NoSQL – Back to the Future or Yet Another DB Feature?

Slide 1

Slide 1 text

“For those of you who think we are engaged in some sort of darwinian processes that make things better for us, it’s actually quite the opposite.” – Alan Kay, 2011 http://bit.ly/AlanKay2011

Slide 2

Slide 2 text

infinipool NoSQL – Back to the Future or Yet Another DB Feature? A deconstruction of NoSQL – all carried out by an arrogant guy.

Slide 3

Slide 3 text

Slide 4

Slide 4 text

Martin Scholl, inﬁnipool GmbH martin@inﬁnipool.com @zeit_geist Disclaimer: What follows are opinion statements by an otherwise unimportant guy. Pictures are copyrighted by their respective owners. On Database History and NoSQL Studied all the different *SQL-Systems out there. Still having data issues. (Dr. Faustus)

Slide 5

Slide 5 text

Slide 6

Slide 6 text

No content

Slide 7

Slide 7 text

Arrogant guy ‛

Slide 8

Slide 8 text

Arrogant guy ‛ with a history book ‛

Slide 9

Slide 9 text

Arrogant guy ‛ with a history book ‛ ...now making bold statements ‛

Slide 10

Slide 10 text

NoSQL Technology is a step back. Claim #1:

Slide 11

Slide 11 text

NoSQL Technology models Applications more closely than traditional Databases. NoSQL Myth #1:

Slide 12

Slide 12 text

No content

Slide 13

Slide 13 text

Graph- based Data + Counters

Slide 14

Slide 14 text

Graph- based Data + Counters Recommen- dation Data

Slide 15

Slide 15 text

Graph- based Data + Counters Recommen- dation Data Journal-like Data

Slide 16

Slide 16 text

Graph- based Data + Counters “real-time” Data Recommen- dation Data Journal-like Data

Slide 17

Slide 17 text

infinipool NoSQL vs Reality • Data is scattered all over NoSQL land! • No (simple) way to ensure various quality domains of data • timeliness and appropriateness • correctness and consistency • Data Integration and Data Quality assurance becomes a full-stack concern!

Slide 18

Slide 18 text

There are no transactions in NoSQL because transactions do not scale. NoSQL Myth #2:

Slide 19

Slide 19 text

infinipool Calvin: Fast Distributed Transactions for Partitioned Database Systems[1] [1] http://cs-www.cs.yale.edu/homes/dna/papers/calvin-sigmod12.pdf

Slide 20

Slide 20 text

infinipool Calvin: Fast Distributed Transactions for Partitioned Database Systems[1] [1] http://cs-www.cs.yale.edu/homes/dna/papers/calvin-sigmod12.pdf No Excuses!

Slide 21

Slide 21 text

NoSQL Data Stores are faster. NoSQL Myth #3:

Slide 22

Slide 22 text

Yes, they are fast. Except the other guys do their homework, too.

Slide 23

Slide 23 text

infinipool MySQL Cluster 7.2 (preview) • 30 node MySQL cluster • sporting 19.5m update transactions per second[1] • yet, it’s Oracle and we all know its a benchmark business. [1] http://mikaelronstrom.blogspot.co.uk/2012/05/mysql-cluster-727-achieves-1bn-update.html

Slide 24

Slide 24 text

NoSQL Technology is a step back. Claim #1:

Slide 25

Slide 25 text

Good old Pre-SQL Times: The IBM 704 Filesystems and Databases. (c) Lawrence Livermore National Laboratory

Slide 26

Slide 26 text

Pre-SQL Databases: Files • “Data is stored in files with interface between programs and files” • Separation and Isolation: Every program has its own files and formats • Duplication, Synchronization, Consistency: Programs share data. Data is not necessarily synchronized or in a consistent state. • Weak Security, High maintenance Costs http://www.comphist.org/computing_history/new_page_9.htm

Slide 27

Slide 27 text

infinipool Databases over Files Databases over Files (1960’s) NoSQL DBs (Cassandra, HBase, Riak, etc.) Separation & Isolation Duplication, Synchronization, Consistency, Security, Maintenance Costs Every program has its own ﬁles and formats Every Data Store has its own APIs and Data Models Programs share data. Data not necessarily consistent, synchronized or consistent Content Transferred into Hadoop. Limited consistency by data model Almost no security; manual data processes Almost no security; Specialized personnel required

Slide 28

Slide 28 text

Edgar Frank ‘Ted’ Codd • Landmark Paper: “A Relational Model of Data for Large Shared Data Banks” • Father of Relational Database Management Systems • Basically invented what Twitter and FB run on • Now a +$12B business • we owe him more than an applause.

Slide 29

Slide 29 text

infinipool Relational Database Model: The Good Parts • Key Insight: Separate Logical Data Model from Physical Data Storage • Radical Simpliﬁcation of Data Access • A phenomenal tool was introduced: Joins • great for “single data insert, multiple views of data”

Slide 30

Slide 30 text

infinipool Relational Databases and NoSQL Relational Databases NoSQL DBs (Cassandra, HBase, Riak, etc.) Logical & Physical Data Model Duplication, Synchronization, Consistency, Downsides Separated Complected Normalization; Constraints for improved data quality Denormalization; Data Quality an Application- level Problem Scalability Issues; some DBMSs quite expensive Almost no security; Specialized personnel required

Slide 31

Slide 31 text

1964 It’s Mainframes all over Software is not a product Databases over Files

Slide 32

Slide 32 text

1964 It’s Mainframes all over Software is not a product Databases over Files 1980 It’s Minicomputers all over Software becomes a product Relational DBMS + SQL

Slide 33

Slide 33 text

1964 It’s Mainframes all over Software is not a product Databases over Files 1980 It’s Minicomputers all over Software becomes a product Relational DBMS + SQL 2012

Slide 34

Slide 34 text

Slide 35

Slide 35 text

1964 It’s Mainframes all over Software is not a product Databases over Files 1980 It’s Minicomputers all over Software becomes a product Relational DBMS + SQL It’s Cloud Computing all over Software becomes a Service NoSQL? 2012

Slide 36

Slide 36 text

Slide 37

Slide 37 text

infinipool Changing Issues in Data Management • Scalability of data storage and transactional access is solved. Everybody can (soon) rent the perfect data storage system in the cloud. • Issue #1: Data-Integration an open task • Issue #2: Data-Quality an open task • Issue #3: Push-based execution model: where are thou?

Slide 38

Slide 38 text

infinipool Changing Issues in Data Management • Scalability of data storage and transactional access is solved. Everybody can rent the perfect data storage system in the cloud. • Issue #1: Data-Integration an open task • Issue #2: Data-Quality an open task • Issue #3: Push-based execution model: where are thou? • The new competitive frontier: Timeliness, Data Integration and Quality

Slide 39

Slide 39 text

NoSQL Technology is a step back. Claim #1:

Slide 40

Slide 40 text

Claim #2: NoSQL Technology is a step back. Claim #1: NoSQL will become yet another DB Feature and/or Cloud Computing Service.

Slide 41

Slide 41 text

Claim #2: NoSQL Technology is a step back. Claim #1: NoSQL will become yet another DB Feature and/or Cloud Computing Service. PostSQL Databases will be indistinguishable from Data Communication Services. Claim #3:

Slide 42

Slide 42 text

infinipool Thank you. Martin Scholl, inﬁnipool GmbH martin@inﬁnipool.com @zeit_geist