►
From YouTube: IETF92-NFVRG-20150326-1740
Description
NFVRG meeting session at IETF92
2015/03/26 1740
A
B
So
welcome
to
the
second
session
of
the
nav
of
G
meeting
I'm
afraid
that
we
have
not.
We
have
no
blue
sheets
for
this
session
in,
despite
that,
we
were
supposed
to
have
and
what
rank
is
going
to
I
want.
I
was
looking
for
them
and
I
could
not
find
anyone
but
yeah
so
he's
going
to
give
a
second
trial
trial.
In
the
meantime,
just
wrote,
for
let
me
show
you
the
again
the
same
that
road
you
should
we
should
design
a
new
blue
sheets.
I
had
to
remind
you
about
the
air.
B
B
Well,
let's
go
through
the
through
the
and
the
India
gender.
What
we
have
is
a
set
of
presentations
that
are
not
connected
to
current
documents
that
are
under
consideration
of
the
research
group
without
more
photos
or
intended
to
be
a
view
of
the
different
activities
and
recent
results
that
we
have
projects.
B
C
A
B
C
B
C
Okay
and
can
I
get
a
quick
show
of
hands
who
has
been
at
the
lunch
talk
today
by
Chris
DiBona
on
an
open
source?
Okay,
quite
a
few.
So
basically
Chris
talked
a
bit
about
his
views
on
open
source
project
and
some
pitfalls,
and
this
experience
so
open
platform
for
anything.
Opn
fe
is
one
of
these
larger
open
source
projects.
I've
had
and
have
recently
been
being
kicked
off,
and
so
I'm
telling
a
little
bit
what
we
are
doing.
C
There
will
be
getting
some
ideas
how
you
could
get
engaged,
so
the
color
of
my
talk
is
actually
to
make
you
aware
and
motivate
you
to
try
to
think
so
how
you
can
maybe
bring
some
of
the
work
that
you
are
doing
here
to
this
project,
because
I
think
it
fits
pretty
well
in
some
of
the
cases.
Okay
next
is
so.
C
The
OPN
fe
is
a
carrier-grade
integrated,
open
source
reference
platform
for
any
free.
So
what
does
it
mean
so
I
mean,
of
course,
I.
Don't
have
to
explain
anything
to
you,
but
in
general
the
idea
is
that
so
at
the
end
of
phase
one
of
the
HC
na
VI
is
G.
It
was
pretty
clear
that
so
any
v
as
a
concept
and
like
technology
platform
is
here
to
stay
and
I
mean
question
is
ok.
How
can
we
expedite
deployment
take
up?
C
How
can
we
support
future
sensitivities,
and
so
the
conclusion
was
too
sad,
a
open
source
project
that
would
develop
its
weapons
platform
so
to
validate
standards,
to
provide
a
link
to
validate
and
standards,
to
provide
feedback
to
two
different
groups
or
say
regarding
employment
ability,
but
also
to
really
provide
a
usable,
so
Carrie,
a
great
platform
that
can
really
be
used
in
deployments
and
where
other
companies,
in
that
space
can
build
innovative
solution
on
top
or
on
the
side.
Exactly.
A
C
Right
so
I
I
mentioned
this,
so
we
are
doing
this
by
psycho,
bringing
together
as
a
large
set
of
vendors
in
the
space,
but
also
users,
so
operators,
service
providers
and
so
on.
So
they
are
part
of
the
project
and
they
also
contributing
requirements
and
give
us
feedback.
And
so,
as
you
know,
I
mean
in
the
energy
space.
C
So
the
focus
is
on
this
lower
part.
Of
course,
that
means
not
to
build
systems.
You
also
need
to
look
at
the
interfaces,
for
example
to
vnf
management
and
kind
of
yeah,
provide
examples
how
the
system
would
be
used
in
real-world
applications.
Ok,
that's
what
we
are
doing
right
now
and
next
line,
and
so
I
will
mention
that
we
are
say
working
intensively
with
these
upstream
projects,
so
we're
indicating
their
work.
So
we
are,
for
example,
proposing
blueprints
and
OpenStack
and
working
with
their
community.
C
So,
all
in
all,
it's
often
also
the
same
group
of
people
like
we
were
working
in
OB,
AV,
sanggup
companies,
you
and
you
could
say
and
and
OpenStack
so
it
works
pretty
well
to
move
the
requirements
from
like
one
community
and
to
the
other,
okay
and
yeah,
listen
so
the
kind
of
workflow
and,
if
you're
familiar
with
other
open
source
project,
so
bigger
ones
like
I
opendaylight,
for
example.
So
you
will
find
this
familiar.
C
So
there
is
a
proposal
face
if
you
start
like
a
new
subjectivity
within
OPN
fe,
and
we
have
some
incubation
phase-
will
be
kind
of
develop
nothing
further
and
at
some
point
we
kind
of
try
to
make
a
decision.
Okay,
this
is
mature
enough
that
we
can
integrate
it
into,
for
example,
the
next
release,
and
so
in
OPN
MV.
