►
From YouTube: Apache Cassandra and DataStax Enterprise Explained with Peter Halliday at WildHacks NU
Description
Speaker: Peter Halliday, Senior Software Engineer at DataStax
A
Excellent,
this
is
actually
as
a
percentage
much
larger
than
usual.
How
many
of
you
have
heard
about
Cassandra
before
okay,
good,
so
we're
going
to
preach
to
the
choir
a
little
bit?
My
name
is
Peter
had
all
day
I
work
at
data
sets
and
I'm
on
a
tee,
nuh
senior
software
engineer,
that
is
on
a
team
for
a
for
Oksana
we're
going
to
get
into
what
that
is
in
a
little
bit.
But
what
that
means
is
it
wasn't
too
long?
A
Actually,
until
I
was
where
you
are
I
graduated
with
a
master's
degree
in
june
2013
from
Cornell
University
with
a
specialty
in
distributed
systems
and
I
got
hired
datastax
working
on
distributed
systems
programming
in
Python
programming
enclosure,
which
is
a
functional
base
language
on
the
JVM.
If
you
are
interested
in
any
of
those
kind
of
topics,
you
can
certainly
geek
out
with
me
after
the
but
we're
going
to
learn
a
little
bit
about
Cassandra
and
we're
going
to
you
know.
I've
already
asked
a
question
about
six
Cassandra,
but
I'm
not.
A
Some
of
you
might
not
know
that
you
are
using
the
Thunder
already
actually
how
many
of
you
guys
have
filed
your
taxes
with
TurboTax.
Okay,
you
use
a
patch
of
Cassandra.
If
you
ever
play
call
of
duty,
if
you've
played
call
of
duty,
you've
used
Apache
Cassandra,
have
you
ever
done
something
an
Instagram
or
Spotify?
What
about
Netflix?
So,
as
you
can
see,
a
theme
of
these
is:
if
you
have
a
large
amount
of
data,
then
you
might
want
to
consider
a
Pecha
Cassandra.
So
what
is
a
budget
Cassandra?
A
It's
a
massively
scalable,
no
SQL
database,
no
SQL
databases.
You
know
a
lot
of
people
are
starting
to
Paula
a
post,
relational
database.
This
is
unstructured
data
that
and
usually
it's
stuff,
that
you
want
to
distribute
all
over
the
world.
It's
so
it's
big
data.
It's
multiple
data!
Centers!
You
want
no
single
point
of
failure.
You
want
continuous
availability
without
compromising
performance,
so
we
cut
across
a
bunch
of
like
I
mentioned
a
lot
of
different
kinds
of
customers.
A
Here
are
an
example
of
a
bunch
of
workloads
that
our
customer
segments
that
we
have
that
utilize
cassandra
and
anything
from
you
know
ebay,
using
a
ecommerce
to
fraud,
detection
with
air
to
their
networks
to
sensor
data.
You
know,
I
mentioned
call
of
duty
online
games
a
lot
of
different
kinds
of
data.
This
isn't
just
you
know
online
websites.
A
They
use
Cassandra
for
a
lot
of
very
important
reasons.
One
is
they
have
data
centers
all
over
the
world
and
their
customers
all
over
the
world.
They
want
this
data
replicated
automatically.
They
want
a
consistent
linear
scalability.
They
want.
They
want
to
guard
against
failure.
Failure
is
going
to
happen,
and
so
they
want
something
plus
they
want
a
masterless
architecture.
You
know
no
slam
against.
A
You
know
our
friends
at
Microsoft
and
Oracle,
but
a
lot
of
those
relational
technologies,
including
MongoDB,
have
a
single
point
of
failure
in
this
mike
in
this
master,
server
and
you'll
see
as
I
introduce
the
architecture
of
Cassandra
that
we
get
around
that
so
a
little
history
lesson.
Whenever
you
talk
about
Cassandra,
you
really
have
to
give
a
hat
tip
to
the
Google
big
cake
table
paper
and
the
Amazon
dynamo
paper.
If
you're
into
distributed
systems
or
databases,
those
are
two
papers
that
are
publicly
available
that
you
should
read
up
on.
A
There
was
a
engineer
at
Facebook
that
read
those
papers
and
decided
to
program
a
database
Cassandra
for
a
use
case
of
their
messages,
their
inbox
app
at
the
time,
and
so
they
created
Cassandra
and
they
open
sourced
it
and
gave
it
to
the
Apache
foundation,
one
of
our
co-founders
Jonathan
Ellis
of
kind
of
discovered
around
that
time,
and
he
is
currently
serving
as
the
chairman
of
the
Apache
foundation
project.
