►
From YouTube: 2022 RetMkt Landscape - Juan Benet
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
All
right,
everyone
great
to
be
here
with
you.
I
wanna
thanks
for
everyone
for
gathering
and
great
to
be
talking
about
greatly
convening
the
retail
markets
workshop.
What
I
want
to
talk
about
today
is
this:
these
four
things
so
just
quickly
touch
on
the
review
markets,
working
group
and
the
workshop
working
workshop.
Patrick
already
gave
a
phenomenal
intro
to
the
retrieval
markets,
problem
and
the
working
group,
and
so
on.
A
A
This
will
be
useful
for
the
people
actually
building
the
the
tech
to
be
able
to
tune
their
approaches
based
on
the
other
opportunities
that
are
happening
in
the
landscape,
and
then
from
that.
I
want
to
then
kind
of
have
an
open
discussion
about
the
retro
markets,
landscape
in
2022
and
various
different
opportunities
and
from
that
narrow
down
to
some
set
of
network
goals.
A
So
the
two
last
two
sections
will
be
more
conversational,
so
want
questions
and
and
so
on,
from
either
group
here
or
on
on
youtube
and
so
on,
and
then
from
that
I
want
to
kind
of
like
propose
a
very
concrete
network
goal
that
we
should
hit
basically
as
fast
as
possible
to
kind
of
unblock
a
bunch
of
the
the
kind
of
open-ended
innovation
that
we
can,
that
I
think
many
other
groups
in
the
ecosystem
have
gotten
to
great
so
kind
of.
A
In
short,
the
objective
of
the
faculty
of
market
is
to
build
the
world's
best
cdn
leveraging
web3
tech.
So
it's
not
about
just
building
the
best
cdn
in
the
web
3
world.
We
want
to
really
build
the
world's
best
cdn
leveraging
the
web3
tech
underneath
the
hood.
So
we
can.
A
We
can
leverage
content,
addressing
crypt
incentives,
verifiability
permissionless
networks
and
markets
to
build
a
massive
scale
cloud
cdn
that
should
compete
and
beat
decentralized
providers
kind
of
the
a
shorter
version
it
kind
of
like
intermediate
target
building
towards
that
is
to
build
a
worldwide
fast
and
efficient
cdn
for
ipfs
content.
So
you
can
use
that
as
a
stepping
stone
on
the
way
to
building
the
world's
best
cdn,
and
so
you
know,
if
we
focus
on
the
second
goal
as
an
intermediary
goal,
we
can
get
to
get
get
to
the
first.
A
The
there's
been
a
a
working
group
effort
around
building
the
regional
markets
for
a
while.
The
I
won't
go
into
in-depth
there,
because
patrick
already
did
a
phenomenal
job,
giving
giving
us
an
overview.
I
want
to
kind
of
point
out
a
set
of
goals
that
I
think
we
should
be
aiming
for.
A
We
should
be
aiming
for
a
network
of
about
you
know
10
000
plus
devices
at
minimum,
so
that
ideally
hundreds
of
thousands
and
millions,
but
just
getting
a
network
up
of
tens
of
thousands
of
nodes
would
already
greatly
improve
the
the
setup
then
from
there
getting
to
thousands
of
people,
thousands
of
individuals
involved
in
material
mining.
If
we
can
get
to
that
as
fast
as
possible,
so
say
in
a
quarter
or
two.
A
That
would
be
a
that
would
kind
of
unblock
a
lot
of
the
innovation
and
then,
if
we
can
establish
less
than
one
second
on
average,
for
delivery
of
any
cid
across
all
content
that
would
be
transformative
to
the
whole
ipfs
network
and
the
falcon
network.
So
right
now,
retrieval
is
a
big
bottleneck
for
a
lot
of
use
cases
as
I'll
go
in
in
detail
in
a
moment,
and
so,
if
we
can
just
get
to
the
point
where,
for
most
retrieval
queries
for
any
cid
worldwide
users
can
get
a
less
than
one
second
retrieval.
A
A
So,
let's
start
with
like
a
very,
very
concrete
creep
problem
space
of
you
know
not.
You
know
10
000
nodes,
it's
not
that
large
of
a
network.
We
can
totally
do
that.
A
thousand
people
plus
not
that
not
that
hard.
We
can
totally
do
that
less
than
one
second
of
delivery
for
cids
worldwide.
That's
harder,
but
we
can
do
it
if
we
can
focus
on
the
things
the
things
that
are
requested
the
most,
and
I
claim
that
there's
probably
somewhere
between
one
to
ten
petabytes
of
hot
content.
A
That's
what
we
need
to
distribute
great.
So
this
workshop
has
this
just
an
introduction
to
the
true
markets
in
general,
then
I
will
have
a
section
talking
about
falcon
saturn
and
we'll
hear
from
one
project
and
then
we'll
have
an
open
discussion,
especially
around
crypto,
econ
and
various
different
approaches.
A
So
there's
a
lot
of
content
already
about
kind
of
like
why
refuel
markets
in
general
there's
a
set
of
videos
that
you
can
look
at
on
youtube,
there's
a
falcon
valio,
explainer
and
there's
awesome
content
being
produced
through
the
referral
market
builder
demo
days.
You
can
go,
look
at
those
and
see
all
kinds
of
progress
across
across
the
network.
