►
From YouTube: DataStax Startup Panel: Cassandra Summit 2015
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
B
Hello
and
good
evening,
everyone,
my
name,
is
Matt
file
and
when
the
two
co-founders
of
data
sacks
and
welcome
to
the
startup
panel,
one
of
the
things
that
we
do
at
datastax
is
provide
our
software
for
free
to
startups,
and
we
do
that
with
the
goal
of
getting
more
people
familiar
with
the
software
getting
feedback
on.
What's
good
and
bad,
usually
they
tell
us
more
about
the
bad
than
the
good,
which
is
actually
good.
A
great
quote
is
bad
news
is
good
news.
Good
news
is
no
news
and
no
news
is
bad
news.
B
Think
about
it.
What's
good
and
up
on
this
age
say
all
what
everyone
introduce
themselves
in.
A
second
is
a
group
of
people
who
have
worked
very
intimately
as
a
part
of
that
program,
as
well
as
with
the
software
and
we're
going
to
keep
this
relatively
free
flow.
In
terms
of
some
questions
about
what
you've
experienced,
what
you
don't
like
some
recommendations
for
the
greater
audience
and
we'll
see
where
it
goes.
B
A
A
C
Right,
I'm,
Sam
Bisbee,
the
CTO
at
threat
stack
a
boston-based
company.
We
do
a
continuous
security
monitoring
for
one
servers
that
are
typically
deployed
in
clouds
like
Amazon
pretty
much.
A
vast
majority
of
our
engineering
team
had
never
used
Cassandra
prior
to
working
a
threat
stack.
So
that
was
an
interesting
experience.
We
had
also
been
scaling
up
as
a
time
series
data
company
trying
to
shove
over
a
hundred
thousand
inserts
per
second
into
Cassandra
two
terabytes
a
day
and
so
learning
how
to
operationalize
that
and
then
also
remove
delete
data
efficiently
was
pretty
interesting.
E
B
To
put
more
flavor
there,
sebastian
has
an
interesting
job.
Data
sax
is
a
corporation,
is
an
enterprise
company,
so
you
know
we
sort
of
count
our
customers
by
the
hundreds,
if
not
load
thousands,
whereas
with
the
startup
program
you
have
a
large
number
of
companies
come
in
and
he
is
the
only
technical
person
except
for
you
who
recently
joined
who
worked
with
all
of
those
hundreds
of
accounts.
B
So
the
interesting
thing
about
what
Sebastian
gets
that
no
one
else
that
data
sax
gets
is
that
you
see
hundreds
of
customers
on
a
month-by-month
basis
and
you
get
to
solve
all
their
technical
issues,
so
he
needs
to
say,
he's
a
little
crazy
at
from
time
to
time.
So
getting
into
this,
what
I
love
to
actually
learn
a
little
bit
from
each
of
you
is:
what
was
your
original
motivation
for
getting
into
Cassandra
like?
What
could
you
not
do
is
something
else
out
there.
A
So
as
a
one
point,
oh,
the
biggest
reason
that
we
actually
got
into
using
cassandra
is
that
we
had
some
serious,
like
high-velocity
high-volume
injection
ingestion
issues,
rather
that
we
had
to
take
care
of.
So
you
know
we
started
out
with
trying
to
do
counters
and
read
us.
Then
we
started
out.
Then
we
moved
to
you
know
we
looked
at
because
couchbase
we
looked
at
HBase.
We
looked
at
basically
anything
else
that
could
be
used
as
to
count
things.
A
We
did
and
then
quickly
found
out
that
the
hardest
thing
to
do
in
computers
is
count,
and
when
you
talk
to
people
who
are
non-technical,
they
don't
believe
that,
but
that's
actually
I
think
the
reality.
So
we've
spent
the
last
couple
of
years
trying
to
figure
out
how
to
make
Cassandra
count
things
very
well
and
I.
Think
mostly
thanks
to
Sebastian.
B
A
In
terms
of
counters,
we
do
about
at
pink.
We
do
about
400,000
counter
operations
a
second
at
lower.
We
do
maybe
about
a
hundred
thousand
counter
operations.
A
second
and
each
counter
operation
could
be
anywhere
from
you
know.
