►
Description
Ceph Update: Erasure Coding and Tiering features Ross Turk from Inktank.
A
Can
you
hi
hello,
I'm,
a
roster,
tremaine
cake
and
I
going
to
talk
to
you
today
about
what's
coming
up
in
the
next
version
of
ink
thanks,
F
enterprise
and
talk
a
little
bit
about
what's,
F
is
and
how
you
use
SEF
to
solve
some
storage
problems.
First,
a
little
bit
about
Seth
Seth
is
different
from
traditional
enterprise
storage.
In
a
couple
of
big
ways.
First,
traditional
enterprise
storage
is
usually
single
purpose.
A
You
buy
one
thing
for
object,
storage,
one
thing
for
block
one
thing
for
a
particular
use
case
and
the
solution
has
to
usually
match
the
use
case,
whereas
SEF
is
a
multi-purpose
unified
platform
that
allows
you
to
build
a
lot
of
different
types
of
storage
with
the
same
software.
Additionally,
traditional
enterprise
storage
is
usually
a
hardware
solution.
It
involves
buying
something
and
bringing
it
into
your
data
center
for
the
forklift
Seph
is
software
the
software
based
solution
and
it's
a
distributed
software
solution
with
traditional
enterprise
solutions.
A
You
end
up
with
some
vendor
lock-in,
because
if
you
buy
your
first
one
petabyte
from
a
storage
vendor,
you
have
to
go,
buy
your
second
petabyte
from
the
same
storage
vendor
to
expand
your
cluster.
But
with
SEF
it's
open.
You
can
use
CEFs
open
source
software
with
any
hardware
vendor
to
to
do
what
you're
looking
to
do
and
another
difference
between
traditional
storage
and
stuff
is
that
traditional
stores
usually
has
a
limit.
A
Most
traditional
storage
products
have
a
limit
of
how
big
they
can
get
and
stuff
has
no
practical
limit
in
it
scales
up
to
the
X
of
light
and
beyond,
although
we
don't
have
proof
of
that,
because
no
one's
taking
it
beyond
exabyte.
Yet,
but
theoretically,
there's
no
single
point
of
failure
and
and
and
everything
scales
out,
the
stuff
architecture
is
broken
up
into
sort
of
three
different
layers
at
the
bottom.
In
the
storage
layer
is
what
we
call
ray
dose,
which
is
the
ceph
object
store,
and
it's
made
up
of,
monitor
nodes
and
object.
A
Storage
demons
above
that
are
the
interfaces
where
there's
an
object
of
block
and
a
file
system
interface
that
allow
you
to
take.
What's
in
that,
set
that
object
store
and
expose
it
to
a
variety
of
applications
like
applications
using
s3
or
Swift,
api's
hosts
and
hypervisors,
needing
storage
for
virtual
disks,
I
scuzzy
targets,
sips
and
NFS
clients,
and
an
SDK
that
you
can
use
to
talk
directly
to
the
storage
cluster
with
OpenStack
you
essentially
a
keystone
and
Swift
are
integrated
into
OpenStack
for
object.
A
Storage
through
the
SEF
object
gateway,
which
we
call
the
radius
gateway,
cinder
glance
and
Nova
are
integrated
with
Steph
through
the
SEF
block
device,
which
allows
storage
of
images
volumes
snapshots.
It
allows
you
to
run
virtual
machines
off
of
storage,
that's
distributed
through
an
entire
storage
cluster,
and
I'm
not
sure
if,
if
anybody
has
heard
the
news
but
ink
tank,
the
company,
that's
the
principal
sponsor
stuff
was
just
recently
acquired
by
red
hat,
which
expands
Red,
Hat's
footprint,
software-defined
storage
and,
as
a
result,
Steph
is
also
well
integrated
with.
Well.
A
Actually,
it's
not
even
as
a
result.
This
happened
months
ago.
This
stuff
was
integrated
with
Rallo,
SP
and
and
certified
for
use.
So
a
little
bit
about
chefs
roadmap.
Pardon
me
ink
tank
stuff
enterprise.