There's
also
a
bunch
of
really
good
youtube
videos,
in
short,
the
the
kind
of
like
the
crisp
view
of
this
is
that
the
internet
is
a
grapevine.
A
Content
must
flow
through
these
very
crowded
branches
to
the
leaves,
and
you
want
to
tune.
You
want
to
create
open-ended
protocols
that
enables
any
participant
around
the
internet
to
spot
an
opportunity
to
deliver
content,
leverage,
content
addressing
and
provide
the
best
possible
delivery
of
that
content.
A
A
But
you
can
do
way
better
if
you
make
it
open,
open
and
permissionless
right
now,
all
kinds
of
groups
that
are
in
a
great
position
to
add
a
content
cache
very
close
to
users
and
where
the
user
demand
is
can't
do
it,
because
there
are
all
these
private
networks
that
are
centralized
and
prevent
their
participation
so
think
of
being
an
isp
or
even
like
a
like
a
cell
tower
self-hour
operator,
or
something
like
that,
and
you
can't
there's
all
kinds
of
traffic
flowing
through
that
network.
A
Oftentimes
the
same
kind
of
content,
the
same
movies
and
so
on.
You
could
totally
deploy
a
content,
cache
right
there
to
provide
content
to
the
users,
but
you
really
need
the
content
flow
to
move
into
a
content
addressed
model
of
the
world
and
you
need
to
be
able
to
join
a
network
that
lets
you
kind
of
participate
in
that
way.
A
So
that's
that's
why
the
falcon
retro
market
can
greatly
speed
up
the
cdns
of
the
world,
the
the
way
sort
of
like
map
search
providers
and
retrieval
providers
to
the
traditional
cloud
so
kind
of
think
of
this
sort
of
layout,
where
you
have
storage
providers
in
the
center
and
then
you
have
virtual
providers
sort
of
as
a
cloud
around
them
which,
as
you'll,
hear
from
saturn
like
maps
out
maps
well
to
the
to
why
the
project
name
is
out
and
so
on.
A
I
should
think
of
storage
providers
as
coming
in
many
different
scales
from
just
a
few
set
of
drives
and
so
on
in
the
home,
to
kind
of,
like
small
racks,
to
very
large
operations
with
you
know,
full
data
centers-
and
we
already
see
this
across
the
network.
There's
massive
scale
search
providers
with
and
then
there's
also
a
ton
of
smaller
scale
storage
providers,
so
the
data
is
finding
its
way
to
really
great
homes
across
across
this
network.
Now
we
need
to
build
out
the
virtual
provider
network.
A
This
will
tune
better
for
smaller
devices
and
smaller
deployments
that
are
much
more
evenly
distributed
around
the
world
close
to
where
people
want
to
retrieve
the
data
I
would
add
on.
This
is
not
on
the
explainer.
I
need
to
add
a
an
image
for
a
mobile
device.
I
think
that
mobile
networks
can
can
also
be
good
for
retrieval,
but
it's
actually
a
much
harder
problem,
so
I
would
not
encourage
groups
to
kind
of
start
with
that,
because
getting
retrieval,
cdns
and
so
on
and
mobile
is
way
harder.
A
But
if
you
sort
of
target
like
home
machines
and
maybe
one
one
single
machine
in
a
in
a
data
center
or
potentially
like
a
full
rack-
that's
about
the
slice
of
water
fuel
providers
that
I
think
will
be
will
be
really
successful.
A
Those
are
great
spots
for
one
of
these,
like
larger
caches
that
you
know,
can
do
a
lot
of
has
a
lot
of
bandwidth
capabilities
or
isps
so
any
kind
of
like
either
hardwired
isp
or
a
wireless
asp.
Those
are
great
places
to
deploy.
These
kind
of
you
know
one
single
rack
of
drives,
for
example,
just
for
clarity,
the
retail
market
kind
of
operates
off-chain,
so
you
can
use
the
the
blockchain
to
set
up
a
bunch
of
the
transactions
and
to
lean
on
payment
channels
or
identities,
for
providers
and
so
on.
A
But
all
of
this
operation
has
to
work
off
chain
and
you
have
to
get
delivery
to
be
extremely
fast
and
with
very
low
onboarding
costs.
So
you
can't
create
a
structure
where,
like
the
end,
users
and
consumers
of
some
content
have
to
have
some
like
onboarding
hurdles
that
prevent
them
from
accessing
the
content.
So
I
think
those
kinds
of
approaches
will
be
harder
to
make
work
very
possible,
but
but
harder
to
make.
A
May
work,
and
one
of
the
images
that
I
want
to
leave.
Everyone
with
is
like
really
try
to
solve
the
kind
of
speed
of
light
problem.
So
a
key
component
in
all
approaches
has
to
be
how
to
reason
about
how
the
requests
are
coming
into
the
various
machines
and
route.
Those
requests
to
machines
that
are
close
to
where
the
users
are
and
try
to
push
content,
as
close
as
you
can
to
where
it's
being
requested
a
lot
so
over
time.
A
Over
the
long
periods
of
time,
you
can
gain
predictive
ability
over
what
kind
of
content
gets
requested
where
and
so
on,
or
have
an
adaptive
network
that
over
time
moves
the
content
to
the
right
spots.