One
increment
of
two
hundred
cells
to
maybe
I,
don't
know
a
couple
thousand
increments
on
a
couple
thousand
cells
on
a
single
row
and
that's
just
on
one
table.
C
So
you
know
we're
a
multi-tenant
platform,
so
we
implemented
multi-tenancy
at
the
column,
family
level
and
we
just
completely
drowned
our
cluster
in
files
and
tables.
We
quickly
learn.
We
never
ran
into
the
counting
problem
because
we
had
other
solutions
for
that,
but
pretty
much
everything
we
were
doing
was
pre
materialized
views
and
that's
actually
been
pretty
successful
for
us.
C
So
far,
we've
had
great
success,
either
very
wide
and
shallow
schemas
or
very
narrow
and
deep
schemas,
so
either
a
schema
that
drives
a
very
specific
widget
or
a
schema
that
will
drive,
for
example,
go
look
up
a
process
for
event
on
a
given
environment
and
then
contextualize
that
against
everything
else
in
that
process
tree
every
network
event
in
that
process
tree
at
that
point
in
time,
but
then
also
historically
over
the
last
30
days
and
go
pull
that
all
together.
Yeah.
D
D
Nobody
should
ever
do
that
and
they
should
use
something
like
RabbitMQ
that
we
learned
quickly.
We've
done
other
crazy
things
like
attempt
to
store
sessions
in
Cassandra.
For.Net
probably
should
also
not
do
that.
So
it's
definitely
been
a
learning
experience
since
point
8
and
for
us
you
know.
Originally,
we
started
looking
at
Cassandra
when
we
were
first
building
our
startup
purely
because
we
knew
we'd
be
processing
a
ton
of
data
with
a
ton
of
updates,
as
we
try
to
index
like
content
online.
B
E
That
kind
of
has
heard
about
those
experiences
and
read
about
them
and
are
being
proactive
and
building
a
system
that
that's
going
to
need
to
scale
it
doesn't
need
to
yet-
and
you
know,
those
folks
usually
have
a
lot
more
time
to
figure
out
how
to
do
things.
The
right
way.
Cassandra
surprises
me
all
the
time
right.
Actually,
yesterday
something
I
learned
something
about
details
and
SS
tables.
That
I
wasn't
aware
of
specifically
that
if
you
change
TTLs
on
an
SS
table
it
only,
it
only
affects
rights
that
are
coming
after
the
fact
right.
E
So
little
things
like
that
are
will
always
keep
coming.
But
one
of
the
one
of
the
important
things
I'd
like
to
kind
of
make
everyone
aware
of,
is
just
that.
Cassandra
and
DC
are
both
very
tunable.
They
have
a
lot
of
levers
and
one
of
the
ways
to
be
really
successful
with
the
products
and
get
things
to
work
well,
is
to
find
out
what
levers
to
use.
One.
B
So
we
talked
a
little
bit
about
Cassandra
and
what
you
guys
are
doing
there,
but
at
data
Sachs,
there's
an
enterprise-grade
edition
of
cassandra
that
we,
you
guys
have
access
to
for
free
that
software
includes
things
like
search
capabilities
through
solar
technology
analytics
through
the
hoop
stack
as
well
as
spark.
It
also
includes
things
like
performance
package
by
doing
things
with
the
main
memory,
security
and
an
OP
Center.
Can
you
guys
talk
briefly
about
which
of
those
features
if
any
are
using
and
what
your
motivation
was
for
using
those
over
other
things
in
the
field.
A
A
What
yeah
I'm
getting
there?
Let's
try
to
try
to
think
while
I
talk
it
doesn't.
I
am
not
good
at
multitasking,
so
the
reason
we
use
spark
is
that
one
of
the
things
that
is
a
you
know
an
issue
with
counters
and,
like
I
said,
we
count
a
lot
of
things,
whether
it's
counting
page
views
or
counting
unique
users,
and
we
count
unique
users
by
using
something
called
HL,
l
plus,
which
is
hyper
log
log
plus,
which
means
we
store
bits
of
data
in
very
large
sequences.
A
Also,
something
Cassandra
was
not
built
to
do
an
anti-pattern
that
we've,
you
know,
fought
that
Cassandra
fights
us
on
regularly.