It's
not
structured
in
different
categories,
so
documentation
requirements,
integration
and
testing
and
quality
and
development.
So
in
the
first
phase,
of
course,
we
mainly
trying
to
get
something
out.
C
C
Yeah,
so
these
development
projects,
that
is
the
bats,
will
software
development
on
particular
topics,
and
so
it's,
for
example,
new
approaches
to
service
training,
something
that
would
be
a
topic
that
OPN,
if
we
would
do
by
itself,
take
ahead
integration
revocation.
This
is
basically
the
groundwork
that
is
required
for
a
big
open
source
project,
so
to
kind
of
automate
the
fabrication
processes
next
light
and,
of
course,
there's
also
lots
of
documentation
required
yeah
go
on
right.
C
So
again,
I
can't
do
anything
about
him
about
this,
so
then,
okay,
so
so
the
the
main
activity
at
the
moment
is
basically
to
tilt
kind
of
bootstrap
the
projects
it
started
in
the
end
of
last
year
and
boot
shopping
here
means
we
are
basically
selecting
components
and
trying
to
fit
them
together,
trying
to
ordinate
their
deployment
and
so
on.
So
so
you
see
a
range
of
components.
We
think
that,
but
at
the
same
time,
basically
the
work
also
has
started
on
these
other
type
of
projects
of
development
and
requirement
projects.
So
there's
a
set.
C
D
C
So
this
is
the
complete
list
of
projects
for
your
reference.
You
can
see,
there's
only
a
lot
going
on,
and
so
one
of
the
reasons
why
these
are
so
many
is
actually.
This
is
not
a
say,
monolithic
activity
that
tries
to
rule
out
competition
and
kind
of
slow
down
invasion
or
something,
as
sometimes
people
think
when
they
see
our
big
open
source
project.
So
some
of
those
activities
are
actually
competing,
which
is
other.
So
there
are
different
approaches
to
service
training,
for
example,
different
approaches
to
suffer
if
I
networking
so,
for
example,
I.
C
Could
this
example
example:
here's
an
have
actually
three
approaches
to
sdn
control,
so
open
daylight
project
a
contrary
project
and
it
may
be
a
owners
activity
in
the
future,
and
so
the
way
how
it
works
is
that
we
let
them
go
and
cannot
see
what
comes
out
and
then
at
some
point
do
a
decision.
So
what
of
those
is
actually
mature
enough
and
useful
enough
to
be
adopted
for
for
Avery's
all
right?
So
I
just
give
you
two
examples
of
like
different
kinds
of
projects.
We
get
an
idea
and
so
what
what
open?
C
We
call
that
openstack-based
viendo
forwarding
graph
project,
so
that's
taking
the
current
work
that
is
done
in
onf
and
an
open
daylight.
So,
like
the
the
open
flow
based
service,
training
approach
and
the
the
current
intent-based
northbound
interface
activities
in
onf
and
build
a
service
training
solution
that
has
integrated
was
open
flow
for
for
opening
for
you.
So
this
is,
you
could
say,
a
bit
more
forward-looking,
so
second,
an
ambitious
activity.
C
C
C
So
we
are
working
largely
on
a
six
months,
cadence
for
releases,
and
so
the
expectation
is
that
so
after
this
first
released
and
the
upcoming
ones,
what
like
contain
more
functionality
and
some
more
output
from
these
projects,
and
with
that.
Thank
you
very
much.
Please
check
it
out
and
contact
me
if
you
want
to
learn
how
to
contribute.
B
A
C
C
B
D
So
yeah
don
clark,
gable,
Absalom
I've
managed
many
teams.
Actually
my
R&D
career,
that
I've
been
heavily
involved
in
the
ITF,
but
I've
not
personally
attended
the
IETF
at
all
in
in
my
many
years
in
the
industry
and
I
was
forewarned
that
you
guys
could
be
pretty
rough
right
on
people.
So
I
was
very
careful
not
to
include
any
technical
content
in
my
slides
in
order
to
be
immune
from
from
harassment.
So
next
slide,
please.
A
D
D
D
So
k
will
add.
Some
cable
novices
are
an
interesting
entity.
I
joined
cablelabs
about
a
year
ago.
It
is
wholly
funded
by
the
cable
operators,
so
we've
got
global
membership,
cable
operators,
56
clabaugh
operators
spanning
the
United,
States,
Europe
and
and
Asia,
and
we
are
we've-
got
a
program
called
cable
cloud
which
is
looking
to
see
how
cable
operators
can
virtualize
the
network
infrastructures
and
leverage
the
benefits
of
open
networking
and
everything.
D
Is
this
that's
good?
That's
cool,
okay,
great
thanks
virtual
home
networking
is
the
first
project
that
I
was
I
was
given
to
manage.
Good
news
is,
I
don't
manage
anymore.
