in Real-time Big Data
● Concepts & Techniques “Thinking with Lambda”
● Case studies in Practice
Trieu Nguyen - http://nguyentantrieu.info/blog or @tantrieuf31
Lead Engineer at eClick Data Analytics team at FPT Online
All contents and thoughts in this slide are my subjective ideas and compiled from Open Source
Just a little introduction
● 2008 Java Developer, developed Social
Trading Network for a small startup (Yopco)
● 2011 worked at FPT Online, software engineer
in Banbe Project, Restful API for VnExpress
● 2012 joined Greengar Studio in 6 months,
scaling backend API mobile games (iOS, Android)
● 2013 back to FPT Online, R&D about Big Data
& Analytics, developing the new core
Analytics Platform (on JVM Platform)
● Big Data means big logs storage ?
● I just installed Hadoop, and it works! Do we
really get a big data solution ?
● We have lots data, so let’s play with cool big
data technologies x,y, z! Do we get profits
from that ?
● We can hire or outsource a professional
team to build big data solution, but do they
answer what problem we get ?
Contents for this talk
● A little introduction about Lambda in history
● Trends of Now and the Future
● Why lambda architecture is correct solution
for big data?
● Lambda in Practice, case studies from
Greengar Studios and eClick
● Questions & Answers
The best way to predict the future is
looking at the past and now ?
Lambda is the symbol to denote:
● Half-life game ?
● Anonymous function, aka: Closure ?
● functional computation/programming?
● scalable system ?
When I study “lambda” ?
I studied Haskell in 2007 with Dr.Peter Gammie http://peteg.org/ when
internship at DRD (a non-profit organization).
● Imperative programs will always be vulnerable to data races because
they contain mutable variables.
● There are no data races in purely functional languages because they
don't have mutable variables.
How did Google scale their search engine ?
How does Hadoop really work ?
running by billion websites
the Lambda Architecture:
● apply the (λ) Lambda philosophy in designing big data
● equation “query = function(all data)” which is the basis
of all data systems
● proposed by Nathan Marz (http://nathanmarz.com/), a
software engineer from Twitter in his “Big Data” book.
● is based on three main design principles:
○ human fault-tolerance – the system is unsusceptible to data loss or
data corruption because at scale it could be irreparable. (BUGS ?)
○ data immutability – store data in it’s rawest form immutable and for
perpetuity. (INSERT/ SELECT/DELETE but no UPDATE !)
○ recomputation – with the two principles above it is always possible to
(re)-compute results by running a function on the raw data.
proposed by @nathanmarz
We, at FPT Online, have applied
the lambda architecture since
Lambda In Practice
2 case studies from my experiences
Case Study 1:
API Backend Monitor + Statistics
Backend System at Greengar Studio
The data and the size, not too big for a small
Where is the lambda ?
I used Groovy + GPars (Groovy Parallel Systems) + MongoDB for fast
parallel computation (actor model) on statistical data
The GPars framework offers Java developers intuitive and safe ways to handle
Java or Groovy tasks concurrently.
● Dataflow concurrency
● Actor programming model
● Agent - an thread-safe reference to mutable state
● Concurrent collection processing
● Composable asynchronous functions
● STM (Software Transactional Memory)
Mobile Apps => Backend APIs =>
Statistics => Find the Trends & Insights?
Case Study 2:
● Real-time Data Analytics
● Monitoring Stream Data (Reactive)
at eClick we have
30~40 GB Logs in Stream
10~20 GB Bandwidth
just for tracking user
in ONE day !
at eClick we must
check campaigns in
at eClick we have many types of log (video, web,
mobile, system logs, ad-campaign, articles, … )
Our big-data system
Leverage Open Source Projects
● Netty (http://netty.io/) a framework using reactive programming
pattern for scaling HTTP system easier
● Kafka (http://kafka.apache.org/) a publish-subscribe messaging
rethought as a distributed commit log.
● Storm (http://storm-project.net/) a framework for distributed
realtime computation system.
● Redis (http://redis.io/) a advanced key-value in-memory NoSQL
database, all fast statistical computations in here.
● Groovy for scripting layer, dynamic query on Redis + RDBMSs
● Hadoop ecosystem: HDFS, Hive, HBase for batch processing
● RxJava https://github.com/Netflix/RxJava a library for
composing asynchronous and event-based programs
Some new ideas for the future:
Connecting the active functor pattern + reactive programming
+ stream computation + in-memory computing to make:
● real-time data analytics easier
● better recommendation system
● build more profitable big data solutions
● http://activefunctor.blogspot.com/ (a special case of Lambda
that actively search best connections to form optimal
topology) - from ideas when internship at DRD with my
● Can a function be persistent (stored as data), distributed in
a cluster (cloud), reactive to right data (best value in
We can't solve problems
by using the same kind of
thinking we used when we
Think more Lambda and Reactive
How could we see "user interest graph" in our user's database ?
● Social Graph
=> Keep the connection
● Interest Graph
=> Make new connection
What I have learned from Lambda and Big
What I have learned
● Keep it as simple as possible, but no simpler !
● Ask right questions=> deep analytics=>Profit
● Reactive and Lambda for your data products
● Implement it! Just right tools for right jobs.
● Turn your data into the things everyone can
"look & feel"
How to build profitable big data solutions?
=> read these Behavioral Economics Books
Stay focused, keep innovating
Big Data is not profitable if you do not know
what you want and ask right questions
“Logic will get you from A to Z;
imagination will get you
everywhere.” - Albert Einstein
Use your imaginationwith data analytics, not