►
From YouTube: TeleStax: Processing Billions of SMS Records Daily
Description
Speaker: Amit Bhayani, Co-Founder
Cassandra is commonly perceived as a write heavy NOSQL database. However at TeleStax, Cassandra is used for even read operations! It processes millions of SMS per day and writes billions of CDR's. The case study shows how TelScale SMSC Gateway leverages Cassandra's scalability and high availability without compromising performance. We will talk about architecture used to make sure SMSC can read records in milliseconds and achieve linear scalability. This new generation SMSC GW offers 100x better performance / price ratio than traditional SMSC from incumbent NEPs.
A
B
For
some
of
these
messages
coming
from
friends
and
of
funny
messages,
maybe
don't
care
if
it's
from
the
airline
telling
you
that
the
flight
was
delayed,
maybe
you
would
care,
the
gate
was
changed
in
it
arrives
at
the
core
an
hour
later.
Even
that
may
be
kind
of
annoying.
If
you
don't
get
that
message,
so
the
guess
is,
nobody
really
has
an
idea
about
that
information
if
the
messages
are
dropped
or
delayed.
So
on.
B
The
reason
being
is
that
a
lot
of
the
services
are
running
on
infrastructure
that
has
been
built
accumulated
for
20-30
years
and
it
was
extended
somewhat
modernized.
That's
telecom
world
pace
which
is
not
the
fastest
pace,
but
it
was
designed
mostly
to
try
to
pass
through
messages
from
hop
to
pop
from
a
tower
to
the
next
hour
to
the
data
center
to
the
other
operator,
if
you're
roaming,
and
if
any
of
these
intermediate
cup
decides
that
they
cannot
push
the
message
forward
because
of
congestion
or
some
other
reason.
B
B
Even
at
this
node,
you
know
at
these
dates.
It
is
considered
kind
of
exotic
fancy
to
implement
infrastructure
nodes
that
have
the
logic
of
storing
a
message
and
reach
rank
for
some
time
until
it
gets
through
and
even
more
exotic
to
help
that
throughout
the
infrastructure,
it's
considered
a
super
expensive,
very
fancy
feature
to
buy
from
traditional
taco
operators
of
telecom
equipment,
vendors.
B
But
things
have
to
change
because,
as
the
price
of
setting
a
message
is
virtually
zero
nowadays,
it
used
to
be
on
ten
cents
fifteen
cents
some
years
ago,
not
virtually
nothing
right,
we're
all
using
OTT
apps,
what's
up
line
depending
which
part
of
the
world
you
live
in
and
a
lot
more
messages
are
going
that
direction
for
good
reason.
You
know
you
get
a
confirmation.
The
message
was
delivered
right
away
or
you
know
it's
still
not
delivered
retry.
You
have
an
idea
of
what's
going
on
lot
better
than
a
regular
text
message.
B
However,
the
simple,
no
frills
text
messaging
still
is
the
most
ubiquitous
way
to
reach
every
mobile
phone.
Maybe
I'm
not
using
the
same
messaging,
app
that
you're,
using
depending
on
what
our
preferences
are,
but
most
definitely
all
of
our
mobile
phones
can
receive
basic
plain
text
messages
which
make
them
still
very
very
appealing
for
businesses
that
tried
to
reach
to
their
audience
either
for
discount
coupons
or
some
sort
of
marketing
campaigns,
or
it
said
earlier,
Airlines
reminders
or
medication
compliance
reminders.
You
know,
Hospital
follow
up
all
kinds
of
use.
B
Cases
are
now
surfacing
and
becoming
economically
feasible
because
it's
so
inexpensive
to
send
messages
and
they're
so
ubiquitous,
it's
almost
guaranteed.
You
can
reach
anybody,
but
because
of
that,
it's
also
important
businesses
to
know
these
messages
are
getting
across
and
what
percentages
are
not
and
do
better
at
delivering
these
messages
so
like
to
introduce
a
terrific
who
is
with
you
phone
pakistan.
It's
a
company
part
of
the
it's
a
lot
group
with
hundreds
of
millions
of
subscribers,
both
in
south
east
asia
and
middle
east,
and
he's
in
charge
of
innovation
for
the
operator
I.
C
Earlier,
as
you
phone
has
always
focused
on
the
services
or
the
density,
only
wise
for
call
and
talk
and
SMS
just
to
send
messages,
you
phone
has
always
focused
on
empowering
its
customers
by
offering
them
different
kind
of
services
like
for
their
businesses
and
their
business
communication
and
marketing
sort
of
stuff.
