►
From YouTube: Ross Turk on Citrix TV Speaking about Ceph
Description
Ross Turk, VP of Community at Inktank, providing an overview of how Ceph provides massively scalable storage on Citrix TV as part of Inktank's partnership with the Citrix CloudPlatform (which is based on Apache CloudStack).
A
So
before
I
talk
about
Seth
I'd
like
to
talk
about
the
current
state
of
the
storage
industry,
it
breaks
down
roughly
into
two
large
categories:
traditional
enterprise,
storage
and
open
source
solutions.
Sure
there's
many
more,
but
these
are
the
two
kind
of
big
areas
of
Technology
today,
with
traditional
enterprise
storage.
What
you
end
up
with
is
very
large
hardware-based
appliances
that
are
expensive
and
there's
single-use
appliances.
So
when
you
buy
a
large
storage
appliance
you're
buying
it
just
for
storage,
it's
not
going
to
be
used
for
anything
else.
A
It's
just
single-use
hardware
and
it's
very
expensive.
Another
thing
that
we've
noticed
is
that
a
traditional
hardware
based
storage
tends
to
have
been
built
for
a
scale
up
world,
not
a
scale-out
world
and
the
scale
out
offerings
that
a
lot
of
the
traditional
storage
vendors
have
seemed
to
be
kind
of
an
awkward
retrofit.
They
don't
seem
to
really
be
built
for
a
scale-out
world.
A
They
seemed
still
to
be
built
for
a
scale
up
world
and
that's
they're,
very
good
at
scale
up,
but
they're
not
really
built
for
scale
out
and
as
a
result
of
that,
particularly
when
we're
talking
about
the
cloud
as
a
sort
of
topic,
the
scalability
becomes
limited
very
very
quickly.
In
addition,
these
technologies
were
completely
not
developed
with
cloud
scale
in
mind.
A
They
were
developed
with
enterprise-scale
in
mind,
or
you
know,
web
or
Internet
scale
in
mind,
but
not
with
cloud
scale
in
mind
and
cloud
scale
has
got
this
massive
scale
out
sort
of
sort
of
idea
and
a
lot
of
the
traditional
storage
systems
haven't
really
thought
about
it.
That
way.
On
the
other
hand,
we
have
a
bunch
of
open-source
storage
solutions
like
cluster
and
lustre,
and
things
like
that-
and
these
were
usually
built
to
solve
a
particular
problem.
For
example,
Gluster
and
lustre
were
built
as
distributed
file
systems
to
solve
that
particular
problem.
A
There
are
other
open-source
storage
systems
that
were
built
to
solve,
object,
storage
needs,
but
they
were
really
just
built
to
solve
a
particular
problem
and
there
are
usually
some
challenges
in
how
you
manage
them
and
how
reliable
they
are
and
how
you
can
get
them
to
perform
way.
You
want
them
to
perform
so
we're
finding
that
there's
still
a
lack
of
maturity
in
open
source
storage
solutions,
so
to
introduce
SEF
I'd
like
to
talk
about
a
few
things
that
created
the
environment
where
SEF
can
thrive,
and
the
first
is
an
explosion
of
storage
needs.
A
People
are
storing
so
much
more
information
than
they
were
storing
before
by
just
a
completely
different
scale,
and
the
cloud
is
driving
this.
The
Internet
in
general
is
driving.
This
and
economic
factors
are
also
driving.
This
virtualization
in
the
cloud
is
another
thing
that
makes
Seth
really
necessary
because
you
have
again
pushing
this
scale
out
idea,
as
opposed
to
the
scale
up
idea.
When
you
start
virtualizing,
you
have
all
kinds
of
virtual
machine
disk
images
to
worry
about,
and
things
like
that.
It's
a
totally
different
type
of
storage
need.
A
On
top
of
that,
you
have
the
efficiency
required
by
the
cloud.
You
need
to
be
able
to
spin
up
and
tear
down
this
storage
instantly
or
near
instantly
in
order
for
it
to
work
for
the
cloud.
It
also
needs
to
be
very
easy
to
operate.
If
it's
not
super
super
simple
to
operator
super
simple
to
script
and
get
your
DevOps
things
working,
then
it
won't
work
in
the
cloud
environment
and,
of
course,
there's
always
cost
pressure,
especially
now
the
things
are
scaling
out,
there's
always
cost
pressure.
