►
From YouTube: CSCON[0] Molly Mackinlay - IPFS & libp2p in 2020
Description
Molly Mackinlay from Protocol Labs gives a year in review for IPFS and libp2p in 2020.
At The First Ever CSCON[0] Virtual Event!
B
B
Well,
excited
to
chat
with
y'all
today
wanted
to
give
a
fast,
lightning
tour
of
all
the
really
exciting
stuff.
That's
happened
with
ipfs
and
lib
p2p
in
the
past
year.
I
think
probably
a
lot
of
folks
last
engaged
more
heavily
with
ipfs,
maybe
around
the
april
may
timeframe
when
we
had
our
big
ipfs
0.5
launch.
So
I'll
recap
that
briefly
and
then
also
dive
into
some
of
the
cool
stuff.
That's
happened
in
q
or.
B
So
for
those
of
you
who
aren't
super
familiar
with
ipfs,
it
stands
for
the
interplanetary
file
system
and
it
aims
to
take
kind
of
the
centralized
http
web.
Where
many
different
end
users
must
all
connect
to
the
same
server.
In
order
to
get
the
data
to
a
more
distributed
peer-to-peer
system
where
you
can
have
subnets
operating
that
are
partially
offline,
you
can
have
a
much
more
flexible
system
of
nodes
directly
accessing
content
on
each
other.
B
The
way
it
does
that
is
by
instead
of
addressing
location
data
by
where
it's
located
the
network
using
location.
It
instead
uses
content
addressing
in
a
global
namespace
to
be
able
to
move
data
around
a
network
and
uniformly
verify
that
the
data
you're
fetching
from
any
random
node
in
the
network
is
the
data
you're.
Looking
for,
and
this
aims
to
address
a
whole
ton
of
different
problems.
B
Ipfs
is
part
of
kind
of
this.
What
I've
been
colloquial,
calling
the
interplanetary
stack,
but
the
set
of
protocols
that
build
on
top
of
each
other
in
order
to
make
kind
of
a
distributed,
peer-to-peer
web
possible,
and
so
ipfs
builds
really
heavily
on
top
of
lib
p2p.
Also
talk
about
a
few
folks
who
are
using
justly
p2p,
say
to
power
various
blockchains
that
are
have
been
launching
over
the
past
couple
of
months.
B
There
are
a
whole
ton
of
people
who
are
building
on
top
of
ipvs.
Now
we
we
actually
have
an
open,
sigma
diagram
where
anyone
can
can
add
new
applications
to
our
ipfs
ecosystem
view,
there's
a
ton
of
new
ones
here,
all
from
kind
of
the
content
creator
side
with
folks
who
are
doing
video
or
audio
or
other
kind
of
content.
B
So
lots
of
expansion
here,
even
just
in
in
the
past
year-
and
we
also
last
year-
saw
a
lot
of
growth
in
the
the
wider
network
as
well,
and
we're
we're
excited
to
see
that
many
of
those
nodes
have
also
upgraded
to
our
most
recent
release,
which
or
maybe
since
0.5,
which
is
our
our
may
release.
A
number
of
folks
have
upgraded
since
then.
B
A
lot
of
amazing
tooling
here
made
massive
gains
and
and
ease
access
to
putting
websites
and
web
tools
on
top
of
ipfs,
and
so
we'll
talk
about
that
briefly.
Q2
is
when
we
made
a
whole
ton
of
performance
improvements
to
ipfs
released
our
biggest
upgrade
to
the
protocol
in
a
long
time
which
brought
a
lot
of
performance
improvements
to
all
different
parts
of
the
ipfs
data
cycle.
B
You
could
work
much
more
simply
with
the
protocol
in
a
really
nice
to
really
nice
to
manipulate
way,
and
actually
one
of
the
groups
that
came
on
the
scene
and
drove
huge
adoption
here
and,
and
I
think,
reached
a
whole
ton
of
folks
in
this
wider
ecosystem
was
fleek,
who
built
a
very,
very
smooth
deployment
platform
where
you
effectively
connect
any
any
content
address
website
on
github
and
it
will
publish
to
to
ipfs
and
just
smooth
every
every
new
pr
to
that
site
gets
auto
deployed
it.
Really.
B
B
That
can
reach
out
and
support
a
set
of
use
cases
and
developer
flows
that
just
have
been
very,
very
difficult
previously,
because
we've
been
spending
so
much
time
on
the
low-level
infrastructure
and
capability
side
of
things
that
we've
neglected,
that
end-user
development
flow
and
so
fleek
has
come
in
and
kind
of
helped
solve.
