►
Description
A
B
Yeah
good
evening,
its
yeah,
a
clerk
here
in
Cape
Town
yeah
the
first
time
I'm,
giving
a
toy
like
this,
so
I
thought
I'd
just
check
can
I
see
this
Brian
and
Josh.
Can
you
guys
see
him?
Can
you
guys
hear
me?
Okay
am
I
gonna,
give
you
like
a
couple
seconds
just
respond
in
case.
You
can't
hear
me
but
I
assume
you
can
hear
me.
I'm
gonna
start
sharing
my
during
my
screen.
Yeah.
B
B
Great
play
from
start:
okay
cool.
Can
you
see
my
for
my
opening
slide?
Yes,
cool,
great
okay,
so
yeah
hi,
my
name
is
I
know
this
is
being
recorded
for
posterity,
so
I'm
going
to
seem
that
I'm
talking
to
people
in
retrospect,
but
yeah
so
I've
run.
My
name
is
Thomas
Thomas,
Bennett
and
yeah
greetings
from
the
southern
tip
of
Africa.
First
of
all,
thank
you
to
Mike
Paris
and
the
staff
foundation
for
inviting
me
to
talk
to
you
on
the
safe,
safe
tip.
B
You
can't
hear
me-
or
you
can't
see
sliders
having
please
just
obviously
stop
in.
Let
me
know
because
there's
no
point
in
them
talking
to
you
guys
if
you
can't
see
what
I'm,
what
I'm,
what
I'm
talking
about
yeah.
So
this
is
now
my
titles,
my
title
slide
and,
as
you
can
see,
yeah
I
mean
that's.
I
mentioned
a
few,
a
few
words,
so
I
thought
I'd
just
kind
of
delve
in
a
bit
and
talk
a
bit
about
the
other
words
I'm
using
so
yeah.
B
So
meerkats
is
the
science
instrument
of
interest
here
and
is
currently
the
most
powerful
radio
telescope
in
the
world,
and
then
you
see
the
word
safe
and
obviously
that
just
makes
sense.
Of
course
I
mean.
Why
would
you
want
to
use
anything
else
and
then,
of
course,
is
radio
telescope
data?
So
we
attach
to
the
radio
telescope
and
I'll,
be
talking
a
bit
more
about
radio
astronomy,
maybe
a
bit
about
the
history
of
radio
astronomy,
talk
about
it,
but
they
make
a
telescope
and
then
I.
B
It
includes
hundred
organizations
across
20
different
countries
at
the
South
African
Radio
Astronomy
Observatory.
It's
a
bit
of
a
mouthful.
It's
brief.
Yet
SRL
is
one
of
these
organizations
and
of
course
that
means
that's.
Africa
is
one
of
these
countries
involved,
thus
arouse
responsible
for
managing
all
the
great
astronomer
initiatives
and
facilities
in
South
Africa.
This
includes
the
meerkat
radio
telescope,
which
is
located
in
the
crew
at
try
and
arid
region
in
the
kind
of
yeah
out
in
the
about
90
kilometres
north
of
Cape
Town.
B
That's
basically,
who
we
are
so
maybe
that
gives
you
a
bit
more
context
as
to
as
to
you
know
what
this
case
and
what
make
it
is.
Okay,
Who
am
I,
as
I
said
before,
I'm
Thomas
Bennett
I
mean
you
can
see.
Is
the
nice
newbie
photo
of
me
looking
proudly
on
upon
my
first
two
notes
that
I
installed
so
fun
as
you
can
see
I'm
a
new
me
because
I've
only
got
two
notes
there
and
we
all
know
that
you
need
to
this
replication.
You
need
these
three
notes.
B
Do
to
start
a
sort
of
safe,
safe
cluster
and
its
default
mode
without
reporting
errors.
Yeah
I've
been
working
on
the
maquette
project
for
the
last
12
years
and
my
jobs
have
included
being
a
software
developer.
I've
been
a
systems
engineer
its
systems
engineer
and
now
in
the
last
way
of
storage
engineer,
which
is
why
I'm
presenting
here
today,
I
guess,
we
started
using
safe
in
the
latter
days
of
jewell,
with
our
first
production
cluster
going
out
in
luminous
in
about
April
2018.
So
we
were
to
be
new
to
the
safe
community,
but
we
yeah.
B
We
are
obviously
hoping
to
be
able
to
contribute
wherever
we
can
and
yeah
at
some
point
say
well,
yeah
I
mean
yeah.