This
is
how
modern
cdns
work,
but
decentralized
networks
in
pure
peer
networks
and
block
networks,
especially
don't
pay
close
attention
to
the
hops
going
around
and
the
where
people
are
kind
of
requesting
things
from
and
so
on,
and
this
is
not
something
that
retro
markets
can
there's,
not
a
luxury
we
can
take.
A
Here
we
have
to
be
very,
very
precise,
with
which
machines
every
single
request
is
going
to
go
to
to
try
and
share
off
tens
of
milliseconds
at
a
time.
That's
how
you
get
to
extremely
fast
retrieval.
So
if
your
approach
as
you're
thinking
about
it,
has
any
kind
of
open-endedness
where
a
user
request
might
accidentally
hop
across
an
ocean,
that's
not
a
good
approach.
Redesign
like
really.
A
You
want
to
be
extremely
careful
to
make
sure
that
users
are
able
to
kind
of
plot
their
their
virtual
pathway
to
nodes
that
are
going
to
ideally
have
the
content,
ideally
in
the
data
at
their
closest
data
center
possible.
Now,
of
course,
that
won't
always
be
the
case.
A
You
want
to
kind
of
like
fall
back
gracefully,
but
you
want
to
kind
of
exploit
the
the
landscape
here
to
minimize
the
speed
of
light,
distance
and
minimize
any
kind
of
round
trip,
and
so
on
also
important
here,
any
kind
of
protocol
that
requires
full
round
trips
to
operate,
also
problematic.
You
want
to
get
to
a
spot
where,
as
the
round
trip
is
going
as
the
request
is
going
out,
you
can
get
the
data
back
and
be
very,
very
careful
about
round
trips.
A
You
can
afford
round
trips
when,
when
you're
going
close,
so,
for
example,
you
can
afford
round
trips
when
a
client
is
talking
to
another
node,
you
know
five
or
ten
milliseconds
away
like
that's,
not
a
big
problem,
a
round
trip
there
is
okay,
but
if
you're
dealing
with
round
trips
with
hundreds
milliseconds,
your
host
do
not
do
that.
A
A
So
you
know
things
lean
on
things
like
noise
and
so
on
and
lean
on
routing,
end
user
clients
to
nodes
very
close
to
them
and
then,
from
there
proxy
over
the
request
over
an
already
set
up
channel
to
be
able
to
kind
of
minimize
the
minimize
the
setup
latency.
A
So
modern
cdns,
like
netflix
and
others,
for
example.
Have
these
like
persistent
connections
where
your
client,
when
it
wakes
up
immediately,
connects
to
to
a
set
of
nodes
that
are
very
close
to
them?
Does
the
connection
set
up
and
from
there
we
use
that
channel
to
not
have
any
kind
of
additional
setup
times
for
any
further
requests.
A
You
want
to
exploit
these
kinds
of
techniques
to
avoid
putting
users
in
a
position
where,
like
they're
routing
around
and
talking
to
dozens
or
hundreds
of
machines
and
each
time
setting
up
new
crypto
handshakes
like
that,
will
totally
kill
any
kind
of
ability
speed.
So
here
we
want
to
kind
of
shave
the
latency
down
as
much
as
possible
and
get
to
again
tens
of
milliseconds
in
the
best
case,
hundreds
of
million.
A
You
know
100
or
200,
maybe
300
milliseconds,
in
the
kind
of
like
average
bat
case
and,
of
course,
in
the
worst
case,
when
the
content
is
not
hot
or
something
like
that,
then
you
get
into
one
or
you
know
one
to
five
second
retrieval
or
something
like
that,
but
that
should
be
very
much
the
exception
and
for
very
very
few
few
cases.
A
A
Speed
of
light
totally
ignore
reuse
of
connections
totally
ignore
crypto
handshakes,
and
they
just
sort
of
assume
that
you
have
the
capability
of
connecting
to
a
ton
of
nodes
in
the
network
and
that
you're,
you
know
for
privacy
or
security
reasons.
You
you're
fine
with
that,
in
reality,
the
modern
web
and
all
video
viewing
will
not
move
over
to
centralized
networks.
A
If
that's
how
we
operate,
we
have
to
totally
shift
the
behavior
here
so
that
end
user
clients
spend
as
little
time
in
connection
setup
and
then
from
there
like
can
make
extremely
fast
requests
so
again,
lindsey
and
the
kind
of
like
best
case
wanted.
We
need
to
make
one
to
ten
millisecond,
retrievals,
possible
and
kind
of
get
to
them,
get
the
norm
to
be
kind
of
50
to
500
milliseconds,
also
something
that
complicates
this.
The
number
of
objects
is
very
large
when
you
think
about
hundreds
of
petabytes
of
content.
A
Those
hundreds
of
petabytes
will
have
tons
of
little
pieces
of
data,
a
lot
of
files,
little
data
structures
and
so
on.
So
you
have
a
lot
of
objects
that
people
want
random
access
to.
So
you
need
to
be
able
to
do
routing
towards
these
objects
quickly.
You
can
exploit
all
kinds
of
things
there
to
be
able
to
amortize
the
lookup
costs
or
do
predictive
looking
at
lookups
and
so
on.
It's
a
whole
other
problem
for
another
time.
A
At
the
end
of
the
day
we
again
kind
of
want
to
have,
and
we
you
need
to
have
a
network
that
has
hundreds
of
thousands
to
tens
of
millions
of
nodes
like
that's
kind
of
the
where
we're
headed.
