►
From YouTube: Retrieval Enthusiasts Unconf Session with Mike Goelzer
Description
This is a space for anyone interested in retrieval markets to discuss ideas. Mike will give a brief overview of the teams who are currently working on retrieval, but anyone is welcome to come and share thoughts, ideas, or questions.
Join us for Filecoin Liftoff Week, an action-packed series of talks, workshops, and panels curated by the web3 community to celebrate the Filecoin mainnet launch and chart the network’s future. https://liftoff.filecoin.io/
Events take place all week, October 19-23, 2020.
For more information on Filecoin
- visit the project website: https://filecoin.io/
- or follow Filecoin on Twitter: https://twitter.com/Filecoin
Get Filecoin community news and announcements in your inbox, monthly: http://eepurl.com/gbfn1n (edited)
A
Hello
everyone,
so
my
name
is
mike
gelser
if
you
are
a
particularly
interested
in
falcoin,
retrieval
markets,
you're
in
you're
in
the
right
place.
If
you
want
to
get
involved
with
any
of
the
projects,
oops
you're
in
the
right
place,
we
had
a
discussion
on
tuesday
night
u.s
time
be
wednesday
morning:
china,
asia,
pacific
and
middle
middle
of
middle
of
the
night
on
tuesday
in
eu.
So
I
know
some
people
could
make
it
some
people
couldn't,
if
you
couldn't
make
it.
A
A
A
A
So
what
I
I
was
just
gonna
do
a
highly
abridged
version
of
what
I
said
on
tuesday
night
just
and
then
I
was
going
to
turn
it
over
to
peter
who
I
was
hoping
to
do
the
same
and
tamar
and
david
if
they're.
Here
I
was
thinking
we
just
like
two
minutes
each
or
something,
and
then
that
just
sets
the
everybody's
on
the
same
page.
A
So
really
briefly,
the
the
original
clone
coin
white
paper
had
these
two
concepts:
there's
a
storage
market
and
a
retrieval
market
and
the
storage
market
is
the
thing
that
was
launched
last
week.
That's
the
sort
of
the
crux
of
the
whole
idea.
The
the
the
retrieval
market,
though,
has
a
different
set
of
goals.
It's
intended
to
make
retrievals
fast,
low,
latency
it's
intended
to.
Let
anybody
become
a
retrieval
miner
without
necessarily
having
you
know,
six
gpus
and
and
an
entire
data
center,
or
something
like
that.
A
So
that's
kind
of
what
the
goal
of
the
retrieval
market
is.
It's
not
super
well
specified,
which
I
think
is
intentional,
and
so
we've
got
four
different
teams.
Who've
been
working
over
the
summer
on
projects
that
are
taking
this
basic
idea
and
sort
of
looking
at
from
different
angles.
Consensus
is
that's
up
here,
he's
gonna
present
on
the
architecture
for
the
retrieval
market.
I
can
just
briefly
describe
the
browser-based
retrieval
client
tomah
for
mile.
A
B
Hey
robert's
here
too,
and
so
I'm
going
to
do
compress,
10
minutes
into
about
two,
so
yeah,
so
we're
doing
a
secondary
market.
Retrieval
solution,
and
so
the
primary
retrieval
market
was
all
about
lotus
nodes
connecting
to
storage
miners,
and
so
we
wanted
to.
You
know
like
have
something:
that's
discoverable
in
network
caching,
low
latency
and
massively
scalable
as
well,
and
there's
incentivization
scheme
to
be
involved
in
the
discovery
of
content
and
retrieval
of
content.
B
And
so
we
came
up
with
a
massive
overall
architecture
where
you've
got
a
library
and
that
you
can
integrate
into
applications
and
it
helps
you
find
content.
And
so
you
have
a
pcid
and
a
cid
and
that'll
help.
That
will
be
what
you
need
to
find
any
content
in
the
system
and
the
idea
is
that
you
have
offers
and
an
offer
is
a
fixed
price
offer
for
a
certain
retrieving
content
and
they're
published
by
these
providers.
