►
Description
Originally recorded during the Berlin Developers Meetings from July 9-13, 2018.
A
C
A
All
right,
so
so
really
I
think
we
talked
about
the
current
state
of
DHT.
The
example
that
I
certainly
well-known
know
like
last
couple
days,
has
been
IP
NS
issues
for
publishing
and
resolving
and
the
issue
about
oh
who's
DHT.
And
how
do
you
scale
that
so
we
talked
about
messaging
protocols
as
well
as
this
idea
of
scaling
it
right
now
with
like
peer
pad
you're,
stuck
to
maybe
scaling
to
about
thirty
or
forty,
maybe
a
few
hundred
users,
and
so
the
issue
also.
A
Is
there
multiple
I
think
connections,
so
yeah,
actually
you're
limited
by
the
connection
connections-
and
this
is
what
we
were
talking
to
peer
about.
I
think
is
that
you
know
basically
in
right
now
with
the
his
issue.
Is
that
they
limited
to
about
thirty
peers,
but
so
the
idea,
basically
taking
states
and
being
able
to
report
state
as
an
IP
LD
object
and
integrating
it
signing
it
or
integrating
it
and
creating
a
proposed
state
for
conflict
resolution
and
and
then
that
and
then
reported
back
as
far
as
any
conflicts.
A
A
There's
an
issue
about
membership
of
DHT
s,
basically
how
the
IP
NS
was
able
to
scale
from
Kyle
was
spinning
up
a
much
a
bunch
of
extra
protocol
labs
nodes
in
the
cloud,
but
the
issue
is
like
how
does
that
scale,
two
million,
and
so,
if
he's
protocol
labs
actually
spins
it
up,
but
why
would
I
trust
open,
bazaars,
dhts
over
protocol
labs,
so
the
idea
of
being
a
trusted
super
node
actually
came
up,
and
how
do
you
actually
manage
a
trusted?
Super
node?
Well,
that's
verifiable
claims!
That's
something!
I
know
about
so
that
L.
A
So
the
issue
flash
crowd,
crowds
come
in
and
try
populate
into
the
DHT
and
the
idea
of
actually
basically
how
to
self
heal
your
continuum,
contributing
resources
to
the
DHT
until
it
actually
propagates
and
be
able
to
have
some
way
of
visualizing
that
network,
so
that
you
can
actually
get
some
feedback
or
the
state
of
the
network
as
its
self
healing
and
automagically
scale
back.
So
some
of
the
solutions
we
proposed
was
the
idea
of
proof
of
work
that,
as
you
push
things
to
the
DHT
you're
asked
to
solve.
A
Some
cryptographic
function,
coordinate
with
signals,
so
that's
messaging
gossip.
So
basically,
I
was
talking
about
gossip,
about
gossip,
basically
again
gossiping
about
the
network's,
the
state
as
an
IP,
LD
object
concept,
see
Rd,
T
and
and
signing
it
and
gossiping,
and
actually
like
reporting
back
the
gossip
about
that
that
sub
node
and
then
that's
using
IP
LD
for
the
state
reputation
systems.
So
basically,
if
this
proxy
is
not
known
and
it's
trying
to
sign
things
for
a
whole
bunch
of
new
peers
and
it's
basically
spinning
up
trying
to
flood
the
system.
A
B
So
the
basic
concept
is
when
you,
when
we
end
up
having
a
couple
like
these
mesh
networks,
right
of
hey,
here's
an
established
network,
they
have.
What
we
want
to
be
able
to
do
is
provide
recipes
for
based
on
the
situation
that
this
network
is
going
to
be
in,
so
that
you
can
have
different
configurations
of
how
these
these
kind
of
sub
networks
want
to
operate
so
that
it
can
kind
of
control
how
it
scales.
B
So,
if
you
have
kind
of
your
your
mesh
rings-
and
you
have
this
network,
topology
kind
of
being
shared
across
as
more
and
more
nerds
are
joining
and
demands
are
differentiating,
and
we
can
say:
oh
you
know
what
we
need.
We
need
to
get
another
DHT
scaled
up,
we're
gonna
go
to
one
of
our
trusted
nodes,
tell
it
to
turn
into
a
DHT,
and
then
we
can
watch
the
kind
of
the
mesh
networks
shift
in
transition,
and
then
over
time
has
skills
back
down.
We
can
we
can
change
that
back.
B
A
And
that
basically
got
to
my
question
for
Brian
was
Brendan
is
the
model
of
the
data
of
the
topology
of
the
network,
and
so
actually,
how
do
you
represent
that
and
to
make
sure
they
actually
here's
the
network
that
we're
trying
to
represent
and
here's
the
proposed
date
for
actually
the
conflict
resolution?
And
at
what
time
point
and
then
lastly,
yeah
I,
guess
see
our
DTS
for
rebalancing
that
so
I
think
defeat.
We
also
had
a
couple
different
things.
You
have
anything
else
to
interject.
