►
From YouTube: Peter Chadwick @ Ceph Day Paris
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
I
wanted
to
talk
a
little
bit
today
about
how
we're
using
souza
are
using
Seth
and
Susan
cow.
We've
got
a
number
of
customers
are
looking
at
and
rolling
out
an
openstack-based
cloud
infrastructure,
the
private
cloud
and
want
to
use
Seth
as
the
principal
storage
in
the
cloud.
So
I
thought
we'd
talk
a
little
bit
about
that
sometime.
How
many
of
you
are
familiar
with
kind
of
OpenStack
as
a
technology
and
okay?
A
So
OpenStack
is
a
open
source
project
designed
to
help
customers
deploy
infrastructure-as-a-service
clouds
and
it's
got
a
number
of
components,
but
just
like
a
linux
distribution
like
like
red
hat
enterprise,
linux,
razoos,
enterprise,
souza,
linux,
enterprise,
server,
the
the
district
components,
you
know
you
need
more
than
just
the
Linux
kernel
to
do
to
deploy
Linux.
You
know
packages
and
tools
around
that
same
thing
with
OpenStack.
So
on
this
in
this
picture,
the
the
boxes
here
in
orange
are
the
components
of
OpenStack.
A
So
you
can
see,
there's
a
compute
controller
called
Nova,
there's
a
essentially
an
identity
service
that
does
token
authentication
across
all
the
components
need
a
place
to
store
virtual
machine
images.
It's
called
glance
there
actually
is
an
object.
Storage.
That's
that's
integrated
as
part
of
OpenStack.
On
top
of
that,
you've
got
user
interface,
you
have
a
set
of
api's
native
api's
as
well
as
Amazon
ec2.
A
What's
considered
software
software-defined
data
centers,
so
that
you
can
undersell
for
control,
setup,
we're
connections
for
your
virtual
machines
or
define
virtual
devices
to
attach
your
virtual
machines
now
to
make
all
this
work,
you
need
a
message
queue
to
let
the
let
the
actual
components
talk
to
each
other,
and
you
also
need
a
database
to
store
to
store
the
the
state
of
your
clout
and
state
of
the
cloud
that
is,
that
is
both
the
physical
infrastructure,
the
virtual
infrastructures
users,
things
of
that
nature.
So,
essentially
you
describe
all
the
information.
A
The
cloud,
obviously
in
an
operating
system,
you
need
a
hypervisor
to
run
it
on
and
physical
infrastructure,
and
then
we
think
that
there's
a
management
layer
on
top
that
you
can
use
to
to
get
additional
levels
of
control
in
your
infrastructure.
Now
we
were
one
of
the
first
to
cover
the
idea
of
using
what
we
call
an
installation
framework
to
actually
orchestrate
the
physical
installation
of
OpenStack
itself.
A
The
reason
being
is
that
I
think
the
last
count
there's
something
like
1500
parameters
that
you
need
to
specify
when
you
deploy
OpenStack
and
you
have
to
deploy
it
in
specific
order.
If
you
get
it,
if
you
get
a
deployment
wrong,
starting
all
over
is
not
fun,
so
we
kind
of
focused
on
that
I.
Think
most
of
the
major
distributions
now
have
some
sort
of
an
installation
framework
to
simplify
that.
A
Now.
If
you
look
at
OpenStack
I've
highlighted
in
yellow
here,
the
components
of
OpenStack
that
are
related
to
storage,
so
first
of
all
you
have
AP
is
that
are
specifically
storage
API.
So,
for
example,
if
you're
familiar
with
Amazon
s3,
a
set
of
restful
api
is
to
define
objects,
retrieve
objects
from
Amazon,
there's,
actually
an
s3
interface
up
there.
You
store
images
in
the
in
glance,
so
you
have
image.
Repository
operating
system
templates,
you've
got
Swift
for
object,
storage,
then
you've
got
this
block
storage
device
and
cinder.
A
So
what
I
thought
I'd
do
is
just
drive
down
and
talk
about
when
you,
when
you
look
at
cloud
stores,
there's
a
couple
different
use
cases
for
storage
in
a
cloud,
so
the
top
one,
what
we
call
storage
as
a
service
and
you're
all
familiar
with
this,
because
almost
everybody
is
using
Dropbox
or
drive
or
there's
an
open
source
project
called
own
cloud
to
build
your
own
web-based
interface
to
store
folders.
