OLTP vs OLAP | Apache Cassandra Tutorial | Edureka Video Lecture | Cassandra: Learn and Master - Database Management

let's start with trying to understand
what is oil a be an OLAP so OLTP is more
of an online transaction processing
system or a data storage system it is
more a representation of data store
where you do lot of online transactions
with that particular data store so more
ad hoc reads and writes are more reads
and writes happening at the real time so
more more of more of a real-time data
store which falls into online
transaction system when it comes to OLAP
OLAP is more of a offline data store but
it is accessed once or many many times
in an offline fashion say for example
you have a bulk log files you are
writing all the bulk log files one time
and then you keep reading that log files
for analytics or for some processing and
then write back that data to the data
store so this is a bad jobs all the log
mining data with data warehousing they
all fall into the category of OLAP
systems Cassandra is more offer or will
be P kind of a data store where you
access the data at real time and process
the data at real time for your real time
applications Hadoop is more of a OLAP
kind of a platform where it is used for
analytical processing and basically it
uses bulk writes and reads the
information for analytics purpose in a
more of a offline or a batch mode so
that's a basic difference okay now why
do I need to integrate these two things
the first benefit is say for example if
you have a huge data set in your
Cassandra say you have a few petabytes
of data store which you have accumulated
for all your hotels application or all
your analytics application now on top of
this say for example if you have to do
some sort of recommendations on your
real time database itself say picking on
say cheapest day
let's pick some hotels database
application what we have built over the
period of love last few sessions say in
the last session we build a hotel
details will be stored static and
dynamic data right say let's assume we
have to figure out what is the cheapest
price in the next three 365 days and I
want to run some promo based on the
price whether if it is lowest or not so
I have to run this promo based on some
condition so if I have to it I trade all
the records and keep on doing analytics
on top of it it's a huge offline kind of
a job which has to kick-start every
every say 3 hours or 6 hours or every
day and it has to do some sort of
offline kind of a thing that's where
Hadoop kind of analytical systems can
really come into picture and help us out
basically to do some bulk data crunching
and try to get us some insights into
what is the data so that is the first
benefits or that could be one of the
benefits what you would get if you are
trying to use analytics jobs over the
Cassandra dataset itself the other
benefits could be say for example if you
don't want to say for example run a
different analytics cluster or a
different Hadoop cluster you wanted to
run only one cluster it is also possible
we'll see in a while how you can do that
you can go ahead and deploy or work with
one single cluster which is cassandra
and hadoop together so the other benefit
is basically you can avoid lot of cost
or you can reduce the operational cost
and the business development cost if you
run this in the single enrollment I just
give few examples
practically you could find enormous
examples of what could
the benefits so I just provided one use
case of why it could be potentially
helpful in leveraging Kassandra data set
over a Hadoop kind of a platform the
other benefit could be say for example
if I am very efficient with say Hadoop
ecosystem projects like hive or Pig
Latin and I want to integrate the data
into hi or Pig Latin
that is also could be an benefit say for
example if I have to leverage the data
what is already existing in HDFS or
Hadoop then you have to plug in some
data source from Cassandra and try to
run MapReduce jobs or data jobs that is
also absolutely possible so these are
the few benefits what you would gain by
integrating these two technologies
together okay so as I just briefly
explained so there is a pattern between
the OLTP and OLAP systems so in OLTP
systems there is a less number of writes
what is happening say let's say let's
take our hotels or flights information
if you take hotel's information the
price changes does not happen very
frequently rather I wouldn't say that it
happens every 5000 times per second
right but the reads happen every 5000
times per second or even more or theta
based on the popularity of your
application so the writes could be say
once in say maybe max one one one write
per second but there could be several
reads or thousands of reads per second
happening so I would read this as more
than one is 2000 kind of a ratio
pattern so one write versus maybe
hundreds or thousands exponentially it's
multiplied by thousand times of Fried's
so typically your data access pattern in
your data store would be once 2,000 kind
of I think one writes or multiple reads
so Cassandra can fit into this model
pretty easily and also it can fit into a
model where your rights and Reeves are
equally balanced safe you tease even one
to one ratio which say thousand rights
happening per second and even thousand
rates happening per second with my
personal experience I could tell you
Cassandra can scale pretty smoothly
without even at least I've been running
Cassandra clusters for last four to five
years which which was like a great
success story we'd never had any issues
having we have roughly around four
thousand rights happening per second and
around fifteen to twenty thousand beats
happening on the cluster so there was no
issues at all so my personal take it
works pretty well and also when it comes
to say ole DB kind of applications even
though Cassandra offers you the eventual
consistent model even if you get into a
tunable or strong consistency models you
could see only milliseconds of gaps or
differences between eventual consistent
model to the strongest consistent model
so Cassandra can fit into a or will t be
a data pattern pretty pretty easily and
it can do all the requirements of a
real-time application data store but it
has some gaps the core system it has
some gaps it can not by default process
the analytical jobs or it cannot run any
data jobs something like MapReduce by
deep by core system or base system
eventually Cassandra introduced all
these things in the later versions but
we will see basically what what are
those enhancements so on the other side
the right section you would see there is
a different pattern on the OLAP kind of
systems or data stores the basic
difference is basically there are
several rights happening parallely
or in a bulk fashion say for example in
any typical overlap kind of application
what you do is you dump all the data in
one shot right you dump all the log
files or huge gigabyte or a terabyte
files into the data store and then start
processing and read just maybe once or
twice whenever thank you whenever they
use it or the application is required to
move what is the outcome or result of it
so the data pattern or the access
pattern is exactly the opposite of the
OLTP kind of application so here
the Hadoop or the MapReduce with HDFS
kind of concepts would really be very
useful so the reason why we have to be
very careful and understanding the data
pattern is basically we have to
understand when we have to leverage
Cassandra and when we have to leverage
her do and in case even Cassandra
started offering to run the MapReduce
jobs and kind of distributed file system
you should be able to figure out what is
the difference between these kind of
systems when you adapt a hybrid kind of
a model
just a quick summarization of the
advantages and use cases so maybe you'd
require to analyze the whole TB data or
Ola platforms which is already existing
in your organization's or in your
environment and maybe times you would
require to process the OLTP data as well
as OLAP data just in case if you have
some data set in Hadoop and some data
set in Cassandra and you wanted to run
some MapReduce jobs without importing
the data into column families you have
to do all of these things it would be
efficient that you've made leverage this
kind of integration models and also
leverage
the existing data store for analytics
which is like more of a merged model in
a variant Cassandra or in Hadoop or in
external sources which is directly
integrated into the Hadoop ecosystem
which is not possible some of the things
with Cassandra say for example you could
directly integrate lot of log data
automatically into the Hadoop file
system by using some of the echo
ecosystem projects which is not possible
currently in Cassandra you can leverage
those kind of things as I said the use
cases could be fine hotels with our
previous application the same price in
real time and run digital marketing
campaigns and promos on the lowest price
dates real time inventory indicators on
flights and hotels on specific dates so
those are all the just two examples or
use cases what I could think of