So
in
order
to
you
know,
double
check
ourselves,
we
use
spark
spark
is
a
is
a
framework
that
allows
us
to
run
against
a
large
amount
of
raw
data,
and
you
know
check
our
aggregated
data
to
make
sure
we're
approximately
correct.
A
And
you
know
those
things
have
enabled
us
to
those
things
a
bit
amazing
and
made
our
life
easier,
because
the
operational
administration
of
solar
and
of
spark
which
would
normally
require
Hadoop
is
actually
kind
of
a
pain
in
the
butt.
So
because
we
already
know,
cassandra
is
actually
the
devil.
We
know
in
this
case,
and
we
know
the
headaches
that
come
with
administering
it
I'm
not
having
to
administer.
Another
set
of
you
know,
distributed
systems
or
two
other
sets
or
three
other
sets
or
distributed
systems
has
certainly
alleviated
quite
a
few
headaches.
C
Great,
so
our
usefulness
for
op
center
has
acted
as
kind
of
a
bell
curve.
Actually
it
was
not
useful
in
the
beginning,
because
we
were
running
op
center
on
the
same
cluster.
We
were
trying
to
monitor
pro
tip.
Don't
do
that.
We
were
also
the
number
of
tables
just
wasn't
making
it
happy,
and
then
we
fixed
those
things
that
we
saw
a
lot
of
value
from
it.
But
then
we
were
large
librado
and
Gravano
users.
C
C
You
know
even
today,
I
question
very
heavily
our
usage
of
leucine
and
potentially
solar,
just
because
all
of
our
data
is
highly
structured.
We
don't
do
analysis
with
was
seen
as
just
kind
of
unnecessary,
so
we
might
actually
end
up
implementing
inverted
index
through
a
sparse
matrix
on
Cassandra.
Instead,
we
won't
be
doing
Hadoop
because
we
banned
Hadoop.
We
also
banned
zookeeper
and
we've
already
got
spark
running.
We
basically
take
data
off
RabbitMQ
10-minute
batch
into
s3
and
then
pull
that
into
spark
because
sparks
driven
was
just
looming
and
waking
file.
C
Why
do
we
Bend
zookeeper
I'm,
trying
to
say
whether
or
not
to
do
that
Athena
chrous,
Ponce
or
the
actual
one
but
yeah
the
actual
one?
So
the
actual
one
is
we
have
very
few
pieces
of
our
infrastructure,
they're,
actually
master
slave
and
we've
had
too
many
prior
experiences
of
putting
zookeeper
into
a
split
brain
situation
and
having
associated
pain,
and
it's
just
it's
completely
unnecessary.
All
the
compute
in
our
stack.
C
B
D
So
we
use
outside
Cassandra,
we
use
spark
and
we
use
solar
when
we
were
introducing
Cassandra
at
Channel.
Like
you,
a
lot
of
people
existing
employees
at
you
like
you
like.
Well,
we
shouldn't
use
solar.
We
should
use
elastic
search
which
I
was
like
well.
Why
we
manage
another
thing
when
Moses
functionality
is
built
into
DSC
over
time,
they
kind
of
saw
those
ways,
and
you
know
we
stuck
with
solar
and
it's
worked.
D
E
Alright,
so
I
don't,
I
don't
use
any
of
these
things
in
production,
but
the
lot
of
the
DC
integrations
when
you,
when
you
think
about
the
apache
suite
of
open-source
products
you
can
you
can
why
all
these
things
together,
and
sometimes
it
takes
duct
tape
and
sometimes
virgins
change
and
one
of
the
things
that
these
he
gives.
You
is
just
time
to
value
and
simplicity,
and
so
you
know
with
a
couple
api
is
like
you
guys
saw
this
morning.
E
B
A
I
actually
want
to
make
a
point
of
this,
because
I
don't,
I
don't
think
anybody
really
unless
you've
gone
through
hell
as
as
a
start-up
and
when
I
say
gone
through
hell,
I
mean,
like
some
startups,
have
a
little
bit
of
trouble.
A
They'll
get
like
you
know
a
website
outage
and
like
it's
annoying
and
it
might
be
the
end
of
the
world
if
your
startup
is
just
a
website,
but
if
you
have
like
serious
back-end
infrastructure
and
you've
had
downtime
or
really
big
problems
with
that
with
with
serious
infrastructure,
you
know
you,
you
actually
have
to
sit
in
question
a
lot
of
times.