We're
certainly
prevalent.
That's
a
lot
as
we've
as
we've
gone
on
yeah.
So
cosmology
is
a
science
which
has
only
a
few
observable
facts
to
work
with.
That
was
a
statement
made
by
Robert
Woodrow
Wilson
he's
an
American,
astronomer
and
co-discoverer
of
the
cosmic
microwave
background
radiation,
and
it's
always
nice
to
include
an
xkcd
in
your
slides,
because
yeah
I
mean
it's
this.
B
I
guess
it's
quite
fun
to
have
a
look
and
just
see
what
what
I'm
trying
to
I
mean
by
this,
and
if
you
look
at
this
this
photo.
This
is
a
optical
photo
of
the
of
the
night
sky
and
I
believe
that
that
bright,
spot
and
kind
of
just
off
this
off
the
center
to
the
right
is
tentura,
say
and
typically,
this
picture
is
shown
in
much
much
higher
detail.
This
is
taken
from
a
military
scope,
which
is
a
optical
telescope.
B
That
slaves
with
to
Mir
kept
pointing,
and
the
day
of
this
is
that
you
want
to
you
kind
of
want
to
do
optical
pictures
of
the
other
night
guy
to
see
where
meerkats,
observing
and
but
typically
that's
a
galaxy,
Centaurus
A
and
it's
usually
saying
slightly
high
resolution.
But
the
reason
for
doing
this
is
kind
of
overlaying
that,
with
the
picture
from
their
cat-
and
this
is
now-
you
can
see-
there's
a
lot
more
detail
and
structure
there
that
you
otherwise
couldn't
see
at
the
optical
regions.
B
There's
large
structures
on
either
side
or
our
radio
jets
that
are
being
created
by
the
supermassive
black
hole
at
the
center
of
this
galaxy,
and
you
can
see,
as
I
said,
I've
been
cast
on
one
of
the
most
powerful
radio
telescope
in
the
world,
and
we
can
create
these
really
beautiful,
high-resolution
images.
And
so
finally,
raid
astronomy
has
become
a
bit
more
a
bit.
Cooler
than
it
used
to
be
typically
was
the
optical
astronomers
that
that
had
the
beautiful
pictures,
but
we
starting
to
catch
up
with
them.
Now.
B
You
just
get
this
going
so
here
guys.
Yes,
just
a
yeah
high
speed,
anime
husband
for
you
of
what
what
the
dishes
look
like.
Just
give
me
a
job
at
where
we
are,
as
you
can
see,
it's
a
it's
a
semi
desert
region
and
be
far
away
from
any
radio
transmitters,
which
obviously
a
big
problem
for
us.
Rfi
is
yeah.
B
If
you
haven't
yeah,
you
know
you
have
it's
a
is
a
big
is
a
big
problem
for
us
when
you
built
away
from
many
urban
structures
and
human
activity-
and
we
also
we
have
government
as
passed
legislation
to
protect
this
area,
so
that
people
can't
actually
build
radio
transmitters
in
the
surrounding
area.
Sam
yeah.
So
basically
in
2006,
seven
government
Commission's,
a
team
of
scientists
and
engineers
to
start
the
design
work
for
the
SKA
precursor
telescope.
B
This
Pete
kirsta
telescope
was
intended
to
demonstrate
that
Africa
could
not
only
play
in
the
global
s
care
project,
but
could
also
demonstrates
Africa's
technical
and
engineering
abilities
at
the
visual
telescope.
A
design
consisted
of
20
dishes
and
became
known
as
the
courier
a
telescope
or
cat
named
after
the
South
African
desert
black
region.
The
career
in
which
it
was
to
be
built
sometime
later
from
government
made
more
budget
available
to
build
a
bigger
system.
B
As
of
July,
twenty
twenty
eighteen,
the
meerkat
was
commissioned
and
is
now
the
most
powerful
radio
telescope
in
the
world,
as
I've
said
before,
this
will
only
be
surpassed
when
it's
absorbed
into
the
SKA
Phase
one
telescope,
which
is
going
to
happen
in
a
few
years
time
yeah.
So
it's
as
I
said
arrayed
astronomies
we've
started
making
these
really
beautiful
pictures
and
make
a
telescope
operating
a
number
of
modes.
One
of
those
modes
is
an
imaging
mode.
B
So
it's
nice
to
show
the
pictures
and
show
some
of
the
stuff
that
we
get
up
to
is
a
radio
image
of
the
center
portions
of
the
Milky
Way
galaxy.
