►
From YouTube: Iroh Cloud overview - @ramfox - Connecting IPFS
Description
Iroh Cloud overview - presented by @ramfox at IPFS þing 2022 - Connecting IPFS - https://2022.ipfs-thing.io
A
My
name
is
casey,
also
known
as
ramfox
on
the
internet,
and
I
work
at
number
zero.
I
work
with
these
lovely
people.
Some
of
them
are
in
the
room
today
and
we
are
building
iro.
Iro
is
efficient,
ipfs
that
can
scale
up
to
the
cloud
and
down
to
a
mobile
device,
and
I'm
going
to
talk
to
you
about
iro
cloud
and
also
some
guiding
principles
we
had
while
building
it
that
we
feel
really
aided
in
our
development.
A
So
the
first
guiding
principle
is:
we
want
to
have
the
flexibility
of
an
architecture
that
allows
us
to
use
different
approaches
for
different
platforms.
They
have
different
needs,
let's
figure
out
how
to
address
those
separately,
and
the
second
is
you
are
what
you
measure.
We
are
a
very
measurement,
focused
organization
that
has
paid
off
immediately.
Every
time
there's
been
a
new
measurement.
I
can
vouch
for
this
one
and
I'll
give
you
a
couple
examples
of
how
that's
really
helped
us.
A
So
the
idea
for
iro
started
b5
and
I
flew
out
to
berlin
to
meet
dig
in
person
for
the
first
time.
We
knew
we
had
a
lot
of
values
in
common.
A
Like
our
dedication
to
distributed
software
and
also
ideas
on
how
we
wanted
to
run
a
company,
for
example,
you
are
what
you
measure,
but
we
wanted
to
make
sure
that
we
had
something
new
to
bring
to
the
table
and
one
morning
in
a
haze
of
like
coffee
and
like
intense
discussion,
sort
of
like
how
I
feel
like
we're
gonna
get
at
this
conference.
A
Digg
drew
us
this
lovely
picture
and
in
case
that's
not
clear,
we
did
some
design
magic
to
make
it
look
like
this.
So
iro
is
not
just
a
single
binary.
Iro
is
separated
out
into
areas
of
concern
all
running
as
their
own
binary
is
speaking
over
our
pc.
A
What
does
this
give
us
gives
us
composability?
If
you
need
something
special
to
be
happening
here,
you
can
use
our
rpc
language,
or
you
know,
be
able
to
talk
to
our
rpc
connections
and
compose
this.
How
you
want
or
put
things
in
between
it,
you
can
do
things
like
things
that
you
would
expect
to
scale
independently
can
now
scale
independently.
A
A
Let's
say
you
just
need
to
add
some
networking.
You
can
just
pull
out
the
p2p
node
and
use
that
on
its
own,
the
last
bit
I'm
going
to
get
to
is
the
gateway,
so
we
have
our.
The
gateway
was
our
first
like
big
push
our
first
initial
offering,
because
we
wanted
to
make
sure
that
we
again,
you
are
what
you
measure
we
needed
something
that
could
be
compared
to
the
different
things
that
are
out
there
in
the
ecosystem
already
make
sure
that
we
are,
you
know,
keeping
up
so
that
so
yeah.
A
So
that
was
our
our
last
piece
that
we
felt
like
was
important
to
build.
Let's
talk
about
these
pieces
a
little
bit
more
in
depth.
The
data
store
is
essentially
an
object
store.
It's
an
opinionated
immutable
key
value
store
the
underlying
database.
We're
using
is
roxdb
written
by
facebook,
big
tech,
but
it's
working
for
us,
because
we
know
that
it
can
scale
up
to
a
data
center
and
down
to
a
mobile
device.
A
Our
plans
for
scaling
this
in
the
future,
if
you
need
to
add
more
data,
add
a
box.
It
is
now
a
shard
in
the
database
when
fetching
can
use
a
hashing
function
to
end
the
number
of
shards
that
we
know
are
in
the
database
to
figure
out
where
your
data
should
live
or
where
it
needs
to
be
pulled
from
we're
talking
about
using
weighted
rendezvous
hashing.
If
you
guys
want
to
talk
about
those
special
properties,
we
can
talk
about
it.