This
is
an
open
source
project.
We
took
aspects
of
both
of
their
of
you
know:
I
Amazon,
dynamo
and
Google
BigTable,
combined
it
into
both.
A
So
in
order
to
talk
a
little
bit
about
some
of
the
terminology
of
Apache
Cassandra,
so
that
we
and
then
we're
going
to
dive
into
like
a
read
and
write
takes
both
on
the
cluster
level
on
the
node
level.
So
what
is
it
Cassandra
note?
Cassandra
node
is
just
you
know,
a
server,
that's
running
Cassandra
software
and
it
could
be
real
or
it
could
be
virtual
in
some
cloud
instance
somewhere.
So
Cassandra
cluster
is
a
group
of
these
nodes
that
are
working
together.
A
data
center
is
obviously
a
group
of
clusters.
A
A
So
the
Cassandra
cluster
allows
you
to
take
data
and
spread
it
all
around
the
cluster.
So
we're
going
to
talk
a
little
bit
about
how
that
is
partitioned.
These
are
two
there's
two
partitioning
strategies.
We're
going
to
talk
about.
One
is
the
random
partitioning,
which
is
the
default
and
recommended
strategy.
This
strategy
takes
a
hash
key
and
assigns
that
randomly
basically
equally
around
the
cluster.
The
other
is
an
order
partitioning
which
allows
you
to
sort
it
in
order
in
in
an
order
way
that
you
would
define.
This
is
like
many
things,
you'll
learn.
A
This
is
configurable.
All
these
choices
are
configurable,
so
indebted
partitioning
I
talked
a
little
bit
about
in
the
random
partitioning
that
we
take
your
data
and
you
assign
some
sort
of
partition.
Key
that
partition
key
is
applies
a
hashing
algorithm
that
by
default
we
use
a
murmur
three
partitioner
which
takes
the
partition
key
and
creates
a
hash
out
of
it.
A
A
For
example,
it
is
a
number
that's
assigned
to
them,
and
the
hash
algorithm
means
that
it's
assigned
a
bunch
of
numbers
that
it's
responsible
for
from
the
token
value
to
one
more
than
the
token
value
of
the
previous
node,
so
its
able
to
decide
on
which
know
that
data
is
owned.
So
that's
how
data
is
partitioned
in
Cassandra.
That's
we
get
that
from
the
Amazon
dynamo
paper.
The
second
part
that
we
get
from
amazon
is
the
replication
strategy,
and
one
of
the
central
parts
to
the
replication
strategy
is
what's
called
replication
factor.
A
Replication
factor
is
how
many
copies
of
the
data
do
you
want
to
have.
This
is
something
that's
also
configurable
replication
factor
of
one
obviously
is
just
one
copy
of
the
data.
We
don't
recommend
that
you
could
have
a
reputation
vector
to
we
recommend
three.
This
is
something
that
is
controlled
at
the
keyspace
level,
so
each
key
space
you
can
choose
a
different
replication
factor,
there's
two
kinds
of
strategies:
this
will
control
which
replicas
in
your
cluster
will
get
that
data.
We're
going
to
talk
about
the
simple
strategy
and
the
network
topology
strategy.
A
Again
these
are
configurable.
So
in
this
example,
the
first
node
has
a
simple
strategy.
Disk
cluster
has
a
simple
strategy
or
replication
factor
of
two,
and
the
first
note
is
probably
the
owner
of
that.
That's
the
one
that
has
done
assignment
token
and
simple
strategy
means
the
next
replica
will
be
the
one
that
will
get
the
second
copy.
If
you
have
a
reputation
of
factor
of
three
the
one
after
that
will
get
that
third
party,
so
that's
simple
strategy,
it's
pretty
simple!
It's
the
default
strategy!
A
Another
useful
one
is
network
topology
and
if
you
have
multiple
data
centers,
which
cassandra
is
optimized
for
out
of
the
box,
love
doing
anything,
you
can
control
the
copies
ASAP.
So
you
can
say
if
you
have
a
data
center
in
London,
you
have
a
data
center
in
new
york.
You
want
to
make
sure
you
want
to
have
two
in
one
location,
three
in
the
in
the
other
and
we'll
show
how
that
happens
on
the
right
level
in
a
couple
screens,
but
network
topology
is
isn't
just
useful.