It
was
used
for
an
hour
race
on
image
in
July
of
2018.
The
plane
of
the
galaxy
is
marked
by
a
series
of
bright
features
which
she
can,
which
is
basically
exploding
stars
and
regions
where
new
stars
are
being
born
and
runs
higher
horizontally
through
the
image.
B
The
black
hole
at
the
centre
of
the
Milky
Way
galaxy
is
hidden
in
the
brightest
of
these
extended
regions
yeah
and
now
what
we
have
never
said,
never
even
seen
before
other
radio
bubbles
that
were
discovered
by
the
meerkat
a
telescope,
and
these
extend
vertically
and
vertically
out
of
the
top
and
the
bottom
of
the
plane
of
the
galaxy.
You
can
see,
there's
kind
of
like
bubble
structures,
I'm,
hoping
that
the
resolution
sign
up
for
you
guys
to
actually
actually
see
that
they're,
the
two
men
ideas
we
had
the
cause
of
these
bubbles.
B
You
can
kind
of
see
the
Galactic
plane
running
over
there
I
think.
No!
No,
they
left
it
then
actually
must
be
running
down
three
over
there,
because
what
we've
gone
and
done
is
we've
superimposed,
the
Galactic
neglect
to
clean
us
up
over
there.
You
can
see
it
over
there.
That's
the
Galactic
plane
running
through
the
top
right
hand,
corner
of
the
image,
and
you
can
see
that
the
central
galaxies-
and
you
can
see
these
yeah
these.
B
B
The
fidelity
of
the
instruments
is
limited
by
the
brightest
source
in
your
field,
and
the
goal
of
of
this
next
experiment
was
to
find
mercy,
find
field
accessible
to
me,
a
cat
for
the
fuse
brightest
sources
and
the
faintest
buy
just
patch
in
the
southern
sky,
as
opposed
to
previous
image.
What
you
see
here
is
that
is
the
last
best
image
of
this
part
of
the
sky,
which
is
part
of
the
SAMS
catalogue,
and
you
can
see
the
moon
to
the
left
to
give
scale.
B
So
the
idea
was
to
observe
the
sky
now
for
about
130
hours,
collect
data,
so
it
took
about
a
year
to
collect
all
this
data,
and
the
resulting
image
is
what
I'll
show
you
next,
which
is
this
image
over
here
yeah?
What
is
what
is
truly
amazing
about
this
image.
Is
that
you
have
all
those
bright
sources
which
are
these
supermassive
black
holes
in
these
galaxies,
but
kind
of?
In
the
background?
All
these
other
dots
and
those
dots
are
actually
galaxies
and
those
are
Milky,
Way
type
galaxies
that
are.
B
That
we've
got
there,
that'd
be
basically
imaged
and
what
is
amazing
is
see
obviously
about
the
Israeli
spot.
Any
witches
images
is
that,
as
you
SEC
further
and
further
wasting
further
further
back
in
time,
so
there's
a
time
component
easiest
to
this
image
and
there
are
small
dots
in
the
background.
There
are
galaxies
that
they're
started
forming
millions
and
millions
of
years
ago
or
billions
of
years
ago.
So
the
idea
is
that
this
image
is
kind
of
a
snapshot
of
the.
B
Is
this
it's
a
single
image
that
captured
that
captures
the
star-forming
galaxies
in
to
the
early
history
of
the
universe,
yeah.
So
the
in
the
story.
We
can
merge
the
evolution
of
galaxies
from
early
universe
atoms
all
present.
Obviously
it
makes
a
cool
picture.
That's
the
actual
science
behind
this.
Is
these
two
graphs
over
here,
and
this
is
a
plot
of
the
histogram
of
the
source.
Count
by
Doppler
shift
with
dr.
ship
represents
time,
and
the
left
is
basically
further
back
in
time.
This
has
now
been
published
in
astrophysics
journal.
B
So,
while
the
two
plots
one
is
the
different
normalizations,
when
is
your
cleaning
and
the
bottom
line
is
practice
normalized,
so
the
cheapest
of
the
plot,
where
we
are
constraining
the
source
count,
is
the
the
boxes
of
that
higher
brightness
degrees
with
the
previous
VLA
data,
which
is
over
there
and
I,
have
a
mere
ket
is
more
sensitive
and
then
the
VLA,
and
therefore
we
can
see
past
that
the
PC
version
data,
which
indicates
which
is
aniket
by
these
green
boxes
on
the
left.
Yes,
so
we
have
to
see
deeper
and
further
back
in
time.
