►
From YouTube: IETF 114: Host Speaker Series
Description
Elad Safshi leads the unified core and access network teams that drive innovation, reliability, scalability, and development across Comcast Cable's network capabilities as Executive Vice President, Chief Network Officer.
A
A
B
You
so
much
for
having
me
jason.
Thank
you,
everyone
for
being
here,
both
in
person
and
virtually
it
is
so
good
to
see
people
again,
and
it's
so
good
to
you
know,
to
be
able
to
interact
in
these
forums,
and
I
really
appreciate
everyone,
both
virtually
and
especially
here
in
the
room
for
taking
the
time
and
joining
us.
B
I'm
gonna
be
describing
our
path
to
what
we
call
10ge,
which
is
really
the
continuous
progression
on
the
network
as
we
continue
to
to
work
and
grow
our
connectivity,
business.
Okay,
I
will
have
time
for
questions
at
the
end,
so
you
know
feel
free
to
ask
away
and
in
the
meantime,
let's
just
get
right
into
it.
A
B
Excellent
well,
first
and
foremost,
welcome.
Welcome
to
philadelphia,
welcome
after
you
know
very
long
pandemic,
where
we
all
spend
way
too
much
on
looking
at
video
screens.
Comcast
is
proud
and
happy
to
to
be
sponsoring
this
event.
I
want
to
thank
jason.
Livigood
who's
really
been
absolutely
instrumental
in
in
everything
that
we
do
here
and
in
other
forms,
and
it's
super
exciting
and
and
candidly
an
honor
and
a
privilege
to
be
able
to
to
host
this.
You
know
great
event.
B
I
also
want
to
thank
everyone,
each
and
every
one
of
you
for
all
the
hard
work
over
the
years
in
making
internet
delivery
what
it
has
been
right.
If
you
really
take
a
step
back
and
think
about
this
march
13th,
which
was
a
friday
of
2020,
we
all
went
home,
okay
and
the
network
in
a
day
needed
to
adapt
to
a
scenario
that
none
of
us
really
expected
right.
Who
would
have
thought
that
in
our
lifetimes
we
all
would
go
home,
jump
on
teams
and
zooms?
B
Europe.
Other
parts
of
the
world-
not
here-
and
I
take
quite
a
bit
of
pride
in
that
I
really
do-
is
having
the
opportunity
to
lead
the
team,
as
we've
done
that,
and
I
had
a
great
discussion
earlier
with
stuart
at
apple.
Who
commented
about
the
improvements
that
we've
done
in
latency
throughout
all
of
this
and
that
I
think
again.
That
speaks
volume
to
the
technical
leadership
that
you
know
comes
from
this
organization,
and
many
thanks
for
that.
B
Okay,
let's
start
talking
about
our
actual
roadmap,
so,
first
and
foremost
for
the
folks
who
are
not
familiar
with
comcast
or
you
know-
maybe
familiar,
but
don't
quite
understand
how
big
our
network
is.
It's
really
big.
It's
really
really
big.
When
you
really
look
at
the
network
end-to-end
over
a
million
miles
of
fiber
and
clocks
plant
spread
across
the
entire
u.s
39
states
in
the
district
of
columbia,
where
we
service
our
customers
throughout
that
footprint
in
terms
of
the
delivery,
how
many
transactions
are
we
actually
carrying
through?
B
And
I
think
that
that
is
such
I'm
trying
to
you
know
balance
between
all
these
screens
so
that
you
could
still
see
the
screen,
and
I
could
you
know
not-
you
know,
get
out
of
view
for
the
lens
1.3
trillion
dns
lookups
every
day,
1.3
trillion
capacity,
literally
doubling
on
the
access
network,
every
two
and
a
half
years
so
think
about
it.
This
way,
comcast
has
been
in
the
high
speed
business
since
late,
the
late
90s
right
and
every
two
and
a
half
years.
B
Our
customer
needs
continue
to
evolve
and
my
team's
job
is
to
make
sure
that
the
network
will
always
be
infinitely
scalable
and
will
be
able
to
never
endure
business
growth.