12
is
about
to
come
out
in
a
couple
of
weeks,
keep
your
fingers
crossed
and
what
it
what
it
brings
support
for
l7,
which
is
a
big
deal,
two
features
which
I'm
going
to
talk
a
little
bit
more
detail
about
erasure
coding
and
cached
hearing
and
some
additional
quotas
and
on
the
calamari
side.
Calamari
is
the
interface
force
F.
A
It's
the
management
console
for
Seth
on
the
calamari
side,
we're
introducing
more
management
features.
Let's
talk
about
cheering
a
little
bit
with
the
stuff
storage
cluster.
You
have
the
ability
to
define,
pools,
different
pools
of
storage.
The
latest
version
of
stuff
allows
you
to
define
or
to
designate
one
pool
as
being
a
cash
for
another
pool
which
gives
you
this
cash
steering.
A
You
can
also
put
this
cash
into
a
read-only
mode
where
it
will
speed
up
reads,
but
writes
will
still
go
through
to
the
original
Parent
pool.
So
using
this
cash
steering
feature
in
sep,
which
is
new,
you
can
define
different
different
types
of
speed
for
different
types
of
storage
needs,
particularly
it's
it's
good.
If
you
have
a
back-end,
that
is
a
razor
coded,
for
example,
and
then
a
front
ash
to
your
pool
in
front
of
that.
That
is
on
solid
state
drives,
so
you
can
get
speed
and
cost-effectiveness.
A
The
second
major
feature
that's
coming
out
in
ink
tanks.
F,
Enterprise
12
is
erasure
coding.
The
normal
stuff
storage
cluster
pool
that
has
been
defined
of
all
up
to
this
point
is
a
replicated
pool,
which
means
that
when
you
put
an
object
into
that
pool
to
store
it,
it's
going
to
make
n
number
of
copies
of
that
object.
The
default
is
three
so
that
you
essentially
need
to
have
three
times
as
much
storage
in
your
in
your
cluster
to
store
that
data
in
order
to
get
the
durability
that
it's
f
provides.
A
The
new
erasure
coded
pool
does
a
calculation
where,
instead
of
storing
three
three
copies
of
the
object,
it'll
split
the
object
up
into
a
bunch
of
pieces
and
then
add
a
few
of
parity
objects
as
well,
so
that
you
end
up
with
you
know,
depending
on
the
code
that
you're,
using
and
depending
on
the
setup
you'll
end
up
instead
of
needing
three
times
the
storage
requirements.
You'll
need
one
point
four
times
the
storage
requirements
or
1.6
times
or
whatever,
depending
on
the
on
the
code.
A
But
the
result
is
you
can
put
more
data
in
stuff
and
have
the
same
amount
of
durability
with
less
total
rock
opacity
required.
The
trade-off
here
is
that
when
you
recover,
you
actually
have
to
do
calculations
and
so
there's
a
bit
more
compute
required
on
the
recovery.
So
it's
a
trade-off
that
you
have
to
have
to
consider,
but
it
allows
you
to
have
a
much
more
cost,
effective
stuff
buster.
A
So
another
thing:
that's
kind
of
interesting-
is
cached
hearing,
plus
erasure
coding
as
a
combination
and
I
alluded
to
this
before,
which
is
an
erasure,
coded
pool
that
is
your
backing
pool
and
then
an
SSD
pool
in
front
of
that
which
allows
you
to
have
very
high
performance
on
the
on
the
hottest
data
and
erasure
coded
a
durability
on
what's
underneath
that
that
allows
you
to
have
a
very
cost-effective
storage
mechanism.
So
at
this
point,
I'm
going
to
break
and
show
you
the
latest
version
of
calamari.
A
And
drag
it
that
way:
hey
there
we
go
so
this
is
calamari.
Calamari
is,
as
I
said,
before
the
interface
force.
F
kallomar
used
to
only
be
available
as
part
of
ink
tanks,
f
enterprise,
and
it
was
proprietary.
But
since
ink
tank
was
acquired
by
Red
Hat
we're
planning
to
open
the
code
for
calamari,
which
is
really
super
cool,
so
everybody
will
be
able
to
look
at
calamari
and
see
what
it
does
and
use
it.
This
shows
me
that
I
have
a
cluster,
that
the
health
is
ok
as
of
a
minute
ago,
which
is
good.