A
If
you
have
multiple
data
size,
let's
say
you
have
a
bunch
of
racks
in
the
same
data
center
and
one
of
the
rafts
go
down
because
you
have
a
switch
problem.
This
happens
all
the
time
in
the
real
world.
You
can
use
network
topology
to
make
sure
that
the
data
is
stored
on
separate
tracks
and
that
way,
you're
dedicating
to
survive.
That
kind
of
failure,
so
that
you
might
be
asking
yourself
I,
can
tell
you
are
actually
how
does
piss
on
don't
know
about
Rex,
it's
a
EST.
A
No,
actually
it's
this
thing
called
the
stitches,
so
snitches
is
a
bunch
of
our
code
that
basically
informs
it
tells
on
each
other
about
the
topology
of
your
application
and
there's
several
snitches
that
you
can
use
again.
This
is
configurable
based
on
your
application,
if
you're
running
it
easy
to,
for
example,
there's
a
special
easy
turn
snitch
that
we
can
use.
For
example,
you
can
see
the
last
one
ec2
multi,
beacon
snitch.
This
can
be
used
to
make
sure
that
you
have
Shepherd
data
in
separate
regions.
A
So
if
one
region
goes
down,
you
can
survive
that
kind
of
region.
So
we
don't
just
tell
on
each
other.
The
note
don't
just
telling
each
other
about
the
network
topology.
It
uses
gossip
to
inform
the
other
nodes
based
on
what
other
nodes
are
down.
That
information
which
notes
are
down
and
which
nodes
are
up
is
spread
through
gossip,
along
with
other
messages
that
are
spread
through
gossip,
so
another
sort
of
technology
I
mean
a
terminology.
Is
load
balancing
and
so,
like
I
said,
we
have
a
masterless
architecture.
A
You
can
read
and
write
from
any
node.
So
the
strategy
that
the
client
uses
to
connect
to
the
nodes
are
one
of
these
three
strategies,
and
this
is
something
as
a
developer
you
get
to
choose.
You
can
say
that
you
want
to
use
the
round
robin
strategy,
and
that
means
it's
going
to
connect
to
a
random,
neither
the
local
data
center
or
the
remote
data
center.
In
this
case,
it
just
so
happens
to
use
local,
but
if
you
have
a
remote,
it
might
take
a
little
bit
longer
for
your
of
connection.
A
So
we
often
advise
to
use
the
DC
aware
their
data
center
way
around
robbing,
and
what
that
does
is
it
gives
preference
to
the
local
data
center,
which
will
be
faster,
but
if
there's
a
failure,
then
it
will
automatically
switch
to
the
next
data
center.
So
you
can
survive
that
kind
of
failure
and,
lastly,
like
I
said
there's
token
aware,
because
we
use
because
the
client
knows
the
kind
of
petitioner
that
you
use
you're
using
this
hashing
algorithm.
We
know
what
requests
you
should.
A
In
the
token
aware
strategy
we
know
which
node
you
should
be
connected
to
and
this
optimizes
for
faster
connections-
and
this
gives
you
more
power
as
a
developer,
to
choose
the
strategy
that
best
fits
your
applications
use.
So.
Lastly,
last
part
of
terminology-
we're
not
going
to
get
into
a
lot
is
virtual
nodes.
A
If
you're
interested
in
this,
you
can
certainly
talk
to
me
after
or
you
can
look
at
planet
Cassandra
for
details
on
this
too.
So,
let's
get
into
reading
and
writing
on
a
cluster
level
like
I
said
this
is
them.
This
is
a
location
independent
of
that.
So
you
don't
have
to
worry
about
connecting
to
one
particular
master
in
this
case,
and
whites
are
automatically
partitioned
and
they're
automatically
replicated
for
you
there's
nothing
that
your
application
needs
to
do
so
in
writing
data.
A
In
this
case,
it's
not
one
of
the
relatives,
so
the
replicas
whoa
overeager
there
yeah
the
coordinator
for
the
updates
to
the
replicas
and
the
replicas
acknowledged
that
the
data
was
written
back
to
the
coordinator
and
a
coordinator
sends
a
successful
response
to
the
client.
But
that's
not
the
real
world
right,
like
accidents
kind
of
happen.
So
when
only
two
new
boots
respond,
what
should
happen?