D
We
don't
know
a
lot
about
what
candidate
solutions
and
like
what
we
could
implement
first,
so
one
of
the
shortcomings
or
one
of
the
problems,
it
all
comes
down
to
give
better
primitives
for
applications
to
like
give
hints
to
a
peer
to
peer
of
what
they
need.
So
when
this,
hopefully,
when
we
talk
about
membership,
management's,
it's
really.
Okay,
like
I'm
a
peer
path
and
I'm,
just
really
interested
all
talking.
We
nodes
that
are
on
the
same
path
as
I
am
so.
Why
am
I
connecting
to
the
rest
of
the
network?
D
And
so
if
there
is
something
that
connects
exposed
by
a
p2p
API?
That
can
give
people
a
way
to
then
give
that
information
to
the
people
peer
that
only
peer
to
peer
can
make
better
judgments.
The
other
thing
is
really
the
coordination
protocol,
which
is
kind
of
like
very
cheap
protocol,
like
it's
just
like
notes
telling
each
other
hey.
Here's
like
what
I
see
from
a
network.
It
seems
that
there's
a
lot
of
a
load
or
like
not
a
lot
of
loads.
Of
course
like
these
can
be
attacked
like
someone
can
be
lying,
but
it's.
D
This
is
just
like
to
help
notes,
make
better
decisions.
If
some
note
is
telling
they're,
like
oh
I,
see
three
million
notes
joining
like
we
should
like
calm
down
and
like
increase
the
size
of
our
crypto
puzzle
and
then,
like
ten
other
notes,
tell
you
oh
I
just
see
they're
not
joining,
then
you
can
like
lower
the
reputation
of
that
note
right,
and
so
you
can
use
all
of
these
to
make
better
decisions
and,
in
the
end,
like
a
lot
of
these
networks,
like
you,
don't
really
need
more.
D
Like
than
hundred
to
two
hundred
connections,
and
this
is
already
a
lot
like
you
can
do.
You
can
access
pretty
much
the
entire
network
with
just
like
100
to
200
and
in
some
runtimes
like
the
browser,
we
really
don't
even
want
to
have
more
than
10
or
20
connections.
Otherwise
we
might
just
kill
it
so
yeah,
like
I,
guess,
I'm,
just
like
repeating
some.
E
This
reminded
me
a
lot
of
coral
and
it's
a
great
that
you
are
looking
at
it.
I
think
I
think
there's
a
whole
area
of
protocol
designed
to
be
explored
there
on
what
these
self-organizing
clusters
should
do.
One
part
is
routing
and
coral.
That's
what
coral
does,
but
there's
a
whole
bunch
of
other
stuff
that
that
could
happen.
I
think
you're,
you're,
starting
to
uncover
I,
would
worry
about
with
a
lot
of
these
systems.
It
has
to
be
used
to
get
a
really
simple
construction
for
how
these
systems
grow.
E
Merge
split,
come
back,
and
in
the
house
we
can
I'm
pretty
stable
when
once
you
ship
the
same
protocol
to
like
millions
of
nodes,
and
so
we
save
like
reducing
the
complexity
of
that,
it
is
important
and
so
I
think
coming
up
with
very
simple
ways
in
which
these
clusters
form,
or
these
subnets
form
and
how
they
do
something
different
would
be
important,
but
just
more
of
a
design.
Thinking.
Okay,.
D
Okay,
one
of
the
things
that
multiple
groups
disgusts
and
we
talk
about
a
lot
about
having
fun.
So
one
thing
that
we
really
want
to
have
and
multiple
people
got
excited
about.
It
just
have
a
contest
like
let's
create
a
and
multiple
people
build
different,
topologies
and
I
know
it's
coming
with
all
these
use.
Cases
like
we're
streaming
and
like
forums
and
blogs
and
social
networks
and
like
just
what
people
go
wild
like,
let's
create
a
huge
contest
it
can
be
in
person
can
be
online.
D
C
We
have
time
for
one
more
question
and
I
saw
a
coal
raising
his
hand.
Oh
yeah,
just
echoing
one
sentiment
about,
like
simplicity,
I
think
something
that
I've
like
already
gotten
a
bit
of
feedback
about
is
like
the
kind
of
back
to
the
like
who's
DHT
question
like
oh,
should
I
be
running
my
own
separate
network
or
is
there
one
global
network?
I
know
that
we
tend
to
focus
on
like
the
single
global
network
and
I.
C
Think
to
that
end,
we
we
really
do
need
a
way
to
decline
initially
like
participate
in
everything
and
then
kind
of
slowly.
Let
the
things
that
you
don't
care
about
time
out
and
like
leave
your
purview.
It
feels
like
something
like
that,
like,
rather
than
being
like
super
analytical
about
what
you're
involved
in
and
like
subscription
management,
which
can
get
super
complicated
yeah.
That
just
seems
like
that's,
like
an
insight
worth
worth,
came
forward.
E
The
thing
I
was
thinking
about
immediately
we're
not
immediately,
but
throughout
the
conversation
is
there's
a
very
interesting
parallel
one
strategy
for
buildings.