B
You
can
just
talk
to
one
entity
in
the
system
and
it
will
be
able
to
give
you
an
offer
immediately
on
content,
so
it
should
be
lower
latency,
and
so
the
idea
is
you've
got
these
gateways
in
the
system
and
you
might
have
payment
channels
set
up
with
a
couple
of
them
and
then
they
will
give
you
offers,
and
so
you
could
you.
Your
client
retr
asks
multiple
gateways
as
well.
B
If
a
gateway
doesn't
have
a
cid
offer
that
you
in
particular
had
need
for
your
cid,
then
it
will
use
a
dht,
and
so
we've
got
to
come
up
with
a
very
special
dht,
which
is
different
to
most
dhts
you've
ever
come
across,
because
it's
using
the
fact
that
you've
got
stable
nodes
that
are
long-lived
and
to
give
you
single,
hop
lookup.
So
it's
again
aimed
at
low
latency
but
full
coverage
of
all
the
pcid
range
and,
as
I
said,
there's
also
the
idea
of
in-network
caching.
B
A
Excellent
excellent
compression
ratio.
That
was
very
good.
Thank
you.
Thank
you
peter.
So
let
me
see
I
want
to
go
next
to
david
from
chainsafe
and
then
yeah.
Then
then,
let's
just
go
into
kind
of
an
open
discussion
so
david.
Can
you
kind
of
do
a
two-minute
overview
of
what
you
know
what
you
guys
have
built
and.
A
A
A
Oh
and,
of
course
this
is
the
repo
we
I
think
over
the
course
of
the
summer
we've
had
about
10
contributors,
there's
probably
five
active
people
right
now,
but
instead
of
showing
trying
to
show
you
a
demo
which
would
take
beyond
the
amount
of
time
I
have
here,
I
would
just
show
you:
this
is
a
screenshot.
The
idea
is
you.
This
is
running
in
your
browser.
You
can
type
you
you
can
put
a
cid
into
this
query
box,
search
for
it.
A
You'll
get
all
of
the
offers
from
other
instances
of
the
browser
extension.
So
all
of
the
browser
extensions
form
a
peer-to-peer
network.
You
can
see
at
the
time
this
shot
was
taken.
There
were
a
total
of
three
on
the
network,
it's
just
just
a
testing
network
and
then
you
can,
you
can
buy
a
cid
from
another
browser
extension
and
payment
channel
will
be
open
and
you
will
exchange
data
and
sign
vouchers
with
the
other
browser
accord.
A
So
if
you
wanna,
if
so,
here's
a
video
of
the
actual
demo,
it
walks
through
a
little
more
detail.
It
shows
you
what's
actually
happening
in
the
logs
so
that
that's
one
thing
to
check
out
and
also
I
just
remind
people
if
you,
if
you're
watching
this-
and
you
don't
have
any
previous
exposure
to
retrieval
markets
this.
This
is
the
place
you
want
to
go.
This
was
the
discussion
we
had
on
on
tuesday
night
so
that
that
was
all
oh.
I
want
to
say
one
one
brief
comment
about
this.
A
Just
I
think
this
diagram
might
make
it
a
little
clearer,
and
I
do
think
that
this
architecture
is
basically
the
same
as
what
tomah
is
using
he's,
not
targeting
the
browser
as
his
deployment
platform.
But
it's
the
same
he's
been
basically
the
same
concept,
so
I
think
this
is
applicable
also
to
his.
He
couldn't
be
here
tonight,
but
so,
like
I
said,
the
browser
retrieval
clients,
those
are
the
green
boxes,
they
form
their
own
peer-to-peer
network.
A
B
E
Yeah
yeah,
this
is
awesome.
I'm
super
excited
about
this.
I
was
also
really
hoping
that
there
would
be
some
browser-based
retrieval
market
stuff,
because
it's
just
where
I
imagine
ingesting
a
lot
of
information
and
playing
with
it.
