
related
topic
in
another
draft,
and
I
haven't
talked
to
the
authors
there,
but
maybe
that
could
be
a
a
new
some
some
discussion
here
and
and
also
these
other
thoughts
about
the
the
indication
to
to
the
indication
of
the
signal
degrading
to
used
in
the
performance
measurement
yeah.
But
there
are,
there
are
minor
discussions
besides
the
proposals
in
the
draft
okay.
Next,
please.
G
Yes,
I
actually
we
we
up.
We
up.
We
have
already
observed
some
suggestions
from
the
other
from
the
other
offline
discussions
or
from
the
previous
meetings,
and
we
would
like
to
to
have
more
discussion
and
to
collaborate
with
other
with
the
the
authors
from
other
related
topic,
related
topics,
jobs,
and
we
also
want
to
hear
more
comments
about
this
new
draft
and
that's
it.
Thank.
B
You
thank
you
loa
or
nick.
Do
you
want
to
take
the
question
on
the
vfd
versus
mpls
or
let
me
know.
C
I
can,
I
can
say
something:
at
least
I
don't
see
this
as
more
complicated,
more
complicated
cooperation
with
another
working
group
than
on
the
previous
two
drafts.
So
I
think
it's
good
if
we
can
keep
the
document,
it's
one
single
document
and
we
need
to
involve
the
vfd
working
group.
C
G
B
Okay,
great
I'll
move
on
to
the
next
presenter.
E
C
I
thought
greg
was
in
the
cube.
B
E
Yeah,
actually,
the
question
is
probably,
I
hope,
that's
simple,
so
you
mentioned
that
you're
proposing
to
allocate
a
bit
so
change
the
size
of
the
diet
field.
E
G
E
Yeah,
not
the
gal
label.
So
if
I
understand
correctly,
you
are
proposing
to
change
the
size
of
the
diet
field.
E
So
why
it
cannot
be
one
of
the
values,
because
there
are
plenty
of
values
available.
G
Yeah,
actually
I
didn't
want
to
change
the
yeah.
I
just
want
to
apply
for
a
new
code
for
the
bfd
dark
note,
diagnostic
field
yeah.
G
A
G
Build
is
actually
for
the
other
proposal.
Yes,.
B
Okay,
anyone
else
in
the
queue
okay,
we
don't
have
anyone
else.
Okay,.
G
B
B
Yeah
we
can
hear
you
yo
go
ahead.
H
H
Please
mpls
can
be
used
to
realize
service
function
path.
So
far,
there
are
two
methods:
one
is
based
on
sr
service
programming,
where
each
service
function
is
related
with
an
mps
label.
Another
is
defining
ifc8595
an
npr's
label
stack
is
used
to
carry
the
logical
presentation
known
as
edge.
The
basic
unit
of
the
label.
Step
comprises
two
labels.
H
One
provides
a
context
within
the
sfc
scope
and
the
other
carries
the
label
to
show
which
service
function
is
to
be
enacted
and
in
nprs
the
iosp
ping
is
used
to
verify
the
data
plan
against
the
control
plan,
and
this
draft
proposes
extensions
of
lsp
ping
to
show
to
allow
verification
of
the
correlation
between
the
control
or
management
plan
and
the
data
plan
state
in
mps
based
sfc
next
slide.
Please.
H
H
H
H
H
And
and
generally
the
packet
processing
functions,
support
supported
by
service
functions
are
limited
service
functions
main
such
as
fio
may
not
support,
nprs
or
and
protocols
like
lsvp,
so
service
function
for
orders
are
responsible
for
ampere
cycle
request,
processing
and
and
a
service
function
for
order
sends
an
sfc
echo
requests
to
its
control
plan.
When
the
receiver
is
a
terminal
sff
for
an
sfp
or
the
mpl's
ttl
expires
in
ifc8595.
H
H
Folder
sends
better
reply
message,
including
sff
and
sf
information
recorded
in
sfc
info
subtly,
as
we
defined
before
as
we're
defining
this
document
and
after
all,
service
function
for
orders
on
the
service
learning
path
send
back
appears
equal
reply.
The
sender
collects
information
about
all
the
transverse
sffs
and
sf's
on
the
random
service
parts.
H
Next
slide,
please,
and
so
this
is
a
summary,
and
this
draft
proposes
extensions
to
ampere
amperes
as
lspp
mechanics,
including
a
new
fax
of
trv,
for
mps
based
on
sh,
a
new
sfc
validation
tv,
including
sub
trvs
for
osr
service
programming
and
nps,
based
on
sh
and
there's
also
an
update
of
ic
or
ifc8595
next
slide.