You've
also
got
web-based
object.
A
Storage,
amazon,
s3
is
the
most
prevalent,
but
again,
customers
are
looking
at
setting
up
things
on
their
own
and
then
you
have
storage
for
compute
as
a
service.
So
the
idea
is
I'm.
Doing
infrastructure-as-a-service
I'm
standing
up
virtual
machines
I
need
a
variety
of
storage
components
to
make
all
that
work.
So
the
first
I
need
to
I
need
some
place
to
store
my
virtual
machines
in
the
case
of
OpenStack.
A
That's
glance
you
need
to
have
disks
that
are
attached
to
the
virtual
machines
and
you
can
think
of
Amazon
EBS
from
elastic
block
storage
is
designed
to
do
that.
So
when
you,
when
you
spin
up
a
virtual
machine
and
ec2
you
can,
you
can
attach
a
disk
to
that
and
then
under
here
this
long
term,
which
is
referred
to
as
ephemeral
disk.
So
the
idea
is
that
when
you
start
up,
you
have
some
amount
of
disk
space,
but
it
goes
away
and
then
you
have
long-term,
so
virtual
block
devices.
A
So
that's
kind
of
the
three
areas
that
we
look
at
when
people
talk
about.
We
talk
to
customers
that
what
they
want
to
do
with
with
storage.
So
this
is
kind
of
a
high
level
view
of
the
OpenStack
architecture.
So
you
see,
you've
got
the
dashboard
up
top,
which
is
a
user
interface
for
both
the
end
user
and
for
the
administrator.
You've
got
networking
components,
you've
got
compute
components
and
you've
got
identity
timeless
all
together,
and
then
you
have
block
storage,
which
is
where
you
get
volumes
that
are
attached
to
the
compute
nodes.
A
A
It
only
was
used
for
for
backing
up
glance,
and
then
you
could,
but
you
can
also
use
it
as
I
talked
about
the
object
stores
as
a
service,
so
that
kind
of,
as
a
background,
still
a
little
bit
more
go
through
some
of
the
components
in
more
detail,
and
so
Swift
is
an
optic
storage.
It's
a
key
core
component
of
OpenStack.
If
you
know
the
history
of
OpenStack,
it
was.
The
project
was
kicked
off
by
nasa
in
the
US
and
rackspace,
which
was
a
US
service
provider.
A
Swift
was
actually
the
major
contribution
that
Rackspace
made
to
the
original
code.
It
is
object,
storage.
Only
there's
no
other
interfaces
to
it.
It
is
fully
integrated
with
the
rest
of
OpenStack.
So
as
as
soon
as
you
register
a
user
for
OpenStack
to
create
virtual
machines
to
run
to
run
workloads,
they
also
have
access
to
a
oops
Swift.
If
you've
turned
that
on,
as
we
show
the
previous
slide,
it
can
be
used
as
a
back-end
for
storing
images.
So
you
can
store
the
images.
Virtual
machine
images,
operating
system
templates,
water
called
instance.
A
Types
are
flavors
actually
in
OpenStack,
so
you
define
the
size
of
the
virtual
machine,
how
much
memory
it
has
things
like
that.
That's
all
stored
in
glance.
It
is
redundant
and
similar
to
SEF.
It
just
writes
objects
to
multiple
multiple
servers
in
the
in
the
Swift
cluster
replications
across
the
act
of
notes
and
it
again
similar
to
the
SF.
It
scales
horizontally.
It's
it's
based
on
commodity
hardware
and
is
designed
to
be
inexpensive.
A
It
is
accessed
by
by
a
restful
api
and
there's
two
available
there's
either
the
native
OpenStack
API,
the
native
Swift
API,
and
then
it
also
supports
amazon
s3
and
the
key
here
is
it's
accessible
from
anywhere,
so
it
doesn't
require
you
to
have
a
virtual
machine
in
the
cloud
to
get
access
to
the
objects
that
are
stored
there.
So
you
can
actually
roll
it
out
as
a
standalone
storage
cloud
once
you
create
an
object,
its
persist
until
it's
deleted
by
the
by
the
user.
A
You
need
there's
no
real
limit
on
the
size
of
the
object
or
on
the
on
the
number
of
objects
that
you
can
store.
You
know
similar
to
Amazon
s3
you
you,
you
set
up
a
container,
then
you
fill
up
that
container,
but
you
determine
how
how
big
that
is,
and
the
user
can
fill
up
up
to
their
to
their
total
quota.