I
was
curious
like
so,
if
it
doesn't,
if
it
doesn't
use
a
lotus,
node
or
like
a
client
that
is
directly
running
on
the
host
like
computer,
how
does
it
actually
tap
the
network
and
how
are
payments
handled
and
everything
like
that?
Is
it
like
a
similar
like
a
meta
mask
where
it's
tapping
out?
A
Yeah
I
mean
I
can
I
can
take
that
one.
So
if
I
just
sorry,
let
me
go
back
to
sharing
my
screen
and
then
I'll
share
a
window
that
isn't
just
slides.
So
here's
here's
what
the
the
browser
client
that
we've
built
so
far
looks
like,
and
I've
got
two
instances
of
it
running
one's
in
chrome,
one's
in
firefox,
the
chrome
one
has
a
cid
in
it
already.
So
it's
like.
A
Yeah
it's
stored
in
the
browser's
local
storage.
So
if
you
want
to
get
it
onto
your
desktop,
you
actually
have
to
click
a
second
download
button,
but
this
is
just
a
local
download
from
your
own
machine
cool,
but
but
the
way
the
transfer
itself
works
is
you
first
you
send
out
a
query
for
the
cid.
It
goes
to
every
node
in
the
network
right
now,
there's
only
one
other
node,
which
is
the
firefox.
A
A
Yeah
yeah
exactly
yeah
yeah.
I
think
the
biggest
difference
with
peter's
system
is
that
they're
trying
to
design
an
architecture
for
low
latency
retrievals,
whereas
as
you're
going
to
see
we,
we
could
talk
about
this
for
the
rest
of
the
for
the
next,
probably
two
to
three
minutes,
and
this
four
kilobyte
file
will
only
barely
have
finished
transferring
so
I
wouldn't
exactly
call
that
low
latency,
but
you
can
see
you
can
see
all
the
stuff
that's
really
going
on.
If
you
go
into
the
logs
on
the
two.
A
So
this
this
guy,
the
requester
of
data
sets
up
a
filecoin
payment
channel.
Where
does
he
do
that?
Yeah
create
payment
channel?
He
submits
a
message
onto
the
chain,
I'm
not
going
to
walk
you
through
the
whole
execution
phrase
unless
you
really
want,
but
but
it
goes,
the
two
of
them
do
the
dance
of
the
whole
exactly
the
same
way
that
one
lotus
node
would
retrieve
from
another.
E
A
Yeah
well
sort
of
okay,
so
you
have
to
put
in
a
wallet
and
obviously
you
have
to
put
in
the
wallets
private
key.
The
reason
is
all
of
your
all.
The
messages
you're
gonna
publish
on
chain
are
signed
locally
here
in
your
browser
and
then
they're
pushed
out
to
a
public
cloud
lotus,
which
is
just
like
it's
a
lotus
instance
where
you
can.
You
can
publish
messages
on
the
chain,
but
it
doesn't
have
your
wallet.
C
A
Yeah,
yes,
yeah,
so
this
isn't
trying
to
be
a
wallet,
and
I
think
it
might
work
better
with
something
like
metamask
or
supporting
like
ledger
hardware
wallets
or
but
this
is
just
the
design
right
now
to
make
it
work
at
a
fundamental
level
and
yeah
so
on.
On
this
instance,
the
chrome
one
that's
doing
the
download
the
wallet
is
the
one
the
payer
funds
are
coming
out
of
this
wallet
and
then
this
guy
also
has
his
own
wallet
and
his
own
private
key,
and
he
will
be
the
recipient
of
the
funds.
A
B
A
Yeah
they
use
lib
p2p's
web
rtc.
It's
a
it's
a
separate
package
that
you
can
install
within
jslib
p2p
and
it's
just
and
then
we
run
a
webrtc
signaling
server,
it's
like
the
rendezvous
point
and
when
they
start
up,
they
all
connect
to
the
signalling
server
it
informs
them
of
how
to
connect
to
each
other.
And
then,
after
that,
they're
just
it's
a
direct
connection
between
yeah.
