►
From YouTube: Apache Cassandra - Colin Clark
Description
Last 1st of April we hosted a very interesting presentation of Apache Cassandra by Colin Clark.
A
A
A
A
A
A
E
A
F
A
A
A
A
A
A
G
Thank
you.
My
name
is
khan
farah
khan.
The
solution
architect
from
data
stack
a
little
bit
about
my
background.
With
this
Andhra
I've
only
been
the
biggest
after
about
six
weeks
six
weeks.
I
think
that
put
40,000
hermana
tonight
my
frequent
flyer
program,
I'm
happy
to
be
here
the
identical
cells
and
to
fetch
area
and
a
big
guy,
good
notice,
and
so
the
bed
and
the
shop
are
really
important.
G
My
approach
to
these
stocks
of
the
older
than
maybe
you've
been
exposed
to
high
Khan
from
of
a
very
proprietary
software
for
me
plus
series
of
the
building
Florida
distributed
systems.
You
know
the
stock
exchanges,
so
I
worked
on
the
Paris
stock
exchange,
entre
Stock
Exchange,
New,
York,
Stock,
Exchange,
Boston,
Stock,
Exchange
stock-exchange.
G
Fidelity
and
most
recent
ads
and
I
see
your
onyx,
which
is
the
new
york
stock
exchange.
You
don't
ask
me,
run
about
28
exchanges
globally,
I
think
there's
one
here,
and
so
what
I'm
going
to
talk
about
is
a
big
system
that
we
built
a
call
direct
star
which
is
on
Cassandra
I'm,
going
to
talk
a
little
bit
about
why
we
chose
Cassandra
some
of
the
issues
that
we
have
coming
from
a
little
proprietary
software.
G
G
Problem
we're
trying
to
solve
their
own
bunch
of
problems
from
a
business
point
of
me,
boo
Wendy
build
things,
and
we
said
it
was
cloud.
What
we
really
meant
was
an
ASP
model
and
what
that
meant
was
developing
a
customer
and
we
would
stand
up
hundreds
of
thousands
of
dollars
with
the
hardware
and
make
all
the
internet
where
they
cut
the
network
cross,
connects
and
hope
that
they
stayed
with
this
long.
As
that
we
could
make
some
money
stock
exchanges
at
the
people.
G
They
should
make
money
off
of
our
customers,
and
but
that
model
was
getting
old.
It
was
getting
part
and
we
wanted
to
do
something
that
was
really
hold
a
tenant
that
was
really
cloud-based
so
that
every
time
you
brought
a
new
customer
on,
we
didn't
have
to
recreate
the
balls
back
and
the
specific
problem
that
we
had
yourself
to
that.
Because
we
wanted
to
detect
fraud.
G
But
the
first
is
that,
if
you're
going
to
monitor
somebody
in
comparison
to
the
market,
even
if
you
just
had
one
customer
you
not
to
monitor
the
entire
market,
and
so
if
we
were
doing
that,
yes,
keeper
in
Reverse
say
that
Hardware
a
database
for
each
customer,
we
would
have
been
duplicated
to
drop
them
over
and
over
and
over
again,
the
network
bandwidth
would
have
been
just
the
hardware.
The
network
everything's
just
do
not
going
to
make
any
money
right,
so
that
was
from
the
business
22
the
problem
from
the
technical
services.
G
What
I
was
interested
in
was
walking.
How
do
we
do
this
and
so
I'm
going
to
talk
about
Cassandra
I
was
gonna
talk
about
stream
processing
a
little
bit
or
event
processing
and
coming
match.
Those
up
do
some
some
neat
things.
We
actually
persistent
very
vast
amounts
of
data
in
the
internal
time.
We'll
talk
about
that.
We
had
a
100-percent,
much
time
versus
an
SLA
and
in
the
world
that
I
come
from
slowest
down
and
slowly
spider.
G
G
B
G
G
Anybody
here
direct
start
on
the
Cassandra
and
I.
Thank
you.