That
problem,
which
helped
catalyze
this
amazing
explosion
of
d
websites,
so
examples
of
folks
who
took
advantage
of
that
ethereum.org
went
and
put
their
site
on
ipfs
and
ens.
B
So
you
can
view
that
at
ethereum
link,
origin
created
a
whole
set
of
of
d
web
stores
where
you
can
go
and
buy
swag
from
various
different
providers.
So
brave
has
put
their
swag
store
up
on
origin
using
ipfs,
which
is
pretty
cool.
There's
also.
This
awesome
explosion
of
d5
sites
who
are
using
ipfs
to
have
a
fully
distributed
front
end
that
doesn't
rely
on
any
single
party
to
keep
that
website
up
which
really
helps
with
the
decentralization
of
sites
like
uniswap,
kyber
ether,
wallet.
B
Things
like
that,
where
you
know
again
the
the
core
app
like
back
end
of
the
application
lives
in
the
blockchain.
The
front
end
is
distributed
on
ipfs
and
you
have
no
central
party.
That's
a
bottleneck
or
fully
in
control
of
that
platform
and
kind
of
the
this
awesome
movement
of
sites
into
the
d
web
culminated
with
ipfs
actually
getting
default.
B
Looking
at
q2,
this
was
our
major
ipfs
release
quarter
and
so
a
lot
of
a
lot
of
time
spent.
This
quarter
was
the
upgrade
and
and
release
of
these
performance
improvements
and
then
also
making
sure
that
the
tools
we
used
in
order
to
validate
and
benchmark
and-
and
you
know,
optimize
the
improvements
we're
making.
The
protocol
also
became,
accept
accessible
to
the
wider
ecosystem,
so
kind
of.
B
We
definitely
hit
a
number
of
scaling
challenges,
and
so
this
refactor
helped
clean
out
a
number
of
nodes
within
the
network
that
were
not
doing
a
good
job
serving
queries
and
and
update
some
of
the
the
logic
within
the
dht
to
better
identify
where
content
was
and
and
search
the
network
better.
So
one
of
the
first
challenges
was:
we
had
lots
of
connectivity
gaps
where
people
were
joining.
B
So
if
you're,
behind,
like
a
home
local
area
network
and
you're,
not
able
to
serve
query,
you
wouldn't
join
the
network
as
a
server
and
therefore
you
wouldn't
end
up
being
a
like
a
gap
or
a
someone.
Who's
not
able
to
effectively
forward
on
queries
to
their
correct
destination.
And
so
this
helped
us
in
many
ways
remove
a
set
of
nodes
who
were
previously
slowing
down
our
queries,
our
queries
and
and
having
dead
ends.
B
Then
the
next
thing
we
did
was
we
made
sure
for
the
good
nodes
you
were
connected
to.
You
stayed
connected
to
them,
and
so
this
was
a
lot
of
improvements
both
to
autonat
and
to
the
routing
tables
within
our
dht.
So
understanding
which
peers
are
are
valuable,
where
you
have
open
connections
and
that
you
are
prioritizing
peers
and
keeping
them
in
closer
buckets.
B
B
So,
as
I
was
mentioning,
we
needed
to
do
a
lot
of
testing
and
a
lot
of
simulation
in
order
to
make
these
changes.
These
are
pretty
fundamental
changes
to
our
dht
implementation,
and
you
know
deploying
those
to
a
network
of
millions
of
end
users
and
hundreds
of
thousands
of
nodes
and
crossing
our
fingers
that
it
worked
was
not
going
to
cut
it.
B
You
know
a
live
real
network
with
hundreds
of
thousands
of
nodes,
and
so
we
we
ran
our
new
dhd
improvements
through
this
over
and
over
again
and
allowed
us
to
optimize.
For
example,
one
of
the
really
important
and
valuable
improvements
we
got
here
was
20
to
30
x,
faster
improvement
in
providing
data
to
the
the
ibm
sdhd,
still
a
lot
of
room
to
go
here.
B
So
a
lot
more,
we
can
do
to
make
this
even
faster,
but
we,
these
sorts
of
massive
changes
when,
when
viewed
within
our
testing
framework,
were
you
know,
helped
us
validate
that
we
were
on
the
right
track.
Similarly,
when
we're
finding
providers
of
the
content
we're
looking
for,
we
were
able
to
make
a
two
to
six
x,
faster
improvement
for
fetching
data
from
ipns,
which
is
our
mutable
naming
system.