B
A
B
I
fully
understand-
and
I
think
I
mean
this
is
so
we
thought
about
you-
know
using
an
rpc
mechanism.
You
know
in
the
boot
nodes
or
something
for
file
coin,
for
our
system
and
ended
up,
saying,
look
we'll
use
smart
contracts
so
and
whether
they
be
ethereum
on
ethereum,
main
net
or
on
file
coin
main
net
you
know,
will
depend
on
when
the
evm
is
up
and
running
on
file
coin
but
yeah.
B
I
I
and
the
reason
why
we
ended
up
not
going
with
our
pc
was
we
thought
that
the
it
meant
that
those
boot
notes
would
be
potentially
hammered
and
also
a
dose
attack
point
as
well,
whereas
if
it's
in
a
smart
contract,
then
you're
just
attacking
or
talking
to
your
local
node
with
its
local
copy
of
the
distributed
ledger?
So
that's
how
you
know
we
aim.
That's
that's
how
we
got
to
where
we
got
to
with
how
to
do
node
discovery,
yeah.
A
A
So
then,
if
the
webrtc
rendezvous
point
is
knocked
off
the
internet,
you
still
have
these
look
other
local
peers
and
you
can
use
them
to
get
other
to
find
to
discover
other
peers
on
the
on
the
peer-to-peer
network,
and
this
is
all
like
built
into
lib
p2p.
This
wasn't
invented
for
this
extension,
but
it
was
part
of
our
sort
of
robustness.
Goal
for
lupita
p
that
you
live.
Pdp
networks
should
be
very
resilient.
You
shouldn't
be
able
to
just
knock
them
offline.
Of
course
everything
has
you
know.
A
B
A
A
A
There
will
be
a
new
cid
for
it,
and
a
clever
miner
could
could
take
this
browser
code
and
basically
automate
it
to
always
check
check
for
that
and
then
be
the
the
lowest
cost
provider
of
whatever
software
package
update,
and
so
I
don't
think
we're
going
to
get
browser,
retrieval
miners
who
just
want
to
mine
every
possible
cid
and
provide
them.
It's
it's
not
realistic.
It's
not
possible
yeah.
B
F
Yeah,
so
they
definitely
we
we
do
have
you
know
and
sort
of
our
final
report
is
when
we
will
go
through
that
with
you
guys
some
thoughts
about
what
sort
of
ongoing
scaling
on
the
payment
system
is
going
to
be
needed,
because,
right
now
the
payments
on
the
retrieval
payment
channels
are
competing
against
the
storage
deals,
whereas
the
volume
of
the
retrieval
to
storage
deals
may
be.
You
know,
orders.
F
Orders
of
magnitude
higher
on
the
retrieval
side,
so
we
need
to
offload
this
with
some
layer.
Two
systems,
both
state
channels
and
roll
up
systems,
are
likely
going
to
be
needed.
One
thing
that
I
thought
would
be
interesting
to
share
we've.
Also,
given
you
guys
mike
finalized
version
of
a
document
that
this
table
comes
from.
This
is
a
table
that
I
did
some
analysis
early
on.
F
You
know
looking
at
what
sort
of
the
objectives
that
we're
trying
to
get
and
part
of
what
we
focused
on
part
of
the
reason
why
you
know
what
peter
put
together
with
the
ring
dht
and
the
two
structure
system
was
that
you
know
we
looked
at
what
you
know.
What
does
it
mean
to
actually
have
a
retrieval
system,
and
you
know
juan
has
showed
this
diagram
hold
on.
Let
me
actually
bring
up
the
diagram
that
he
shows
from
his
presentations
sometimes,
and
it's
this
one
right
and
it's
sort
of
a
nice
graphic.
F
It
shows
like
okay,
here's
storage
miners,
here's
retrieval
miners!
This
is
what
the
cost
and
the
capacity
look
like,
and
you
know
you
see
these
kind
of
dotted
lines
that
are
glacier.