B
Basically,
basically,
that's
also
the
deepest
l-band
continuum
image
ever
made.
This
is
pretty
exciting
India.
Once
again,
this
is
a
yeah.
This
is
now
a
published,
published
data,
so
this
is
kind
of
hot
off
the
press.
This
is
a
another
continuum
image,
so
one
of
the
modes
that
that
meerkat
has
now
supports
is
30,
TK
mode,
where
we
channelize
data
into
32
thousand
channels.
B
You
can
imagine
it's
as
basically
a
snatch
image
as
a
snapshot
in
frequency
and
there
32,000
of
these
images
and
what
enormities
we
compress
all
those
images
to
make
a
single
image,
which
is
this
image
over
there.
However,
as
I
say,
we've
recently
started
capturing
32,
32
thousand
channels,
and
this
is
a
new
mode
which
is
being
supported
and
what
is
pretty
cool?
Is
we?
If
you
look
in
the
center
there,
there's
there's
a
galaxy
there.
B
I'm
just
going
to
kind
of
zoom
in
on
that,
and
what
we've
done
in
by
capping
is
32
thousand
channels
is
we're
able
to
resolve
the
the
Doppler
in
this
galaxy
and
what
it
means
is
that
we
actually
get
a
three-dimensional
view
of
this
of
this
of
this
galaxy
and
if
you
view
it
by
by
by
channels
basically
I
will
just
get
this
going.
You
can
see
here
that
you
can
see
that
the
spiral
arms
are
kind
of
being
being
imaged,
and
this
is
basically
Doppler
and
so
Doppler
represents
distance.
B
So
the
stuff
on
the
left
is
further
away
than
the
stuff
on
the
right,
and
you
can
see
we
kind
of
iterating
through
there
through
the
channel,
so
it
makes
it
quite
a
cool
video,
but
in
your
mind's
eye
you
can
kind
of
see
that
universe.
Imagine
that
the
stuff
in
the
left
is
in
in
the
it's
moving
away
from
essa
and
the
as
the
galaxy
spinning
and
the
Stephanie
and
other
actors
moving
towards
us,
especially
what
it's
anyway.
This
is
well,
it's
quite
cool
about
this
is
this:
is
this
image
was
created?
B
It's
a
fully
automated
image
creation
and
this
yeah,
it's
basically
kind
of
where
we
heading
towards.
Is
these
disability
to
do
automatically
image
these
at
this
data
and
yeah
I
mean
set,
there's
a
central
role
and
doing
that,
so
it's
also
pretty
cool
we've
got
some.
We've
also
got
some
mysterious
stuff
that
we
that
we
bet
you
don't
know
what
it
is,
and
this
is
pretty
cool.
This
isn't.
B
The
original
Sam's
catalog
can
see
on
the
left
there
and
then
right
there,
it's
kind
of
resolved
with
the
naked
telescope,
and
we
still
don't
know
what
this
is.
This
is
apparently
in
the
in
the
in
the
local
galaxy,
but
there's
still
debate
as
to
what
this
actually
is
adjusting
film
Texel.
It
look
like
an
octopus
which
seemed
pretty
cool
yeah
and
that's
basically
comes
the
science
that
I
wanted
to
talk
to
you
about
and
yeah
they're
turning
to
set
related
matters.
B
So
we
have
a
number
of
SIF
classes
at
Sara,
name
of
the
class
I
give
some
detail
about
where
the
cluster
has
been
deployed.
Classes
co-located
with
the
telescope
are
prepended
with
the
name
yeah
cat,
just
to
identify
them
as
clusters
at
on-site
classes
located
at
a
dead
stood
here
in
Cape,
Town
are
called
circuit
and
she
caters
the
Afrikaans
word
for
octopus.
So
it
was
nice
nice
way
to
kind
of
just
make
it
all
make
sense.
So
it
literally
means
the
XS
see
cat
or
octopus
yeah,
so
class
is
located.
B
An
office
are
prepared
with
sweat
cat,
so
that
kind
of
gives
in
it
about
where,
where
those
are
and
yeah
yeah,
so
it
can
this.
This
table
I
just
want
to
give
energy
about
the
number
of
what
variants
we're
running,
we're
currently
running
full
running
or
luminous.
They
just
at
some
point
upgrade
this
all
of
our
main
production
clusters
to
Nautilus
I'm
in
the
process
of
upgrade
paths
for
us
to
be
able
to
move
onto
onto
Nautilus
yeah
and
give
an
idea
about
sizes,
meerkats
clusters.