My
colleague,
Michael
clobber,
dancing
like
was
around
manages
it,
and
my
cliff
is
not
around
to
actually
getting
a
demo
ready
for
virtual
the
bits
and
buy
something
this
evening
got
a
blue
tinge.
Yes,
okay,
great
next
slide,
please!
D
So
many
of
you
recognize
the
challenges
for
home
networks
right,
so
increasing
complexity
of
end-user
devices,
multiple
routers
appearing
in
the
home
to
do
all
kinds
of
things,
and
this
is
causing
problems
for
service
providers
right
they've
not
got
complete
visibility
to
the
end
user
device,
but
those
end-user
devices
are
consuming
the
service
providers
services.
So
when
those
services
go
wrong,
the
service
provider
has
got
very
limited
visibility
to
actually
do
troubleshooting
across
that
home
network.
D
So
the
idea
here
is
to
try
and
use
virtualization
techniques
to
give
a
service
provider
better
ability
to
manage
that
home
network
troubleshoot
and
ed
to
offer
new
services.
So
next
slide
interesting
idea
for
within
cable
Abbas's,
we're
using
another
a
presentation
early
today
on
DevOps
and
DevOps
and
Lean
Startup
methodologies
are
very
interesting
techniques
that
we've
stolen
from
the
software
industry
to
to
increase
the
accelerator
R&D.
D
So
we,
instead
of
spending
many
many
months
of
detailed
analysis
coming
up
with
some
big
report
at
the
end
of
it
we're
actually
taking
the
problem
and
we're
dividing
it
into
individual
challenges
and
we're
analyzing
those
and
then
basically
feeding
results
into
our
service
provider.
Customers
asking
giving
the
rationale
and
asking
them
whether
they
agree
that
the
result
we've
come
up
with
is
an
approach
they
want
to
move
forward
with.
So
this
this
gets
learning
very,
very
quickly
across
a
very
broad
spectrum
of
conté
to
influence
with
that.
With
our
work,
we're
focused.
D
Solutions
and
standards,
how
many
of
you
were
in
there
Krista
Bono's
lunchtime,
talk
to
their
an
open
source.
That
I
was
a
little
bit
disappointed
with
that
with
that
session,
because
it
really
emphasized
more
to
do
with
licensing
than
the
real
advantages
for
me
in
open
source.
For
me,
open
source
offers
a
huge
number
of
benefits,
such
as
deduplication
of
core
development.
We
can
all
use
the
same
bits
of
code
that
are
the
basis
for
all
our
products,
but
we
can
then
innovate
on
top
of
those.
D
D
This
is
this-
is
the
the
architectural
vision
that
we're
talking
about
virtual
home
networks
in
the
cable,
labs
or
cable.
Loosely
context
is
not
about
virtualizing
the
cpe
right,
so
we're
not
thinking
about
how
can
we
get
rid
of
that
piece
of
hardware
from
the
home
we're
looking
at
virtualizing
the
functions
that
are
necessary
for
delivering
connectivity
and
then
seeing
how
that
impacts
both
our
devices,
notably
the
cable
gateway,
the
cable
modem
transport
system
and
and
the
MSO
the
local
systems
operator?
D
Okay,
we'll
provide
Internet
service
provider,
virtualization
platforms,
so
there's
a
kind
of
bunch
of
fairly
randomly
named
virtual
functions
in
the
cloud
and
a
couple
of
fairly
randomly
named
virtual
functions
on
the
Gateway.
What
we
basically
saying
with
this
project
is
we
want
to
give
a
cable
operator
service
providers
and
indeed,
in
an
industry,
standard
context,
anybody
that
wants
to
engineer
a
tailings
and
into
a
network
the
option
to
instantiate
functions,
we're
either
in
the
cloud
environment
or
in
some
intermediate
place
and
that
intermediate
place
could
be
with.
Is
it
wherever
there's
a
physical
node?
D
So
if
your
mobile
operator,
that
might
be
at
the
radio
base
station,
if
your
cable
operator
it
might
be
at
the
fiber
node
and
also
for
the
point
of
view
of
home
networking,
what
might
be
instantiated
on
the
cable
gateway
itself
and
what
kind
of
resources
would
be
necessary,
the
capital
gateway
and
to
do
that?
And
then
we,
if
we
virtualized
things
on
the
cable
gateway,
what
happens
when
the
access
goes
down
so
because
I
forgot
the
routing
functions,
virtualizing,
the
in
the
MSA
environment
and
and
the
wine
goes
down.
D
Does
that
mean
we
lose
routing
within
the
home
environment?
So
I
can't
see
my
printer
from
my
pc.
Well,
those
are
the
kinds
of
things
that
we've
analyzed
in
the
project
and
we've
come
up
with
approaches
that
we
believe
are
realizable
and
can
be
deployed
within
a
reasonably
short
period
of
time
and
we're
trying
to
figure
out
ways
of
which
this
enables
the
the
cable
industry
to
come
up
with
a
set
of
scenarios
which
call
out
standardized
functions
from
over
the
open
source
community,
as
well
as
the
the
traditional
quote.