So
there
are
couple
things
that
attracted
me
to
Cassandra
and
we're
going
to
go
through
that
because
it
could
offer
me
another
percent
of
time
and
it
can
offer
me
replication
of
the
box
how
many
people
got
master-slave
replication
right
about
babies,
because
you
different
size
themselves
or.
H
D
G
Object,
it's
a
tactic,
and
so
this
is
a
pie
at
100
of
CBS
for
the
day
in
the
market,
35
equals
E
is
the
order
type
by
on
her
shares
of
CBS
right
when
the
problem
is
that
big
business
standard
is
a
suggestion.
Fix
is
a
standard
at
the
the
network
level
for
establishing
two
machines
that
has
to
be
pretty
much
spot
on,
but
the
content
of
these
messages
is
different
for
each
each
potential
participant
in
the
financial
world
they
might
have
additional
tags,
so
they
can
relate.
G
Information,
in
fact,
is
something
internally
that
they
don't
want
to
share
at
this,
and
so
now,
I
have
the
requirement
of
always
up
replication
and
dynamic
schema
that
building
all
stores
before
a
tech,
data
kenai
other
databases
before
that
wasn't
going
to
work.
The
volume
of
you
had,
because
we
also
want
to
be
more
like
that.
G
So
what
about
the
market
we
need
to
compare?
This
is
why
I
said
earlier
that,
even
if
you're
doing
this
you're
one
person,
one
company,
we
would
need
to
look
at
all
the
words
all
the
markets
and
that's
about
40
or
50
execution
that
use
in
the
united
states
or
about
athlean
message
sustained
burst
into
so
many
messages
per.
Second,
that's
just
the
market
data
that
was
going
out
my
market,
so
that
that's
the
gift
that
he
directed
that
number
further
more
memorable
messages.
High
velocity,
the
pics
orders
below
is
a.
G
About
EVPs,
before
you
get
one
of
those
and
they're
wider
messages,
and
so
it's
all
to
put
about
the
system
we
have
to
do
over
a
million
per
second,
we
had
to
do
three
standard
deviation
on
confident
the
animal
market
event.
We
had
to
have
a
hundred
percent
of
time.
We
have
replication
and
we
had
a
definite
dynamic
you
to
keep
trying
a
different
message
formats.
G
Let's
talk
about
our
external
bit,
Cassandra
was
the
club
we
went
with
for
the
persistence,
their
direct
service,
the
event
processing
engine.
Does
anybody
have
any
experience
with
speed
processing
s
for
business
words
to
that
event,
processing
language?
What
we
did
it
director
be
des
Courtney,
been
event,
processing,
language
or
thinking
that
is
at
the
main
language
and
we're
going
to
give
some
examples.
We
had
continuous
cursor,
so
when
I
select
from
the
street,
you
might
get
a
continuous
result.
B
G
Back
with
the
conditions
that
I
was
asking
for
satisfied
examples
of
that
time,
when
windows
so
I
want
the
average
stock
price
group
by
simple
over
the
last
30
minutes.
Tell
me
what
that
is
very
in
it
right
now,
I'd
be
pattern
matching
across
windows
when
somebody
trades
on
this
side
of
the
market
and
then
trades
on
this
side
of
the
bracket.
G
That
does
that
a
number
times
unless
someone
to
call
monastic
way
to
it,
because
you
have
to
breaking
ball
there
here
are
some
examples:
processing
language
that
we
use
in
Dark,
Star,
select,
select
star
from
event
where
x
is
greater
than
100.
So
this
is
a
sequel
like
language
that
looks
like
sequel
to
be
the
only
governments
here.
Is
it
continuously
right?
And
so,
if
I
said
that
in
from
a
java
program,
every
time
a
new
event
occurs
on
the
commercial
thermoset,
it's
like.
G
Temperature,
in
that
event,
is
over
I'm
going
to
be
my
program,
get
over
it
support
versus
a
pole.
So
let's
talk
about
that
Aquino
so
that,
let's
say
on
the
average
temperature
from
a
thermostat
over
the
last
30
minutes.
I
don't
want
that
every
time
the
temperature
is
measured
at
want
you
to
give
it
to
be
plus,
what's
a
minute
about
the
average
over
the
last
13
pence.