B
You
know
the
time
for
petabytes
and
that's
all
SSD
storage
and
then
we've
got
our
spending
just
storage.
We
got
a
stake
at
C,
1
and
C
at
C,
3
C
3
is
our
current
production
cluster
IC
1.
Instead,
it's
being
decommissioned
and
it's
kind
of
a
test
cluster
and
we
all
yeah.
We
plan
to
do
some
experiments
on
that
before
we
decommission
it
and
absorb
it
into
the
circuit,
C
3
cluster.
B
So
how
we
meet
data,
so
we
validate
Iran
on-site
at
the
telescope
and
it's-
and
this
can
then
accumulate
over
a
number
of
days
if
it
needs
to.
If
we
have
a
link
break
between
the
link
between
captain
and
site,
it
means
that
we
can,
we
can
saw
the
telescope
can
still
operate
normally
and
once
the
link
is
fixed,
we
can
then
save
the
data
down
to
our
15
petabytes
archive
and
kept
on
yeah.
B
So
the
miacca
cluster
has
has
two
pools:
there's
a
cold
buffer,
which
is
where
we
store
stuff
to
to
sink
down
to
Cape
Town,
and
you
have
a
hot
buffer
that
just
built
out
of
SSDs,
and
the
deer
of
that
is
that
this
is
for
our
imaging
cluster,
and
so
we
color
cat
our
our
EP
use
with
our
SIF
cluster.
So
we
have
a
whole
bunch
of
nodes
with
each
20s
is
season
four
or
GPUs,
and
this
last
iteration
of
30k
married.
B
We
actually
had
some
problems
with
that,
because
the
images
and
the
ancef
were
grabbing
too
much
memory
at
the
same
time,
and
so
the
whole
installation
became
quite
unstable
and
we
spent
a
fair
bit
of
time
just
having
backing
things
down
so
that
we
could
get
the
memory
management
under
control,
but
yeah
so
basically
generated
on
site.
It's
this.
This
hot
buffer
then
creates
all
these.
It's
the
data
cycle,
the
governor
of
again
and
eventually,
once
it's
created
images.
Their
debtors
then
delete
because
there's
no
like
it
either.
B
This
intermediate
data
for
creating
the
imaging
yeah.
So
another
important
important
aspect
is
how
users
access
data,
so
users
log
into
our
data
service
portal
and
authenticate
using
sam'l
once
they
authenticated
they
can
search
for
the
data.
They
can
then
generate
a
JSON
web
token,
which
then
provides
access
to
an
s3
bucket
that
contains
the
observational
data
and
observation
is
typically
made
up
of
a
small
number
of
s3
buckets.
B
A
header
packet
containing
telescope
metadata
like
antenna
pointing
another
sensor
data
and
then
the
data
stream
packets,
which
contain
data
captured
from
the
telescope
and
typically
have
to
the
order
of
one
200
million
objects.
The
user
will
access
this
data
using
respect
data
access
layers
using
like
that
access
layer,
called
cat
tell
which
lazy
layer
it's
the
data,
so
this
can
be
the
denigration
can
be
formed
from
your
laptop
or
from
a
supercomputer
and
scales
accordingly
to
the
connectivity
and
the
number
of
resources
that
you
have
so
so.
Typically,
this
is
kind
of
yeah.
B
Those
images
are
showed
you
that's
per
typically
generated
by
by
astronomers,
however,
say
we
are
tiny,
build
up
this
automated
mode
of
doing
anything,
these
images
as
well,
and
that
that
last
continuing,
which
are
going
to
showing
you
where
we
are
shared,
the
galaxy
you
are
showing
the
doll
from
the
galaxy.
That's
all
that's
all
automatically
generated
images
there,
and
so
this
is
really
trying
to
head
towards
it's
creating
these,
these
automated
images
and
so
plays
a
central
role
in
in
that
in
that
system
yeah.
So
yeah,
okay,
so
that's
yes,
I!
B
Guess
what
have
we
been
up
to?
We
focus
if
metrics
for
own
purposes,
we
actually
stopped
at
all
the
the
dashboards
and
get
have
got
them
working
for
us.
Lots
of
detail
of
most
of
the
panels
require
a
bit
of
head-scratching
to
provide
some
good
information,
you're
still
looking
to
integrates
the
SoundCloud
IP
my
tools,
happily
my
total
export
into
this
as
well,
because
now
PMI
provides
useful
information
that
would
otherwise
be
quite
difficult
to
get
off
the
system
and
provides
information.