A
Of
course
it
also
has
to
be
readily
available
if
you're
doing
an
internal
cloud
deployment,
for
example-
and
you
need
you
know,
ten
petabyte,
n,
terabytes
of
storage-
it's
much
better
if
you
can
go
build
that
storage
network
without
talking
to
a
vendor
without
having
to
sign
contracts
or
ndas
or
anything
like
that,
you
need
readily
available
technology
and
that's
in
part
pushing
this
demand
for
open
Technology,
open
source
technology
is
readily
available
and
that's
what
makes
it
so
powerful,
so
intercept.
Sef
is
an
open
source
solution.
A
It's
what
we
call
software-defined
storage,
which
is
kind
of
buzz
word
compliance
people
are
talking
about
software-defined
networking.
This
is
software-defined
storage.
It's
a
storage
solution
that
works
based
on
just
its
software,
which
means,
of
course,
that
it
runs
on
commodity
hardware
and
by
commodity
hardware.
I
mean
every
day
hardware
hardware
you
can
get
from
a
from
a
vendor
like
Dell
or
HP,
or
a
vendor
like
that,
and
it
is
efron's
on
top
of
this
commodity
hardware.
In
addition,
seth
is
a
universal
storage
system.
A
It's
not
just
a
distributed
file
system,
it's
not
just
scale
out
as
or
or
I
scuzzy.
It's
not
just
an
object
store,
it's
all
of
that
its
object
block
and
file
in
a
single
storage
system,
which
makes
it
really
powerful.
It's
also
super
easy
to
provision
when
you
want
to
add
new
plus
new
nodes
to
your
cluster.
You
spin
up
new
software
onto
commodity
hardware
on
commodity
hardware.
A
It's
it's
a
lot
simpler
than
it
is
to
say,
for
example,
double
your
your
net
up
capacity
and
it's
self
managing
when
nodes
go
up
and
nodes
go
down,
SEF
understands
how
to
rebalance
the
network.
It
understands
how
to
keep
operations
up
in
a
degraded
fashion.
It
understands
how
to
how
to
grow
a
cluster,
to
take
up
all
available
hardware
and
still
get
a
distribution
of
data
that
makes
sense
and
it's
massively
scalable.
There
is
no
part
of
the
SEF
architecture
that
was
not
built
to
scale
out
and
that's
something.
A
That's
really
important
when
you're
looking
at
a
cloud
scale
distributed
storage
system
and,
of
course,
in
order
to
accomplish
this,
it
has
to
also
have
no
single
point
of
failure.
Now,
this
all
kind
of
sounds
too
good
to
be
true
right,
software-defined,
storage
on
commodity
hardware,
Universal,
object,
block
and
file
storage,
rapid
provisioning,
self-managing
and
completely
scalable
with
no
single
point
of
failure
sounds
like
a
really
tall
order
and
that's
what's
F
is
trying
to
try
to
accomplish
and
and
has
so.
A
This
is
the
SEF
Universal
storage
system,
it's
a
universal
distributed,
object
store,
and
on
top
of
that
object
store,
we
have
object,
interfaces,
block
interfaces
and
file
interfaces,
so
we
can
really
target
six
types
of
data
uses.
The
first
is
object.
Storage
set
provides
object,
storage
that
is
compatible
with
the
s3
and
Swift
api's,
which
allows
you
to
build
applications
that
can
store
assets
really
efficiently
without
without
a
whole
lot
of
developer.
Overhead
cloud
storage
is
another
thing
that
Seth
does.
A
It
has
a
virtual
block
device
that
allows
you
to
put
virtual
machine
images
inside
Seth
and
use
it
to
spin
up
VMs.
It
can
also
do
enterprise
storage
again,
with
virtual
block
devices
for
the
I
skazhi,
also
a
distributed
file
system
that
allows
you
to
have
sort
of
work
group
servers
and
that
sort
of
thing
it
can
also
support
high-performance
computing.
It's
got
a
drop-in
replacement,
shim
for
HDFS
and
Hadoop,
for
example,
and
a
lot
of
people
that
were
talking
to
about
separate
using
it
in
hpc
context.
A
Big
data,
as
well
again
with
the
Hadoop
plugin,
can
also
handle
sort
of
big
data
for
massive
amounts
of
data
operations
and,
finally,
application
storage,
and
this
is
something
not
a
lot
of
people.
Think
about.
What's
F
is
that
seth
is
also
an
application
platform.
You
can
build
applications
on
top
of
SEF,
using
the
underlying
object
store
as
a
powerful
storage
mechanism.
The
same
way,
you
would
think
about
my
sequel.
You
can
think
about
SEF
as
part
of
your
application
stack,
so
SEF
is
integrated
into
cloud
platform
with
multiple
storage
methods.
A
At
this
point,
you
can
integrate
the
block
storage
as
one
of
your
primary
storage
pools
for
a
44
cloud
platform.