Okay,
so
I
can
do
things
with
this
way.
Did
my
average
stock
price
just
jump
a
bunch
in
one
minute
versus
30
minutes?
G
G
B
G
Going
to
stockton,
then
the
symbols
IBM
and
the
price
goes
over
80
in
60
seconds
from
when
I
submit
this
parade.
This
is
occurring.
I
want
you
to
tell
me,
or
your
family
example,
find
events
that
occur
all
by
themselves
over
the
last
10
minutes.
It
don't
we
don't
repeat
themselves
within
the
10
minutes,
so
this
is
good.
Be
an
art
betray.
Tell
me
what
my
heart
beats
go
down
between
you,
two
different
systems
so.
G
G
B
G
So
this
man
is
persisting
all
the
inbound
data
and
all
the
directivity
the
drive
data
on
the
aggregated
data
or
completed
data,
one
of
the
things
that
we're
doing
community
this
starts
in
a
picture
we're
listening,
two
million
tonne
per
second
we're
only
interested
in
right.
Now
we
made
out
hundreds
of
thousands
of
clients
connect
to
the
system.
G
I
can't
send
me
one
of
those
clients
of
doing
a
mess
per
second,
so
I've
got
to
find
a
way
to
filter
an
aggregate
aggregate
them
down
into
meaningful,
but
make
them
completely
revived
us
and
the
send-off
things
that
they're
interested
in
and
so
Cassandra
help
with
that,
as
well
as
the
garden
star.
So.
G
This
is
a
young
cluster,
Dark
Star,
so
that
the
darker
p.m.
virtual
machine
with
dishonor,
you
had
a
number
of
fixed
beads
and
then
market
data
eating
the
system,
and
then
we
had
a
client.
So
with
a
potential
client
api,
a
global
clients
for
operations,
the
system
was
maintained
in
ran
for
the
three
countries:
New
York
das,
in
Ireland,
vanilla.
G
So
this
feather
are
there
any
questions
on
this
and
this
personal
her.
We
want
to
talk
about
it,
we're
in
the
deployment
system
like
this.
When
you
get
replication
hundred
percent
on
and
dynamic
cielo,
you
might
think
the
laws
of
physics
has
changed,
they
haven't
been
and
you
can
still
take
a
lot
of
mistakes
so
how
many
people
have
downloaded
Cassandra
just
to
start
paying
anybody
again
and
so
start
with
the
probably
go
through
some
things.
You
learn
start
with
queries
used.
It
falls
under
load,
etcetera,
let's
just
start
in
there.
G
If
you
want
to
have
a
really
bad
experience
with
any
North
sequel
database
for
your
relationship,
our
border,
okay,
load,
300,
3000
tables,
start
milling.
Client-Side
joints
it'll
be
lovely,
it's
not
going
to
work
because,
wherever
and
Buster
for
distributed,
okay
so
select
star
from
huge
distributed
table,
I
did
you
guys
have
no
cluster
another
note
Buster
each
one
of
those
notes,
the
Senate
everything
back
to
you
and
you're,
going
to
overwhelm
another
memory
and
40mm,
ok,
so
use
partition,
keys
and
cluster
competition
keys.
Help
me
find
a
row
that
have
insisted
that
distributed
database.
G
G
I'm
waiting
at
all
that
data
back
when
I
specified
that
partition
is
going
to
be
sorted
for
we
use
an
address
example.
So
we
have
people
in
our
customer
database.
Who's
changing
addresses
over
time,
and
you
want
to
add
an
address
to
that
and
you
had
a
customer
for
15
years
and
they've
changed
address
at
seven
times
and
so
using
white
rose
and
Cassandra.
You
cluster
on
the
date
of
the
address
change,
and
then
you
can
has
changed
when
I
do
select,
cleanest
address
where
partition
key
equals
custom
and
I'm
interested
in
women.
G
G
That
some
people
do
because
I've
been
using
it
for
so
long
to
me
as
a
distributed
tool.
Kit
gives
me
storage
application
and
dynamic
scheming
for
free
I.