B
While
the
host
is
down
as
well
as
safe
deployments,
we
use
sip
ansible,
but
yeah
we're
kind
of
looking
towards
building
our
own
a
deployment
method,
you're,
probably
using
surf
ansible
or.
But
you
want
to
look
at
a
whole
bunch
of
other
stuff
as
well
to
see
you.
How
are
we
gonna
deploy
going
forward,
yeah,
so
yeah?
So
that's,
that's!
Pretty
much
working
up
you've
also
got
a
datum
use,
which
is
all
bespoke
software.
That's
written
to
move
the
data
to
cluster
in
head-turn,
there's
just
a
view
of
our
SIF
matrix
yeah.
B
What
you're
spending
a
lot
of
time
doing?
Is
configuring
a
read,
a
scarecrow,
doing
a
whole
bunch
of
benchmarking
to
see
what
kind
of
sizes
we
need
it?
As
you
can
see,
I
I've
turned
off
that
Emily
sharding
and
we
even
hit
a
number
of
times
by
performance
issues,
because
when
you're
trying
to
be
short
stuff
and
write
into
text
at
the
same
time,
it's
not
not
not
ideal.
So
instead,
what
I
do
is
I
know.
B
I'm
gonna
have
these
buckets
with
masses
of
data
rather
just
set
ten
shots
and,
and
then
yeah
object
size
as
well.
To
avoid
creating
a
lot
of
shared
objects,
we
reset
our
max
object
size
to
20
megabytes.
This
is
basically
because
we
have
control
over
how
big
objects
are,
and
so
so
you
know
that's
that
it's
never
gonna
be
bigger
than
pretty
megabytes.
B
So
it's
it's
an
fall
for
the
majority
up
for
99%
of
the
data
to
be
no
time
to
be
an
object
is
not
to
be
a
big
lean,
20
megabytes,
so
it
makes
sense
to
kind
of
tune
our
entire
cluster
to
to
those
values.
We
did
some
performance
testing
and
we
pretty
confident
that
we
don't
lose
anything
by
by
doing
this,
so
it
kind
of
made
sense,
and
so
we
that's
what
we
ended
up.
Setting
yeah,
I,
guess,
safe
lessons
learned.
We
had
a
number
of
interesting
hardware
failures.
B
We
had
a
bad
batch
of
and,
of
course
it
kind
of
failed
silently,
and
then
we
suddenly
we,
but
our
STIs
is
falling
out
because
they
can't
log
anymore
and
trying
to
restart
which
is
being
pretty
interesting,
and
this
is
the
other
reason
why
I
want
to
IP
my
tools
install
just
make
sure
we
can.
We
get.
You
know
stats
outside
of
the
the
machine
itself
habit,
yeah,
very
basically,
the
the
lesson
learned
there
was
that
don't
use
cheap
drives
rather
use
some
some
use,
a
better
driver
for
the
phalaris
drives
yeah.
B
B
Yeah
not
deleting
objects
when
at
the
restarting
well
the
index
is
not
eating
initiatives,
and
so
we
had.
Our
index
is
running
on
nvm
ease,
nor
does
happening
Suzy
Amis
we're
getting
bloated
and
we're
falling
over
the
whole
time,
and
it's
it
created
some
really
weird
strangeness,
but
yeah
we
were
able
to
once
we
overcome
that
and
we
sorted.
The
asset
was
so
it's
all
working
great
yeah
yeah,
so
we
kind
of
all
moved
to
all
the
classes
on
to
twelve
to
twelve
now,
and
there
was
to
then
start
using
the
balancer.
B
One
of
our
machines
is
we're
running.
The
balancer
was
particularly
under
SPECT
and
the
best
you've
seen
a
such
a
big
cluster
with
Romeo's
DS.
The
manager
needs
a
bit
more
CPU
power
to
be
able
to
be
able
to
operate
pretty
effectively
yeah.
We
all
say
that
some
finest
unnoticed
cluster,
we
got
one
notice
cluster
in
our
history,
it's
running
through
proxmox,
but
this
is
kind
of
like
a
Franklin
class
term.
B
You've
got
iced
teas
with
200
gigabytes,
2,
STS
or
5
terabytes,
and
it's
squinting
trying
to
see
how
we
kind
of
tear
this
thing
apart,
so
that
we
can
make
it
work
effectively
with
such
despaired.