So
we
have
got
a
larger
volume
in
terms
of
SMS
and
voice
that
we
need
to
get
her
on
daily
basis.
So
it
was
a
challenge
as
we
have
got
for
SMS.
C
We
have
got
500
to
600
million
CD
as
per
day
for
SMS
and
that
all
was
handled
through
the
legacy
systems.
Where
I
mean
the
hardware
that
is
quite
expensive
and
that
was
not
flexible
enough
to
meet
future
needs.
So
we
were
little
bit
worried
about
that
and
there
was
a
need
that
there
should
be
some
system
that
should
be
robust
and
cost-effective
sure,
and
so
we
can
deploy
and
we
can
use
it
to
meet
our
needs
and
to
satisfy
our
customers
for
their
businesses.
C
So,
as
I
told
you
that
number
of
CDRs
it
was
very
expensive
to
handle
and
to
control
that,
because
there
were
a
lot
of
licenses
involved
in
that
for
CDI
processing,
their
storage
and
stuff
like
that,
so
which
is
next
so
legacy
systems
was
complex
and
it
was
not
flexible
enough
to
be
integrated
with
the
campaign
management
solutions,
and
a
lot
of
cost
is
also
involved
to
go
for
a
certain
kind
of
changes
over
that,
and
it
was
not
recommended
do
so.
It
was
very
expensive
for
us
and
we
were
stuck
there.
C
So
a
prime
objective
was
at
that
time
to
just
come
up
with
a
solution
that
is
very
cost-effective,
robust
and
reliable,
because
reliable
reliability
was
mandatory
at
that
time,
because
if
somebody
know
it
doesn't
know
that
whether
his
SMS
has
been
reached
or
not
that
that
is
something
that
is
not
a
customer
does
not
want
that
kind
of
situation
from
a
telecom
provider.
So
it
was
like
that
it
should
be
also
integrated
in
future
products
like
future
innovation
project.
So
we
can
integrate
that
platform
with
the
future
products,
so
that
was
the
objective
next
fuse.
C
So
eventually
we
have
got
Talis
tax,
SMS
e.
We
have
got
their
product
initially,
it
was
very
difficult
for
us
to
decide
for
telus
x
or
for
software
based
SMS,
see
because
it
has
never
been
deployed
on
that
region,
and
we
worked
on
that
and
we
were
thinking
that
it
could
be
like
we
have
to
be
sure
about
its
reliability
and
robustness.
C
So
we
have
tried
that
first
time
in
Pakistan,
so
we
started
it
with
the
limited
capacity
and
limited
functionality
initially.
But
as
we
deployed
it,
we
have
got
very
good
results,
so
we
decided
to
expand
it
now.
We
are
now
handling
all
our
business
SMS
through
this
SMS
e,
because
it
is
highly
reliable
and
we
have
replaced
legacy
expensive
s
embassy
with
this
one,
because
it
has
got
very
good
results
with
higher
reliability
and
the
virtual
infrastructure,
which
is
quite
cheap,
but
it's
inexpensive
for
us
to
expand
it.
C
It
is
scalable
and
it
is
again
highly
reliable
later
on.
We
added
store
and
forward
feature
into
that
smsc
and
that
work
again
well,
and
that
gave
us
very
good
results.
So
we
have
moved
all
our
campaigns
to
this
SMS
see
and
we
have
integrated
IBM's
unica
platform
with
this
SMS
see
to
cater
all
business
needs,
and
hence
we
added
multiple
revenue
streams
with
this
platform
good,
it
is
so
now.
C
C
So
we
are
going
to
defy
for
these
kind
of
services
and
also
we
would
be
allowing
application
to
access
analytics
over
that
that
platform
via
a
pas,
so
we
can
provide
analytics
to
our
customers
and
to
the
users.
So
then
I
would
say.
Thank
you.
Thank
you
to
tellus
tax
and
cassandra
for
that,
because
we
are
now
doing
better
just
because
of
these
guys.
Thank
you.
Thank
you.
They.
B
Kept
are
so
not
as
that
book
of
the
conversation
so
far
was
about
how
much
pain
it
is
to
actually
do
anything
with
legacy
telecom
infrastructure
and
the
chances
are
that
almost
everybody
here
has
unless
you're
in
the
telecom
industry
for
a
while,
you
have
anyone
thought
about
how
to
integrate
telephony
features
into
an
actual
business
application
into
a
website
or
into
a
smartphone
app,
because
it's
been
virtually
impossible
to
even
get
to
that.
So
how
many
of
you
have
actually
done
that
right?