Okay,
so
let's
go
into
what
2021
was
and
it's
funny.
The
summarized
2021
were
almost
in
august
of
2022,
but
it's
I
thought
it
would
be
interesting
to
kind
of
take
a
step
back
and
look
at.
B
You
know
where
we
come
from
before
we
talk
about
where
we
head
into
as
much
as
we
saw
internet
growth
in
2020,
and
we
saw
explosive
internet
growth
and
added
millions
of
subscribers
to
our
network,
which
we're
very
happy
to
do
in
2021.
We
continue
to
grow
usage,
okay,
both
downstream
and
upstream
so
downstream.
Continued
to
increase
at
peak
at
11
upstream
increased
by
five
percent.
B
Okay,
and
so
the
reality
of
it
is
that
the
network
is
still
very
much
a
downstream
delivery
network.
Okay,
it's
not
to
say
that
upstream
is
not
important.
It
is,
and
we've
seen
some
pretty
significant.
Obviously,
traffic
increases
on
the
upstream
driven
by
video
telecommunications,
but
it's
still
very
much
driven
by
downstream
consumption,
primarily
entertainment
services.
That's
not
a
surprise
right
and,
as
you
see,
those
ott
services
continue
to
proliferate.
B
B
Last
but
not
least,
while
video
conferencing
share
of
peak
network
traffic
literally
doubled
during
the
pandemic,
okay
doubled
one
day,
it's
still
only
five
percent,
actually
less
than
five
percent
of
the
total
bits
carried
on
the
network.
Okay.
Now,
if
this
guy
has
anything
to
do
with
it,
you
know
they'll,
you
know
they'll,
you
know
we'll
all
be
broadcasting
at
8k
out
of
our
home,
but
you
know
the
reality
is
that
you
know
for
the
you
know
for
the
current
subscriber
trajectory.
B
It's
you
just
put
that
into
a
perspective.
Okay,
let's
talk
about.
You
know
what
it
means
to
the
network
and
what
I
try
to
show
here
is
both
kind
of
the
physical
topology,
but
also
the
spectral
topology
of
what
the
comcast
network
represents,
so
60
million
homes.
The
vast
majority
of
the
network,
which
I'll
start
you
know
down
below,
is
rated
at
750
megahertz
of
of
spectral
capacity,
okay
of
which
the
small
part
of
which
you
see
down
below
on
the
left-hand
side
between
5
and
42
megahertz
is
dedicated
to
upstream.
B
Everything
else
is
dedicated
to
downstream
and
is
really
shared
by
the
various
formats
of
docsis
over
cable
delivery
and
traditional
video
channel
deliveries.
While
all
of
our
lineup
does
is
carried
in
ip,
we
still
deliver
channels
to
traditional
set
of
boxes
are
quant
based,
okay
in
terms
of
the
physical
plant,
you
know
what
what
it
looks
like
for
high-speed
data
delivery.
Traditionally
we
relied
on
separate
pipe
delivery.
B
Okay,
the
vast
majority
of
delivery
is
over
docsis,
but
we
do
have
parts
of
the
network
that
are
fed
with
fiber
and
what
you
see
here
is
a
typical
cable
architecture
which
relies
on
analog
fiber
from
the
cmts,
which
resides
in
a
local
data
center,
typically
closer
to
the
subscriber
because
of
the
analog
optics
limitations,
feeding
an
analog
node.
The
reason
for
that
is
historical.
B
It's
the
technology
that
existed
when
the
two-way
cable
plant
was
built,
and
then
we
used
amplifiers
to
basically
amplify
and
extend
the
rf
reach
all
the
way
down
to
the
home.
What
you're,
seeing
here
in
cable
lingo,
is
something
called
n,
plus
five,
the
node
plus
five
amplifiers
on
this
cascade,
okay,
and
this
is
kind
of
a
what
a
traditional
cable
network
looks
like
not
just
for
comcast,
but
you
know
among
our
other
cable
brethrens,
and
the
challenge
with
this
is
really
twofold.