Just
take
those
for
granted
notes
and
what
I
want
to
do
is
use
the
specific
capabilities
within
kassandra's
make
my
data
modeling
like
easier
in
the
description
growth
you
skinny
Rosanna,
make
sense
so
secondary
indexes
bad
idea
right.
Why?
Because
I
create
an
index
and
distributed
across
the
cluster
and
then.
G
G
Cassandra
belief
that
I
could
have
10
million
59
records
of
email
address,
so
user
logs
in
rather
than
use
a
secondary
index
to
the
other
table,
just
scan
the
email
table
get
the
idea.
Is
this
partition?
Even
the
user
table?
You're?
Not
that's
going
to
be
a
very
fast
free,
so,
instead
of
using
secondary
indexes,
flat-out
look-up
tables-
and
this
applies
really
to
any-
no
sequel-
not
just
asunder
with
the
white
/
document
indicates
those
types
of
things.
G
There's
things
that
dishonored
us
via
the
sequel,
databases,
group
or
no
sequel
to
it-
is
to
do
with
the
wagons
use
the
vaults
their
work,
a
lot
of
people
that
have
this
honor
clusters
and
before
the
application
is
done
or
before
the
data
modeling
is
done.
They
want
to
mess
with
configurations,
wait
now,
you're
done
use
the
defaults,
and
then
there
are
schools.
G
G
B
B
G
You're
testing
simulator
a
simple
like
you're
reading,
trigger
compaction
you,
if
you're
using
this
number
because
of
their
application,
because
the
up
not
because
the
dynamic
schema
then
science
that
buster,
so
that
goes
down
at
two
o'clock
in
the
morning.
You
schedule
it'll,
replace
them.
You
don't
have
to
get
up
at
two
thirty
and
replace
it.
You
can
do
it
in
your
schedule.
So
sighs,
sighs
your
clusters
so
that
you
have
done
these
things
to
them,
break
it,
break
it
again
and
practice
pacing
it
so
that
you
don't
have
fix
it.
G
Is
monotonous,
its
mind,
betting,
its
hours
of
fun
again
same
thing
right,
so
unless
once
you
get
your
applications
dies
once
you
once
you
tested
it
under
various
scenarios,
you
know
now
start
taking
a
look
at
the
parameter.
Okay,
so
you
run
those
fools
Cassandra
buster!
You
see
that
we
got
some
walking
me
blush
Raiders.
This
is
an
example
here,
I
might
I
might
increase
the
bus
riders
I've
got
some
far
new
GC
right,
so
maybe
my
team,
that's
too
big,
maybe
my
new
deep
sizes
too
big
that
would
be
test.
G
One
change
one
at
a
time
run
the
tests
may
keep
track
of
it.
Don't
don't
want
to
use
interval
bunch
of
things
at
the
same
time,
because
you've
got
so
much
going
on
together,
compassionate
going
on
the
deck.
That's
not
11
a
database
okay,
so
you
want
to
make
sure
that
you're
just
testing
one
thing
at
a
time.
G
Fraction
compared
using
up
resources
again
the
ba
physics,
we're
replicating
data
and
cluster
so
that
it's
a
bit
old
and
we
get
a
lot
of
congestion
and
people.
Reading
random
data
make
sure
you
have
some
memory.
Any
application
is
just
a
cash.
That's
the
way,
I
think
applications,
the
first
roll,
the
cash
is
the
member.
G
The
second
level
of
the
program
and
access
might
have
some
specific
assets
in
hand.
There's
the
OS
page
cache
I
saw
my
naked,
keep
her
note
for
Cassandra,
but
linux
that
blue
tube
is
really
smart
about
using
memory
for
us
page
catch.
Ok,
so
don't
change
memory
cheap
course,
a
good
number!
The
second
thing
to
the
first
thing
you
can
do
after
forty
year,
data
model
from
verbal
is
running
on
Sam,
because
then
what
you're
going
to
need
to
take
as
a
stupid
cluster
based
system
and
aggravating
all
the
eye
on
that?
One
bucks.
G
G
So
we
did
some
things
with
Cassandra.