Good,
yes,
right!
B
Nobody
in
this
room,
but
then
the
use
cases
for
virtual
endless
once
you
get
a
flavor
of
what
is
possible.
Let
me
give
a
real
quick
example
of
what
is
possible.
We
talked
about
the
airlines.
Let
me
get
back
into
that
again,
so
you
want
to
change
your
flights.
How
do
you
do
that?
Call
one
eight
hundred
number
first
of
all
go
look
for
it.
Where
is
it?
Ok
does
the
number
there
you
get
ask
Who
am
I
like
who
ru
Hui
corner?
What's
your
account
ID?
B
What's
your
confirmation
number
and
then
you
get
transferred
a
few
times
between
call
centres
and
then
like
15
minutes
later,
maybe
you're
talking
to
somebody
who
actually
can
answer
the
question
and
help
you
with
the
risk
scheduling
and
you
start
searching
with
them.
It's
a
frustrating
process.
Anybody
enjoying
changing
their
airline
ticket
I,
don't
see
any
hints
okay,
so
some
airlines
daughter
more
progressive
and
working
with
companies
like
you
phone.
What
are
they
doing?
B
They're
pushing
their
smartphone
apps
like
most
other
airlines,
but
on
top
of
that,
once
you
open
the
app,
it
will
show
you
just
like
on
Airbnb
or
modern
type
of
applications.
Do
you
need
to
know
anything
about
your
five?
Do
you
want
to
change
it?
Okay,
yes,
immediately
a
conversational
video
call
starts
with
an
agent
who
contextual
is
aware
that
you
are
calling
about
that
ticket.
They
know
who
you
are
on
the
other
end
and
the
conversation
continues
from
that
point
on.
Do
you
want
to
change
your
ticket?
B
B
Adventure
we'll
get
to
the
big
data.
That's
the
whole
point
of
this
talk,
but
in
the
telecom
industry
there
is
history
there,
it's
a
few
dinosaur
years
behind
quickly
by
the
company.
Our
mission
in
life
is
to
build
middleware
that
abstracts
that
legacy
infrastructure
craziness
from
bunch
of
legacy
vendors,
who
don't
wanna
talk
to
each
other,
don't
want
to
open
up
your
black
boxes,
to
mainstream
developers,
to
people
that
are
used
to
programming
everything
they
want
to
program
the
way
they
want
to
program
it
I.
B
Let
customers
in
57
countries
and
our
parents
are
based
on
an
open
source
community
that
we've
been
fostering
for
over
ten
years,
and
it's
been
a
journey.
A
lot
of
work
has
been
done.
A
lot
of
millions
of
lines
of
code
have
been
written
and
more
than
likely,
nobody
in
this
room
has
ever
heard
about
these
open
source
projects,
but
now
that
it's
a
new
year
and
we're
getting
more
to
the
point
of
yes,
it's
possible.
It's
programmable
to
the
communication.
Rich
contextual
communication
features
in
your
apps.
B
Also,
so
going
back
to
the
specific
scenario
here
with
with
your
phone,
what
were
you
trying
to
do?
You're
trying
to
sell
new
businesses,
get
new
revenue
from
messaging
services
and
they
couldn't
do
it
with
black
boxes
way
too
expensive
economically
impossible
to
enable
this
type
of
services.
So
what
are
the
options?
You
know
they
came
to
us
and
we
have
to
learn
our
lessons
because
it's
a
relatively
new
and
unheard
of
stuff
to
do
in
the
telecom
industry
and
of
course
we
made
our
mistakes.
B
A
lot
of
our
first
generation
products
were
were
based
on
normalized
nice,
equal
structures,
beautiful
well
understood,
everybody
gets
it.
The
poem,
of
course,
is
that
with
a
mobile
operator
that
has
hopes
in
many
cities
and
multiple
countries,
it's
hard
to
scale
how
to
replicate
hard
to
to
make
it
he'll
available,
and
you
can
write
from
many
locations
to
reach
for
many
locations.
B
So
you
know
tough
luck,
learned
that
lesson.
Like
everybody
else
in
this
room,
he
has
probably
they
will
move
up
to
to
the
quick
answer
Cassandra
and
it
has
sickle.
So
if
use
the
same
sequel
but
just
change
Cassandra,
the
database
is
modern.
You
everything
will
work.
Just
fine
obvious,
not
everybody
here
knows
better
than
me
now,
but
it
didn't
work
out
that
that
that
way,
so
you
know
without
understanding
the
intrinsic
way
that
Cassandra
works,
the
data
structures,
the
theory
behind
it.