Please.
H
B
Thank
you
so
currently,
I
think,
we'll
track
the
spring
working
group.
Have
you
presented
this
work
in
spring
as
well.
H
B
Okay
yeah,
my
feedback
on
this
is:
it
has
an
mpls
aspect
and
you
are
presenting
it
in
the
mpls.
We
are
interested
in
in
keeping
you
know,
updating
and
getting
this
work
done
here,
but
but
this,
given
that
you
are
also
allocating
specific
service
function,
sids
or
that's,
my
understanding
is
a
spring
working
group
might
be
interested
in
this
work
as
well.
B
We
can
continue
for
now
to
keep
it
in
mpls.
This
is
my
say,
and
the
other
working
group
chairs
can
comment
and
you
we
advise
that
to
keep
the
spring
working
group
in
the
loop
for
now
lower
nick
feel
free
to
comment
as
well.
C
I
have
one
comment:
we
are
still
a
little
bit
worried
about.
C
Keep
of
lsd
so
since
we
are
getting
new
theories-
and
I
think
you
are
allocating
them
for
number
one,
if
that's
correct.
H
But
I
can't
hear
it
clearly
and
could
you
send
the
question
to
my
list?
I
think
what
what
you're
trying
to
is
it
you're
asking
that's
the
we,
the
extension
of
the
fact
sub
theory
here.
H
Oh
yes,
we
we
defined,
we
defined
the
effects
of
tav
not
for
collecting
the
sfp
message,
and
this
fact
sub
tv
is
used,
as
other
subject
means
in
kind
of
the
same
way
it
it
carries.
The
information.
H
Gets
from
the
control
protocols
such
as
bgp,
so
the
sender
may
be
know
the
that
she
can
get
from
bgp
the
rd
and
the
sf
type.
It
is
advertised
so
bgp,
so
he
it
can
use
flex
up
to
v
to
check
if
the
data
plan,
if
the
the
the
data
get
from
the
control
plan
is,
is
the
same
from
the
control
plane
and
so
that.
C
Part
is
understood,
I
think,
but
what
I'm
saying
is.
If
you
have
a
sound
theory,
you
can't
take
the
sub
tv
on
its
own
into.
H
D
B
Let
me
check
if
there
was
other
other
questions
appending.
Oh,
that
was
lower.
Okay,
lower
you're,
okay
from
your
side,
can
I
proceed
to
the
next
all
right?
Okay,
thank
you,
and
next
we
have
regis
or
thomas
he's
presenting
on
larp
labelled
arp
yeah
hi
tariq
hi.
Can
you
hear
me?
Yes,.
I
I
Yeah
so
from
the
last
working
group,
there
has
been
significant
progress
on
the
implementation
front.
We
have
a
working
prototype
on
tungsten
fabric
for
an
sdn
solution
where
we
use
the
actually
extend
the
mpls
fabric,
all
the
way,
also
our
mbls
fabric
in
the
data
center.
I
We
also
have
a
working
prototype
of
lr
client,
as
well
as
a
basic
server
on
linux,
and
in
this
case
the
lab
trigger
is
actually
from
the
packet
path,
in
contrast
to
let
the
earlier
case
where
the
trigger
is
actually
on
the
control
plane
via
route
download.
I
I
Yeah,
so
there
are
some
of
the
to-do
items
we
need
to
actually
update
on
the
draft.
So
one
of
the
assumptions
which
we
had
was
on
the
lr
replay
today
we
only
send
an
alar
and
the
there's
an
implicit
assumption
that,
on
the
specific
interface
we
have
a
single
lab
server
or
a
gateway,
and
so
this
lap
label,
which
is
returned
by
the
lab
server,
is
bound
to
that
specific,
the
ip
address
of
that
gateway.
I
I
So
again,
we
need
to
update
a
draft
on
the
label
refresh
part
from
the
client
side,
as
well
as
the
label
withdrawal
from
the
server.
I
I
Yeah,
so
we
do
have
some
of
these
things
to
iron
out,
but
we
believe
that
the
overall
approach
is
solid,
considering
we
have
like
a
working
prototypes
and
there
has
been
interest
from
service
providers
on
this,
and
based
on
that,
we
believe
that
our
document
is
ready
for
a
networking
group
production
code-
yeah-
okay,
that's
it
yeah.
Thank
you.
B
Let
me
see
if
there
are
questions
and
there
are
no
questions
in
the
queue.