I
won't
go
into
detail
here.
This
is
kind
of
well
covered
in
a
bunch
of
the
other
videos
and
so
on.
A
There's
a
lot
of
great
tools
and
tech
at
our
disposal,
there's
all
kind
of
components
that
a
lot
of
us
have
been
building
for
five
to
eight
years
now
and
there's
you
know
this
broader
approach
that
we're
taking
of
you
know
really
tuning
for
typical
cdn
product
experience
and
aiming
to
build
a
profitable
and
easy-to-run
retrieval
provider
software
and
aiming
to
kind
of
wire
those
components
into
really
good
products
cool.
A
A
Thank
you
thanks
for
the
extra
time,
so
this
is
kind
of
a
map
of
the
falcon
ecosystem
right
now,
sort
of
made
this
relatively
recently
or
sorry.
A
The
filecoin
broader
kind
of
data
system,
the
ecosystem,
is
larger
and
includes
a
lot
of
other
participants
now
on
this
kind
of
like
a
zoomed
in
view
to
the
data
flow,
so
think
of
kind
of
the
data,
starting
with
the
storage
clients,
that
data
gets
is
usually
kind
of
associated
with
some
use
case
or
some
pattern
and
the
the
end
users
of
that
use
case
or
the
developers
of
that
use
case.
A
We'll
want
to
tune
the
access
pattern
and
we'll
want
to
tune
the
the
onboarding
flow
and
so
on
to
that
particular
type
of
data
and
so
on.
So
that
means
that
these
sort
of
data
on-ramps
become
extremely
useful.
These
are
services
like
nft
storage,
web
studio,
storage,
estuary,
fission,
fleeq
and
so
on.
You
can
think
of
those
as
use
case-specific
applications
that
tune
the
interface
and
tune
the
parameter
space
to
fit
the
data
access
patterns
for
that
particular
type
of
data.
So
you
can
think
of
like
the
interfaces.
A
The
platform
provides
as
lower
level
infrastructure
or
lower
level
plumbing,
and
you
can
think
of
the
developer
on
ramps
and
so
on,
as
a
specific
kind
of
like
use
case.
Porcelain
tuned
to
to
a
particular
data
flow,
so
that's
working
super
well,
and
each
of
these
on-ramps
is
scaling
significantly
and
a
ton
of
usages
coming
in
through
those
systems
we'll
see
it
in
a
moment
that
and
kind
of,
like
the
data
onboarding
pipeline
is,
is
very
large
already
and
growing.
A
A
Three
space
is
kind
of
like
somewhere
between
two
and
five
petabytes,
so
like
we're,
adding
way
more
content
to
falcoin
in
one
week
than
the
rest
of
the
western
world,
three
world
has
in
total,
and
that
means
there's
all
kinds
of
users
coming
from
what
to
and
from
other
more
traditional
approaches
that
are
starting
to
use,
use
filecoin.
A
Now
that
data
today
is
not
retrievable
quickly,
it's
pretty
slow
and
spotty,
and
so
on.
As
we
saw
in
the
last
talk,
one
key
piece
that
was
missing
was
ability
to
do
indexing,
so
sps
were
not
advertising
all
their
content
into
the
iphone
chd,
because
that
would
be
horrendously
slow
and
and
a
huge
burden
and
a
new
network
indexer
has
been
deployed
in
the
last
few
weeks
and
that
indexing
service
now
can
index
all
the
content
from
the
various
source
providers.
A
So
just
right
now,
something
that
wasn't
possible
a
few
weeks
ago
is
that
now
material
networks
can
lean
on
that
indexer
to
find
all
the
any
and
all
of
the
content
stored
on
filecoin,
so
great
opportunity
to
then
create
these
kind
of
retrieval
prior
networks
that
can
take
a
user.
Retrieval
request.
A
Look
up
the
in
the
index
quickly
kind
of
like
an
of
one
access
pattern
where
these,
where
to
go,
get
it
go,
get
the
content
and
now
provided
to
the
user,
so
we're
now
kind
of.
Finally,
in
that
in
that
spot,
and
so
the
devon
ramps
are
working
really
well
now
the
what's
not
working
well
yet
is
the
ritual
provider
networks.
We
need
to
greatly
improve
that
the
one
that's
most
used
today.
A
You
can
think
of
the
ipfs
gateway
and
the
ipf
spinning
services
as
the
kind
of
like
best
retrieval
networks
today,
and
we
want
to
do
way
better
than
those
so
we'll
zoom
into
this
model
a
little
bit
more
later
for
context.
The
number
of
sps
globally
scaling,
there's
growth
in
in
a
ton
of
places
in
the
ecosystem,
the
this
kind
of
like
a
chart
of
like
falcon
plus,
which
has
been
extremely
useful
in
scaling
up
data
usage
and
so
on.
A
You
can
see
the
graph
in
the
bottom
right
of
the
amount
of
data
being
committed
daily
to
to
the
network.
The
y-axis
is
in
terabytes
so
like
the
top
is
like
a
thousand
1.2
petabytes
and
and
that
you
know
that
continuing
all
of
that
data
needs
fast
retrieval
so
as
I'll
cover
some
use
cases
in
a
moment
and
all
of
those
are
hurting
for
faster
tool.
A
So
nfc
storage,
you
know,
has
on
the
order
of
like
it's
about
to
cross
60
million
nfts,
which
is
huge.
