►
Description
Lightning Talk: Hybrid/Multi Cluster Storage with Erin Boyd of Red Hat.
Filmed on October 28th 2019 in San Francisco.
A
Thank
You
Diane,
so
my
name
is
Aaron
Boyd
I'm,
a
senior
principal
engineer
for
Red
Hat,
I
work
in
the
office,
the
CTO
currently
I'm,
working
on
hybrid
and
multi
cluster
storage.
In
just
the
multi
cluster
story,
we
have
coming
out
of
Red
Hat
thought
it
was
a
thank
you
diamante
for
setting
it
up.
I'm
talking
about
hyper-converged
infrastructure.
That's
one
of
our
main
focuses
that's
a
powerful
message
for
why
we
enter
the
realm
of
hybrid
and
multi
cluster
storage.
A
So
why
would
you
want
a
hybrid
type
of
setup
or
even
multiple
clusters?
Well,
especially
in
terms
of
AI
and
ml,
you
might
have
performance
considerations
where
you
want
to
run
some
of
your
workload
on
a
very
specific
cloud.
You
need
fault
tolerance,
you
want
to
back
it
up,
you
know
maybe
between
different
zones,
and
maybe
you
need
some
specialized
hardware
to
run
some
of
your
workloads.
A
The
problem
with
that,
though,
is
if
you
happen
to
choose
one
vendor
for
all
of
those
services,
you're
locked
in
and
then
there's
the
idea
of
regulation
and
you
want
to
collaborate.
So
you
want
to
be
able
to
run
your
workload
where
you
have
those
services,
wherever
they
might
be,
they
might
be
in
gke,
they
might
be
an
AWS
and
due
to
regulation
you
might
want
to
run
them
on
Prem.
A
So
those
are
the
considerations
where
you
want
to
look
at
hybrid
cloud
or
multiple
clusters
within
your
kubernetes
deployment,
to
be
able
to
facilitate
running
your
workload
where
it
runs
best
so
whit.
There
are
lots
of
different
applications,
of
course,
in
AI
and
ml,
and
how
do
we
share
data
within
those?
You
know
they're
the
typical
four
things
we
share,
our
an
object
store
or
it
could
be
a
database,
a
file
system
or
a
queue.
Almost
every
slide.
A
I've
seen
you
know,
evidence
of
that
I've
seen
Postgres
I've
seen
Kafka
I've
seen
spunk
I've
seen
a
lot
of
these
already,
showing
that
this
is
what
we're
using.
This
is
how
we
use
them
between
clusters
and
I'm.
Today,
I
really
want
to
focus
just
on
how
objects
store
and
what
we're
doing
that's
a
little
bit
different
and
in
the
community.
Today.
A
So
object
storage
is
convenient
in
an
AI
ml
presence
and
that
you
can
have
a
bucket
within
something
like
three
and
have
many
different
applications
either
feeding
to
or
reading
from
that
bucket.
It's
pretty
easy
to
use
the
API
lots
of
different
cloud
vendors
support
that
API
and
you
can
write
your
application
easily
to
get
and
put
from
that
bucket
and
provide
policies.
A
A
So
within
that
you're
able
to
not
only
be
tied
to
the
you
know,
s3
version
that
you're
using
on
AWS
you're,
able
to
abstract
that
out
and
use
nuba
as
an
interface
between
any
s3
provider
so
on
the
backend
nuba
also
provides
the
ability
to
do
dupe
the
data
get
sections
of
the
data
to
move
over.
So
you
have
your
basically
your
replication,
snapshotting
and
backup
as
well.
All
this
is
still
tied
into
kubernetes
because,
as
I
mentioned
before,
we
don't
really
have
a
consistent
API
in
kubernetes
for
object
storage.
A
How
many
of
you
guys
use
persistent
volume?
Precision
volume
claims
today,
yeah,
a
lot
of
people
are
very
familiar
with
that.
So
what
my
team
at
Red
Hat
did
has
come
up
with
a
CR
D
that
allows
you
to
create
an
object
bucket
in
an
object
bucket
claim.
These
concepts
are
just
like
you're,
persistent
by
persistent
biome
claim,
but
specific
to
the
very
needs
of
object,
storage,
and
this
way
it
can.
It
provides
a
consistent
control
path
so
that
I
can
create
an
object
bucket
claim
just
like
I.
A
So
what?
If
I'm
not
using
object
data?
What,
if
I,
want
to
keep
my
application?
The
way
that
it
is
I
want
to
use
all
of
you
guys
that
raise
your
hand
when
you
said
you're
using
persistent
data.
Rook
is
another
great
open
source
product
that
project
that
allows
you
to
automate
the
installation
of
things
like
SEF
or
mini,
oh
and
soon
to
be
long
form,
and
so
what
it
does
is
it
provides
the
ability
to
have
an
operator
to
deploy
the
come.
A
Take
the
complexity
out
of
the
storage
and
come
provide
this
consistent
backbone
across
many
different
clusters.
So
I
believe
in
the
discover
thing
we
heard
earlier
today
they
were
talking
about
using
shared
storage
using
things
like
NFS
or
EFS.
Rook
also
provides
now
a
plugin
force
ffs.
So
then
you
have
the
shared
storage.
So
even
if
you're
not
using
object
data,
you
still
can
use
the
power
of
an
operator
than
to
deploy
your
storage
system
consistently
across
many
different
clouds.
A
So
what
are
we
doing
in
the
community?
I'm
also
part
of
the
kubernetes
Storage
sig
and
the
CN
CF
Storage
sig
and
the
community
is
working
to
enhance
how
we
use
data
more
agile
e
in
the
community.
Things
like
snapshots,
cloning
and
volume
transference
are
soon
coming
out
to
be
able
to
be
deployed
and
used
to
leverage
the
ability
to
have
hybrid
cloud
because,
as
you
know,
kubernetes
is
always
claimed,
we're
completely
stateless,
we're
agile.
We
can
move
anywhere
and
then
that
all
fails
when
we
start
talking
about
purses
storage.
A
So
look
forward
to
many
of
these
features,
helping
improve
the
way
that
we
can
manage
our
data
within
the
systems
and
then
lastly,
I
think
this
is
great.
This
actually
ties
back
to
what
do
you?
Monty
was
talking
about
when
you
have
hybrid
cloud
and
when
you
have
these
challenges
of
Manning
managing
different
clusters,
you
need
this
consistent
administration
across
all
this.
When
you
talk
about
your
applications,
you
have
to
have
network.
If
you
don't
have
network,
you
don't
have
the
ability
to
have
the
distribute
distributed
storage.
A
It
simplifies
administration,
allows
you
to
enforce
things
like
quotas,
I
believe
that
discoveries
are
talking
about
either
and
then
apply
those
policies
across
all
of
them.
So
with
that
hopefully
I
met
the
under
five
minutes.
They
did
an
excellent
job
of
that.
Thank
you
very
much,
and
so
we're
going
to
get
Kyle
to
come
up.