S3
and
cloudfront
and
they're,
referring
referring
to
the
you,
know
most
prevalent
cloud,
distributed
storage
system
that
everybody
uses,
which
is
amazon
s3,
and
so
you
know
this,
we,
you
know
we
started
off
our
project.
We
didn't
kind
of
jump
into
coding.
As
you
guys
know,
we
were
doing
architecture
work,
and
you
know
this.
F
This
sort
of
was
the
the
my
best
you
know,
because
amazon
has
the
doc.
The
information
at
the
bottom
is
not
something
you
just
find
in
one
spot
for
amazon.
This
is
like
a
collection
of
two
or
three
dozen
documents,
pieced
together
and
other
things,
but
you
know
I
was
trying
to
put
together
an
idea
of
what
do.
F
What
do
those
kind
of
different
levels
of
storage
look
like
and
it
was
useful
I
thought
to
actually
show
like
cold,
warm
and
hot
and
actually
mike
right
before
our
session
yesterday,
the
guy
who
won
the
space
race,
challenge
right.
The
miner
who
won
that
was
actually
talking,
and
he
made
a
comment
about
this,
which
is
correct,
which
is
that
you
know
the
file
point
system
without
a
strong
retrieval
system
in
front
of
it
really
is
very
much
like
glacier.
It's
cold
storage
right.
So
what
you
need
to
do
to
actually
get
it.
F
It's
not
coldest
right.
The
reason
why
for
amazon,
glacier
archive
and
glacier
are
kind
of
like
this
coldest
storage
is
they're
actually
on
tape
and
tape
takes
many
minutes
inherently
to
unpack.
And,
of
course
you
know
filecoin,
you
know
it's
on
hard
drives,
so
you
can
retrieve
it.
You
know
in
seconds
right
it's
a
very
quick
retrieval
in
theory,
assuming
you
find
it
negotiate
the
payment
and
everything
else.
So
you
know
what
peter's
described
in
the
architecture.
That
really
does
you
know,
I
think
addresses
this
is
recognizing.
F
If
you
want
to
get
to
the
range
of
s3
and
cloudfront,
you
know
you
don't
have
the
time
necessarily
to
go
through
a
large
multi-hop
distributed
hash
table
of
the
type
of
codemlia
or
others
where
there's
very
kind
of
low
radix,
where
you
only
have
six
or
eight
neighbors,
and
you
may
go
through
many
many
hops
in
order
to
get
there.
You
need
to
both
discover
who
actually
has
the
file
that
the
cid
points
to,
and
then
you
need
to
connect
with
that
person,
which
in
condemnedly
is
actually
two
different
lookups.
F
F
It
has
kind
of
a
key
idea
in
there,
which
is
you
know,
because
when
providers
are
looking
at
where
they
want
to
be
in
network
they're,
finding
hey,
I
see
parts
of
the
ring
dht
right
now
that
don't
have
a
lot
of
providers
covering
it.
So
I'm
going
to
insert
myself
into
this
part
of
the
network
where
there's
kind
of
low
density
and,
if
you
add
in
looking
at
latencies,
you
can
also
say:
oh
there's,
not
a
good
provider
in
this
part
of
the
network
in
australia.
There's.
F
But
if
I
sit
myself
in
australia
and
there's
a
good,
you
know
retrieval
flow
coming
through
I'll
make
money.
So
it's
really
kind
of
looking
to
drive
incentives
for
the
network
to
get
stood
up
in
the
right
direction.
For
this
one
or
two
hop
latency,
there's
a
lot
more
work.
That
needs
to
be
done,
but
it's
trying
to
kind
of
do
that.
First
step
in
that
direction
because,
like
you
were
saying
mike,
you
know
we're
going
to
want
to
be
able
to
compete
against
amazon
s3
in
order
to
make
file
coin.
A
Yeah
yeah
one
thing
by
the
way,
good
to
see
you
hannah
thanks
for
thanks
for
joining.
I
just
want
one
thing
in
response
to:
that
is
the
way
I've
kind
of
been
thinking
about
this,
and
I
mean
none
of
this
is
like
decided
yet
or
whatever.