Like
am
I,
going
to
be
able
to
survive.
This.
Is
my
company
going
to
be
able
to
survive
this
and
and
sebastian
has
spent?
A
A
You
know
late
nights
in
in
in
our
office
and
we're
just
one
of,
however
many
hundreds
of
startups
that
he's
had
to
work
with,
and
I
would
say
that
on
the
whole,
you
know
the
talk
that
I'm
giving
tomorrow
is
really
95,
plus
percent
a
result
of
of
his
being
able
to
be
a
part
of
like
a
full-time
employee
of
my
company,
as
well
as
datastax
and
and
at
least
these
two
guys
and
everybody
else.
A
B
We
do
not
actually
lists
him
as
a
feature
of
the
startup
program
on
the
website,
but
you
know
we
could
put
it
on
there
all
right.
It
happens
all
right.
So
why
don't
we
actually
go
through
and
just
briefly
talk?
Is
it
this?
The
point
of
this
was
a
little
bit
like
the
good
stuff
is
stuff
that
you
we're
not
up
here
to
preach
the
good
stuff.
You
can
weed
the
marketing
material
for
that
this
is
actually
which
I'm
trying
to
convey
things.
B
That
would
be
lessons
learned
of
things
to
either
pay
attention
to
if
it
ever
happens
to
you
or
things
to
avoid
altogether.
So
you
know
and
that's-
and
we
we've
done
a
panel
like
this
at
each
of
the
startups,
where
it's
basically
like
the
horror
stories,
this
one
just
got
appropriately
titled
slightly
nicer
than
what
it's
been
in
the
past.
Why
don't
you
go
through
and
talk
about?
B
D
A
Great
question:
he
asked
it
if
I
have
so
many
bad
stories.
Why
am
I
still
using
well?
The
answer
is
that,
like
a
lot
of
the
technical
problems
that
my
company
has
to
solve,
they
just
there
aren't
really
solutions,
and
the
thing
is:
is
that
like
in
order
to
make
a
company
in
order
to
build
a
company
that
is
a
technical
company?
A
Sometimes
you
have
to
solve
problems
that
don't
really
have
solutions
and
that's
hard
and
you
take
the
best
of
what's
available
and
you
make
do
and
you
hope
that
it
works
and
like
there's
a
lot
of
that
and-
and
you
know,
there's
a
reason
that
Matt
and
I
have
become
such
good
friends
over
the
past
four-plus
years,
because
we've
we've
hit
some
really
hard
times,
and
you
know
I
could
tell
the
story
of
counters
up
here
and
like
the
hell
that
it's
been
through
for
us
for
many
years,
but
if
you're
really
interested
in
how
we
solve
that,
like
I'm,
going
to
do
40
minutes
on
that
tomorrow
morning,
I
think
it's
much
more
interesting
to,
rather
than
tell
a
very
specific
horror
story,
to
kind
of
convey.
A
A
But
after
four
plus
five
plus
years
of
breaking
rules
left
and
right,
I
can
tell
you
that
the
data
that
note
you
have
to
know
the
rules
in
order
to
break
them
and
it
takes
a
company
to
push
the
limits
like
the
way.
Datos
taxes
I
feel
like
fanboy,
in
a
very
weird
way.
Considering
how
much
like
bad
stuff
I've
been
saying
over
the
years,
you
have
to
learn
how
to
learn.
A
You
have
to
learn
how
things
are
done,
the
right
way
in
order
to
learn
how
to
do
things
the
wrong
way
and
we've
dumped
enough
stuff
the
wrong
way
in
order
to
make
it
the
right
way.
It's
counterintuitive
but
I
feel
like
anybody
up
here.
It
would
just
nod
their
head
and
be
like
yeah
that
makes
sense
and
until
you've
been
through
the
fire,
it
actually
really
like
it
doesn't
make
sense
until
you've
been
through
the
fire.
C
No
it's
about
Cassandra,
so
we
originally
had
this
brilliant.
So
I
worked
with
a
lot
of
other
databases.