You
can
also
have
object,
storage
via
the
s3
api's
or
programmatically
through
the
language
bindings
thats
FS.
So
we
have
both
block
and
object
inside
cloud
platform
integrated.
Today
we
also
have
file
system
storage
available.
If
you
want
to
provide
a
shared
file
system
to
the
virtual
machines
in
your
cloud
or
to
other
clients
in
your
cloud
and
again,
it's
all
open
source
software
on
commodity
hardware.
A
This
technology
has
been
validated
by
ink
tank
and
Citrix.
Both
the
block
storage
capabilities
are
built
into
cloud
platform.
Today,
the
services,
support
and
training
are
available
from
ink
tank
and,
of
course,
we're
Citrix
ready
and
verified
for
cloud
platform.
So
going
into
a
little
bit
of
detail
about,
what's
F
actually
is
Seph
is
at
its
most
base
level,
ray
dose,
which
is
the
reliable
autonomic
distributed,
object
store.
A
So
we've
taken
the
opposite
approach
from
a
lot
of
universal
storage
stacks,
in
that
we
built
the
object,
store
first
and
built
everything
else
on
top
of
the
object
store.
So
one
of
the
things
we
built
on
top
of
the
object
store
is
liberate
us,
which
is
language
bindings
for
C,
C++,
Java,
Python,
Ruby
and
PHP
that
allow
you
to
get
really
rich
access
to
the
underlying
object
store,
but
that's
not
all
that
we've
built
as
an
object
store
in
the
library,
that's
cool,
but
that's
not
everything.
A
On
top
of
that,
we've
also
built
rados
gateway,
which
is
a
bucket
based
rest
gateway,
that's
compatible
if
s3
and
Swift,
so
it
has
buckets,
it
has
accounting.
It
has
everything
you
would
want
for.
Essentially,
building
an
s3
competitor
and
the
case
that
I'll
be
presenting
later
is
somebody
who
did
exactly
that.
They
built
an
s3
competitor.
The
next
piece
is
rbd,
which
is
short
for
the
ratos
block
device,
greatest
block
devices,
a
reliable,
fully
distributed
block
device
with
a
linux,
kernel,
client
and
boot,
support
in
q,
mu
and
kvm.
A
So
the
upshot
of
this
is,
you
can
take
a
disk
and
you
can
split
it
into
a
bunch
of
pieces.
A
bunch
of
tiny
blocks,
spread
it
throughout
the
object
store
so
that
you
have
massive
parallelism
on
reads
and
writes.
Then
you
can
map
that
to
a
linux
device
and
pass
it
to
a
virtual
machine
and
boot
off
of
it.
A
Two
different
metadata
servers
instantly,
so
sefa
FS
is
probably
the
most
advanced
distributed
file
system
out
there.
It's
definitely
the
one
that's
been
built
from
the
ground
up
to
scale
best.
So
why
SEF
SEF
is
sepsis
is
again
an
open
source
project,
so
we
don't
know
who's
using
it
because
everybody
already
has
it.
People
come
to
ink
tank
all
the
time
and
they
say:
where
can
I
buy
SEF
and
I
say
to
them?
You
can't
you
already
own.
A
It
because
the
whole
world
already
owns
it,
it's
open
source,
but
we
do
know
of
a
couple
of
installations
that
are
that
are
really
powerful.
First
off
chef
is
powering
dream
host,
which
is
a
service
provider
based
in
Los
Angeles
with
their
s3
competitor,
competitive
product
they're,
also
powering
the
products
of
piston
cloud
and
meta
cloud
and
a
lot
of
other
cloud
service
providers.
So
SEF
is
currently
seeing
a
lot
of
adoption
amongst
those
who
are
building
public
and
private
clouds.
A
Another
thing
that's
interesting
about
Seth.
Why?
Why
why
you
should
pay
attention
to
SEF?
Is
that
Mark
Shuttleworth
I?
Don't
know
if
you're
familiar
with
him,
but
Mark
Shuttleworth
is
the
co-founder
of
canonical
he.
He
went
to
space
and
he
gave
us
a
million
dollars
to
make
sure
that
Seth
grew
as
a
community
and
that
the
software
became
more
more
mature.
So
we
have
a
lot
of
investment
from
the
open
source
community.
A
lot
of
investment
from
open
source
luminaries.
A
Also
451
research
had
this
to
say
about
us:
CEFs
scale-out
architecture
and
its
ability
to
provide
file,
block
and
object
storage
should
attract
interest
from
service
providers
and
enterprises.