It's
a
ton
of
individual
objects.
People
want
really
fast
access
to
those
nfts
and
good
news.
They're
not
that
large,
when
you
think
of
all
of
those
put
together,
it's
not
actually
that
big
of
a
data
set
so
that
could
be
replicated
in
a
ton
of
places
and
observed
very
quickly
to
to
end
users.
A
Webinar
storage
is
a
similar
kind
of
developer.
On-Ramp.
That
also
has
a
ton
of
data
flow
in
is
growing
at
a
similar
scale
to
nfc
storage
and
all
of
those
objects
need
random
access.
A
You
can
think
of,
like
other
large-scale
clients
and
and
onboarding
funnel
there's
a
ton
of
different
groups
that
are
in
in
some
stage
of
a
trial
of
falcoin,
like
some
users
are
early
they're
finding
out
about
it,
and
so
on.
Others
are
kind
of
running
proofs
of
concepts
in
the
moment
and
in
tons
of
in
all
those
conversations,
retrieval
becomes
a
big
point,
and
so
there's
a
ton
of
providers
that
want
to
make
sure
they
can
retrieve
their
data
and
so
on.
The
access
patterns
will
vary.
A
So
some
of
these
are
large
data
users
that,
when
they
retrieve
something
they'll
want
terabytes
like
they
have
their
own
local
operations.
They're
using
these
networks
for
archival
or
long-term
storage
or
resilient
storage,
and
when
they
retrieve
something
they
might
want,
full
terabyte
drives
to
be
sent
to
them
and
so
on,
though,
that's
not
the
norm.
The
norm
is
that
people
do
want
kind
of
random
access
to
their
objects.
A
So
again,
nfcs
are
growing.
A
ton
metaverse
and
games
is
a
big
one.
These
are
very
much
right.
Limited
by
by
retrieval
mona
is
a
great
example,
they're,
a
platform
that
enables
anybody
to
make
metaverse
rooms
and
whole
spaces
and
then
sell
them
as
nfts
and
then
wire
them
up
together.
So
you
can
explore,
go
into
one
room
and
jump
into
another
and
so
on.
Each
one
of
these
objects
is
gigabytes
in
size.
A
So
when
you,
especially
the
higher
quality
ones,
that
sell
more
because
they're
like
much
more
beautiful,
are
way
bigger
so
like
the
textures
are
pretty
big
and
so
on.
Plus
then,
all
kinds
of
techniques
that
you
might
use
to
kind
of
do
faster
loading
in
the
client
side
blow
up
the
data
size.
So
if
you
want
to
do
like
careful
texture,
resizing
and
so
on,
you
end
up
with
like
a
larger
overall
overall
size
of
a
of
an
asset.
A
So
this
needs
really
really
fast
retrieval
for
the
end
users,
especially
if
you
think
of
like
exploring
a
space
that
has
a
lot
of
links
to
many
other
rooms
and
you
have
to
sort
of
pre-load
the
other
spaces
so
that
when
you
walk
around
a
corner-
and
you
enter
another
space
that
other
space
has
now
been
loaded
and
can
be
stitched
together
cleanly
right
now,
if
you
hop
into
a
mono
room,
you're
hit
with
a
progress
bar
where,
like
the
majority
of
the
weight,
there
is
loading
and
just
downloading
the
thing.
A
So
that's
kind
of
like
a
very
concrete
problem
itself
and
there's
going
to
be
a
bunch
of
other
new
kinds
of
nfts
emerging.
So
I
think
video
nfcs
are
around
the
corner.
They
also
need
really
fast.
Retrieval.
Videocoin
now
called,
I
think,
vivid
and
lifebear
are
going
deeply
into
this
they're
those
tools
just
got
deployed
in
the
last
quarter
quarter
or
two
and
they'll
follow
a
pretty
fast
growth
rate.
A
I
would
expect
that
within
two
to
three
quarters
we
have
millions
of
those
and
they'll
need
really
really
fast,
retrieval
the
and
then
there'll
be
other
kinds
of
nfcs
that
emerge
too.
So,
when
you
think
of
like
spaces
like
these
think
of
being
able
to
create
full
experiences
within
those
within
those
environments,
it's
not
just
a
space.
You
can
like
program
the
space
to
have
a
game
or
have
some
other
kind
of
artistic
experience,
and
that
kind
of
stuff
is
probably
maybe
three
to
four
quarters
out.
A
So
I
would
expect
like
next
year
to
see
a
lot
of
kind
of
like
the
combination
of
3d
spaces
on
art,
where
you
have
like
full
experiences
being
developed
and
sold
as
nfts,
and
those
will
need
super
fast
retrieval.
So
a
ton
of
a
ton
of
opportunity,
just
in
general,
media
and
so
on.
Of
course,
like
video
and
audio,
are
huge.
A
The
vast
majority
of
the
energy
traffic
on
the
network
is
video
on
the
broader
internet
I
mean,
and
so
being
able
to
tune
the
retro
market
to
serve
video
extremely
well
and
better
than
other
networks
faster
than
other
networks,
and
so
on
is
like
a
really
really
great
direction.
So
I
sort
of
expect
some
of
the
retro
market
networks
to
evolve
into
a
spot
where
they
can
deliver
this
video
extremely
quickly.