Thank
you
regis,
and
I
think
this
was
presented
multiple
times
now
to
the
working
group
and
yeah.
We
can
progress
it
to
the
process
of
getting
it
adopted.
You
know
a
review
from
the
mpls
review
team
and
then
yeah.
We
we
can
progress
the
document
if
there
we
encourage
the
working
group
to
give
feedback.
B
C
Yeah,
thank
you
so
terrible
question
that
actually
struck
me
now
and
I
haven't
thought
about
really
it
might
be
stupid.
But
what
are
their
scaling.
I
J
Yeah,
I
know
what
I
what
reggie
said
is
that
I
think,
typically
in
the
data
center,
a
compute
server
is
connected
to
via
an
l2
connection,
but
only
to
the
nearest
store
and
after
that,
it's
all
l3.
J
So
the
number
of
arp
requests
that
that
it
would
send
would
be
exactly
for
the
servers,
the
other
compute
servers
it
needs
to
connect
to
so,
as
he
said,
it's
on
demand.
J
J
Those
are
the
only
things
that
you're
going
to
allow
for
so
I
mean
there
is,
of
course,
always
a
problem
with
arp
in
general,
in
terms
of
who
do
you
connect
to,
but
if
this
server
needs
to
connect
to
a
hundred
other
servers,
it
should
be
sized
for
that
if
it
needs
to
connect
to
a
thousand
other
servers.
The
arp
entry,
I
don't
think,
is
the
big
problem.
It's
the
whole
bunch
of
other
issues
that
there
would
be
internal
scaling,
but
it
is
a
very
much
constrained
thing.
J
C
J
Sorry
I
mean
you
have
both
running
and
your
regular
arc
will
always
be
there,
but,
as
I
said
since,
the
tor
is
typically
connected.
Your
point
to
point
with
this
sorry,
the
compute
service
connected
point
to
point
with
the
tor.
J
They
could
be
a
land,
but
it's
unusual,
but
it's
limited,
the
arp
regular
arc
will
be
limited
to
whatever
is
on
that
land.
The
larp
goes
beyond
because
the
car
is
acting
as
a
proxy
server
for
all
the
remote
computes
that
this
this
particular
server
needs
to
talk
to.
So
in
fact,
the
larp
table
will
probably
be
be
bigger
than
the
regular
arp
table,
especially
if
you
have
compute
connected
point-to-point
with
the
tor.
B
Okay,
any
other
questions,
and
I
don't
see
in
any
further
questions.
Kiriti
I
have
since
you're
on.
You
have
the
mic
now
and
you're
asking
question.
I
have
another
request
for
a
presentation
on
the
agenda
for
no
further
pastor,
you're
out.
J
J
In
this
case,
we
we
would
really
like
an
a
special
purpose
label,
not
an
extended
special
purpose
label.
So
what
you
know,
what
should
we
do
to
get
that
going?
And
the
second
thing
is
in
this
draft
also,
I
think
it's
I
mean
it
has
not
been
a
whole
lot
of
discussion,
but
it
is
reasonably
stable
and
I
think
we've
put
in
the
in
the
document
several
use
cases
for
the
no
further
fast
readout
label.
B
Okay,
wow,
I
can
do
you
want
to
take
that
lower.
C
C
I
think
we
need
to
start
looking
at
what
we
expire,
what
what
type
of
argumentation
and
documentation
before
we
actually
assign
a
new
base
special
purpose
labels,
but
I'm.
J
Not
ready
to
respond
just
now.
No,
no,
so
I
I
guess
the
question
I
had
was
I
mean
I
do
want
that
label.
The
question
I
had
is:
how
do
I
proceed?
What
is
the
process
for
getting
an
extended
special
purpose
label?
Do
I
make
a
request
to
the
list?
Do
I
do
it
via
you
know
and
update
to
the
top?
The
draft
already
asked
for
one
do
I
I
don't
know
what
the
process
is.
J
Of
course
it's
an
individual
contribution
right
now,
so
yeah
there
is
that,
but
but
it's
in
the
iona
section
to
your
other
point
about
you
know
we're
getting
low
on
these
labels.
I
don't
believe,
there's
been
a
special
a
base.
Special
purpose
label
requests
in
the
last
several
years.
J
So
in
fact,
since
the
draft
was
written
since
the
rfc
was
written,
so
I'm
not
sure
what's
different
today
from
I
mean
if,
in
the
last
three
years
or
four
years,
we've
had
like
two
or
three
such
requests,
that
would
be
a
quarter
of
the
space.