D
Unquote
stands
community,
including
the
ITF
I,
think
my
last
slide
is
so
we've
got
this
said
earlier
on.
This
is
really
a
bit
of
an
invitation
really
to
to
the
vendor
community
to
participate
in
helping
us
know.
The
project
for
we've
got
a
number
of
vendors
involved
already
in
order
to
be
involved,
there's
obviously
a
nondisclosure
agreement,
that's
really
about
protecting
the
discussions
and
then
helping
them
to
be
fairly
open
and
free,
but
that's
not
to
say
that
everything
we
do
is
all
secret.
We
are
going
to
be
pretty
open
about
our
results.
D
We're
going
to
be
coming
into
the
ITF
into
the
relevant
projects
with
with
contributions
and
we're
going
to
be
contributing
code
data
models
end
api's
into
the
open
source
community,
so
I
have
access
to
a
team
of
developers,
pretty
experienced,
Lee
they've
already
pretty
active
in
the
open
source
community
and
they
are
resource
and
that
I'm
very
pleased
to
have
and
we've
got
a
prototype
under
development.
I
think
Michael
is
going
to
demonstrate
something
in
bits
and
bytes,
but
I'm
not
sure
about
now.
B
A
Surely
going
to
Valley
Cisco,
so
you
said
one
thing:
so
the
virtual
home
is
different
from
the
virtual
cpe
art
pictures
that
cover
their
force
that
are
happening
in
the
industry,
so
I
am
associated
with
that
project
on
virtual
CP,
from
what
I
understand,
at
least
in
the
world,
that
we
are
doing
at
Cisco.
What
you'll
see
p
is
essentially
the
ability
to
move
certain
functions
from
the
cpe
and
run
it
in
a
cloud
right.
A
D
Isn't
I
mean
the
I
just
try
to
emphasize
that
we
know
we
know
this
is
not
about
simplifying
the
cpe.
That's
the
main
message
from
this:
it's
about
managing
the
home
network
better
by
virtualizing
the
functions.
So,
of
course
here
you
could
virtualize
pretty
much
all
the
CP
functionality
and
put
that
in
the
cloud
I'm
just
basically
emphasizing
so
that's
not
it's
not
the
focus
with
project.
Okay,.
A
So
one
additional
comment
that
you
said
done:
the
increased
visibility
into
the
home,
that's
or
at
least
that's
how
I
took
it.
One
comment
second
point
you
said
was:
essentially
what
diagnostics
are
something
in?
Essentially
the
ability
to
diagnose
whatever
is
something
goes
wrong
in
a
home.
How
is
what
is
the
architectural
assumption
here
to
realize
that
the.
B
D
Apple,
thank
you
very
much
trying
to
trap
the
is
absolutely
about
choices
for
customers
and
improving
the
customer
experience.
If
we,
the
thing
is
it's
a
virtuous
circle:
isn't
it
if
you
supply
customers
with
services
they
enjoy
and
when
things
go
wrong,
they
prefer
that
even
though
they've
gone
wrong,
then
you
sell
more
services.
So
is
very
much
like
improving
customer
experience,
but
but
how?
What
what?
What
specifically
is.
D
For
example,
be
able
to
see
beyond
gateway
in
home
beyond
building
that
gateway
into
the
home,
beyond
the
routers,
potentially
by
loading
functionality
on
the
routers,
so
that
we
can
diagnose
problems
and
see
it's
either
connectivity
beyond
that
gateway.
If
we
can
see
the
connectivity
beyond
the
gateway,
then
we
could
verify
that
the
router
router
is
actually
connected,
and
maybe
it's
a
problem
with
the
end
user
device.
Wi-Fi
configuration
that
kind
of
that
kind
of
thing,
but.
D
B
B
B
So
before
before,
I
start
this
aaron
aaron
from
the
from
day
with
continues
to
see
here,
could
you
stand
up?
Please
I
meet
it
before
the
horrible
mistake
of
having
our
own
making
a
question
on
the
on
the
mic
lying.
He
is
the
winner
of
the
applied
networking
research
price
from
I
soak
this
year,
so
I
would
like
to
avoid
is
very
much
related
to
this.
We
are
talking
here,
so
they
would
like
to
well
to
greet
the
room
with
a
great
applause
from
there
from
the
group,
because
he's
making
a
really
outstanding
contribution.
B
Well
now
now
this
said:
let's
go
to
the
unimportant
things.
This
is
the
first
well.
This
is
the
first
public
presentation
of
a
banana
which
is,
as
I
said,
there
is
the
data
playing
really
open
source
and
if
the
man,
you
stack
that
we
have
built
in
our
labs
well,
basically,
the
reason
for
building
this
is
is
this
is
that
when
you
move
to
the
data
playing
and
when
you
try
to
visualize
functions
in
there
in
the
data
plane,
you
realize
that
it's
not
simply
moving
the
d
function
ality
to
the
club.