B
B
B
So
essentially,
every
10
minutes
would
scan
through
these
rows
and
all
attempted
messages
to
certain
numbers
with
brought
in
and
reach
right.
And
if
that's
retry
failed
a
few
times,
then
we
moved
on
to
a
different
sort
of
table
which
is
set
aside
for
greater
delays
and
it
the
ones
that
succeed.
It
will
be
moved
on
to
yet
another
data
structure
for
longer-term
storage.
B
Probably
one
of
the
bigger
lessons
learned
here
was
that
deletes
are
horribly,
horribly
slow
if
they're
not
designed
for
the
most
part,
these
ssl
tables,
with
you
know,
dump
stone
markings
can
create
a
lot
of
junk,
because
this
is
such
a
temporary
data.
It
comes
in
it
most
of
the
time
it
goes
away
quickly,
but
sometimes
it
stays
a
couple
of
hours.
So
just
marking
for
deletion
creates
a
lot
of
garbage
and
it's
very
spaced
inefficient.
B
So
what
we've
learned
is
to
create
data
tables
by
the
day,
so
every
day
is
for
itself,
and
once
we
pass
time-wise
on
to
the
next
day,
whatever
is
left
out,
unsane
from
the
previous
day
would
be
retried
again
and
if
it
felt
it'll
come
into
the
new
days,
data
table
and
the
previous
day
would
be
soon
afterwards
drop
the
hotel
with
dropped.
So
it's
one
big
chunk
of
data
structure
just
goes
away
and
frees
up
space.
B
With
this
lights,
what
kind
of
show
that,
in
intense
in
a
sense
of
robustness
and
high
availability
in
photo
runs?
There
are
multiple
layers
of
that
in
a
telecom
network,
the
database
is
extremely
important
of
performance,
tell
/,
etc,
but
it
eventually
comes
as
one
of
the
layers
of
high
availability
before
we
even
get
to
that
there,
the
layers
of
integrating
with
the
core
network
nodes,
the
layers
of
of
the
middleware
logic,
of
being
able
to
understand
multiple
protocols
for
messaging,
be
able
to
translate
and
be
able
to
to
handle
failures
at
that
level.
B
The
application
server
level,
because
the
on
one
side,
there
multiple
applications,
said
their
marketing
campaigns
of
different
kinds
with
different
business
logic
that
would
interact
with
a
messaging
layer,
so
that
itself
has
to
be
robust
and
fault
tolerant,
and
this
kind
of
background
is
somewhat
simplified.
But
it
does
you
straight
on
the
right,
far
right
side,
the
core
as
a
seven
network
elements,
the
messaging
middleware
is
kind
of
in
the
middle
red
boxes
in
the
applications
which
are
in
telecom
terminology
is
called
TSM
in
I
forget
what
it
stands
for.
B
And
then
we
arrived
to
the
data
storage,
the
photons
for
that
and
the
implementation,
which
is
where
finally,
Cassandra,
does
make
life
a
lot
easier.
Once
we
figure
out
how
to
correctly
use
it,
it
did
make
it
almost
transparent
to
scale
to
help
notes
in
different
cities
and
different
regions,
and
not
worry
about
how
they
actually
sync
up
with
each
Ahrendts
and
fell
over.
B
B
Just
a
complexity
of
the
technology
of
what
we've
done
is
not
anywhere
near
some
of
the
latest
cutting-edge,
a
big
data
operational
use
cases
that
we
see
in
the
eye
to
domain
for
the
previous
topics
session
here
was
on
security,
professional,
real-time
indexing,
a
real,
interesting
advanced
cases.
The
telecom
world
is
a
few
generations
behind.
So
this
this
implementation
was
sort
of
a
baby
step,
but
it
did
open
the
the
minds
of
a
lot
of
people
in
the
industry
to
what
is
possible
to
come.
B
I've
asked
this
earlier,
but
I'm
kind
of
reminding
it
again.
Anybody
in
telecom
industry
ask
them.
How
do
you
get
data
out
of
an
alcatel
box,
CD,
RS
or
any
kind
of
interesting
information
aggregated
over
time?
The
answer
is,
who
knows
like?
Definitely
not
through
a
stressed
aps.
Maybe
there
is
some
slide
where
somewhere,
that
shows
it's
possible,
but
now
it's
with
transparency,
it
open
source
implementation
and
tons
of
documentation
and
communities
of
out
there,
demonstrating
the
tremendous
innovations,
windows,
new
ways
of
doing
analytics
and
big
data
and
connecting
that.