That's
left,
but
right
now
we
actually
haven't
seen
any
new
requests.
I
am
not
tracking
it
really
closely,
but
at
least
if
I
go
to
ayana,
I
don't
see
anything
there.
B
Yeah,
I
I
think
you
know
from
my
side.
I
think
that
we
should
you
know,
progress
the
document
to
get
adopted
and,
since
you're
highlighting
implementations,
existing
implementations,
kiriti
right.
This
is
what.
J
I
I
will
report
on
the
status
of
that
you
know
at
the
next
ietf
or
interim
meeting
for
mpls,
but
so
so
or
I
can
do
it
on
on
the
email
list
and
based
on
that,
we
can
decide
to
progress
this
to
working
group
document.
C
J
An
answer
to
that
the
early
request
earlier
than
last
call
because
we
want
to
implement
it
and
see.
You
know
actually
use
that
and
see
how
that
does.
But
I
mean
that
that's
the
point
of
an
early
request.
J
But
but
if
certainly
I
will,
I
will
send
a
mail
to
the
list
about
the
status
of
implementation
and
then
based
on
that
you
guys
can
decide
how
to
move
that
forward
to
a
work
move
document
at
which
point
we'll
come
back
and
revisit
the
question
of
early
allocation.
B
That
that's
fair
enough!
I
got
disconnected
and
came
back
now,
oh
and
so
so
I
I
think
it's
a
fair
point
that
if
you
think
the
document
is
in
a
good
condition
to
be
undergoing
the.
F
B
Adoption
we
can
progress
with
that
and
take
it
from
there
as
long
as
so.
B
Sounds
good,
I
I
see
that
I
think
ac
is
trying
to
ask
a
question.
J
J
K
One
thing
one
thought
I
had
you
know
for
assuming
assuming
we
allocate
this
best
first
label:
do
you
think
this
label
should
be
a
label
that
used
for
a
form
of
mpls
looping
where
the
use
case
has
to
be
described,
and
this
is
the
first
use
case.
I'll
put
this
comment
out
when
corey
asks
for
allocation
as
well.
J
Much
looping,
but
it's
that
if
you've
done
a
fast
readout
and
you
do
another
fast
layout,
then
you
could
get
yourself
in
trouble
because
you
might
cause
a
loop.
You
might
cause
other
things.
So
it's
not
you're
trying
to
prevent
looping
per
se
you're
trying
to
prevent
a
second
pass
data
which
could
have
a
bad
result
and
and
the
the
thing
is
it
could
arise,
and
the
document
says
multiple
ways
that
it
could
arise
and
it
can
arise
also
in
the
in
the
context
of
a
label
stacked
with
adjacency
says
so.
J
Having
this
to
be
a
special
purpose
label
would
be
much
more
efficient
than
having
needing
two
labels
for
each
thing,
but
to
your
process.
I
I
no.
K
I'm
not
debating
that.
I'm
just
saying
you
know
terminology.
When
we
allocate
this
label,
could
it
be?
Could
it
be
used
for
detecting
this
type
of
loop
condition?
It
seems
to
me
when
I
maybe
maybe
maybe
it
maybe
it
couldn't,
but-
and
this
would
be
the
first
use
case
of
using
this
label
for
detecting
looping.
J
J
K
So
yeah,
I
know
I
understand,
I
understand
the
use
case,
I'm
just
trying
to
see
I'm
just
trying
to
see
if
we
can
get
more
mileage
out
of
these
limited,
limited,
special
purpose
labels,
not
the
extended
ones.
Okay,
yeah.
J
I
appreciate
that
and
I
mean
if
you,
if
you
have
an
example,
I
think
it's
worth
thinking
about.
But
the
thing
is
the
semantics
is
you've
done.
I
mean
this
thing
already
has
done
a
fast
fair
if
you
were
planning
to
do
a
password,
don't
and-
and
so,
if
you
add
to
that
semantics,
then
you
can
use
a
label
for
multiple
things,
that's
great,
but
I
I
think
then
you
might
actually
not
achieve
the
first
case.
I.
K
B
Okay,
thank
you
keniti
and
yeah.
We
expect
you
know
an
update
on
the
mailing
list
with
with
what
you
promised
and
we
will
take
the
adoption
further
perfect
and
with
this
this
was
the
last
presentation
on
our
agenda
today.
We
thank
you
for
attending
the
session
and
looking
forward
to
seeing
you
in
the
next
ietf
meeting
before
we
adjourn,
if
nick
or
loa
want
to
say
anything
feel
free
right
now,.