A
There
are
other
uses
which
have
a
lot
less
data,
but
require
a
lot
more
random
access,
so
things
like
individual,
you
know
personal
file,
storage
and
so
on.
There
are
a
lot
of
these
systems
already,
and
there
are
you
know
some
set
of
source
virtual
networks
are
gonna
tune
for
these.
A
These
the
benefit
of
these
ones
is
that
it's
not
that
much
data,
the
files
are
really
small
and
you
can
cache
them
everywhere,
but
it
just
sort
of
tunes
your
approach
to
deciding
where
to
put
stuff
and
and
so
on,
because
here,
like
you,
have
like
a
very
long
tail
of
files
that
you
access
very
rarely.
But
when
a
user
wants
to
access
them,
they
want
them
very
quickly.
A
But
it's
you
know
very
long
tail
of
files
and
it's
very
hard
to
predict
exactly
which
files
will
be
used
and
then
there's
a
whole
bunch
of
other
really
cool
data
sets
that
people
are
putting
on
falcon
already
large
scale
data
archives.
You
know,
including
things
like
the
internet
archive
where
you
know
the
interarchive
itself
is
like
doing
a
pilot
of
putting
you
know
large
chunk
of
data
into
the
falco
network,
but
imagine
being
able
to
we
we
can
actually
host
the
entire
archive
many
times
over.
A
A
A
Those
are
prime
candidates
for
for
retail
markets,
because,
especially
something
like
brave
with
bat,
you
can
hook
up
the
brave
the
user's
bat
wallet
to
pay
for
the
retrieval,
the
content
directly
and
like
that.
That
would
actually
close
the
loop
and
do
kind
of
like
the
old.
A
You
know
even
1960s
vision
of
hypertext,
where,
like
the
actual
user,
is
paying
for
the
tool
of
the
data
as
they're
viewing
it,
and
that
would
be
pretty
cool,
but
you
know
again
anything
that
sort
of
like
requires
user
onboarding
on
a
per
page
basis,
won't
work.
This
has
to
be
completely
automated,
underneath
the
hood
from
the
users
like
that
wallet
automatic
swaps
with
defy,
and
then
you
know
paying
for
paying
for
their
tool.
A
Yeah.
Here's
like
a
view
into
the
network,
indexer
project.
I
think
this
blog
post
was
last
week,
so
you
can
go
check
it
out
and
that
kind
of
like
really
closes
this
really
important
loop
here
of
being
able
to
kind
of
index
all
of
the
content
to
be
able
to
retrieve
it
quickly.
A
The
last
thing
I'll
mention
is
fvm,
so
fvms
around
the
corner.
This
will
add
smart
contract
computation
to
the
falcon
network,
which
will
bring
a
ton
of
smart
contracts
into
it
and
so
on.
All
of
those
will
need
a
lot
of
those
need
access
to
data
and
that
will
also
create
the
opportunity
for
computer
data,
which
is
a
whole
other
set
of
things
which
is
you're
able
to
kind
of
orchestrate
and
organize
large
scale,
computational
pipelines
around
the
data
on
falcon
so
think
of
large
data.
A
Having
gravity
you
bring
the
data
onto
falcoin,
think
of
large
scale,
data
sets
and
so
on.
Then
you
want
to
run
computation
over
that
data,
so
you
want
to
ship
the
computation
to
wherever
that
data
is
that's
less
of
a
use
case
for
retrieval,
because
in
that
case,
you're
trying
to
move
the
computation
to
to
that
data.
But
that
is
likely
to
generate
a
bunch
of
results
that
that
are
way
smaller
than
the
original
data
that
do
need
very
fast
retrieval.
A
So
that
will
likely
generate
all
the
computer
data
stuff
will
likely
generate
all
kinds
of
intermediary
artifacts
that
need
to
be
retrieved
quickly,
great,
so
so
so
that's
kind
of
like
an
update
view
across
the
whole
network.
So,
let's
now
kind
of
talk
a
bit
about
the
ritual
landscape
in
2022.
So
going
back
to
this
this
model,
we
need
to
boot
up
these
retrieval
market
networks
that
try
to
localize
regions
to
fight
the
speed
of
light.
A
A
We
can
sort
of
start
up
with
the
region
size
of
the
cloud
centralized
cloud
has
built,
which
is
kind
of
like
somewhere
in
the
order
of
five
to
20
regions
in
the
world
and
then
subdivide
from
there
as
we
as
we
need,
and
we
kind
of
want
to
lean
into
leveraging
the
fact
that
there's
two
basic
problems.
One
problem
is:
how
do
you
deal
with
the
massive
amount
of
requests
that
are
going
to
come
from
retrieval
clients?
A
So
how
do
you
deal
with
the
tons
of
connections
and
initial
connection
setup,
which
introduces
latency
and
so
on
and
deal
with
that
level
of
bandwidth
and
provide
good
user
experience,
especially
to
phones
that
you
know
you?
You
want
to
have
really
really
fast
access
and
you
don't
want
the
phone
to
do
a
lot
of
stuff
and
then
on
the
other
side.
How
do
you
do
the
routing
of
the
request
to
find
where
it
is
and
fetch
the
data
and
ideally
pre-load
it
into
a
region
so
think
of
it
like
solving
two
problems?
