Big Data use cases over Hadoop have slowly evolved from batch analytics that rely on long running MapReduce jobs to include more real-time analytics that rely on quick data retrieval through technologies such as NoSQL and Interactive SQL. With so many solutions available in the market, how should one go about deciding which technologies to pick? This session helps the audience answer that question. The session discusses how enterprise architects can take a holistic approach to real- time Big Data problems and look at NoSQL and SQL technologies as complementary pieces of the same puzzle. The session will delve into SQL queries over Hadoop and the various ways Hadoop NoSQL database and SQL work hand in hand to provide the best real-time results for the business user. Specifically, technologies such as Apache HBase and Apache Drill will be discussed.

1. NoSQL and SQL Work Side-by-Side
to Tackle Real-time Big Data Needs
Allen Day
MapR Technologies

2. Me
• Allen Day
– Principal Data Scientist @ MapR
– Human Genomics / Bioinformatics
(PhD, UCLA School of Medicine)
• @allenday
• allenday@allenday.com
• aday@maprtech.com

3. You
• I’m assuming that the typical attendee:
– is a software developer
– is interested and familiar with open source
– is familiar with Hadoop, relational DBs
– has heard of or has used some NoSQL technology

4. Big Data Workloads
• ETL
• Key-value store
• Lightweight OLTP
• Model creation & clustering & indexing
• Classification & Anomaly detection
• Web Crawling
• Stream processing
• Batch reporting
• Interactive analysis

5. What is NoSQL? Why use it?
• Traditional storage (relational DBs) are unable
to accommodate increasing # of observations
– Culprits: sensors, event logs, electronic payments
• Solution: stay responsive by relaxing storage
requirements
– Denormalize, loosen schema, loosen consistency
• This is the essence of NoSQL

6. NoSQL Impact on Business Processes
• Traditional business intelligence technology
assumes relational DB storage
– Scaling solution is to use MPP (Aster, Greenplum)
• However, collected data aren’t in relational DB
– Data volume still increasing
– Technology still in flux
• Decoupled data storage and decision support
systems
– Very high opportunity cost to business

7. Ideal Solution Features
• Scalable & Reliable
– Distributed storage
– Parallel processing
• SQL application support
– Ad-hoc, interactive queries
– Real-time responsiveness
• Flexible
– Can accommodate rapid storage and schema
evolution
– Can accommodate new analytics methods and
functions

8. From Ideals to Possibilities
• Migrate NoSQL data/processing to SQL
– High cost to marshal NoSQL data to SQL storage
– SQL systems lack advanced analytics capabilities
• Migrate SQL data to NoSQL
– Breaks compatibility for legacy business functions, e.g.
financial reporting requirements
– Limited relational support (joins) & high latency
– Technology still in flux
• Other Approaches?
– Yes. First let’s consider a SQL/NoSQL use case

9. Impala
Interactive Queries & Hadoop
low-latency

10. Example Problem: Marketing
Campaign
• Jane is an analyst at an
e-commerce company
• How does she figure
out good targeting
segments for the next
marketing campaign?
• She has some ideas…
• …and lots of data
User
profiles
Transaction
information
Access
logs

11. Traditional System Solution 1: RDBMS
• ETL the data from
MongoDB and Hadoop
into the RDBMS
– MongoDB data must be
flattened, schematized,
filtered and aggregated
– Hadoop data must be
filtered and aggregated
• Query the data using
any SQL-based tool
User
profiles
Access
logs
Transaction
information

12. Traditional System Solution 2: Hadoop
• ETL the data from
Oracle and MongoDB
into Hadoop
• Work with the
MapReduce team to
write custom code to
generate the desired
analyses
User
profiles
Access
logs
Transaction
information

13. Traditional System Solution 3: Hive
• ETL the data from
Oracle and MongoDB
into Hadoop
– MongoDB data must be
flattened and
schematized
• But HiveQL is limited,
queries are slow and BI
tool support is limited
User
profiles
Access
logs
Transaction
information

14. What Would Google Do?
Distributed
File System
NoSQL
Interactive
analysis
Batch
processing
GFS BigTable Dremel MapReduce
HDFS HBase ???
Hadoop
MapReduce
Build Apache Drill to provide a true open source
solution to interactive analysis of Big Data

15. Apache Drill Overview
• Interactive analysis of Big Data using standard SQL
• Fast
– Low latency queries
– Columnar execution
• Inspired by Google Dremel/BigQuery
– Complement native interfaces and
MapReduce/Hive/Pig
• Open
– Community driven open source project
– Under Apache Software Foundation
• Modern
– Standard ANSI SQL:2003 (select/into)
– Nested/hierarchical data support
– Schema is optional
– Supports RDBMS, Hadoop and NoSQL
Interactive queries
Data analyst
Reporting
100 ms-20 min
Data mining
Modeling
Large ETL
20 min-20 hr
MapReduce
Hive
Pig
Apache Drill