Well,
that's
really
up
to
you
as
a
developer.
A
What
should
happen
and
that's
what
I'm
going
to
do
something
called
blight
consistency,
so
you
might
have
heard
of
a
term
called
eventual
consistency.
Apache
Cassandra
Dennis
X.
We
prefer
something
called
a
tunable
consistency
that
as
a
developer,
you
get
to
choose.
How
consistent
do
you
want
your
data?
Do
you
want
to
have
strong
consistency
or
you
out?
We
consistently
consistency.
This
is
something
that
is
tunable
by
you.
Kurt
a
patient
/
read
/
write
across
multiple
data
center
operations,
we're
talking
about
for
consistency
levels.
A
One
of
them
is
any
so
that
means,
if
you
write
as
long
as
it
happens,
to
one
of
the
replicas
it
works.
That's
any
corn
is
a
majority.
A
majority
across
all
the
games
and
local
forum
is
the
just
at
local
data.
Centers,
just
a
forum
and
the
quorum
is
define
it
as
51%
or
more
for
all.
That's
the
strongest
consistency
very
similar
to,
like
you
know,
more
of
a
relational
model.
So
an
example
is
in
the
failed
notes
from
earlier.
A
If
you,
if
you
can
face
questions
to
the
end,
if
that's
okay,
so
in
the
failure
example,
will
this
successfully
succeed
if
you
have
a
quorum?
Well,
yeah,
of
course,
because
the
forum
is
a
majority,
whereas
if
you
have
two
failures,
obviously
that's
not
going
to
succeed
right
because,
like
that's,
not
a
majority
Oh
as
an
aside,
there's
actually
a
solution
to
this,
because
we
have
one
there's,
actually
a
configuration
setting
that
allows
settings
if
you're
using
a
data
sex
driver.
A
So
in
the
download
scenario,
how
does
the
node
eventually
learn
of
the
day
the
the
node
learns
of
the
data,
because
the
coordinator
stores
in
memory
a
hint
of
the
data
for
that
node?
And
when
the
note
comes
back
online,
so
the
data
is
replayed
to
the
hints
that
is
replayed
to
that
node
and
there's
actually
we're
going
to
talk
about
in
read
path
in
just
two
slides
or
so
we're
going
to
talk
about
another
scenario
in
case
that
note
doesn't
get
that
data
so
about
reading
gate.
It's
very
similar.
A
A
client
connects
to
a
random
know
that
node
becomes
the
coordinator,
and
that
coordinator
sends
the
read
request
to
the
replicas,
the
replicas.
It's
Emma
data
that
and
the
data
gets
sent
back
to
the
client.
So
in
the
case
of
the
failure,
I
talked
about
in
the
right
path
where
one
of
them
failed.
So
when
the
read
comes
in
that
we
might
be
stale,
maybe
because
it
didn't
get
that
hint
for
some
reason
right.
In
that
case,
how
does
the
you
are
duplicate
nodes?
How
do
you
dispute?
How
do
you
settle
the
dispute?
A
The
coordinator
sees
that
one
of
the
nodes
have
old
data
and
it
sends
the
new
data
back
to
the
plan
and
then
issues
a
request
of
to
the
out-of-date
node
to
refresh
their
data
from
one
of
the
other
nodes,
and
so
that
that's
another
method
that
we
repair
itself,
there's
also
a
percentage
chance
that
this
will
happen.
Randomly
there's
a
configuration
option
that
will
let
you
tunas
the
percentage
chance
that
a
reed
repair
will
be
done
automatically.
So
the
cluster
is
really
designed
to
help
repair
itself.
A
This
is
also
a
procedure
that
you
can
have
done
manually
on
the
cluster.
Obviously
there's
some
performance
in
back
to
this.
If
you
want
to
do
this
on
a
cluster
wide
level,
so
you
might
be
thinking
to
yourself
it's
kind
of
inefficient
right.
You
have
replication
factor
of
three
and
a
reed
consistency
of
two.
Why
do
you
send
three
read
requests
out?
Well,
the
answer
is
we
don't
really
we
send
two
out
automatically
and
if
those
two
don't
come
back
quickly,
then
we
sent
out
an
additional
one.
A
We
can
be
play
these
commit
logs
and
make
sure
that
data
isn't
lost
and,
at
the
same
time,
it's
written
to
these
men
tables
which
are
in
memory
versions
of
these
SS
tables
and
when
mm
tables
filled
up,
then
it's
written
to
disk
in
a
permanent
fashion
called
an
SSD
table.