B
Right
number
one
is
notice
that,
while
the
network
is
asymmetric,
there's
obviously
a
very
small
sliver
of
capacity
that
is
available
to
be
sent
upstream.
Those
are
not
easy
things
to
fix,
because
they're
really
limited
by
rated
devices
in
the
network
that
are
able
to
pass
or
not
pass
frequencies
in
order
to
change
the
number
of
lanes
that
are
dedicated
to
to
upstream
and
downstream
delivery.
B
Okay.
Now
the
challenge
that
we
have
is
analog
optics
in
a
world
of
you
know
everything
digital,
a
cmts,
which
is
an
appliance-based
architecture
in
a
world
that
is
all
web
scale
and
web
speed.
How
do
we
evolve
the
architecture
to
to
be
able
to
authorize
the
network
challenge
and
about
five
years
ago?
I
got
this
amazing
opportunity
to
really
take
this
on
at
comcast
and
lead
this
virtualization
transition
across
the
board.
So
why
don't
we
go
to
the
next
slide?
B
Please,
and
what
you're
seeing
here
is
the
continuous
evolution
and
innovation
on
the
access
network
and
I'm
gonna
start
with
the
spectrum
and
you'll
see
that
what
we're
doing
is
we're
doing.
Two
things
number
one
is
we're
increasing
the
size
of
the
pipe
okay
to
either
one
gigahertz
or
1.2
gigahertz.
B
Why
are
we
doing
this
because
the
customer
needs
continue
to
evolve
right?
You
know:
how
do
we
continue
to
have
the
pipe
that
is
there
whether
the
capacity
is
needed
today
or
whether
it's
going
to
be
needed
10
years
from
now
right?
Those
are
very
large,
lengthy
upgrades
that
we
we
need
to
go
through.
The
second
piece
is
that
notice
that
we're
increasing
the
amount
of
capacity
that
is
dedicated
to
upstream
by
executing
on
something
called
mid-split.
B
Okay,
there's
an
another
split
called
high
split,
which
extends
the
upstream
frequencies
all
the
way
up
to
204
megahertz.
The
reason
why
we
chose
a
mid-split
is
you
see
that
garband
that
white
space
in
between
it's
there?
You
know
so
that
we
could
continue
to
deliver
video
services
to
cable
cards
and
other
retail
devices
uninterrupted
okay?
And
it's
really
important
for
us
that,
as
the
network
progresses
we're
not
interrupting
or
disrupting
the
customer
delivery
from
a
service
standpoint,
even
on
legacy
video
quan
video
not
just
legacy.
B
What
we're
also
doing
is
that
we're
virtualizing
the
entire
delivery
system,
and
so
we
have
launched
several
years
ago
what
we
call
our
virtual
cmps
by
disaggregating
and
separating
the
hardware
from
the
software
we're
able
to
innovate
in
web
scaling.
Web
speeds
we're
running
on
intel-based
platforms
instead
of
proprietary
appliances.
B
You
know
put
that
in
perspective,
we're
on
our
second
generation
of
that
platform
about
to
transition
to
the
third
generation
of
that
platform,
all
riding
on
intel's
moore's
law
coattail.
You
know
with
effectively
enabling
to
double
the
density
and
capacity
writing.
Moore's
law,
rather
than
you
know,
reliant
on
proprietary
chipset
applications
we're
transitioning
to
brand
new
digital
fiber
capabilities,
which
enable
us
to
extend
the
distance.
You
know
on
the
physical
plant,
which
enables
us
to
invest
more
into
what
we
call
the
critical
infrastructure.
B
B
The
legacy
video
delivery
will
continue
to
function
and
all
of
that
will
be
able
to
do
on
our
coaxial
cable.
I
had
the
privilege
of
demoing
this
technology
at
cablelabs.