My
prior
company
was
a
database
company
I
had
a
very
fundamental
concern
coming
in
to
doing
time,
series
on
Cassandra,
where
everybody
was
telling
me
hey,
use,
ttls
and
like
well.
No,
the
whole
point
of
the
database
is
to
not
delete
my
data
I.
Don't
trust
Cassandra
because
of
all
the
horror
stories.
C
I've
heard
much
like
this,
because
I
don't
trust
when
the
Cassandra
community
implements
a
feature
for
the
first
time
and
yeah,
and
so
I
just
was
going
to
basically
go
in
and
do
data
cleanup
myself
and
reclamation.
That
was
a
very
bad
idea.
We
got
into
a
situation
where
we
were
running
our
36,
no
cluster
of
I
to
to
xls
at
like
85
to
90
percent
disk
utilization
per
node
for
a
very
long
time.
C
A
C
So
part
of
the
moral
story
was
to
set
TTLs,
but
that
doesn't
actually
fully
fix
the
problem
because
of
all
the
sessions
of
this
week
on
dtcs
problems.
We
were
also
incised
here:
compaction.
That
was
another
problem.
The
moral
that
story
was
it's
very
hard
to
pick
out
just
one,
because
there
was
like
many
things:
there's
a
hundred
micro
fixes
to
go
and
address
the
fundamental
problem.
I
think
one
of
the
things
it
Sebastian
with
it
to
earlier
was
it.
C
There
are
so
many
tunable
in
Cassandra
itself
and
then
tossed
in
the
JVM,
which
is
black
magic,
and
you
know
you're
in
a
world
of
pain,
because
you
know
we're
a
start-up
of
people
who
haven't
done
large-scale
operations
of
Cassandra.
Before
and
literally
it's
like
peaches,
lock,
mr.
dev,
ops
and
me-
and
you
know,
the
rest
of
the
people
are
doing
what
they
should
be
doing,
which
is
building
product
and
value,
and
it's
the
two
of
us
just
you
know
making
it
work
and
we
way
understaffed.
C
Cassandra
I
mean
anybody
going
into
the
Cassandra
they're
going
in
with
less
than
like
two
or
three
full-time
engineers.
Who
would
really
understand
data
model
JVM
and
operations
at
the
cluster
I
can't
recommend
it?
We
have
come
out
the
other
side
of
that
tunnel
through
all
the
fires
and
everything,
and
now
Cassandra
is
one
of
the
most
stable
happy
parts
of
our
polyglot
data
platform.
C
B
B
The
first
one
had
about
145
people,
so
the
ecosystems
gotten
bigger
and
one
of
the
really
nice
things
about
a
bigger
ecosystem
is
there
is
a
partner
network
now
or
then
we
phrase
important
work,
there's
an
ecosystem
of
people
of
companies
whose
sole
job
is
to
actually
run
other
people's
Cassandra
deployments
as
a
service,
whether
that's
in
a
public
cloud
or
manage
on
your
own
hardware,
if
I
would
highly
recommend
unless
you
really
want
to
don't
run
it
yourself.
Well,.
C
There's
that,
but,
as
part
of
this
sense,
does
it
start
up
and
like
that's,
not
affordable
and
it's
very
hard
as
a
start-up
where
you
know
the
ratio
of
engineers
to
databases
is
already
bad
enough
for
us
to
then
go
out
and
bring
in
contract
to
hire
or
just
contractors
or
ms
peters
or
anything.
It's
really
not.
It's
really
not
a
cost
effective
model
for
us
I.
C
There's
way
too
much
content
in
there,
I
would
say
you
get
really
fine
grain
in
the
metrics,
so
there's
just
general
JVM
stuff
like
get
really
fine.
Grain
of
the
metrics
really
understand
all
the
different
eden,
space
etc.
Pärnu
like
if
you're
going
to
go
roll
out
java,
8
or
you're
going
to
go
roll
out
g.
C
A
Anything
you
download,
you
have
to
download
separately
with
just
cuz
Oracle,
like
that's
your
choice,
but
I
think
the
huge
take
away
from
that
is
monitor
and
instrument
every
single
possible
thing
and
worry
about
whether
or
not
you've
over
instrumented
or
over
monitored
later,
because
that
will
probably
never
be
a
thing.