B
That's
built
on
top
of
ipfs
we're
able
to
get
approximately
like
5x,
faster
and
then
actually
for
ipns
over
pub
sub,
which
was
something
that
we
added
as
part
of
0.5
and
then
stabilized
kind
of
the
the
release
after
we
were
able
to
make
putting
and
getting
with
ips.
Also,
a
ton
ton
faster,
still
more
improvements
there
as
well,
but
able
to
like
refactor
a
system
that
you
know
had
had
some
significant
performance
hiccups
into
something
that
that
works
pretty
reliably.
B
So
in
all,
we
were
able
to
look
across
ipfs
odot
4.23,
which
is
our
release
ahead
of
ipfs
1.5,
and
this
one
we
were
able
to
see
really
really
massive
performance.
Improvements
for
finding
data
in
the
network
went
from
like
somewhere
from
8
to
42
seconds,
which
was
just
unbelievably
slow
and
very,
very
painful
to
sub
second,
in
most
cases,
and
now
we're
seeing,
I
think,
95th
percentile.
As
of
now
as
more
nodes
in
the
network
upgrade
to
the
latest
versions
of
ipfs.
This
continues
to
get
faster
and
faster.
B
So
I
think
we're
seeing
even
better
metrics
than
this,
where
95th
percentile
content
routing
is
now
below
three
seconds,
which
means
that
the
vast
majority
of
queries
in
the
network
are
completing
very
fast
and
are
able
to
yeah
again
still.
We
would
like
to
get
see
these
in
the
millisecond
time
frames.
For
you
know
many
use
cases
like
loading
pages,
but
huge
improvements
here
from
where
we
were
at
at
the
beginning
of
the
year.
B
In
addition
to
finding
content
in
the
network,
another
area
we
collaborated
with
netflix
actually
to
improve
was
transferring
data
in
the
network.
So
you
know,
depending
on
your
configuration,
you
may
as
an
ipfs
node,
be
spending
a
lot
of
time,
actually
syncing
data
between
other
nodes
or
fetching
data
from
many
nodes
at
once,
in
order
to
benefit
from
having
this
kind
of
peer-to-peer
architecture
and-
and
that
was
the
the
use
case
and
challenge
that
the
the
netflix
team
was
working
to
solve
from
kind
of
a
ci
cd
pipeline
platform.
B
We
also
added
some
very
early
support
for
graphsync
graphsync
is
the
the
data
transfer
algorithm
that
is,
was
developed
kind
of
in
part
thinking
directly
about
blockchains
and
how
blockchains
want
to
exchange
content
in
the
network
where
you
have
very
deep
chains
of
parents,
and
you
want
to
make
sure
that
as
you're
fetching
data
you're
able
to
kind
of
pull
each
incremental
block
not
with
a
round
trip
in
between
where
you
like.
You
know
get
the
parent
and
then
you
ask
for
its
parent
and
that's
parent.
It's
parent!
B
You
want
to
make
one
direct
query
all
the
way
through
the
history
of
the
network.
Graphsync
has
a
selector
language
that
allows
you
to
effectively
do
like
a
send
me
all
parents
of
this
node
sort
of
request
and
then
sync
all
of
those
back
and
forth,
and
so
there's
ongoing
work
in
progress
to
work
on
bringing
graph
sync
and
bitswap
closer
together.
B
We
also
added
a
number
of
important
things
from
an
ad
performance
perspective
and
from
a
hardening
perspective.
So
some
of
the
folks
here
may
have
heard
about
the
improvements
to
gossip
sub,
which
is
the
kind
of
peer-to-peer
gossip
exchange
protocol.
That's
used
by
many
blockchains
with
who
are
using
lib
p2p
as
their
peer-to-peer
networking
layer,
and
this
also
used
a
ton
of
the
the
testing
using
test
ground
in
order
to
simulate
a
number
of
attack
scenarios
and
validate
and
make
improvements
to
gossip
sub.
B
B
Looking
at
q3,
this
was
a
quarter
of
of
standards,
and
so
we
took
a
number
of
protocols
that
had
either
been
experimental
or
had
been
kind
of
on
their
way
to
being
phased
out
and
started,
making
them
default.
So
the
quarter
of
defaults,
we
added
quick
support
by
default.
This
had
been
kind
of
experimental
in
ipfs
for
a
long
time
and
is
now
the
default.
So
you
know
the
quick
spec
had
been
evolving.