A
A
A
So
now
I
want
to
talk
about
image,
storage
for
compute
as
a
service
and
again
it's
repository
for
cloud
workloads.
So
if
you
think
about
it,
when
you
go
on
Amazon
ec2,
when
you
create
an
image,
use
you
get
in
your
dashboard
and
you
can
see
all
the
different
instances.
You've
defined
you
launch
instances
they're
coming
out
of
an
image,
repository
same
same
kind
of
interface,
that
you
have
the
same
interface
when
you
when
you're
running
OpenStack
and
there's
two
ways
that
there's
two
types
of
things
that
get
stored
in
there.
A
There's
images
that
are
created
by
the
the
administrator,
so
its
operating
system,
templates
or
operating
system
images
plus
templates
on
how
big
the
MSR,
how
many
virtual
CPUs,
how
much
RAM,
how
much
how
much
temporary
describe
and
then
as
users
create
virtual
machines,
they're
actually
stored
back
into
into
glass
as
well
glance
is
the
service
that
manages
it
and
you
can
have
multiple
different
backends
from
local
disks.
All
the
way
to
in
the
latest
release
of
OpenStack.
A
You
can
use
VMware
vCenter
store
your
images,
so
we've
got
quite
a
lot
of
flexibility
into
what
you
use
behind
glance.
What's
really
what
I
really
like
about
it?
Is
it
inherently
as
they
builded
snapshot
copy
clone
images,
so
that
you
can
really
maintain
your
image?
Isn't
up
in
a
pretty
in
a
pretty
straightforward
way:
lots
of
capability?
A
It
also
is
responsible
for
connecting
images,
moving
images
from
from
the
repository
on
to
compute
nodes
when
you
actually
want
to
go
run
when
you
want
to
go
run
an
instance,
it
does
maintain
image
made
it
metadata,
which
is
important.
When
you
want
to
go
start
launching
things,
you
may
have
a
specific
image.
That's
only
you
only
want
to
run
on.
You
know,
for
example,
Intel
hardware,
because
intel
has
specialized
instruction
sets
you
want
to
take
advantage
of.
So
you
need
to
understand
the
you
know
kind
of
more
information
about
about.
A
What's
actually
on
the
image
and
there's
an
API
available
that
you
can,
you
can
use
to
query
and
control
the
the
images
directly
now.
Let
me
talk
about
so
that's
kind
of
sort
of
administrative
use,
I
think
of
the
supporting
virtual
machines,
but
if
you
look
at
virtual
machines
and
you're
running
them
in
a
cloud,
you
want
them
to
access
storage.
There's
three
different
ways:
they
can
do
that.
A
So
on
this
side
you
can
use
a
standard,
swiftor,
s3
interface,
to
access
the
optic
store
and-
and
this
looks
just
like
any
other
virtual
machine,
your
kind
of
actually
going
out
of
the
public
network
or
out
on
the
internal
network,
internal
IP
network
and
coming
back
into
your
object,
store
to
get
that
I'm,
not
sure
that
a
lot
of
people
look
at
using
it
that
way.
But
it's
it
is
possible
to
do
that.
The
next
month
was
called
in
femoral
the
femoral
drives
and
those
are
typically
disk
space.
A
That's
on
the
physical
server
that
you're
deploying
the
virtual
machine
on
and
it's
it's
it's
a
sign
as
the
as
part
of
the
image
definition.
So
if
you
say
you
know,
your
image
is
defined
to
have
a
10,
gig,
ephemeral
disk.
When
you
start
your
image,
you've
got
10
gigs
there,
that's
typically
where
your
operating
system
is,
but
the
key
difference
is
once
you
shut
down
the
virtual
machine,
you
lose.
You
lose
that
disk
drive
once
you
just
the
virtual
machine.
The
disk
drive
goes
away,
then
the
other.
A
The
other
way
to
get
access
is
through
persistent
block
devices.
So
cinder
is
as
a
as
another
service
as
part
of
OpenStack
that
you
can
think
of
as
block
device
as
a
service.
So
the
way-
and
this
works
is
the
virtual
machine
or
the
you
know
says:
I
need
a
you
know:
I
need
a
disk
drive
with
60
gigabytes,
there's
a
plug-in
that
can
talk
to
an
external
array.
Configure
the
60
gigabyte
drive
give
that
information
back
to
the
nvm
then
connects
to
it
directly.