B
It's
not
applying
cloud
two
covers
it's
something
that
is
a
little
bit
different
them,
and
essentially
because
two
things
that
kind
of,
but
not
necessarily
it's
not
the
size
of
the
of
the
workload
the
size
of
the
world
can
be
very,
very
high.
In
other
cases,
if
you're
doing
a
hadoop,
mapreduce
can
be
extremely
high,
but
the
characteristics
of
the
world
are
different.
First
and
second,
you
had
to
follow
the
shape
of
the
network.
B
This
is
very
important
as
well,
because
you
cannot
forget
that
you
have
to
attach
the
functions
to
two
endpoints
and
you
have
to
provide
you
the
functionality
according
to
searching
what,
where
the
network
is,
and
you
have
to
put
the
different
functions
in
the
in
the
appropriate
places.
So
that
means
that
you
cannot
follow,
and
this
is
something
that
we
have
been
experimenting
with.
B
You
cannot
simply
follow
the
cloud
placement
cloud,
orchestration
procedures
and
you
need
to
go
a
little
bit
fatter
father,
not
only
in
the
orchestration
but
as
well
in
the
implementation
to
deal
with
overload.
That's
one
place,
for
example,
something
that
we
have
found
is
that
to
have
network
functions
that
work
in
the
data
plane
that
not
only
perform
at
a
high
rate
but
as
well
that
are
able
to
provide
us
a
predictable
performance.
Do
you
need
to
apply
something
that?
Well,
we
call
them?
B
We
call
it
in
several
ways,
but
some
time
ago
the
people
for
an
inter
came
with
that
with
a
way
of
calling
it
is
a
enhancement
platform
and
awareness,
essentially,
is
to
apply
techniques
that
have
been
applied
before
in
high
performance
computing.
Like
non-uniform
memory,
access
like
the
possibility
of
using
the
what
the
memory
in
unconventional
ways
the
possibility
of
trying
to
avoid
or
shortcut
in
the
by
bus
communications,
with
different
processors,
etc.
So
you
move
from
the
view
that
the
cloud
has
a
provide
surf
to
the
beautiful
machines
that
is.
B
Essentially,
you
have
a
uniform
memory.
You
have
a
uniform
CPU.
You
have
a
uniform
way
of
dealing
with
the
input
output
in
a
way
in
which
you
can
make
some
tweaks
to
optimize
that
performance.
For
example,
you
are
able
to
use
the
pay
the
pages
in
the
memory
in
a
different
way
and
to
locate
locate,
searching
a
set
of.
A
B
B
But
if
you're
running
something
like
a
bit
or
rather-
and
if
you
want
to
give
us
the
same
performance,
it's
really
important
that,
for
example,
you
don't
require
to
go
through
all
the
stack
of
the
of
the
bov
switch
and
they
on
the
operating
and
they
hot
the
guest
operating
system.
Just
to
do
that,
so
you
rely
on
specific
server
libraries
that
are
able
to
provide
direct
access
to
the
to
the
to
the
hard
work
that
has.
B
To
be
allocated
so
well,
yes,
thats,
Sarah,
I
know
you
know
that
sums
up
what
you're
killing
virtualization,
no
no
we're
not
going
to
Vegas
is
a
software
image
that
you
can
locate
freely
on
the
on
the
on
the
club
like
infrastructure.
What
you're
not
doing
is
that
we're
sharing
it
using
the
common,
the
common
platform
that
it
is
provided
by
to
you
by
the
Victrola
station
platform,
you're
running
on
tuba
on
top
of
a
virtualization
platform,
a
yo-yo
shortcutting,
what
affects
your
performance
and
that
makes
your
performance
not
consistent
across
different
deployment.
B
A
B
And
it's
essentially
could
be
open
data
plane
as
well,
but
we
have
a
collaboration
limit,
so
it
is
the
PDK
next
one
please
so
this
is
this,
isn't
it
that
we
are
recently
demonstrated
in
barcelona
in
the
mobile
world
congress,
where
else
close
to
home?
We
were
playing
at
in
hall
at
home
and
just
to
show
there
are
two
beautiful
three
beautiful
daughters
showing
that
well
using
a
traditional
with
the
same
VN
apps,
the
same
servers,
the
same
switches
using
different
way
of
the
point,
the
BN,
apps
and
different
ways
of
playing
with
infrastructure.
B
In
one
cases
we
were
using
opens
up
playing
OpenStack.
In
the
other
case,
we
were
using
our
work.
Our
plus
enhanced
are
aware,
vim
and
orchestration
stack.
That
was
platform
aware
and
knowing
about
this
optimization,
that's
one
place.
These
are
the
results.
This
is
the
results
with
packets
of
more
than
1k.
Well,
you
see
that
the
important
two
important
things
well
when
we
are
talking
about
the
green
is
our
is
our
setup
is
matt
is
over
performing
it,
but
the
important
thing
is
not
about
performance.
Frankly,
we're
not
talking
about
that.