These
ads
us
tables
are
immutable
and
can't
be
changed
and
the
only
way
to
change
them
is
through
what
we
call
compact
and
so
like.
A
As
you
can
see,
as
data
is
moved
around
the
cluster
and
becomes
less
relevant,
we
could
have
duplicate
copies
of
data
or
we
can
have
data.
That's
in
the
SS
tip
tables
that
that
cluster
is
no
longer
responsible
for
so
compaction
would
does.
Is
it
creates
new
SS
tables
from
the
old
SS
tables
and
then
deletes
the
old
SS
tables?
So
this
is
the
way
that
knows
update
these
SS
tables.
So
that's
what
happens
on
a
right?
It
read
is
a
little
bit
simpler
depending
on
your
application.
A
The
first
thing
we
do
is
look
in
memory.
Look
in
these
men
tables.
If
you
have
a
high
right
application,
these
might
actually
still
be
in
memory
and
then
it's
very
quick.
Otherwise
we
look
at
what's
called
a
bloom
filter,
I'm
not
going
to
get
into
what
a
bloom
filter
is,
but
if
you're
a
student
I
definitely
highly
encourage
you
to
look
up
on
Wikipedia
about
a
bloom
filter,
but
basically
it's
a
high
probability
index
that
allows
you
to
tell
with
high
probability
whether
something
is
in
one
of
these
SS
tables
or
not.
A
So
you
don't
have
to
scan
through
all
the
SS
tables,
and
so
then
we
look
in
the
bloom
filter
and
if
we
still
can't
find
it,
there
is
a
chance
it
still
on
the
SS
tables,
and
then
we
do
a
sikh,
obviously
across
the
SS
status.
So
usually,
if
it's
in
the
database
that's
caught
in
the
bloom
filter
and
it's
returned
rather
quickly,
and
so
that's
what
that's
the
repast.
A
So
this
Apache
Cassandra
and
I'm
going
to
talk
very
quickly
about
what
makes
datasets
different
datastax
is
a
company
that
supports
Apache
Cassandra
and
what
makes
us
different
as
a
company
as
as
opposed
to
companies
like
mondo
or
or
Oracle
or
sequel.
Server
is
a
couple
things.
One
is
Dennis
X
enterprise,
which
is
a
production,
a
certified
production
version
of
apache
cassandra.
This
is
something
that
we've
tested
internally
and
it
is
its
run
with
customers,
like
I,
said
like
eBay
and
Netflix,
and
it's
what
they
use.
A
It's
used
across
a
variety
of
use
cases
not
just
knee
Road
segments
of
the
market
is
integrated,
oltp,
integrated,
analytics,
integrated
search,
it
has
in-memory,
OLTP
and
analytics.
It
has
strong
data
protection.
It
has
management
tools,
we'll
talk
about
several
of
those.
We
also
saw
a
lot
of
different
languages
and
drivers.
In
many
cases
our
developers
are
the
ones
who
are
writing
these
drivers,
some
of
which
tastes
like
a
ruby
and
no
Jas
which
are
very
new
and
Python
our
open
source.
You
can
use
these
currently
for
free.
A
Other
versions
of
these
are
available
as
a
data
sex
customer
and
so
one
feature
that
you
should
use
data
sex
enterprise
for
security.
We
have
like
a
sex
security
as
a
part
of
Apache
Cassandra.
You
can
use
username
passwords,
you
can
use,
object,
permission
management,
so
don't
allow
a
user
a
gel
to
do
a
delete
on
this
table
because
he's
a
dumbass
or
you
might
have
default
of
client
encryption
between
a
certain
nose.
A
But
if
you're
an
enterprise
customer
you
want
real
enterprise
security,
which
is
things
like
external
authentication
which
allows
you
to
connect
to
Kerberos
or
Active
Directory
ldap.
You
want
a
tres,
you
want
get
a
quick
encryption
for
the
data,
that's
at
rest,
so
that
people
don't
look
at
your
data
in
the
SS
tables.
Unless
it's
encrypted,
you
want
data
auditing.
Those
are
all
features
that
are
part
of
data
steps.
Enterprise.
A
Another
feature
is
absent,
I
told
you
I
was
going
to
talk
about
the
product
and
responsible
for
like
as
a
hacker.
We
all
love
our.