So
it's
not
a
it's,
not
a
data
path.
A
Typically,
it's
used
to
run
the
operating
system
by
default,
that's
kind
of
where
it
gets
where
it
gets
stored
and
it's
accessed
via
a
file
system.
So,
if
you're
running
Linux,
you
can
use
any
of
the
supported
file
systems
available
in
the
linux
distribution
that
you're
that
you're
running
it's
only
accessible
from
within
a
vm
from
one
vm
and
it's
only
accessible
and
and
persists
as
long
as
that,
virtual
machine
is
defined.
As
soon
as
you
terminate
that
virtual
machine,
you
lose
all
that
any
data
that
was
on
that
describe.
A
So
the
virtual
block
device
is
really
where
most
customers,
what
most
customers
use,
they
define
a
block
device
attached
it
to
a
virtual
machine.
It
looks
just
like
if
you
had
a
physical
physical
device
attached.
The
key.
The
key
difference
is
its
defined
by
the
user
through
the
dashboard.
So,
as
a
user
you
can
you
can
log
in
and
say:
ok
I
want
to
set
up
a
you
know
a
20
gigabyte
drive
and
you
can
attach
it
to
any
virtual
machine.
A
You
know
one
virtual
machine
at
a
time,
but
you
can
set
it
up
on
a
virtual
machine,
run
it
for
a
while.
If
you
terminate
that
virtual
machine,
you
don't
lose
the
data.
So
if
you
have,
for
example,
if
you're
using,
if
you
have
a
workload
that
you
run,
you
know
for
a
couple,
you
know
once
or
twice
a
month
you
may
not
want
to.
You
know,
keep
the
virtual
machine
lying
around
all
the
time,
but
you
need
to
have
the
data
available.
You
start
up.
A
A
virtual
machine
attach
the
attached
that
this
to
it
do
whatever
you
need
to
do
and
shut
everything
down,
but
you
keep
that
you
keep
the
disk
drive.
It
does
have
integrated
snapshot
management.
It's
a
very
powerful
feature.
If
you're,
if
you're,
if
you're
running
multiple
virtual
machines,
you
can
take
snapshots,
make
changes
start
them
up
again.
It's
like
access
as
a
block
device,
so
it
looks
like
a
block
device
to
the
to
the
virtual
machine,
and
then
you
can
again
use
it.
Formatted
use
it
through
any
typical
file
system.
A
Again,
it's
only
accessible
from
within
a
virtual
machine,
but
it
does
support
external
storage
through
vendor
provided
plugins.
So
how
does
you
know
why
our
customers
interested
in
Seth?
Why
were
we
interested
in
Seth
is
because
it
can
really
do
all
of
these
things.
So
if
you're
setting
up
a
cloud
it
can
handle
all
three
of
the
use
cases
I'm
not
going
to
go
in
a
whole
lot
of
detail
about
what
Seth
is.
You
know,
obviously
its
object
and
block,
and
it's
in
a
single
system,
ultimately
files
as
well.
A
It
has
long
been
integrated
with
OpenStack.
Most
recently,
it's
added
cinder
black
back
in,
but
it's
always
supported
glance.
According
to
some
analysts,
we've
talked
to
most
OpenStack
deployments,
actually
use
Seth
or
more
youssef
than
use
use
swift
for
object,
store
it
fully
integrates
with
the
compute
environment
for
automated
provisioning.
It
has
the
RESTful
API
that
is
s3
or
Swift
compliant.
So,
if
you
have,
if
you
have
workloads
that
have
been
written
for
OpenStack
Swift,
you
can
actually
migrate
those
directly
on
top
of
on
top
of
SEF
same
thing
with
s3.
A
For
our
perspective,
we
just
released
souza
cloud
4
and
a
fully
integrated
or
fully
supports
f
as
part
of
that
as
part
of
that
product.
So
what
does
this
look
like?
There's
a
couple
different
things
so
same
picture
here,
you've
got
you've
got
cinder.
You've
got
a
femoral
storage.
You've
got
object
if
you've
got
an
object,
storage.
It
just
comes
in
through
rato
skateway
talks
directly
to
rato
stalks,
you
know,
gets
access
to
the
objects
you've
defined
in
the
in
the
SEF
cluster.
A
If
you
want
to
use
it
for
persistent
storage,
you
use
rbd
the
Ronald's
block
device.