A
A
It
was
straightforward
to
implement
and
it
is
just
working
for
us
which
that's
what
you
want
in
your
software,
and
we
know
that
as
they
they
have
things
along
their
road
map
that
will
improve
rustler
p2p
and
we
can
just
draft
off
that
we're
very
excited
so
for
scaling
the
p2p
node.
A
We
do
have
a
lot
of
room
to
grow
here
before
we.
This
is
like
a
crazy
concern.
I've
seen
proof
of
a
wrestler
p2p
node
handling,
20k
connections
with
under
five
gigs
of
memory.
A
That's
pretty
good
and
we
know
that
the
simplest
way
to
scale
this
will
just
be
to
add
more
nodes,
but
there
are
obviously
many
other
arrangements
and
needs
that
people
will
want
and
but
the
first,
the
first
step,
hey,
you
need,
you
need
more
stuff,
you
need
more
connections,
add
another
node.
It
can
talk
to
the
same
database
because
everything
is
speaking
over
an
rpc
connection.
A
It's
fast,
it's
reliable
comes
with
load,
balancing
health
checks
and
then
finally,
our
gateway,
so
we've
been
using
the
gateway,
like
I
said,
to
kind
of
determine
and
if
we're
playing
in
the
right
ballpark,
because
you
are
what
you
measure
and
this
next
section,
I'm
going
to
basically
be
singing
the
praises
of
one
of
our
other
team
members
who
couldn't
be
here
tonight.
His
his
name
is
azmir
and
he's
really
been
leading
the
charge
on
this
he's.
His
role
is
dedicated
to
deploying
and
measuring
our
software.
A
These
metrics
that
he's
been
coming
out
with
they're
immediately
useful.
It
allows
us
to
pinpoint
the
slow
parts
of
our
code
and
it
really-
and
I
think
this
is
the
important
part
it
gives
us
guidance
on
where
we
need
to
spend
our
time,
because
we
know
we
have
evidence
that
hey
this
part
of
our
code
is
slow,
so
it
is
worth
spending
the
time
fixing
it
as
we
built
the
gateway
checker.
A
It
is
a
measuring
stick
to
rank
ourselves.
It
is
not
like.
Oh
we
shaved
two
seconds
off
of
our
time
on
the
gateway
like
that's
not
the
thing
it's
measuring,
it's
measuring,
hey!
Are
we
in
the
right
ballpark?
It's
comparing.
It
runs
these
tests
for
a
well-known
cid.
Does
it
resolve
and
how
quickly
for
a
set
of
new
fcids?
Does
it
resolve?
A
How
quickly
does
it
cash
well
on
repeated
checks,
and
we
are
running
these
tests
and
here's
like
a
little
bit
of
an
overwhelming
view
of
what
the
gateway
checker
actually
looks
like.
We
are
testing
against
cloudflare,
d-web,
ipfs
and
infura.
The
other
boxes
are
like
our
different
experiment
boxes,
for
example.
This
is
our
us
box.
Eu
box
dig
has
like
a
magic
box
in
his
apartment.
That
has
a
little
bit
of
better
hardware,
and
then
these
are
have
load
balancers
in
front
of
them.
A
This
one's
just
like
hey,
let's
see
what
happens
if
we
like
pay
amazon,
a
little
more
for
some
juice
and
see
if
it
actually
makes
a
difference.
So
there
are
a
bunch
of
experiments
that
we're
running,
and
I
got
to
tell
you
the
first
time
that
we
saw
this
gateway
checker.
It
did
not
look
like
this.
We
were
like
two
orders
of
magnitude
out
of
the
ballpark,
where
our
errors
were
way
higher
than
everybody
else's.
A
Our
timeouts
were
two
minutes
long,
which
is
a
full
minute
longer
than
everyone
else's
which
made
for
some
hilarious
duration
numbers.
That
was
a
great
fix.
We
got
100
improvement
just
by
changing
one
stat
and
we
were
just
a
slower
implementation
and
the
way
that
we
were
able
to
focus
on
hey
and
the
way
we
were
able
to
quickly
iterate
on
that
and
focus
on
hey.
What
needs
to
be
checked
is
iro.