Back
in
april,
we
were
able
to
demonstrate
nine
gigabits
down
in
five
gigabits
up
on
this
under
service
and,
what's
really
cool
about
this
is
with
the
virtualization
innovation
that
you
know.
We've
worked
on
for
the
past
four
or
five
years.
B
That's
the
innovation
that
is
the
comcast
commitment
to
continuous
investment
in
the
network
in
our
technology
and
into
products
and
services
that
our
customers
continue
to
demand
of
us,
as
technology
continues
to
evolve,
super
exciting
to
see
it
happening,
but
we're
not
done
there.
Let's
go
to
the
next
slide,
please.
B
I
talked
earlier
about
the
low
latency
delivery
and
what
we
do,
what
we've
done
in
terms
of
active
queue
management
during
the
height
of
the
pandemic,
and
how
and
again
then
rice's
team
here
really
helped
us
to
optimize
the
cmts
delivery
configurations
so
that
we
don't
just
take
care
of
capacity
and
we
don't
just
take
care
of
speed.
Yes,
we
deliver
gigabit
services
to
more
customers
than
anyone
else.
Anyone
else
speed
is
one
element.
B
Still
ahead
of
us,
I
haven't
even
talked
about
how
the
same
virtualized
platform
effectively
abstracts
the
wire
delivery,
and
so
I
mentioned
earlier
the
fiber
to
the
home
delivery
that
we
have,
and
we
with
the
virtualized
platform
and
the
distributed
access
architecture
that
we
put
in
place.
We're
able
to
deliver
those
multiple
gigabit,
symmetrical
surfaces
on
whether
the
fiber
is
coaxial
or
whether
the
fiber
is
fiber
or
whether
the
wire
is
fiber,
because,
again
the
customers
don't
care
whether
there's
clocks
in
the
black
wire
or
whether
there
is
a
glass
in
the
black
wire.
B
They
want
to
make
sure
that
the
products
and
services
that
they
rely
on
they
continue
to
to
deploy
with
the
virtualized
architecture.
With
the
digital
connection.
What
I
am
emotionally
excited
about
is
the
real-time
visibility
and
the
real-time
traceability
that
we
have
to
the
service
end-to-end,
it's
a
digital
connection.
B
It's
now
doesn't
rely
on
polling
technology
anymore.
It
actually
streams
the
real-time
customer
experience
across
this
network
in
real
time
so
that
we
could
act
on
it.
You
know
I
mentioned
the
octave
ai
engine
that
we
built
for
bit.
Optimization.
Imagine
if
you
will
the
same
thing
for
fiber
connectivity.
B
Imagine
if
you
will
service
interruptions
where
we're
able
to
detect
in
five
seconds
instead
of
you
know,
polling
technology
which
is
ultimately
limited
by
at
the
cpu
and
the
appliance
and
so
on
and
so
forth,
and
so
this
architecture,
beyond
the
fact
that
it's
going
to
be
faster
beyond
the
fact
that
it's
going
to
deliver
low
latency
like
no
other,
is
truly
extensive.
It's
truly
broadly
deployed.
B
You
know
we
don't
dig
this
side
of
the
street
or
that
side
of
the
street.
All
of
our
customers
are
able
to
experience.
Gigabit
speeds
today
and
they'll
be
able
to
experience
even
better
in
the
future.
So
that's
my
my
story
for
the
for
the
path
of
10g
and
I'll
pause
here
and
I'm
happy
to
take
any
questions
that
you
you
may
have,
whether
it's
here
in
a
room
and
let
me
start
in
a
room
and
then
you
know
we
could.
C
C
C
A
question
I
just
came
up
to
thank
you
for
your
presentation
to
thank
comcast
for
its
ongoing
support
of
the
ietf
to
thank
all
the
comcast
engineers,
like
jason,
livengood
and
others,
that
I
work
with
for
the
amazing
work
you
do
being
on
the
forefront
of
creating
new
technology
and
I'm
not
going
to
name
other
names.
But
there
are
other
isps
and
operators
who
are
not
here,
who
are
very
clearly
playing
catch
up
and
five
years
from
now,
they'll
be
finally
figuring
out.