B
You
can
get
a
throughput
of
Tang
or
100
packets
per
second
or
whatever
we're
talking
about
to
be
consisting.
If
you
look,
the
line
in
the
case
of
the
Green
Line
is
completely
straight.
The
red
line
has
irregularities,
and
this
is
something
that
is
what
is
really
complicated
to
deal
with
precisely
the
nones
non-sustainable
non
consistent
performance
unless
one
nice
one
place,
but
they
broke
the
curious
things
when
it
comes
to
small
packets.
The
outperformance
is
two
orders
of
magnitude
and
small
spark.
B
A
B
A
B
Because
the
two
orders
of
magnitude
may
really
complicated
to
show
it
next
one
please
so
for
these
to
achieve
this,
we
found
that
the
current
orchestration
systems
were
not
supporting
what
we
leave
it.
So
we
started
working
in
something
that
it
was
originally.
It
was
something
that
we
started
to
do
at
home,
for
our
lab
and
to
demonstrate
is
once
we
went
into
more
detail
and
we
talked
with
some
of
our
partners,
etc.
B
They
told
us
that
it
could
be
reasonable
to
to
try
to
contribute
it
to
the
two
total
contributed
to
the
open
source
to
see
whether
others
could
find
it
usable
and
the
well
if
they,
if
they
are
willing
to
contribute
us
back
even
telling
us
that
is
pure
crap,
and
they
don't
it
at
all,
and
it
would
be
a
good
result.
Well
here
is:
it
has
an
enhanced
awareness
aware.
D
B
That
is
able
to
do
mapping
pinning
or
or
defining
the
deeper
infrastructure
by
connect
directly
connecting
open
flow
to
the
openflow
standard
switches.
It
has
a
performance
aware
orchestrator
and
it
has
a
generic
bien
FM
that
I
must
confess,
is
under
development.
So
far
we
are
running
without
the
nfm.
We
are
en
effet
the
nav
old
us,
whatever
is
required
to
the
lifecycle
management,
we're
working
on
that
it
is
available
during
this
in
the
github
repository
and
it's
caught
up
in
mono.
B
For
those
of
you
who
know
Spanish
should
be
many
mono
means
hands
in
Spanish
and
by
chance
only
for
you
to
know
that
hands.
This
is
one
of
the
main
source
of
the
really
aiming
it
today
we
make
we
take
a
snapshot
of
the
efficient,
so
whoever
that
wants
to
evaluate,
where
is
that
wants
to
contribute,
is
welcome,
and
now
let
me
confess
one
thing:
the
vim
is
not
based
on
OpenStack:
it's
not
open
stock.
Oh
great
thing!
Yes,
why?
Because
the
current
version
of
urban
zagnut
next
slide.
B
Please
doesn't
support
this
apiary
awareness
and
what
we
wanted
to
avoid.
Is
this
this
situation
in
which
you
take
open
sac
and
you
make
some
modules
you
put
them
as
adults
at
song
and
you
contribute
you
don't
contribute,
then
you
leave
them
over
there
and
then
you
are
completely
looking
to
a
version
of
OpenStack,
because
the
extensions
are
for
a
certain
version.
What
did
you
did
we
do?
We
have
contributed
this
to
the
OpenStack
upstream.
We
expect
to
have
those
extensions
as
part
of
the
kilo
release
in
the
future,
but
right
now
open
manner.
B
If
you
don't
load
up
a
man,
you
don't
be
afraid,
has
no
OpenStack
and
you
will
be
the
bold
ones
that
are
dealing
with
opening
a
good
sorry
with
nav
without
using
OpenStack
and
well.
This
is
you
know.
This
is
a
real
select
club,
that's
one
please!
And
is
it
the
final
one?
This
is
a
plan
we
have.
We
started
playing
with
the
idea
of
having
up
orchestrator
on
your
manager
for
vertical
integration.
For
our
lab.
We
found
that
it
was
possible
to
make
it
a
little
bit
better,
let
better
structure.
B
B
May
have
some
sense,
even
in
the
future,
even
if
we
converse
with
with
with
the
OpenStack
may
have
some
sense,
because
it's
much
more
lightweight,
it's
not
a
full-fledged
shopping
sake,
sir.
It's
a
very
tiny
one
and
for
certain
environments
we
believe
that
could
make
sense.
For
example,
we
have
been
talking
with
some
people
that
are
working
on
precisely
what
is
called
built,
ocp
devito
CP
for
enterprises
and
that
they
are
now
trying
to
base
in.
B
There
are
some
ideas
about
having
a
some
kind
of
split
so
to
have
some
part
of
the
cpe
still
in
the
customer
premises
and
to
have
to
some
part
of
the
CP
in
the
inductor
on
the
cloud
and
in
those
cases,
maybe
there
is
a
tiny
orchestration
stack
could
have.
Two
could
have
some
merit
in
the
piece
small
piece
that
you
live
at
the
other
customer
premises.
This
is
something
that
we
are
thinking
about.