But
the
way
I've
kind
of
been
thinking
about
this
recently
is
that
the
whatever
architecture
you
guys
from
consensus
end
up.
A
A
So
if
your
architecture
comes
along
and
kind
of
revolutionizes,
what
you
know
how
retrieval
works
to
the
point
where
maybe
I
don't
have
to
work
with
low
level
payment
channels
all
the
time
I'm
just
I
just
updated
to
work
on
your
stuff,
and
I
would
I
was
sort
of
thinking
david
chan
said
if
you
would
want
to
do
something
similar
like
if
we
had
a
whole
retrieval
infrastructure,
you
would
want
to
leverage
that
and
I'm
guessing.
I
don't
want
to
put
words
in
your
mouth.
A
Yeah,
okay,
so
I
think
we
are
sort
of
more
of
the
user
side
like
we're
trying
to
provide
something
right
now
we're
trying
to
provide
something
kind
of
quick
and
dirty
for
users.
It
could
get
more
much
more
usable
in
the
future,
but
at
the
end
of
the
day,
we're
trying
to
be
the
user
side,
not
the
infrasight,
and
I
think
you
guys
are
trying
to
be
the
infraside
and
then
you're,
not
gonna,
build
a
ui
or
desktop
app
or
whatever.
G
A
B
Sorry,
all
right
for
sure
I
I
think
we
need
both.
I
mean
when
I,
when
I
look
at
you
know,
like
I
mean
we've
all
used,
google
drive
and
how
you
can
just
drag
and
drop,
and
you
know
we
need
that
sort
of
stuff,
the
whiz
bangery
of
it
all,
but
yeah.
I
think
it's
it's
gonna
be
awesome
when
the
infrastructure's
there
and
the
ui
is
there
and
it's
all
just
you
know,
working
really
well,
that'd
be
really
cool.
A
F
You
know
alessandra
brought
up
this
great
thing
of
the
web.
3
torrent
that
the
state
channels
team
had
done.
Currently
the
state
channel
team
at
consensus
has
been
working
with
the
graph
on
a
crypto
payment
system
around
the
graph
indexing
retrieval
and
then
doing
very,
very
efficient
server
wallets
that
can
kind
of
do
three-party
payment
brokering.
F
So
that's
another
part
which
actually
you
know
you
have
sort
of
this
client-side
browser
infrastructure.
You
have
this
infrastructure
around
discovery
and
kind
of
finding
the
deals,
and
then
you
have
this
thing
about
making
it
happen,
and
so
that's
the
part
where
web3
torrent,
you
know,
was
sort
of
a
payment
system.
F
A
Yeah
that
makes
a
lot
of
sense,
yeah
and
I,
like
ethereum
state
channels,
a
lot.
I
think
I've
looked
at
ways
to
incorporate
this
that
into
this
work.
It
didn't
really
make
a
lot
of
sense
this
summer,
yet
I
think
it
was
more
just
like
I
need
to
get
some
kind
of
an
mvp
that
it
is
workable.
You
know
usable
enough
that
people
could
could
try
it,
but
yeah
yeah.
I
think.
F
Yeah
and
the
payment
system
is
a
nice,
it's
a
solid
payment
system
right
and
you
know
having
the
channels
and
being
able
to
like
do
a
research
you
know
claim
or
redemption
on
close
a
whole
bunch
of
channels.
At
the
same
time,
is
it's
a
nice
design,
so
the
the
state
channels
guys
did
an
analysis
of
it.
You
know
there'll
be
extensions
and
some
swapping,
because
state
channels
have
gone
a
lot
farther
in
terms
of
what
you
can
do
inside
of
them.
F
A
A
E
G
I
have
a,
I
have
a
question
so
well.
First
of
all,
I
don't
know
if
I
get
the
sense
that
we're
we're
always
on
the
verge
of
doing
that
thing
that
you
want
mike,
where,
like
essentially
remote,
signing
delegated
signing
of
things
that
would
allow
you
to
use
the
whole
lotus
api,
as
opposed
to
just
like
and
pull
push
message.