So
again,
it
emulates
a
block
driver
sitting
on
top
of
the
SEF
cluster,
fully
integrated
with
cinder
through
through
a
standard
plug
in
so
same
kind
of
processes.
You
would
use
for
any
other
external
array.
Virtual
machine
says:
I
need
a
disk
cinder
talks
to
Seth
talks.
The
rbd
says
give
me
a
disk
once
it's
got
it.
A
The
vm
then
talks
directly
to
two
rbd
through
liberato
and
through
drivers
that
are
integrated
on
the
operating
system,
and
so
this
is,
you
know
again,
just
just
going
back
to
the
to
the
picture
of
the
opens
OpenStack
distribution.
This
is
souza
cloud
four
runs
on
suse
linux,
enterprise
server
11's.
We
support
both
then
and
kvm
natively
on
suse
linux,
for
the
required
services.
A
We
ship
rabbitmq
for
the
message,
queue
and
postgresql
for
the
state
database,
and
then
we
have
an
installation
framework
and
I'll
show
you
a
couple
screenshots
of
that
based
upon
open
source
project,
crowbar
+
chef,
some
other.
You
know
typical
things,
and
then
we
support
pretty
much
any
x86
64
server
for
compute
or
storage
nodes.
We
also
support
pretty
much
all
the
major
Network
and
an
external
storage
vendors.
A
Well,
we
also
support
one
key
differentially.
We
support
VMware
and
hyper-v
as
as
hypervisors
for
compute
nodes,
and
then
we
also
automatically
configure
all
of
the
control
services
for
OpenStack
to
be
highly
available.
So
we
we
deploy
a
pacemaker
cluster
and
then
lay
the
services
on
top
of
that,
most
of
our
customers
were
telling
us
that
doc
going
to
run
they're,
not
going
to
run
a
private
cloud
without
without
high
availability.
A
A
A
So
this
is
what
it
would
look
like.
This
is
a
picture
from
our
installation
framework
and,
as
you
can
see
you,
you
deploy
things
in
the
order
in
which
they
show
up
on
the
screen.
So
the
first
thing
you
put
is
the
database
because
obviously
all
state
needs
to
get
put
their.
You
need
authentication,
because
everything
needs
to
to
be
authenticated
all
trend.
A
Then
once
you've
got
the
once
you've
got
that
set
up.
You
set
up
the
message
queue.
The
next
thing
you
do
is
you
install
you
install
your
on
fixed
or
because
the
next
thing
to
do
is
install
glance
and
glance
depends
upon
which
objects
door.
You
want
to
use
as
the
back
end,
so
you
set
up
Seth.
Then
you
set
up
glance.
Then
you
can
set
up
the
rest
of
the
rest
of
the
functions.
A
So
this
is
an
example
of
setting
up
installing
Seth
as
part
of
this
d'souza
cloud.
Through
the
installation
framework,
you
can
see
that
you've
got
you've,
got
nodes
that
are
available,
haven't
necessarily
being
assigned
to
roles,
and
then
you
just
you
just
do,
drag
and
drop
and
say:
okay,
I
want
these
to
be
active.
Sef
monitors
I
want
these
to
act
as
as
as
Oh
SDS.
Obviously
we're
overlay
overloading
functionality,
because
this
is
done
on
a
on
a
virtualized
environment.
A
A
Once
you
install
the
once,
you
install
set
this
automatically
pops
up,
and
then
you,
you
configure
you
configure
Otto's
for
authentication,
so
essentially
all
of
the
parameters
that
you
need
to
put
into
the
to
put
in
the
environment
to
configure
a
glance
to
work
with
with
SEF
are
populated
and
then
for
cinder
same
thing.
So
when
you
say
I
want
to
add
I
want
to
add
a
dirt
to
cinder
and
you
can
add
multiple
adapters.
A
So
just
as
a
quick
summary,
you
know,
when
you
look
at
deploying
a
cloud,
you
understand
what
kind
of
use
cases
you
want
to
support.
You've
obviously
got
sindhu
got
Swift,
but
we
think
that
most
of
our
customers
are
going
to
be
going
with
SEF,
because
it
it's
a
single
back-end
that
addresses
all
three
use
cases
and
we
also
think
it's
going
to
overtime,
be
the
most
cost-effective
and
the
highest
performance
solution
out
there
and
that's
so
that's
kind
of
what
we've
been
doing
with
the
integration.
So
any
questions.