A
Has
universal
trace
ids,
so
the
trace
ids
follow
the
requests
from
the
beginning
to
end
even
over
the
rpc
connections,
and
this
is
all
common
tooling,
we're
using
rest
tracing
module,
open,
telemetry
grafana.
If
you
run
your
own
iro
cloud
service,
you
can
pull
all
these
things
in
and
make
your
own
lovely.
A
You
know
chart
of
your
own.
So
this
is
what
a
trace
looks
like.
I
want
to
point
out
this
pretty.
You
know
142
microseconds,
the
slowness
I
was
talking
about
before
that
was
our
bit
swap
implementation,
and
this
tiny
little
line
was
a
large
yellow
line.
That
was
two
minutes
of
time
out
and,
as
brendan
said,
the
first
time
we
saw
these
traces
hey.
This
is
so
cool.
We
have
so
many.
We
can
learn
from
this.
A
So
much
it's
literally
a
giant
yellow
line,
saying
drop
everything
and
just
work
on
bitswap,
and
we
did
literally
a
week
later
we
had
four
prs
merged
that
were
just
about
fixing
bitswap
and
that's
our
our
new
time.
So
those
are
just
some
examples
of
how
using
metrics
has
had
an
immediate
impact
on
our
organization.
A
A
And
we
did
that
by
following
these
different
approaches.
Sorry,
these
these
two
different
guiding
principles,
use
different
approaches
for
different
platforms,
and
you
are
what
you
measure.
Irocloud
is
coming
q4
2022..
A
You
know
type
out
that
if
you
want
to
join
our
discord
or
wait
till
these,
these
slides
get
somewhere,
you
could
click
if
you
want
to
just
check
out
our
gateway.
Those
are
those
are
that's
the
address,
so
we
support
ips
and
ipfs
and
yeah.
Does
anyone
have
any
questions
again?
I'm
casey.
A
B
Yeah,
so
the
identity
story,
there
is
hard
yeah.
The
first
feeling
that
we
want
is
the
single
identity
across
multiple
people.
So
basically
you
just
share
the
fight,
because
you
that
that's
the
point
where
you're
like
well,
I
have
this
one
in
anybody.
I
needed
to
have
500
million
connections,
so
I'm
gone
around
space
on
this
one
box
right
the
more
complex
solution
where
you
want
to
have
multiple
identities
in
the
same
kind
of
arrangement.
B
I
would
love
to
support
that
eventually,
but
I
actually
need
to
talk
to
people.
So
if
you
have
that
need,
please
come
talk
to
me
because
I
actually
don't
know
in
what
scenarios
using
his,
because,
at
the
end
of
the
day
that
identity
doesn't
actually
matter
because
the
identity
you
have
different
keys
for
ips.
A
B
B
Run
an
independent
node,
at
least
because
all
the
basics,
all
the
ppp
protocols,
are
they're
conceptually.
They
can
only
understand
why
so
you
have.
There
is
no
like.
Oh
I'm,
two
different
people.
D
B
Or
whatever
you
have
some
thoughts
of
implementing
customers
which
actually,
as
well
as
extended
programs
with
a
lot
of
state,
there's
three
traffic.
So
you
can't
actually
do
nice.
Things
like
being
looking
at
traffic
and
luke
wants
things
like
sleeping
sessions
in
there
but
yeah.
For
now.
We
haven't
hit
that,
but
it's
definitely.
A
A
Yeah
you
can
take
one
of
the
services
online,
it
won't
affect.
Obviously,
if
you
ask
for
something
from
the
store
you
take
the
store
down,
you're
not
going
to
get
anything,
but
it
won't
affect
the
like
uptime
of
the
other.
C
B
I'm
but
the
idea
is
that
the
scalability
layer
already
basically
solves
that
in
the
sense,
as
soon.
C
B
Have
that
you
can
say
I
need
to
take
this
in
the
dockline
and
remove
it
from
the
cluster,
which
means
that
the
data
particular
scientist
is
starting
to
move
automatically
actually
and
then.
B
B
You
just
you
actually
know
the
instance
of
data
onto
another
one
and
then
take
the
original
offline,
and
just
say
this
new
box
is
now
this
exact
moment
and
the
system
is
changed.