G
Yeah,
I'm
curious
the
other
thing
that
that's
like
that
would
take
a
bit
that
would
lower
the
barrier
to
entry
a
lot.
The
other
thing
is
like
I'm
curious
like
do
you
you
guys,
I
think.
Well,
I
guess
chris
chris
chafee
is
implementing
grassland
and
javascript,
which
yeah
like
I
mean
I,
I
wrote
the
grassland
I
like
it.
I
think
it's
a
super
awesome
protocol.
G
G
C
G
A
A
No,
I
know
what
you
mean.
I
mean
I
think
so.
The
one
factor
is
juan,
really
wants
everything
to
be
graph
sync
based
like
all
transfers
of
regressing
face,
so
that
was
why
he
asked
me
to
do
it
and
then
I
started
looking
into
it
and
I
was
like
we
don't
even
have
a
js
implementation
of
that,
so
we'd
have
to
be
compiling,
go
code
to
azum,
there
wasn't
binaries,
and
so
then
I
just
went
to
chris.
A
I
was
like
hey:
do
you
want
to
work
on
this
and
create
this
js
like
okay,
yeah
great?
So
that's
that's
where
that
is
another
factor
I
think
is
graph.
Sync
is
faster.
If
there's
an
element
tell
me
if
this
is
correct,
because
I'm
I
may
be
wrong.
I
believe
grassing
is
faster.
If
there's
an
element
of
oh,
I
see
we're
at
time
over
time.
I
see
we're
over
time.
Well
here
I'll,
just
ask
this
last
question
and
then
yeah.
A
If
anybody
has
to
drop
off
feel
free
and
thank
thanks,
everyone
for
coming
so
graph,
sync
is
faster.
If
there's
an
element
of
deduplication,
that's
happening
and
you
that
you
end
up
transferring
less
data,
but
if
it's
just
a
one-shot
retrieval
like
I
want
to
retrieve
entire
cids,
not
not
ipld
selectors,
not
like
a
hat
on
a
border
transfer,
and
I
just
want
you
to
get
the
second
half
it's
just
like
the
whole
cid
and
it
succeeds
or
fails
like
atomically.
A
G
And
if
you're
using
bitswap,
what
you're
going
to
have
to
do,
because
you
don't
know
the
other
cids
ahead
of
time,
and
you
have
no
way
of
expressing
that
you
want
like,
I
want
the
whole
graph
of
things
underneath
you're
gonna
end
up
doing
a
whole
bunch
of
round
trips.
Where
you're
like
give
me
the
first
block.
Okay,
I
got
the
first
block.
What
are
the
links
in
here?
Give
me
the
next
blocks
and.
G
G
So
that's
I
mean
that's
actually,
honestly,
the
main
speed
advantage
of
grasping
for
point-to-point
transfer
bit
swap
is
in
some
cases.
It
is
better
for,
like
a
multicast
like
I'm,
asking
15
people
for
this
thing,
which
you
know
we're
not
attempting
to
do
with
retrieval
yet
at
all
right
now,
so
in
file
coins
context,
graphic
is
much
more
ideal.
B
F
Saw
that
netflix
at
the
last
ipfs
camp
or
whatever
they
they
were
using
bitswap
because
they
were
sending
around
four
gigabytes.
You
know
large
large
images
and
they
were
doing
the
multicast
and
they
were
looking
to
sort
of
have
the
network
kind
of
self
torrent
and
bit
swap
can
work
well.
In
that
case,.
G
Yeah
they
can
for
point-to-point.
I
mean
honestly,
you
know
you're
gonna
be
doing
payment
channel,
but
you
know
like
sending
back
and
forth
payments
and
you
know
like
there's
a
lot
of
reasons
that
will
slow
it
down
a
little
bit
more
anyway.
So
I
don't
know
if
the
difference
is
that
big,
you
know
it's
the
type
of
thing
you
would
find
out
if
you
ran
both
practice
in
the
same
generic
situation,.