Some
of
these
things
would
be
much
more
secure
might
be
to
do
something
like
probabilistic
checking
of
a
DHD,
so
you
spin
up
a
thousand
nose.
We
distribute
them
in
a
ring
and
then
start
for
these
kind
of
reputation,
systems,
system,
design,
type
of
ideas.
E
You
know
if
you
have
some
kind
of
auto
scaling
and,
like
you
notice,
a
ton
of
people
trying
to
do
an
ongoing
attack,
then
that
just
puns,
a
problem
to
participants
just
delaying
their
their
joining
and
then
join
you
very
slowly,
acting
honestly
for
a
long
time
and
then
attacking
the
network
in
one
moment.
But
a
way
that
that
might
be
averted
is
reputation,
systems,
the
reputation
systems
and
peer-to-peer
pair
really
well,
but
nobody
has
really
build
one
that
works
extensively.
E
Well,
there's
a
bunch
of
like
problems,
I
think
the
torrent
and
tracker
trust
like
tracker
peer-to-peer
systems
and
one
of
the
tracker
reputation,
systems
and
Victorian
like
ratios,
we're
a
very
good
example
of
this
working
very
well.
But
there
are
that's
a
few
and
far
between
success
or
success
is
very
rare.
Compared
to
the
number
of
times
people
have
tried
this
kind
of
stuff,
but
this
might
work
well
with.
E
C
A
A
I
think
just
building
upon
the
reputation
systems,
I,
think
the
identity
issues
and
capability
systems
that
I
think
that
the
D
ID
spec
is
actually
supporting,
and
these
still
early
days
for
creating
a
specification.
But
the
idea
of
basically
yeah
I
have
an
identity
and
I
have
some
nodes
that
are
controlled
under
my
identity.
That
are
share
peers
that
so,
even
though
that
node
is
not
trusted
I'm
trusted
and
so
that
you
can
tie
back
in
ephemeral
nodes.
C
Yeah
I,
like
as
extreme
micro
thought,
but
just
as
far
as
like
the
streams
are
concerned,
I
feel
like
super
cheap
ephemeral
streams
are
like
really
great
as
far
as
protocol
design
and
keeping
complexity
down
is
concerned.
So
I
go
Stephen,
go,
go,
everybody
go,
let's
figure
it
out,
because
yeah
I
can't
imagine
doing
more
intense
work
in
the
line
of
like
pipelining,
without
introducing
like
massive
complexity,.
D
All
right
so
I
feel
a
lot
of
my
ideas.
I've
been
represented
here
by
others,
so
if
I'm
being
redundant,
tell
me
to
stop,
but
one
that
I
would
definitely
would
like
to
bring
is
like
on
this
reputation
systems,
it's
very
important
to
make
sure
that
we
don't
leak
important
information
about
the
users
that
are
using
a
network.
We
typically
think
reputation.
Systems
are
like
the
hammer
that
solves
all
the
problems.
D
They
do
solve
a
lot,
but
they
also
mean
like
they're,
stateful
right
I,
get
something
over
time
like
something
that
you
can
start
analyzing
and
like
just
observing
the
patterns
and
observing
where
that
identity
is
valid.
Where,
and
so
you
need
to
find
another
mechanisms
to
like
have
reputation
but
I'll
click.
That's
easy
same
user.
There's
some
interesting
research
on
that.
There's
a
paper
on
a
non
rep
that
basically
is
a
peer
to
peer
anonymous
reputation,
system.
D
I,
haven't
really
had
the
chance
to
explore
the
construction
and
in
depth,
but
like
a
lotta,
great
people
talk
about
it
and
other
things
like
constructions.
Just
for
a
key
rotation
and
like
this,
the
ideas
around
like
double
ratchet
algorithms
and
just
like
group
coordination
like
in
a
way
not
only
peer
so
join
but
actually
wearing
devices
to
join
into
you
or
other
peers
like
so.
D
If
I
have
a
group
of
people
and
I
am
talking
on
a
chat
room
with
my
desktop
I
want
to
have
my
phone
join,
that
room,
and
so
I
can
do
something
like
I
I
notify
everyone
that
I'm
joining
with
my
device.
I
prove
that
the
device
is
mine,
but
just
to
make
sure
that
there
is
no
malicious
entity.
Trying
to
force
me
to
convince
everyone
at
another
device
is
joining.
Imagine
like
NSA
kind
of
like
adding
one
device
to
me
through
the
network.
D
D
D
A
Just
so
the
issue
of
our
correlation,
we
talked
about
this
in
the
D
ID
spec
and
the
what
we
came
up
with
is
a
is
ephemeral,
identifiers
and
so
which
are
pairwise.
So
this
is
why
255
1
9
is
actually
critically
important
is
actually
you
do
the
de
feel
something
like
the
Tiffany
Hellman
key
exchange
and
you
get
to
a
prime
number,
which
represents
the
pairwise
Association,
and
that
itself
is
actually
the
token
to
move
forward.
So
actually
there's
a
pairwise,
so
they
actually
they'll
never
be
reused
in
less
than
the
same
relationship.