DATA 228

Big Data Technologies and Applications (Fall 2024)

Sangjin Lee
 Overview

• Introduction to Big D t

• H doop

• HDFS

• YARN

• M pReduce

• Sp rk

• High-level APIs

• Sp rkSQL

• Future of Big D t
a
a
a
a
a
a
a
a
 Instructor

• S ngjin Lee (s ngjin.lee@sjsu.edu, https://linkedin.com/in/sjlee)

• H doop project PMC (project m n gement committee) member

• Contributor to sever l open-source projects beyond H doop

• Expertise/experience: Big D t , cont iner pl tforms, tr ic m n gement, developer

experience, distributed systems, J v , GoL ng, nd more
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
ff
a
a
 Class rules

• Ple se refer to the syll bus

• Cl ss will st rt on time

• Homework nd quizzes/tests will use C nv s

• No l te ssignments re ccepted

• Quizzes re in-cl ss

• Ex ms re comprehensive

• TA: Sowmy Kurub (sowmy .kurub @sjsu.edu)

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
Why are you taking this class?

• Poll
Rise of Big Data
 Rise of Big Data

• Exponenti l growth of d t everywhere

• Tr ns ctions

• Internet cr wling

• Impressions nd clicks

• Soci l medi feeds

• IoT (c rs, c mer s, devices, etc.)

• Expect ~ 164 zett bytes of tot l d t by 2025

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Rise of Big Data

• Perfect “d t ” storm

“D t L kehouse in Action”, Menon

a
a
a
a
a
 Rise of Big Data

• Not just size or volume

• D t re diverse

• D t re often produced by di erent te ms, org niz tions, nd comp nies

• D t re often unstructured
a
a
a
a
a
a
a
a
a
ff
a
a
a
a
a
Rise of Big Data
Relational DBs

• “Good ol’ d ys” when d t w s ne tly in rel tion l DBs (Or cle, MySQL, …)

• But…

• Comp nies lre dy h d to ind w ys to sc le beyond gig ntic single-m chine DBs

• How to loc te your d t when the d t is “distributed”

• They were lre dy trying to get to Big D t without the n me

a
a
a
a
a
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Rise of Big Data
Data warehouse

• How bout d t w rehouse (DW)?

• T ble u, Ter d t , etc.

• Highly cur ted w lled g rdens

• They don’t pl y well when they st rt mixing di erent sources of d t

• It’s SUPER expensive ($$$$)

• Side note: OLTP vs. OLAP

a
a
a
a
a
a
a
a
a
a
a
a
a
a
ff
a
a
What is Big Data?
 What is Big Data?

• Store much l rger volumes of d t

• Compute/ n lyze much l rger volumes of d t

• H ndle diverse nd mostly unstructured d t

• … And do it che ply

a
a
a
a
a
a
a
a
a
a
a
a
a
What is Big Data?
Store much larger volumes of data

• Let’s t lk bout typic l stor ge server these d ys (~ 2024)

• ~ 10 TBs of HD or SSD stor ge

• Even sm ll comp nies e sily h ve tens of PBs of d t

• How m ny stor ge servers would you need?

• Not just m ny m chines…

• How do you ind your d t ?

• How do you keep your d t f ult-toler nt nd resilient?

a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
What is Big Data?
Compute/analyze much larger volumes of data

• Let’s t lk bout typic l processing server these d ys

• 32-core CPU (3.x GHz)

• 256 GB RAM

• 100 Gig ethernet

a
a
a
a
a
 What is Big Data?
Compute/analyze much larger volumes of data

• You w nt to ingest nd n lyze 10 TB of d t with single server

• How long would it t ke to complete this t sk?

• How long does it t ke for single server to stre m 10 TB?

• How much d t c n single server hold?

• Do multiple cores help?

• Wh t is the most import nt bottleneck?

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 What is Big Data?
Compute/analyze much larger volumes of data

• It should become obvious m ssive p r llelism is required

• Distribute compute to m ny m chines nd run them in p r llel

• Sort nd collect the results

• This points to some sort of d t compute fr mework th t c n model this beh vior nd lets
you progr m this in n e sy nd e icient w y

• —> M pReduce, Sp rk, etc.

a
a
a
a
a
a
a
a
a
a
a
a
a
ff
a
a
a
a
a
a
a
a
a
a
a
What is Big Data?
Handle diverse and mostly unstructured data

• Need bility to “discover” schem from d t

• Need bility to h ndle m ny di erent d t schem s nd encodings

• Need good d t govern nce th t sc les

• Need strong d t security

a
a
a
a
a
a
a
a
a
a
ff
a
a
a
a
a
a
a
a
a
 What is Big Data?
… And do it cheaply

• Big d t is built on che p “commodity” h rdw re

• Horizont l sc ling

• F ult-toler nt nd self-he ling rchitecture

• Big d t is now incre singly coupled with Cloud

• E.g. use of s3 or GCS

• Compute on dem nd
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
Your “dev” environment

• Poll