16. How Does It Work?
• Drillbits run on each node, designed to
maximize data locality
• Processing is done outside MapReduce
paradigm (but possibly within YARN)
• Queries can be fed to any Drillbit
• Coordination, query planning, optimization,
scheduling, and execution are distributed
SELECT * FROM
oracle.transactions,
mongo.users,
hdfs.events
LIMIT 1

17. Key Features
• Full SQL (ANSI SQL:2003)
• Nested data
• Schema is optional
• Flexible and extensible architecture

18. Full SQL (ANSI SQL:2003)
• Drill supports SQL (ANSI SQL:2003 standard)
– Correlated subqueries, analytic functions, …
– SQL-like is not enough
• Use any SQL-based tool with Apache Drill
– Tableau, Microstrategy, Excel, SAP Crystal Reports, Toad, SQuirreL, …
– Standard ODBC and JDBC drivers
Drill%Worker
Drill%Worker
Driver
Client
Drillbit
SQL%Query%
Parser
Query%
Planner
Drillbits
Drill%ODBC%
Driver
Tableau
MicroStrategy
Excel
SAP%Crystal%
Reports

19. Nested Data
• Nested data is becoming prevalent
– JSON, BSON, XML, Protocol Buffers, Avro, etc.
– The data source may or may not be aware
• MongoDB supports nested data natively
• A single HBase value could be a JSON document
(compound nested type)
– Google Dremel’s innovation was efficient columnar
storage and querying of nested data
• Flattening nested data is error-prone and often
impossible
– Think about repeated and optional fields at every
level…
• Apache Drill supports nested data
– Extensions to ANSI SQL:2003
enum Gender {
MALE, FEMALE
}
record User {
string name;
Gender gender;
long followers;
}
{
"name": "Homer",
"gender": "Male",
"followers": 100
children: [
{name: "Bart"},
{name: "Lisa”}
]
}
JSON
Avro

20. Schema is Optional
• Many data sources do not have rigid schemas
– Schemas change rapidly
– Each record may have a different schema
• Sparse and wide rows in HBase and Cassandra, MongoDB
• Apache Drill supports querying against unknown schemas
– Query any HBase, Cassandra or MongoDB table
• User can define the schema or let the system discover it automatically
– System of record may already have schema information
• Why manage it in a separate system?
– No need to manage schema evolution
Row Key CF contents CF anchor
"com.cnn.www" contents:html = "<html>…" anchor:my.look.ca = "CNN.com"
anchor:cnnsi.com = "CNN"
"com.foxnews.www" contents:html = "<html>…" anchor:en.wikipedia.org = "Fox News"
… … …

21. Flexible and Extensible Architecture
• Apache Drill is designed for extensibility
• Well-documented APIs and interfaces
• Data sources and file formats
– Implement a custom scanner to support a new data source or file format
• Query languages
– SQL:2003 is the primary language
– Implement a custom Parser to support a Domain Specific Language
– UDFs and UDTFs
• Optimizers
– Drill will have a cost-based optimizer
– Clear surrounding APIs support easy optimizer exploration
• Operators
– Custom operators can be implemented
• Special operators for Mahout (k-means) being designed
– Operator push-down to data source (RDBMS)

22. How Does Impala Fit In?
Impala Strengths
• Beta currently available
• Easy install and setup on top of
Cloudera
• Faster than Hive on some queries
• SQL-like query language
Questions
• Open Source ‘Lite’
• Doesn’t support RDBMS or other
NoSQLs (beyond Hadoop/HBase)
• Early row materialization increases
footprint and reduces performance
• Limited file format support
• Query results must fit in memory!
• Rigid schema is required
• No support for nested data
• Compound APIs restrict optimizer
progression
• SQL-like (not SQL)
Many important features are “coming soon”. Architectural foundation is constrained. No
community development.

23. Drill Status: Alpha Available Q2
• Heavy active development by multiple organizations
– Contributors from Oracle, IBM Netezza, Informatica, Clustrix, Pentaho
• Available
– Logical plan syntax and interpreter
– Reference interpreter
• In progress
– SQL interpreter
– Storage engine implementations for Accumulo, Cassandra, HBase and various
file formats
• Significant community momentum
– Over 200 people on the Drill mailing list
– Over 200 members of the Bay Area Drill User Group
– Drill meetups across the US and Europe
– OpenDremel team joined Apache Drill
• Anticipated schedule:
– Beta: Q3

24. Why Apache Drill Will Be Successful
Resources
• Contributors have strong
backgrounds from
companies like Oracle,
IBM Netezza, Informatica,
Clustrix and Pentaho
Community
• Development done in the
open
• Active contributors from
multiple companies
• Rapidly growing
Architecture
• Full SQL
• New data support
• Extensible APIs
• Full Columnar Execution
• Beyond Hadoop

25. Closing Thoughts
• What problems can NoSQL and Drill solve for
you?
• Where do they fit into your organization?
• Which data sources and BI tools are important
to you?

26. Me
• Allen Day
– Principal Data Scientist @ MapR
– Human Genomics / Bioinformatics
(PhD, UCLA School of Medicine)
• @allenday
• allenday@allenday.com
• aday@maprtech.com