We
got
triggers,
we
had
sort
procedures,
but
we
don't
know
we
didn't
contribute
to
open
source.
So
we
actually
maintain
your
own
bird
Cassandra
charlie,
and
it
was
easier
for
us
to
do
that
and
bill
Gregory
specific
pieces
of
functionality,
because
a
very
limited
documents
triggers
so
we
distribute
George
new
store
procedures.
F
F
G
C
B
G
Think
of
the
cluster
is
the
cluster
I
have
a
cluster
in
one
place
in
the
bust
of
another
place?
Think
about
that
as
your
cluster,
located
in
physically
district
area,
it's
a
different
way
to
think
about
what
you're
doing,
rather
than
a
hot
for
a
box
and
warmth
and
light
up.
The
replication
factor
in
the
cluster
23
year
of
3
is
a
minimal
and
red
pen,
and
so
every
right
that
I'm
going
into
the
cluster
at
least
three
copies
of
the
data.
So
I
can
destroy
two
machines.
H
G
Know
we've
been
doing
this
for
about
sixty
five
years,
and
so
it's
not
surprising
that
we're
still
learning
rates
going
back.
We
have
memory.
We
had
mainframe
that
make
this
collaboration
as
the
apple
of
the
database
to
the
home
machine.
Right
now
we
got
my
server
more
functionality
on
the
client
and
an
extra
guitar
push
the
database
down
in
the
work
week
or
departments
order.
Then
we
went
to
end
here
right
now.
We
have
sharding
physical
machines
and
down
the
database.
G
Then
we
get
a
goal
and
we
have
hei
right
and
someone
came
along
with
it.
I'll
make
a
box
and
I'll
put
all
my
databases
in
that
and
I'll
start,
which
is
great,
so
he
NC
interrogated
visa
all
these
companies.
They
solve
the
problem
and
we've
good
at
something
on
all
these
volatile
traits.
We
build
the
assistant
every
day,
but
as
a
fundamental,
fundamentally
different
approach,
I'm
strong
local
copies
everything
is
relative.
Dishonor
there's!
No
one
made
copy
everything:
the
replica
weapon,
aging
factor,
three
serenity
puppies,
I'm.
G
Sorry,
maybe
three
copies
on
my
local
data
center
from
a
local
cluster
baby,
three
copies
remotely.
So
the
things
that
the
sand
gives
me
a
nice
and
most
the
time
I
think
you
detectives
know
me
now
now:
I
can
snap
shot
right,
then
offload
to
mass
or
BBS
on
amazon
or
something
of
that
nature,
but
for
the
operation
of
the
Buster
in
green,
a
production
ass
and
will
give
you
more
headaches
than
that
I'm
good.
This
is
a
different
architecture.
H
C
C
C
G
G
D
B
G
D
G
So
I
do
is
not
real
time
on
the
school
that
I
can't
take
it
back
sister
than
make
me
go
time.
I
can
take
a
real
time
system,
assembly,
hatch,
but
what
to
do
and
I've
got
placed
and
other
companies
we
looked
at
it.
That's
for
some
best
for
some
of
the
ideas
the
MapReduce
can
come
and
find
that
processing
the
DNA
was
overhead
and
what
will
be
required
for
us
to
support
the
queries.
G
So
we
gotta
I
congestion,
two
very
20
milliseconds
per
thousand
rows
on
the
on
the
outbound,
and
that
was
a
sustained
congestion
in
search
for
second,
that
was
just
not
possible
with
with
du
monde.
Louis
is
a
master-slave.
Architecture
is
great
on
one
machine.
I
wouldn't
even
think
you
think
which
I
initially,
but
it's
not
the
types
of
operations,
the
scale
and
the
ingestion
rate
with
it,
with
replication
back
to
the
aj
thats,
not
I
don't
be
best
on
losing
space.
Lisa
didn't
that's
90
day
for
us.
For
that.
G
D
G
G
H
B
G
D
G
The
reason
that
the
sounder
work
with
me
wouldn't
work
well
with
the
stream
processing
and
you're
doing
with
we're
and
Julie
schema
cream.