You
know
our
tools
that
are
on
the
command
line,
but
when
you
have
clusters
that
are
like
thousands
of
nodes
at
hundreds
of
nodes,
you
often
want
a
browser-based
or
something
that
you
can
do
large-scale
operations
across
the
cluster.
This
all
ops
center
also
allows
you
to
do
backups
of
your
cluster
it
even
with
replication,
like
people
can
do
dumb
things
like
delete
old
tables
and
delete,
holds
key
spaces.
A
You
want
to
be
able
to
backup
from
events
like
that.
Opscenter
allows
you
to
do
that.
You
also
want
to
be
able
to
do
things
like
alerting.
So
when
people
like
you
have
certain
spikes
or
certain
other
things,
you
can
be
alerted
of
events
like
that
on
one
of
the
common
of
scenarios
for
using
ops
center.
We
use
this
at
internally.
Data
sex
is
what's
called
provision,
so
let's
say
you
want
to
in
Amazon
ec2.
A
You
want
to
create
10
notes,
you
can
click
on
the
create
new
cluster
button
and
add
in
the
IPS
of
the
machines
that
you've
already
spun
up
and
add
in
the
either
the
connection
or
the
SSL
key.
It
will
automatically
install
automatically
optimally,
configure
these
machines
and
then
suddenly
you
have
a
10
note,
plus
tur.
And,
alternatively,
if
you
have
ec2
of
credentials,
you
can
just
log
in
to
your
ec2
credentials,
and
it
will
automatically
do
all
of
that
for
you.
So
that's
app
center.
A
A
It
won't
have
all
of
the
abilities
of
DSC,
but
one
of
the
big
uses
that
people
use
apache
a
bit
of
sex
and
prices
integrated
search
for
that
we
use
a
technology
called
solar
and
I'm
not
going
to
get
into
those
of
the
search
and
the
analytics
in
fine
detail.
You
have
questions,
please
let
me
know,
but
one
of
the
benefits
of
this
is
it
allows
you.
A
Instead
of
setting
up
a
separate
of
solar
instance,
we
optimally
configure
these
integrated
for
textual
geospatial
of
faceting
searches
of
a
you
know
in
the
cluster,
the
Cassandra
cluster,
and
you
can
actually
segregate
these
so
that
the
search
nodes
are
run
in
a
separate
data
center
so
that
your
application
doesn't
load
down
your
cassandra.
Your
customer,
facing
up
to
sign
their
data.
Also
search
is
one
thing
that
your
application
might
need.
It
also
might
need
analytics
if
you're
doing
no
SQL
so
that
you
don't
use
joins.
A
You
might
have
to
use
something
like
Hadoop,
which
is
what
we
use
for
integrated
batch
analytics
and,
if
you're,
using
the
integrated
Hadoop.
It's
probably
because
you
don't
have
your
own
hood,
or
do
you
want
to
avoid
the
single
point
of
failure
in
to
do?
There
is
a
name
server,
has
a
single
point
of
failure,
and
also
HDFS.
If
you
haven't
connect,
if
you
haven't
tried
to
configure
HDFS,
it's
kind
of
a
pain,
and
so
maybe
you
want
to
avoid
that
by
using
our
integrated
approach.
A
This
is
something
that
we
call
bring
your
own
food
and,
lastly,
what
it's
kind
of
the
new
thing,
which
is
spark
real-time,
Hadoop
I,
mean
real
time
analytics,
and
this
is
a
wig
of
data
section,
a
price
of
4.5,
and
this
allows
you
to
do
things
in
a
virtually
real
time.
So
this
kind
of
offering
we
think
is
very
compelling,
as
opposed
to
our
competitors,
which
have
single
points
of
failure
and
don't
have
linear
scalability
or
when
you
look
at
the
benchmarks.
Our
performance
outlays.
A
Those
performance
as
well
as
developers,
I,
want
to
say
two
things.
One
is,
you
know:
I
encourage
you
to
use
us
for
the
hackathon
or
four
projects
on
you
can
go
to
planet.
Kisan
org,
you
can
go
to
get
sexcom,
look
at
customer
use,
cases
look
at
tutorials
and,
lastly,
we
have
a
booth
upstairs
we
are
hiring
and
you
are.
We
do
have
student
internships,
please
come
up
and
give
us
your
resume
drop
by
and
talk
to
us
a
little
bit,
I'm
going
to
be
up
there
for
a
little
bit.