Two
things
were
mentioned
in
the
451
report
in
this
quote
directly
the
scale
out
architecture
and
the
ability
to
provide
file,
block
and
object
storage
in
the
same
cluster,
the
unified
storage
platform,
so
the
industry's
also
been
fairly
enthusiastic
about
commercial
support
offerings
for
Seth.
When
we
launched
ink
tank
in
May
to
provide
support
for
Seth,
we
were
put
on
a
few
different
lists.
A
So
dreamhost
is
our
parent
company
and
there
also
are
our
best
customer,
which
is
kind
of
like
getting
a
car
loan
from
your
dad.
But
it's
still
a
very
good
validation
of
the
technology.
So
dreamhost
approached
the
industry
with
a
few
challenges
and
opportunities.
The
first
was
they
needed
storage
that
could
be
deployed
at
large
scale
again
with
the
scale
I'm
going
to
talk
about
scale
all
day
scale
scale
scale.
That's
all
it
is.
They
needed
storage
that
could
deploy
at
large
scale.
A
They
also
want
to
make
sure
it
can
be
brought
to
the
market
very
very
quickly
because
they
didn't
feel
like.
There
was
a
lot
of
time
to
waste
getting
into
public
cloud
offerings
today,
so
they
wanted
something
that
was
that
could
be
scaled.
That
could
be
cheap
and
that
could
be
fast
and
you
usually
don't
get
all
three.
They
also
wanted
something
very
particular,
and
this
is
something
special
for
dreamhost.
A
They
wanted
to
make
sure
people
could
deploy
applications
against
dream
objects,
which
is
their
s3
competitive
product,
and
then
they
could
move
those
objects
to
their
own
deployment
if
they
wanted,
so
they
want
to
make
sure
whatever
tool
they
chose
could
be
deployed
by
people
on
their
own
infrastructure
if
they
wanted
to,
and
there
was
no
lock-in.
It's
very
important
to
dreamhost.
So,
of
course,
the
solution
that
they
ran
into
that
they
chose
was
SEF,
so
dream
objects.
Is
there
is
their
product?
It's
a
cost-effective
public
cloud
storage
service,
it's
based
on
SEF.
A
It
has
API
compatibility
of
s3
and
Swift
through
rado
skateway.
So
really
what
dreamhost
deployed
is
just
seth
with
web
load
balancers
in
the
front,
and
then
they
built
some
user
interface
items
to
allow
you
to
manage
your
account
and
check
your
usage
and
utilization
and
browse
through
your
objects
and
buckets
and
deal
with
authentication.
A
But
really
it's
seth,
pretty
much
seth
when
you,
when
you
make
a
request
to
dream
objects,
you're
hitting
ratos
gateway
and
the
result
is
that
they
were
able
to
launch
rapidly
and
they
were
able
to
launch
at
a
very
large
scale.
They
have
a
3
terabyte
working
cluster
for
dream
objects
today
and
they're
able
to
keep
the
cost
very
low.
It's
just
seven
cents
per
gigabyte
per
month,
so
how
to
get
started
with
Seth
I'm
sure
everybody
wants
to
go
download
it
right
away,
and
the
good
news
is
that
you
can.
A
The
technology
is
out
there
and
it's
freely
available,
so
you
can
download
stuff.
You
can
try
it
today.
If
you
visit
SEF
com,
you
can
get
SEF
deployed
and
starting
to
store
data
in
just
a
few
minutes
in
the
documentation.
There's
a
five
minute,
quick
start
guide
that,
quite
literally,
if
you
have
an
Ubuntu
thing
already
installed,
is
a
five
minute.
Quick
start
guide,
it'll
get
stuff
up
and
running
in
five
minutes.
A
Also
we'd
like
people
to
visit,
ink
tank
calm
to
understand
the
company
behind
SEF
the
company
that
will
stand
behind
seven
that
will
provide
supporting
services
consulting
and
training
to
help
you
have
a
successful
deployment
of
Seth.
So
in
summary,
Seph
is
massively
scalable,
open
storage
for
your
cloud,
massively
scalable,
no
single
point
of
failure.
Every
component
was
built
to
be
scaled
out.
It's
open
storage,
meaning
that
it's
software
based
you
running
on
commodity
hardware,
the
software
is
free,
the
software
will
always
be
free.
A
Seph
is
integrated
into
cloud
platform
today
to
provide
block,
object
and
file
storage,
not
just
one
kind
of
storage,
all
kinds
of
storage
in
the
same
cluster,
so
you
manage
it
once.
Instead
of
managing
it
three
times,
it's
all
open
source,
again,
it's
all
on
commodity
hardware
and
it's
backed
by
ink
tank
with
support
and
services
and
it's
validated
by
Citrix.
Thank
you
very
much.