What
you're
doing
and
realizing.
B
C
B
E
Can
you
hear
me
perfect
yeah,
so
amir
virginia,
so
I
so
you
actually
did
a
very
good
like
interesting
job
and,
like
you
know,
congratulations
are
you
looking
to
expand
this
sort
of
work
like
you
know,
to
other
parts
of
the
network,
because
this
is
one
small.
I
mean
it's
not
a
small
section,
but
it's
one
part
of
the
network
so.
F
B
B
So
the
short
answer
is
yes,
but
let
me
give
you
the
long
answer
too.
Okay,
we
don't
do
technology
for
the
sake
of
technology.
We
don't
right,
you
know
as
much
as
you
know,
engineers
like
to
do
that.
That
is
not.
That
is
a
means
to
an
end
in
the
end.
Here
is
a
customer
experience.
It's
reliability,
it's
the
scalability
or
continuous
scalability
and
continuous
growth
of
the
network.
B
Okay,
when
you
look
at
the
other
side
of
the
network,
the
smaller
side
of
the
network,
the
core
network
is
I'm
just
joking,
the
the
when
you
look
at
the
core
delivery
right.
The
core
delivery
today
very
much
relies
on
appliances
up
and
down
the
delivery
for
various
historical
reasons.
Right.
You
know
this
is
being
a
vendor-led
vendor-driven
industry
that
had
very
little
incentive
to
innovate,
disrupt.
B
Okay
and
you
know
it
served
extraordinarily
well
and
continues
to
serve
extraordinarily
well.
But
I
look
at
some
of
the
work
that,
for
example,
18t
has
done
in
terms
of
of
virtualizing
their
core
network
and
driving
both
asic
and
nos
diversity
onto
the
core
network.
I
look
at
the
progressions
in
photonics
across
the
transport
layer
and
when
you
see
innovations
like
zero
optics
and
others,
you
begin
to
to
come
in
and
how
those
get
paired
with
the
aggregation.
B
Routers
and
you
know
the
two
coming
together.
I
think
that
there's
tremendous
opportunity
in
the
core
network
we
are
actively
looking
at.
You
know
what
the
right
technology
decisions
are
for
us,
but
again
to
me
beyond
the
technology
choices
and
beyond
the
hardware
and
software
abstraction
to
me,
it's
the
real-time
visibility
and
how,
for
example,
we
rely
on
gnmi
in
great
scale
right,
not
just
at
the
access
layer,
not
just
at
the
you
know,
close
to
the
customer,
but
really
up
and
down,
and
be
able
to
rely
on
on
real-time
streaming.
B
E
Just
allocate
that
english
as
a
second
language
so,
but
so
you
actually
bring
a
very
good
point
of
like
you
know
ai,
so
industry
is
talking
about
self-healing
networks
for
a
while
all
that
stuff
that
comes
into
picture
is
that
something
that
do
you
think
would
eventually
I
mean
and
everyone's
talking
about
it
for
the
last,
like
you
know,
six,
seven
years.
So
where
do
you
see
in
your
opinion,
where
that
would
you
think
the
self-healing
network?
Would
we
will
see
that
in
the
next
couple
of
years
or
it's
a
ten-year.
B
So
so
I
will
tell
you
that
again
with
the
octave
example
as
it
is
right
now
right,
we
are
effectively
modulating
around
network
impairments,
as
we
speak
millions
of
times
a
day
today,
okay
and
nuts,
on
the
you,
know
the
access
network,
without
going
into
too
many
details.
At
this
point,
we
have
a
platform
that
we
call
net
iq,
which
you
know
operates.
You
could
check
it
out
online.
We
actually
have
a
you
know
a
couple
postings
or
not.
B
B
Dealing
with
kind
of
you
know,
vertical
appliances
is
that
they
don't
care
across
very
well
right.
But
if,
for
example,
the
issue
was,
I
had
a
big
local
metro
area,
fiber
cut.