This
is
something
that,
if
you
download
it
you
play
with
it,
you
have
an
idea
of
how
it
goes.
B
E
E
B
I
bet
by
Caltrans
behavior,
we
mean
here
precisely
something
that
provides
a
performance
that
is
acceptable
for
our
operational
guys
to
substitute
the
physical
functions
for
the
virtual
functions
first
and
second,
and
mostly,
most
importantly,
precisely
so,
then
that
provide
a
consistent
performance
that,
wherever
you
make
a
deployment,
you
know
that
you're
going
to
get
10,
gigs
10
max
1k,
but
always
the
same,
because
our
experience
with
it
with
the
vanilla
cloud
it
was
waterfront
over
to
some
time
was.
It
was
one
key,
something
that
the
clouds
the
management
system
was
telling
us.
B
A
A
B
Far,
the
we
are
starting
to
go
to
the
training.
Do
we
I
mean
very
few?
We
are
very
much
focused
on
a
couple
of
projects
that
we,
we
have
been
assigned
a
very
tight
schedule
to
show
that
this
can
deliver.
So
we
are
very
focused
on
something
that
is
connected
to
this
home.
Brutalization
Don
was
mentioning
before,
and
so
they
set
the
service
training
is
very
limited.
I
would
say
we're
talking
not
more
than
three
four
functions
right
now.
We
have
plans
and
well
right
now.
B
This
teaching
of
the
functions
is
done
through
the
directly
through
the
open
flow
controller.
In
the
end
of
em,
we
have
plans
to
explore
the
deal
for
general,
a
function
chaining
later
on.
So
this
is
something
that
and
well
if
you,
if
you
won't,
do
somehow
things
about
to
help
to
merge
these
with
a
opening
f,
and
this
is
something
that
that
would
be
really
interested
to
talk
to
you.
A
B
A
B
B
B
A
E
A
E
You
can
set
up
a
test
bed
or
you
can
run
simulation
or
you
can
develop
analytical
model.
There
are
varieties
of
way
to
solve
this
problem,
so
the
next
few
minutes
are.
We
talking
about
I
am
going
through
some
of
the
selected
slides,
not
all
of
the
slides.
If
you
are
interested
in
the
details
of
the
development
of
the
model
analysis,
there
is
a
reference
to
journal
papers
which
we
have
published
in
the
GT
communications
mera
jee.
E
So
the
outline
is
what
is
the
difference
between
traditional
telecom
demand
hardware
and
the
common
of
the
shell,
and
we
will
talk
about
the
analysis
framework.
As
I
say,
it
is
simulation
and
also
Markov
chain
model
like
anybody
who
has
done
state
transition
diagram,
you
will
start
with
short
and
then
we
will
keep
on
adding
the
transition
states,
the
state's
explode
at
some
point.
You
have
to
limit
it
because
they
becomes
unmanageable
and
then
we'll
talk
about
the
enhancing
application
reliability
by
our
backups.
E
Traditionally,
people
try
to
do
one
for
one
back
up
14
to
back
up.
But
the
point
is:
if
you
keep
on
adding
the
backup
servers,
it
adds
the
complexity
but
the
marginal
gain.
It's
not
as
I
as
you
expect.
So,
instead
of
141
backup,
we
have
14
to
backup
or
if
you
have
shared
operation,
so
you
can
see
how
these
things
change.
That's
we
also
included
in
the
model
and
then
we'll
talk
about
some
of
the
results.
Next,
like
this,
so
the
traditional
telecom
harder.
E
As
you
know,
they
have
strong
fall,
detection
and
fault
isolation
capability
at
the
other
level,
because
the
harbor's
have
been
hardened
to
operate
in
the
harsh
conditions
and
well-established
tradition
on
software
upgrade
and
patching
anybody
who
has
experienced
on
PhD
and
switches
or
network.
You
know
that
they
have
scheduled
maintenance
window
at
3m
in
the
morning
for
half
an
hour
or
something
like
that
when
the
traffic
is
very
low
and
then.
E
Of
the
central
of
fish
is
assumed.
That
means
five
nines
of
availability
like
call
completion
time
is
within
certain
limit
and
also
the
quality
of
the
data,
plane,
control,
plane
and
also
half
times.
This
has
to
be
significantly
high.
Now,
in
the
course
hardware,
so
they
may
have
smaller
mean
time
between
failure
and
the
mean
time
to
repair
and
the
cots
procedure
for
software
upgrade
and
patching
and
maintenance.
Those
are
not
oil
scheduled
or
oil
establish
time
window,
so
the
different
grade
of
reliability
for
different
data
centers.
E
D
E
Points
basically
is
the
gods
procedure
for
software
upgrade
and
patching.
Those
things
are
the
main
culprit,
because
if
those
things
are
unpredictable,
you
will
see
that
the
side
down
time
for
type
1
type,
2
type
3
and
five
for
data
centers.