Sizes
in
OSD
drives
yeah
every
interesting
learning,
learning
exercise,
but
so
far
I
mean
we're
very
happy.
You're,
safe
and
yeah
I
think
we've
we've
made
the
right
choice
and
yeah,
there's
no
reason
to
yeah
no
reason
to
not
continue
using
it,
and
so
just
give
it
a
day
about
30
nerd
hard
way.
B
We
built
our
own
hardware
from
from
scratch.
Basically,
the
idea
was
to
yeah
builder
I,
know
that
suits
our
purposes.
What
simpson
has
you'll
see
I've
kind
of
highlighted
the
fact
that
we
have
48
or
49
s
DS,
so
48
hard
drives
and
one
in
via
me
as
for
indexes
on
each
machine,
and
you
can
see
our
process
in
the
memory
we
pretty
pretty
under
spec,
so
we
have
to
really.
We
use
a
steam
memory
memory
targets
quite
aggressively
to
tune
those
T's
down
to
2
gigabytes
and
from
all
the
various
testing.
I've
done.
B
We've
we've-
and
this
is
one
of
the
big
tests
I
need
to
do
on
the
echo
cluster
one.
It's
70%
full
now
is
to
take
an
OSD
I,
take
a
host
down
and
see
how
the
recovery
behaves,
but
whenever
we've
we've
had
these
problems
are
asking
production
refund.
If
they're
being
particularly
concerned
that
things
are
going
pear-shaped,
the
recovery
takes
a
bit
of
time
because
you've
tuned
a
whole
bunch
of
parameters
down,
but
it's
all
recovers
and
yeah.
We
seem
to
be
you
to
be
fine
yeah,
so
so
far,
so
good
moving
ahead.
B
We
kind
of
wanting
to
look
at
using
EC
pools
and
we
think
that
this,
because
we
slightly
unrespectful
EC
pools,
we
have
had
to
do
the
more
hardware
procurement
and
we
we
kind
of
leaving
more
greatly.
Yeah
we've
been
more
liberal
with
our
choice
of
processor
and
definitely
want
to
get
that
that
up
a
bit
more
yeah
run
run
EC
Falls.
But
you
know
exactly
where
we
worry
we're,
stamping,
not
quite
sure,
and
one
of
the
tests
wanted
to
now
say:
Katzie
won
before
decommission.
B
It
is
to
actually
go
and
create
some
EC
pools
and
take
some
checks
and
fairly
demands
and
see
how
the
system
behaves
and
just
see
kind
of
how
in
respect
we
are
to
get
a
a
day
about
where
we
should
be
where
she'd
be
dead,
hitting
and
the
outside
of
me.
Akechi
also
involved
in
community
activities.
We've
organized
a
few
years
of
meetups
in
Cape
Town
we've
got
a
get
to
I
am
community.
B
We
presented
at
a
number
of
second
conferences
to
kind
of
get
better
exposure
for
safe
and
yeah,
and
also
just
PR,
trying
to
bring
users
together
to
kind
of
yeah,
create
a
create
a
real
community,
so
yeah.
So
as
part
of
this
process,
I
went
and
I
emailed
how
much
people
to
try
and
find
out
who
was
using
Steph
and
how
are
they
using
safe?
This
is
my
supply
x
and
z,
as
as
their
local
name
for
south
africa.
It
means
the
south
and
closer
and
the
app.
B
So
that's
that's
why
it's
aims
and
Z's
their
clusters
yeah.
Then
this
a
number
of
organizations
there
and
they
all
kind
of
like
to
pull
their
surf
glasses.
No
one
says
biggest
ask
yet,
but
but
yeah
this
attorney
Greg
been
seeing
this
bit
of
interest
generated
in
the
young
in
the
community
to
be
using
safe,
yeah
and
future
plans.
Yeah
decommission
ii
see
one
and
do
some
testing
along
the
way
as
part
of
that
process.
B
We
want
to
upgrade
to
nautilus
and
understand
that
upgrade
past
a
nautilus
and
see
you
make
sure
that
we
that's
that
everything
still
in
spec
and
we've
started
the
design
work
on
our
new
storage
cluster.
So
we've
got
more
funding
to
build
twenty,
more
dishes,
building
twenty
more
dishes,
the
data
problem
scales
with
the
pilot
square,
the
number
of
dishes.
So
you
can
see
we
basically
by
adding
a
few
more
dishes.
We
require
another,
these
20
petabytes
of
raw
storage
and
this
time
us
everyone
to
both
easy
pools.