A
One
problem
is:
how
do
you
handle
the
bandwidth
and
the
large
scale
of
the
connections,
and
then
how
do
you
either
preload
or
go,
find
and
cache
the
data
quickly
to
kind
of
serve
it
from
from
within
that
region?
A
A
The
broader
picture,
though,
is
like
there
are
many
different
use
cases
and
several
different
economic
models,
and
because
of
these
use
cases
and
economic
models,
this
will
necessarily
mean
that
it
will
be
different,
retrieval
networks
with
taking
different
approaches,
and
that
means
like
different
software
distributions,
different
content
and
all
kind
of
stuff.
A
We
want
to
create
an
environment
that
enables
those
different
virtual
networks
to
boot,
up,
run
and
so
on
and
interrupt
as
much
as
possible,
and
if
we
can
lean
on
kind
of
payment
channels
and
so
on
and
a
very
common
interface
on
how
to
retrieve
something
that
could
actually
even
allow
interop
between
between
these
networks
that
I
sort
of
see
that,
as
like
a
you
know,
maybe
late
this
year
or
probably
next
year,
type
of
problem.
A
Let's
first
kind
of
work
on
booting
up
these
these
networks
to
get
to
serving
the
kind
of
most
common
common
one
to
ten
petabytes
of
content
around
the
world
around
the
network
and
preload
that
into
everywhere
around
the
world,
to
then
be
able
to
to
serve
that
extremely
quickly
and
create
the
networks
that
enable
onboarding
of
a
lot
of
people
in
in
those
regions
to
be
able
to
provide
nodes
to
be
able
to
serve
in
those
those
areas
cool.
I
will
stop
here
any
questions,
thoughts.
B
A
So
I
think
right
now,
they're
they're
working
on
a
pilot
to
store
the
archive
ritual
right
now,
in
fact,
is
slower
than
the
internet
archives.
So
once
we
have
faster
retrieval,
then
we
can
do
that
but
build
a
tech.
First
then
do
it.
C
A
Yeah,
so
the
reason
why
mobile
is
harder
is
that
these
are
much
smaller
devices
they're.
The
type
of
connection
changes
a
lot
so,
for
example,
the
profile
of
a
mobile
phone,
it's
very
different
when
it's
at
home
plugged
in
a
wi-fi
network
plugged
into
the
wall.
So
that
means
like
it
has
reliable
power
and
reliable
internet
and
probably
not
a
bunch
of
use
from
the
from
the
human,
because
the
humans
fall
asleep,
so
that
so
then,
at
that
point
like
the
mobile
phone
is
good.
A
But
that
usually
is
at
night
when
other
people
around
are
probably
not
using
the
internet
very
much,
and
so
other
people
elsewhere
in
the
world
might
be
using
it.
A
You
also
have
like,
if
you
now
deal
with
mobile
phones
when
they're
in
your
pocket
and
you're
moving
around
and
so
on
now
you're
dealing
with
like
connection
changing
because
you're
going
across
wi-fi
networks
to
mobile
networks,
mobile
internet
network
networks
and
so
on,
which
means
like
a
bunch
of
like
connection,
tear
down
and
reset
up
and
the
phone
itself,
depending
on
the
on
the
carrier,
may
or
may
not
be
getting
a
different
ip
which
for
overlay
networks
and
peer
to
peer,
like
that's
really
hard.
A
So
you
deal
with
like
so
you
have
like
these
massive
networks
with
like
nodes
with
tons
of
churn,
so
they're
coming
in
and
out
all
the
time
they're,
not
that
big.
You
can't
put
that
much
storage
in
them
relative
to
a
home
computer
or
something
like
that
and
the
internet
connect
and
the
bandwidth
is
not
as
good
as
as
some
home
bandwidth.
It's
also
very
expensive,
like
if
people
sort
of
like
screw
up
their
setup,
they
might
be
like
charged
tons
of
money
for
for
for
serving
this.
A
So
it's
just
like
a
much
harder
product
landscape,
and
so
I
would
say
like
get
it
working
really
well,
first,
with
home,
desktop
type
computers
and
with
like
data
center
small,
like
single
server
machines,
you
know
when
you
put
like
a
ton
of
drives
in
like
one
one
server
or
you
know
a
few
and
like
that's
kind
of
like
the
best
candidate
target
for
qr
markets.
Today,
that's
easiest
get
that
working
then
from
there.
You
can
then
solve
the
hard
problems
of
the
mobile
network.
A
That's
my
guess,
though
it
could
be
that
you
know
someone
like
focuses
on
those
problems
really
well
finds
a
narrow
use
case.
Maybe
it
only
works
when,
when
you
plug
it
in,
maybe
it
only
works.
When
you
know
you
go
to
sleep,
you
open
an
app
plug
it
in
and
like
leave
and
and
then
maybe
it
works.
But
like
again,
I
think
I
think
you'll
get
a
you'll.
You
get
more
storage
and
better
internet
access
from
these
other
kind
of
like
dedicated
machines.
D
Just
sort
of
adding
on
to
that,
because
you
also
mentioned
netflix,
I
was
wondering
if
you
could
add
more
color
to
how
netflix
and
or
youtube
manages
mobile
cdn.
If
you
had
any
insight
there,
yeah.
A
A
A
So
the
kind
of
like
perspective
is
that
you
know
this
is
sort
of
like
a
map
of
the
internet,
and
so
you
have
this
great
grapevine
behavior.