B
B
B
Speaking
of
new
friends,
we
spent
a
good
bit
of
time
documenting
some
of
the
awesome
groups
who
are
building
on
top
of
ipfs
so
that
they
could
share
kind
of
both
what
brought
them
to
ipfs
and
what
their
development
architecture
and
story
looks
like.
We
have
a
lot
of
different
people
who
are
using
and
building
on
ipvs
in
different
ways.
B
So
if
you
are
excited
about
starting
your
own
thing
on
ipfs
and
you
want
to
learn
from
the
the
motivations
and
the
the
kind
of
upgrade
path
of
other
people
building
on
the
platform,
this
is
super
useful
and
there's
a
lot
of
good
information
within
here
about
groups
like
fleeq
and
audience
and
open
bazaar
and
how
they
architected
themselves.
On
top
of
ipfs.
B
I'm
sure
this
is
continuing
to
grow,
and
that
was
amazing
for
us,
as
both
like
a
feedback
loop
of
many
different
teams,
building
on
top
of
the
the
project
all
at
once
and
also
kind
of
finding
their
own
niches
and
being
able
to
help
and
support
each
other.
It's
been
great
to
see
these
teams
come
out,
and
many
of
them
are
kind
of
continuing
and
formalizing.
B
Many
of
those
groups
also
launched
improvements
and
tooling.
That
makes
that
easy
from
a
development
perspective.
If
you
want
to
hear
about
more
about
the
ways
we
contributed
to
or
how
we
see,
this
evolving
really
recommend
watching
the
talks
from
falcoid
liftoff
week
on
the
falcon
youtube
channel
they're
phenomenal.
B
And
there
is
now
four
client
implementations
interoperating
on
on
the
beacon
chain,
so
huge.
Thank
you
to
these
teams
for
like
helping
push
libby
to
be
forward
and
making
it
better
super
exciting
cool,
and
my
call
to
action
for
you
in
my
last
minute
is:
we
are
actively
working
on
ipvs
2021
planning,
there's
a
lot
of
stuff
happening
in
the
ecosystem
and
a
lot
of
ideas
on
how
we
can
continue
to
make
it
better
and
where
we
should
focus
our
time.
B
As
you
know,
both
some
of
the
core
working
groups
and
then
also
this
entire
ecosystem
of
awesome
projects
who
are
trying
to
work
within
the
ipfs
ecosystem
and
make
it
better
and
so
15
proposals
already
awesome
conversations
happening.
Please
jump
in.
If
you
want
to
talk
about
ipves,
1.0
or
content,
permissioning
or
new
dht
improvements
come
join
the
conversation
we
would
love
to
hear
your
thoughts
cool
well.
B
A
B
Yeah,
I
would
say
a
lot
of
this
is
organic,
like
it's
just
honestly
happening
on
its
own
accord
and
it
started.
You
know.
Over
a
year
ago,
lots
of
groups
within
the
ipvs
ecosystem
were
already
looking
to
things
like
falcoin,
or
were
attracted
as
that
as
an
upgrade
path
for
them
for
keeping
data
storage
around
inaccessible,
a
number
of
groups
who've
been
building,
developer,
tooling
and
apis
system.
B
Infuria,
textile
fleeq
others
have
already
dive
deep
into
filecoin
and
are
some
on
how
to
configure
use
it
from
a
developer,
and
so
it
this
is
well
underway.
Definitely
go!
Look
at
some
of
the
videos.
These
these
groups
are
highly
overlapped
and
in
many
ways
using
falcoin
is
using
an
ipfs
node.
It
runs
on
top
of
the
same
protocols.
It
uses
little
b2p,
it
has
ipld
data
to
add,
add
new
data
to
the
falcon
network.
You
first
turn
it
into
ipfs
shape
data
and
send
it
over
liberty
to
people.
B
B
I
spent
a
while
thinking
about
this
last
time
and
I
I
avoid
picking
favorites
in
the
ipfs
ecosystem,
because
I
love
all
of
them
and
they're
doing
really
really
amazing
work.
The
one
that
I'll
I'll
actually
shout
out
to
which
is
not
not
technically
an
ipfs
ecosystem
is
matrix.
I
just
I
guess
now
called
element.
B
I
love
their
team
they're
just
phenomenal.
They
think
really
thoughtfully
about
kind
of
the
the
content,
moderation
and
agency
of
of
nodes
in
the
network
and
they've
actually
started
building
a
p2p,
matrix
or
p2p
element
on
top
of
libpdp,
which
is
super
cool,
and
I'm
really
excited
to
see
where
that
one
goes.