Also,
we
didn't
use
any
cqo
and
Cassandra.
All
our
stuff
is
done
under
covers
with
drift,
so
everything
just
like
a
byte
buffer
us
all.
So
everything
good
trip
today,
but
the
thing
that
I,
like
I,
could,
if
I
processing
like
major
using
Esper
and.
H
G
A
service
which
most
object
and
can
write
itself
consider
less
and
back
in
the
database.
The
the
thing
I
liked
about
that
was,
in
my
main
event
table
in
dark
star
I
had
a
completely
dynamic
schema.
I
had
small
events
at
big
events,
so
I
had
to
conclude
the
dynamic
schema
that
matched
whatever
was.
It
was
inbound
to
the
stream
processor
and
rainy
intermediate
result
produced
by
the
stream,
and
so
that
might
be
something
to
take
a
look
at
the
other
thing.
That
was
awesome
as
the
fact
that
could
keep
up.
G
G
G
On
stress
levels,
the
when
they
talk
about
OCT
one
commodity
right
work
come
on
a
partner
distance
is
not
a
record
of
the
week
discount
stores
and
buy
consumer-grade
stuff
right,
I
consumer-grade
stuff
to
get
to
test
the
recovery.
Some
in
a
box.
We
men
are
typical
note
at
Apple
raid
raid
devices,
192
gig
ram
in
Korean
processors,
of
course,
resolving
a
big
fall
and
again
the
laws
of
physics
to
apply
like
SST
SST.
There
today
don't
use
consumer
grade.
Ssds
BB
great
great,
spend
the
next
row.
Whatever.
F
G
Time,
you've
every
right
and
you
can
set
up
snap
classes,
they
have
a
river
in
horse
or
you
can
even,
and
you
can
also
trigger
snapshots
and
cupcake.
The
snap
is
not
as
soon
as
you
create
are
things
you
can
say,
talk
to
another
San,
whatever
it
is
you're
using
for
that
we
are,
we
gonna,
so
you
grab
the
box.
C
G
D
G
G
G
D
G
G
G
G
G
F
G
G
D
G
B
G
Where
it's
posted,
you
think
you
have
a
c-section,
but
you
don't
try
to
move
all
the
state
out
of
the
out
of
the
app
right.
So
if
we
can
convince
sessions
well,
that's
not
possible
with
all
the
time
is
easier
to
t
and
so
may
be
using
this
under
the
a
small
Buster
that
each
Center
and
using
that
to
relative
septum
Romanian,
so
that
when
a
client
houses.
G
G
G
G
I
with
friendly
and
then
what
I
see
another
don't
think
about
the
ocean
project,
you
pick
something:
that's
small,
don't
don't
make
up
something
you
know
just
to
use
this
under,
but
pick
something
that
you
need
their
application
to
give
you
always
on
either
dynamic
diva
and
use
it
in
a
small
way
and
learn
and
beat
the
traffic
so
that
you
really
understand
the
characteristics
anymore.
So
you
can
get
for.
This
would
be
my
this.
G
B
D
D
G
Sequel,
you
know
what
he
says:
you
know
emulating
in
relational
databases.
Oh
yeah
I
can
see
that.
But
there's
something
else.
That's
really
is
really
happening,
and
that
is
we're
returning
a
row
on
a
white
robe
right
when
you're
doing
is
getting
a
go
thanks
for
taking
out
a
wide
database,
which
is
really
multiple,
tiny
boats.
And
so
when
you
do
a
select
you
don't
start,
the
partition
p
is
equal.
This,
the
cluster
cluster,
either
greater
than
or
less
than
equal
cetera.
G
G
The
separate
argument
right
now
make
it
look
like
reported
people
kind
of
make
it
look
like
relational
David.
You
know
duplicating
there.
You
know
we
were
told
not
to
do
that
and
the
last
second,
but
what's
the
one
thing
that
that
all
the
would
be
boarded
no
sequel
relational
world's
the
rate?
The
one
thing
the
COS
real
estate
is
well
for
Speed
us
to
materiality.