B
You
know,
I've
just
listed
a
couple
of
use
case
that
we've
already
discussing
with
with
your
phone
and
some
other
customers,
most
websites
still
show
1
800
number,
and
even
though
you
have
2,000
companies
that
would
sell
your
intelligent,
very
interesting
information
about
web
marketing,
automation,
marketing,
analytics
and
here
to
convert
visitors
to
paying
customers.
You
know
help
out
and
Eloqua
all
kinds
of
useful
good
products
are
out
there
with
the
phone
calls,
even
though
a
phone
call
has
a
much
much
higher
our
chance
of
converting
to
a
sale
than
a
website
visit.
B
There
is
very
little
that
companies
know
what
to
do
with
that
phone
call.
It
usually
goes
to
1
800
to
some
menus,
some
time
of
frustration
going
through
menus
and
connecting
to
some
agent
who
may
or
may
not
be
aware
of
what
he
calling
about-
and
you
know
this
is
time
wasted.
This
is
opportunity
wasted
with
what
we've
started
doing
it
doesn't
have
to
be
1.
800
number
it
can
be
unique.
Number
for
a
landing
page,
it
can
be
equipped
to
code,
can
be
WebRTC
widget.
It
can
be
within
a
certain
app
link.
B
That
would
establish
conversation,
but
the
point
is
that,
within
that
context,
now
it's
possible
to
merge
information
available
about
the
user
behavior
with
the
cause
with
the
messages
information
and
that
together
becomes
even
more
valuable
to
make
decisions
about.
What's
the
next
step,
how
do
you
help
this
user
to
become
a
customer
so
that
justice
helped
the
sales
kind
of
scenario
with
support
a
similar
situation
right
if
you're,
if
you
have
using
a
web
portal
for
support
tickets,
now
it's
kind
of
contextual,
it's
little
frustrating
to
go
through
login
and
find
a
ticket
open
ticket.
B
But
now
it
works.
It's
relatively
established.
Work
for
most
businesses
would
sell
for
a
higher
premium.
The
ability
to
call
as
opposed
to
open
a
ticket
online
and
that
kind
of
makes
sense,
you're
talking
to
a
live
person,
but
still
most
of
the
time,
you're
calling
1-800
number
some
generic
line
and
it's
to
go
for
a
few
minutes
of
figuring
out
who's
calling
and
what
are
they
calling
about?
That's
not
necessary
that
can
be
improved
dramatically
with
merging
context
into
transparent
data
into
accessible
programmable
aps
and.
B
B
C
They
want
to
reach
to
their
customers
as
well,
so
they
can
I
mean
integrate
with
this
solution
and
they
can
get
analytics
as
well,
where
they
have
sent
their
text
and
how
I
mean
what
is
the
condition
now
and
what
is
the
real
statistics
that
how
many
of
those
text
messages
have
been
delivered
successfully
and
how
many
has
responded
to
those
s?
Embassies,
these
kind
of
stuff
and
actually
SMS-
is
a
very
good
way
to
interact
with
the
your
customers
that
they
can
read
it
and
they
can
respond
it.
C
So
there
is
some
sort
of
interaction
instead
of
just
listening
and
just
ignoring
so
they
opt
certain
options.
We
have
got
a
very
good
experience
with
that
that
we
can
also
throw
a
lot
of
campaigns
of
our
internal
products
to
our
customers
and
get
advantage
of
that
that
they
opt,
and
there
is
a
lot
of
revenue
out
of
that
products
which
we
cannot
sell
through
our
website
or
others.
So
by
this
we
can
add
like
multiple
streams
to
our
revenue.
So
if
you
continue
okay,.
B
So
to
kind
of
wrap
it
up
here
there
is
for
sure
a
lot
of
change
happening
communications
no
doubt
even
though
the
perception
is
and
which
is,
for
the
most
part,
correct
that
the
large
telecoms
for
the
most
part
don't
really
care
about
anything
but
selling
their
data
plans
and
SMS
planting.
Oh,
that
many
telecoms
around
the
world
are
under
a
lot
of
pressure
to
move.
B
Even
though
it's
in
sort
of
a
beginning
stages,
the
acceleration
is
really
faster
people
that
are
observing
this
space
and
pretty
sure
in
the
next,
maybe
12
months
or
so
it'll
become
a
lot
more
visible.
That
Communications
is
becoming
programmable
available
friendly
for
developers
and
not
some
crazy
magical
cell
that
nobody
wants
to
touch.
You
know
nobody
wants
to
call
and
find
out
how
it
works.