Okay,
if
you
everyone,
I'm
sure,
is
called
their
cable
company
to
try
and
get
service.
And
the
first
thing
the
unhold
message
tells
you
is
reboot
your
modem,
okay.
Well,
you
could
reboot
your
model
a
thousand
times.
If
there's
a
fiber
cut
to
your
area,
that's
not
going
to
help
right,
and
so
how
do
we
de-duplicate
the
alarms?
B
Because
think
about
the
alarms
that
when
it
just
went
off
in
our
network,
the
cable
modems
were
streaming
that
they
lost
connection.
Okay,
the
smart
network,
the
amplifiers,
the
nodes
and
the
vcnps
said
we're
not
getting
traffic.
It's
not
coming
through.
Okay,
the
local
network
said
well,
I
lost
something
northbound
and
now
it's.
How
do
you
shoot
the
fiber
in
order
with
an
otdr
in
order
to
pinpoint
exactly
where
the
fiber
disconnect
is,
and
that's
what
we're
working
on
right
now?
B
How
do
you
tie
all
of
these
indicators
together
into
a
coherent
alarm?
That
only
pinpoints
where
the
breakage
point
is,
but
also
then
has
the
intelligence
to
message
to
our
customers,
what
their
expectations
are.
Okay,
and
so
it's
not
just
about
the
self-healingness
of
this
is
what
do
you
tell
the
end?
Customer
okay,
don't
reboot
your
modem,
there's
a
fiber
cut
in
your
area.
B
D
A
A
A
A
G
B
So
sean
it's
a
it's
a
good
question.
Let
me
let
me
answer
it
this
way.
Okay,
comcast
is
not
a
wireless
company.
We
do
offer
cellular
services
but
they're
through
an
nvno
with
verizon.
We
do
have
a
very,
very
large
wi-fi
network.
In
fact
it's
the
largest
in
the
us
which
we
use
for
you
know
both.
You
know
customer
benefits,
but
also
as
a
way
to
offload
solar
minutes,
and
you
know
help
with
the
economics
of
the
nvnl
as
such,
and
really
driven
by
that.
B
Obviously,
you
know
what's
right,
for
the
comcast
network,
you
know,
may
be
different
than
companies
like
rogers.
That
is
actually
a
wireless
carrier
and
has
you
know
a
tighter
integration
and
so
sean?
I
really
think
that
the
answer
to
your
question
is
probably,
but
it
really
depends
on
the
actual
operator
and
the
the
unique
services
that
they
provide.
H
B
And
you're
you're,
100,
right
and
jason.
Thank
you.
I
would
say
you
know
last
comment
on
this
on
the
ipv6
pieces,
it's
so
ingrained
to
everything
that
we
do
from
a
network
standpoint.
We
could
not
scale
at
some
point.
We
were
the
largest
consumer
of
ipv4
addresses
and
and
we
literally
hit
a
wall.
So
you
know
this
has
been
a
instrumental
for
us.
You
know
to
your
point
for
many
many
years.
Yes
last
one
and
then
they
kicked
me
out
hi.
F
Paul
curtis
sienna
in
talking
about
the
monitoring
and
troubleshooting
you
mentioned
gnmi.
I
guess
I'm
curious
to
the
extent
you're
using
gnmi
just
for
monitoring
also
for
all
network
management,
and
if
there's
anything
you
can
say
about
your
progress
there,
where
you
guys
are
at
and
rolling
that
out,
yeah.
B
B
I'm
trying
to
think
what
I
could
say.
I
could
say
that
I
could
say
this:
we're
very
excited
about
the
fact
that
it
is
a
common
standard,
even
though
every
vendor
siena
has
a
you
know,
different
implementation
of
gnmi
right,
you
know
same
for
the
nokia
person
you
know
back
there
and
and
so
on
and
so
forth.
Okay,
and
so
part
of
an
operator
challenge,
is
what
pieces
of
it
can
you
take
that
you
could
fairly
standardize
that
you
could
then
feed
across?