It's
not
very
easy
to
look
at
it
here.
Maybe
maybe
these
diagrams
are
better.
Like
you
know,
this
is
like
a
map
of
the
internet.
You
zoom
in
this
is
you
know,
sort
of
like
how
what
the
grave
line
looks
like
there's
another
way
to
look
at
it,
so
you,
so
you
really
want.
A
A
So
you,
you
kind
of
want
to
be
adaptive,
and
so
then
the
way
that
netflix
does
it
is
that
you
know
they're
they're
kind
of
like
in
the
content
provider
spot
over
here
and
then
their
users
are
consumers
in
homes
or
offices,
and
that
path
is
really
expensive
for
them
and
so
what
they
ended
up
doing
was
they
shipped
for
a
long
time
they
shipped
racks
of
drives
to
isps
I'm
not
sure
if
they're
still
doing
this,
but
they
pre-loaded
these
machines.
A
This
part
of
the
reason
part
of
the
like
reason,
I
knew
that
things
like
fpfs
with
content
addressing
and
things
like
falcon
were
gonna
were
gonna.
Work
is
because
companies
like
netflix
do
this.
They
pre
they
they
would
identify,
which
movies
were
the
most
likely
to
be
watched
in
a
specific
area,
and
they
would
then
preload
them
into
drives
and
ship
that
and
then
do
kind
of
in
the
background
updates
of
those
machines.
A
A
So
that
becomes
like
a
super
fast
hop
right,
like
that
becomes
kind
of
four
to
twenty
millisecond
round
trip
and,
like
that's,
phenomenal
and
and
the
there's
like
a
set
of
great
talks
from
the
netflix
team,
talking
about
kind
of
various
different
things,
they've
done
to
their
their
architecture,
one
of
them.
I
know
that
for
even
for
like
their
apis,
they
do
this
like
they
don't
want
individual
connections
set
up
every
time
and
so
even
for
their
apis.
A
They
kind
of
have
like
this
bastion
node
that
you,
your
user,
your
consumer
like
connects
to
and
then
from
there
they
send
requests
out
and
all
those
requests
get
proxied
from
that
node
out
somewhere
else,
and
that
node
is
like,
depending
on
their
service,
that
might
be
close
to
the
user
or
that
might
be
in
a
data
center
somewhere.
But
it
kind
of
gives
you
kind
of
this
much
more
stable.
A
Look
from
the
from
the
user
as
opposed
to
getting
you
know,
a
phone
to
kind
of
open
tons
of
connections
to
different
machines
and
so
on.
So
they
really
kind
of
took
an
overlay
network
approach
to
this,
where
they
kind
of
like
set
up
a
connection,
maintain
that
connection
and
and
push
as
much
traffic
as
they
can
through.
That
and
another
important
thing.
A
Netflix
very
much
separates
the
the
application
stuff
from
the
content,
so
the
movies
get
distributed
by
a
completely
separate
system
than
the
static
assets
of
the
page
so
like
the
static
assets
that
are
involved
in
the
in
the
web,
apps
or
the
movie
catalogs,
and
so
on.
All
those
little
files
get
distributed
by
a
different
system
than
the
then
the
movies
themselves,
because
it's
just
you
end
up
with
like
different
different
requirements,
and
so
you
want
software
that,
like
is
just
tuned
for
the
specific
use
case.
E
Yep
is
there
anything
working
right
now
or
anyone
working
right
now
on
like
mobile
apps,
that
you
just
live
online
while
you
sleep,
which
would
be
like
awesome.
A
Yeah,
I
do
think
that
this
kind
of
thing
exists.
I
haven't
done
a
review
of
the
app
stores
to
find
them,
but
I
think
somebody
has
done
this.
I
don't
think
we've
been
the
only
because
I've
discussed
this
with
a
few
people.
I
don't
think
we've
been
the
only
people
to
kind
of
think
about
this,
so
I
would
bet
that
some
exist.
I
know
that
some
people
in
our
in
our
network
want
to
go.
Do
this,
it's
just
they
haven't.
F
So
we
talked
about
mobile,
like
cell
phones
being
a
very
like,
not
optimal,
candidate
for
for
delivering
data
and
stations
at
home
being
very
optimal.
What
about
laptops,
which
also
frequently
change
places
are
switched
on
and
off
is
yeah.
A
A
It's
just
like
a
home
desktop
or
like
a
a
nas
like
a
network
attached
storage
device
and
those,
I
think,
are
like
the
ideal
candidates
because
they
have,
they
usually
have
somewhere
between
half
a
terabyte
to
10
30
terabytes
in
in
the
larger
case,
and
then
these
other
you
know,
then
you
have
a
lot
of
people
who,
either
in
in
data
centers
or
at
home,
with
fast
connections,
deploy
like
little
server
racks,
either
kind
of
like
a
small
system
with
somewhere
between
10
to
100,
terabytes
or
some
larger
systems
with
you
know
full
petabyte.
A
So
there's
a
lot
of
people
out
in
the
world
that
have
you
know
that
type
of
storage
and
they're
not
using
all
of
it,
and
so
that's
like
perfect
candidates
to
kind
of
preload
tons
of
content
in
in
there
and
then
serve
it
from
there.
You
just
have
to
kind
of
route
the
request
to
them
well,
which
is
kind
of
why
the
two
layer
architectures,
I
think,
works
really.
Well.