A
It's
just
running
on
this
laptop
here.
So
let's
keep
our
fingers
crossed
that
it
remains
ok
or
actually,
if
it
doesn't
that'll,
be
cool,
get
to
see
what
happens
when
a
cluster
is
not
okay.
I
have
30.
Sds
nos
d
is
the
basic
building
block
of
SEF.
It's
short
for
object,
storage
demon,
and
this
is
the
software
agent
that
provides
the
cluster
access
to
storage
resources.
A
So
if
you
have
100
servers
with
five
disks
in
each
one,
you'll
probably
run
500
OS
DS,
but
in
this
case
I've
only
got
three
running
because
it's
running
in
a
virtual
environment.
On
my
laptop,
I
also
have
three
monitors.
The
monitor
is
the
second
building
block
for
force
F,
so
the
OSD
is
the
main
building
block
for
a
stuff
cluster.
A
A
A
We've
got
an
eye
off
scrap
here
on
the
lower
left,
which
doesn't
really
have
much
in
it.
But
you
can
see
that
a
while
ago
some
stuff
happened
in
this
cluster
I.
Think
I
can
I
actually
do
that
now
can't
select.
You
can
see
that
there
there
was
some
some
heavy
I
ll.
Not
heavy
slide
is
a
little
while
ago
on
this
virtual
machine.
It's
also
a
usage
thing
that
tells
me
I
have
81
gigs
full
of
160
gigs,
and
there
are
three
hosts
in
this
cluster.
This
is
my
dashboard,
an
overview
of
calamari.
A
You
can
also
go
into
the
workbench
and
do
some
drilling
down
its
its
and
see
the
three
different
OS
DS
here
and
see
what
what
the
different
statuses
are
of
each
one
of
these.
So
right
now
we
have
three
OS,
these
that
are
scrubbing
and,
of
course
it
shows
which
hosts
they're
all
on,
but
because
I
have
one
host.
It's
all
the
same
host.
A
This
allows
you
to
sort
of
sort
and
filter
based
on
the
status
of
the
placement
group
and
view
the
cluster
by
status
and
diagnose,
which
hosts
have
the
placement
groups
that
are
in
which
states
and
sort
of
drill
down
into
the
information
about
this
host.
It
also
allows
you
to
get
some
some
graphs
about
your
cluster
as
a
whole.
You
can
see
the
aggregate
I
ops
for
the
entire
cluster
here
over
time.
A
Let
me
zoom
in
on
that,
so
you
can
see
the
activity
and
the
total
disk
free
and
disc
users
face
of
your
entire
cluster
or
drill
down
into
each
individual
machine.
If
you
want
so,
let's
look
at
my
first
SEF
store
cluster.
You
look
at
the
CPU
summary
or
of
any
any
of
these
other
things
like
I,
ops,
for
example.
So
this
is
calamari
and
it's
it's
it's
a
very
new
part
of
Seth.
It's
only
been
around
for
about
about
six
or
seven
months.
So
it's
it's
it's
very
new,
but
we
think
it's.
A
Hdfs
support,
also,
which
allows
you
to
use
set
as
a
drop-in
replacement
for
HDFS
in
Hadoop
is
targeted
for
few
as
well,
and
our
BD
mirroring,
which
allows
you
to
take
a
virtual
block
device
and
mirror
it
to
another
virtual
block
device
and
then
further
in
the
future
is
I.
Scuzzy
got
some
quality
of
service
stuff
going
in
RDMA,
which
a
lot
of
people
are
asking
for,
and
LDAP
integration
for
the
cluster
is
further
on.
So
this
is.
This
is
in
sort
of
big
building
block.
A
This
is
the
duplo
version
of
our
roadmap
for
people
who
need
the
big
blocks,
and
just
finally
I
want
to
tell
you
a
little
bit
about
ink
tank
and
what
we
do
we're
the
principal
sponsors
of
the
stuff
project,
but
we
also
sell
Enterprise,
subscriptions
professional
services
and
SEF
training
to
help
people
deploy
SEF
and
be
successful
with
Seth
in
production,
and
that
was
it.
Thank
you
very
much
for
your
time.
Here's
my
contact
information.