It
can
vary
from
ninety
nine
point:
six,
seven
one,
two
ninety-nine
point:
nine
nine
five,
which
is
almost
59,
which
is
the
traditional
the
carrier
grade
yeah.
So
this
is
the
traditional
141
back
up
next
one
and
next
so
in
this
one
you
can
see
the
system.
E
Availability
is
p1
times
p2,
which
is
the
p1.
Is
this
availability
of
the
left
side,
block
platform
and
the
p24
the
right
side
and
the
connectivity
also
contribute
to
the
reliability
and
availability
next
one.
So
is
it
possible
to
have
servers
with
49
or
even
three
lines
of
reliability
and
ability
to
provide
the
five
lines?
So
the
answer
is
yes,
but
the
marginal
gains
becomes
low
as
you
keep
on
adding
more
servers
next
one.
So
this
one
is
141
back
up
next,
so
this
one
as
you
can
see
one
primary
backup
on
secondary
backup.
E
But
the
point
is
you
have
to
do
the
synchronization
of
the
master
with
the
slave
server,
one
which
is
back
up
one
and
in
addition,
you
have
to
do
the
synchronization
of
the
slave
server
back
up
to
with
the
master
as
well.
So
it
adds
additional
complexity
because
you
have
two
backups
and
the
pepper
I
was
mentioning
is
available
in
this
URL.
Anybody
interested
you
can
go
and
download
the
pepper
and,
as
I
said,
this
paper
is
a
shortened
version
of
the
big
report.
Anybody
interested
send
an
email
to
the
distribution
list.
E
E
E
So
this
is
the
school
markup
straight
through
state
Markov
model,
the
server
is
good
and
the
current
master
and
the
server
is
bad.
So
this
is
typical
to
state
transition
model.
So
this
is
the
simplest
model
and
then,
when
you
add
the
complexity
next,
so
intermediate
States,
the
silent
error
probability.
So
in
this
case
the
s1
and
s2
one
server
is
good
on
several
said,
and.
A
E
Okay,
so,
as
you
can
see,
as
you
keep
on
adding
the
servers
and
if
the
the
server
mean
time
between
failure
and
the
silent
error
probability,
you
increase
this
one.
As
you
can
see,
you
can
see
this
server
with
69,
so
you
can
achieve
those
kind
of
availability,
next,
one
yeah,
so
with
the
mean
time
to
repair.
It's
also,
as
I
said
before,
the
improvement
in
significant.
It
is
not
as
significant
so
with
five
backup.
E
E
Also
from
the
simulation
next
so
again,
improvement
for
1,
plus
to
dwell
back
up,
please,
as
you
can
see,
you
can
achieve
improvement
deteriorates
fast
with
this
island
error.
Not
significant
influence
is
not
as
significant
as
it
was
before
next,
so
this
is
the
before
metal
starts
for
no
fault,
you
have
do
the
synchronization
that
are
simpler,
one
with
the
Sun
to
next
one
yeah.
E
So
this
is
the
dresser
one
where
we
have
the
mush
runs
late
and
the
dead
center
be
could
be
in
maintenance
mode
with
other
application,
and
the
after
maintenance
starts
reverting
when
there
is
no
fault
next.
So
this
impact
of
the
site
maintenance
is
negligible
again.
This
is
based
on
the
state
transition,
diagram
and
developing
on
the
equation
and
solving
them
next
next.
E
So
these
are
the
four
types
of
data
center
we
considered
in
the
first
one
single
non
redundant
and
the
reliability
is
99.6
71
in
the
second
one
we
have
redundant
component.
That
means
the
capacity
has
some
additional
capacity
available.
You
can
achieve
reliability
of
19
and
17
for
one
in
the
third
one
you
have
redundant
component
and
also
redundant
connectivity,
IT
equipment.
So
in
the
fourth
one,
it
means
all
of
the
first
second
and
third
requirement.
In
addition,
it
has
standby
power
supply,
so
in
case
the
power
supply
fails.
E
E
Diagram
model:
you
can
get
the
more
details
in
from
the
pepper
which
we
have
put,
and
this
is
the
service
impact
error
probability
for
various
data
center
next
next,
so
it
is
possible
to
provide
five
nines
of
reliability
and
availability
with
God's
hardware
with
the
application
level
backup.
So
the
impact
of
the
mean
time
to
repair
is
not
as
significant
if
it
is
reasonably
solve,
for
example,
less
than
10
minutes
next,
so
the
impact
of
the
data
center
scheduled
maintenance
is
also
negligible.
So
the
cost
procedure
for
software
upgrade
patching.
E
They
do
not
contribute
significantly,
so
the
thing
is
different
grade
of
reliability
for
data
center
can
be
achieved
and
59
so
available.
Availability
can
also
be
achieved
using
type
3
and
the
type
for
data
centers,
where
you
have
additional
connectivity
capacity
and
in
the
type
for
you
have
a
redundant
capacity,
redundant
connectivity
and
standby
power
supply.
I
think
this
yeah,
maybe.