I
was
going
to
tell
a
story,
but
I'm
gonna
let
ilya
go
because
we're
getting
the
sign-in
and.
B
D
He's
got
a
lot
of
experience
with
the
sequel
side
and
the
mark
logic
side,
and
we
actually
turned
off
all
our
marklogic
instances.
A
few
weeks
ago
we
bought
a
cake
to
celebrate
the
death
of
marklogic
and
nobody
could
be
happier
so
yeah.
But
as
far
as
our
horror
stories,
I
mean
we've
used
DSC
for
years.
I
could
tell
you
that
you
guys
do
a
great
job
coming
up
with
features
and
marketing
them
and
they
never
work.
D
So
I'll
talk
a
little
bit
about
one
recent
case,
so
ops
center
launched
this
feature
of
like
performance
services
and
slow
queries
and
sounds
great
right.
It's
going
to
help
you
find
the
that
makes
your
stuff
run.
Slow
and
I'll
help
you
fix.
It
sounds
amazing.
The
only
problems
when
you
turn
it
on
will
crash
Cassandra,
eventually
and
eventually
became
every
six
hours
for
us
and
when
you
go
turn
it
off,
it
actually
doesn't
turn
off.
D
And
then,
when
you
tell
us
the
Sebastian,
he
goes
no,
it
should
turn
off,
can't
be
that
and
you
plumb
the
source
code
in
DSC,
and
you
tell
them
no
it's
right
there.
It
doesn't
turn
off.
So
yes,
I
would
say
our
biggest
lesson
is
yeah,
even
though
we
wait
for
like
fixes
and
whatever
the
next
version
is
some
of
the
default
stuff
and
what
launches
in
new
versions,
it
usually
doesn't
work.
The
first
time
around
seen
can.
A
I
can
I
can
I
throw
us
out.
They
get
there
a
couple
years
back
right
before
the
New
York
City
Cassandra
summit
I
had
a
I
had
a
node
decommission
like
it
just
decommissioned
itself,
and
I
I
woke
up
in
the
middle
of
the
night
to
the
node,
just
going
like
just
commissioning
and
the
like
poured.
9160
the
the
CQ
up
or
stopping
listening
and
I
was
like
freaking
out
I'm,
like
I,
don't
know
who
my
company
would
decommission
a
node
without
talking
to
me,
I
would
started
going
through
bash
histories.
A
I
was
forming
out
like
what
is
going
on
and
then
I
started
digging
through
the
logs
and
I
found
a
log
entry.
That
said
out
of
disk
space
decommissioning
and
apparently
at
some
point,
someone
in
DC
on
the
DSC
team
thought
it
would
be
a
cool
idea
to
just
when
the
nave
runs
out
of
space
to
decommission
the
node,
like
just
remove
itself
from
the
cluster,
and
so
I
I
woke
up
and
freaked
out
and
I
sent
this
bug
report
to
at
the
time.
I
didn't
remember.
A
Who
was
we've
had
so
many
people
help
us
over
the
years.
Yeah
I
think
I
might've
entire
Hobson
and
option,
and
I
sent
it
to
somebody
and
he
goes
there's
no
way.
We
would
do
something
this
stupid,
so
he
forwarded
it
to
someone
else
and
the
email
thread
was
just
five
or
six
people
saying
that
there's
no
way
someone
else
must
have
done
it
there's
no
way.
We
someone
would
do
this
this
stupid.
Then
someone
found
that
the
patch
and
they
did
a
get
blame
and
remove
the
name
from
it
and
they
were
like.
A
E
Alright,
so
one
thing
with
the
performance
services:
it
does
turn
off
of
you
set
it
to
zero
all
right.
You
have
to
set
it
to
zero
I'm.
Just
saying
just
saying:
yeah
you
do
all
right,
but
no
so
so
to
be
fair.
I.
Do
I
do
get
a
lot
of
I.
Do
get
a
lot
of
cries
for
help,
let's
call
it
that
kind
of
like
I.
We
were
in
a
firefight
help
me
out
it
really.
B
B
A
So
so,
to
be
fair,
the
idea
of
date,
tiered
compaction,
is
actually
really
really
good.
The
problem
is,
the
implementation
is
really
really
bad.