B
They're
going
to
be
quite
clever
about
how
we
have
specified
there
is
there's
that
hardware
purchase
and
and
yeah.
So
we've
got
some
some
really
cool
kids
at
the
moment
that
we're
testing
out
we've
got
some
obtain
that
we
want
you
to
have
a
look
at.
We've
also
got
Tyler
bought
at
hundreds
discs,
machine,
which
is
the
king
interesting,
and
we
have
one
of
those
that
we're
going
to
be
testing
and
yeah.
We've
thought
about
Saul
bunch
of
supermicro
boxes
that
we
have.
B
You
can
see
testing,
so
you
know
at
some
point
we'll
laugh
a
bit
more
and
you
see
stuff
to
say
about
yeah
new
hardware
purchases
and
where
we're
heading
yeah
and
that
kind
of
confused
my
talk
so
there's
my
contact
details.
Tom
said
this
case
today
is
my
email
address.
Welcome
to
email
me.
If
you
have
any
further
questions
or
more,
you
want
to
know
our
websites.
That's
the
way
we
dress
for,
in
fact,
so
I
realized
I've
changed
this.
B
It
should
be
cerrado
so
today,
but
if
you
point
that
it
will
redirect
you
to
this
or
our
website
and
yeah,
just
a
bit
of
advertising,
there's
a
beautiful
picture
of
Cape
Town,
that's
that
they
live.
And
that
concludes
my
talk.
Sorry
I
think
I
see
yeah.
You
have
any
questions.
Please
awesome
now
I
say
well.
Contact
me
you're,
welcome
to
drop
any
comments
or
questions
in
the
chat.
Yeah
yeah.
That's!
Basically,
that's
basically
it
so.
B
The
hardware
is
enclosed,
in
fact,
if
I,
if
I
go
back
to,
there
was
a
picture
of
me
or
at
the
beginning
words
there's
my
newbie
picture.
That's
the
chassis
I'm,
so
sorry,
I,
just
I,
just
put
that
there,
because
it
looks
cooler
having
me
showing
you,
but
just
progenitor
had
the
hydras,
enclosed,
yeah,
so
ya
know,
and
we
we
have
a
hardware
at
the
Center
for
high
performs
competing,
which
is
they
have
a
data
center
here
in
Cape
Town
and
we
they've,
given
us
nine
racks
day
where
we,
where
we
run
our
kit
yeah.
B
B
For
the
you
know,
the
the
writing
headlocks
for
for
the
file
store
and
by
the
time
we
went
into
production,
blue
store
was
around
and
when
we
did
our
benchmarking
using
police
store,
we
got
just
as
good
performance
without
using
the
nvme
drives
for
the
for
the
right
ahead
logs,
and
so
what
we'd
end
up
things
when
they're
fleeing
the
nvme
is
to
be
used
as
a
for
a
different
purpose
and
so
yeah.
So
we
ended
up
implementing
just
yeah.
We
have
all
our
steeds
have
no
offloading
onto
enemies,
those
enemies.
B
B
B
Yeah
so
unfortunately,
Radio
Astronomy
data
by
by
definition
is
noise,
so
it's
actually
quite
difficult
to
compress,
but
that's
it.
He
doesn't
stop
us
from
trying
stuff
out
and
seeing
but
ya
know
we
we
have
the
same
problem
right,
think
they're,
trying
to
take,
because
when
we
eventually
when
the
data
gets
cold
enough,
we
we
back
it
up
on
to
Ted
and
we
track
compression
there
and
we
get
now.
We
could
no
advantage
over
doing
any
any
compression.
Okay.
B
The
bucket
index,
yeah,
yeah,
I,
suppose
yeah
because
you'll
be
offloading
the
yeah.
It's
pretty
much
the
same
thing,
except
that
we
can
now
specify
which
pool
it
belongs
to
as
opposed
to
which
is
DS,
so
I,
guess,
I,
guess
by
making
them
tease
themselves,
they
yeah.
They
know
the
one
thing
which
I,
which
I
haven't
quite
figured
out
yet
is
actually
how
much
of
the
infamies
we
are
actually
using
and
because
obviously
I've
seen
that
I'm
apt
to
get
quite
big
but
they're,
not
they're,
not
filling
up
the
whole.
B
C
B
B
Well
I'd
say
my
contact
details.
There
feel
free
to
drop
me
an
email.
Thank
you.
Yeah
yeah
fully
enjoy
this.
Your
your
Thursday
I
must
I'm
gonna
hit
home
now.