In
the
absolute
best-case
scenario
it's
bad
and,
and
the
folks
over
at
I
can
remember
his
name.
I
can't
remember
the
company
I
said
it
before
you
remember:
okay,
that's
fair!
So
oh
I,
remember
it
now,
so
the
folks
over
at
CrowdStrike
did
a
really
good
job
of
elucidating.
A
That's
not
an
exaggeration
actually
was
like
two
years.
You
know
the
point
where,
if
you
remove
or
add
a
note
or
any
sort
of
streaming
thing
it
tears,
it
triggers
a
major
compaction
everywhere
around
the
cluster,
so
date
tiered
compaction
when
they
get
it
to
work.
Right
is
actually
brilliant
for
time.
Series
data
and,
and,
as
you
know,
he'll
be
able
to
tell
you
like
it
will
massively
reduce
your
I/o,
but
if
you're
in
his
use
case,
it
will
reduce
your
I/o
if
you're
in
our
use
case,
it'll
quadruple
it.
A
So
it
really
just
depends
on
being
able
to
be
smart
about
the
pattern
you
use
and
understanding
the
implications.
Rather
than
saying,
this
is
a
good
idea.
I'm
going
to
go
ahead
and
implement
it
and
then,
like
being
like
I,
don't
know
why
didn't
work,
because
you
don't
really
fully
understand
like
the
the
hell
you
just
imparted
on
yourself.
C
C
Yeah
I
mean
yeah.
So,
like
you
know
for
us
for
our
use
case,
we
insert
only
that's
our
entire
data
model.
We
never
we
never
delete.
We
never
update
bankers,
don't
use
the
racers
right
and
so
date
to
your
compaction
strategy
does
a
very
good
job
of
combining
data
and
then
dropping
it.
The
problem
that
we
run
into
is
that
it
combines
too
much
data
and
the
calculus.
You
need
to
do
to
figure
out
why
it
would
combine
data
the
way
it
does.
I
don't
have
time
for
that.
You
know
it's
in
a
lot
it.
C
It
was
a
very
clever
solution
to
what
should
be
a
very
simple
problem.
Right
you
know:
I
was
able
to
implement
something
on
nodb
mysql
like
I,
went
and
implemented
a
object
store
on
MySQL
like
it
was
1995
right,
but
it
was
very
simple
because
you're
able
to
create
tables
/
days,
you're
able
to
partition,
based
on
time
like
being
able
to
bucket
based
on
insertion
time.
That's
all
that
we
wanted
and
once
it
times
out
to
actually
unlink.
C
The
file
like
I
would
have
loved
to
build
that
on
Cassandra,
but
I
could
not
reliably
reclaim
disk
as
a
time-series
data
company
that
sells
based
on
retention
period.
If
I
sell
you
15
days
of
retention,
every
second
I
whole
past
that
that
16th
day,
that's
just
cogs
and
laying
out
the
window.
That's
all
margin.
B
So
this
is
the
end
we
have
no
more
time
but,
first
and
foremost,
I'll,
say
two
things.
This
group
of
guys
up
here
is
sarcastic,
as
they
are
as
blunt
as
they
can
be
from
time
to
time
also
represent
some
of
the
smartest
people
I've
ever
met
in
my
life.
There's
some
of
my
favorite
people
on
the
planet
and
well.
Three
of
the
four
of
you
are:
you
can
guess
which
one
and
they
they
know
this
stuff
inside
out.
B
I
think
these
guys
would
be
hired
by
any
company
on
the
planet
virtually
any
time
if
it
wasn't
for
your
less
than
award-winning
personalities
here
and
there
with
that
said,
if
anyone
does
have
any
questions,
I
want
to
deep
dive
on
anything.
I
think
you
two
have
speeches
tomorrow
on
the
ends.
I
did
you
ever
wanted?
You
do
no,
no.
B
Didn't
qualify,
I'm,
sorry
and
you
had
one
today
right
so,
but
so
I
would
definitely
recommend
seeing
them.
I.
Think
they're
great
speakers
actually
I
haven't
seen.
You
speak,
but
you're,
a
good
speaker,
yeah
you're
good
in
private,
at
least
so,
but
and
I
think
we're
also
going
to
go
over
to
the
hotel
bar
for
a
drink.