in memory

37
In-Memory Processing for High Performance Analytics We will begin momentarily

Upload: kiscribd

Post on 03-Dec-2015

214 views

Category:

Documents


0 download

DESCRIPTION

NA

TRANSCRIPT

Page 1: In Memory

In-Memory Processing for High Performance Analytics

We will begin momentarily

Page 2: In Memory

2

Ask Questions

• There is a <Q&A> module in the upper right corner of your screen. Ask questions as they come to mind throughout the presentation. We will address as many as possible as time allows.

• There are also additional icons to explore – The Resource Links icon will allow you to download today’s presentation, as

well as other relevant materials

– Find more info about the today’s Speakers.

– And there is also a Help button for assistance with any technical difficulties you might experience. Please hit F5 to refresh your console as a first step.

© 2015 Teradata

Page 3: In Memory

3

Today’s Speakers

© 2015 Teradata

Noel Yuhanna

Principal Analyst serving Enterprise Architecture Professionals

Forrester Research

Noel covers big data, data warehouses, Hadoop, in-memory, databases, NoSQL, data integration, data

virtualization, data archiving, cloud, ETL, information fabric, big data integration, data management, data tools,

and data security for Enterprise Architecture Professionals. His current focus is on Forrester Wave™ evaluations,

customer experience, industry trends, new and emerging markets, and architecture. More

Imad Birouty

Director of Product Marketing

Teradata

Imad Birouty holds the position of Director of Teradata Product Marketing and is responsible for Teradata

software and hardware products including the Teradata Database, Teradata Platform Family, Teradata

QueryGrid, Teradata Unity, Tools and Utilities, and our In-Database Analytics. Prior to this, Imad led the Product

Management team responsible for the Teradata Platforms; setting product strategy and direction.

Page 4: In Memory

© 2014 Forrester Research, Inc. 4 © 2009 Forrester Research, Inc. Reproduction Prohibited

In-Memory Processing for High Performance Analytics

Noel Yuhanna,

Principal Analyst, Forrester Research

August 11th 2015

Teradata Webinar

@nyuhanna

#inmemory

Page 5: In Memory

© 2014 Forrester Research, Inc. 5

Business growth and speed are

changing the need for accessing data

faster than ever… it’s a competitive

advantage.

Page 6: In Memory

© 2014 Forrester Research, Inc. 6 © 2014 Forrester Research, Inc. Reproduction Prohibited 6

Currency Oil

Bacon

Data is the new ….

It has become critical for

every business!

© 2015 Forrester Research, Inc. 3 © 2015 Forrester Research, Inc.

Page 7: In Memory

© 2014 Forrester Research, Inc. 7

of data is on the

public net

However data explosion means more cost, slower

data access and increasing data challenges…

BIG DATA

mobile

social

sensors

IOT

Video

Cloud

Page 8: In Memory

© 2014 Forrester Research, Inc. 8

Challenge

Businesses think of analytics as a set of boring

historical reports and dashboards… they don’t

want yesterdays data tomorrow!

Page 9: In Memory

© 2014 Forrester Research, Inc. 9

We are moving from IT to Business Technology (BT)..

MIS

IT

BT

Self-service Real-time

Automation

Limited

data

Digital

business

Trend

Lots of

data

Batch

© 2015 Forrester Research, Inc.

Page 10: In Memory

© 2014 Forrester Research, Inc. 10

Real-time data access requirements have grown.

Mobile devices – we need data now!

Competitive pressure – to act more quickly

Run OLTP and DW/Analytical apps faster

Need to act quickly with IOT/sensor data

New insights, advanced analytics that need data quicker

© 2015 Forrester Research, Inc. 7 © 2015 Forrester Research, Inc.

Page 11: In Memory

© 2014 Forrester Research, Inc. 11

What technology is helping with real-time?

Falling memory (DRAM) prices – from

over $100K/GB in 1990 to less than

$1/GB today.

Support for SSD/Flash along with Disks

Organizations are already running a few

Terabytes of in-memory and we are

heading petabytes by 2018+

0

100

200

300

400

200

2

200

2

200

3

200

4

200

6

200

6

200

7

200

8

200

9

200

9

201

0

201

1

201

2

Falling memory prices

memory

© 2015 Forrester Research, Inc. 8 © 2015 Forrester Research, Inc.

Page 12: In Memory

© 2014 Forrester Research, Inc. 12 © 2014, Forrester Research, Inc. .

Two tracks support various real-time analytical needs…

Batch

Faster access

Real-time

9 © 2015 Forrester Research, Inc.

Page 13: In Memory

© 2014 Forrester Research, Inc. 13

Software that can filter, aggregate, enrich, and analyze a

high throughput of data from disparate live data sources to

identify patterns to visualize business in real time, detect

urgent situations, and automate immediate actions

Streaming Analytics

11 © 2015 Forrester Research, Inc.

Page 14: In Memory

© 2014 Forrester Research, Inc. 14 © 2014, Forrester Research, Inc. .

Distributed in-memory technology helps deliver real-time high performance analytics

Operations Analytics

Apps

Logs

Other sources

Clickstream Distributed in-memory

(Horizontal scale)

Business Insights

© 2015 Forrester Research, Inc.

Database

Sensors

Tickers

streaming

Traditional sources

14 © 2015 Forrester Research, Inc.

DRAM

SSD/Flash

Disk

Page 15: In Memory

© 2014 Forrester Research, Inc. 15 © 2014, Forrester Research, Inc. .

Use cases typically seen with in-memory processing

Age of the Customer – personalization

IOT – machine analysis, proactive maintenance

Real-time analytics – various business insights

Operational intelligence

Fraud Detection – risk management, online trading

Mobile Apps – Hotel reservations, inventory tracking

And others…

15 © 2015 Forrester Research, Inc.

Page 16: In Memory

© 2014 Forrester Research, Inc. 16 © 2014 Forrester Research, Inc. Reproduction Prohibited 16

Age of the customer is driving the need for a real-time data platform….

17 © 2015 Forrester Research, Inc.

Page 17: In Memory

© 2014 Forrester Research, Inc. 17 © 2014 Forrester Research, Inc. Reproduction Prohibited 17

Consumer personalization has become critical for organizations to succeed…

18 © 2015 Forrester Research, Inc.

Page 18: In Memory

© 2014 Forrester Research, Inc. 18

Age of the Customer

How can you prevent

members from churning?

Page 19: In Memory

© 2014 Forrester Research, Inc. 19 © 2014 Forrester Research, Inc. Reproduction Prohibited 19

Engines Machines Factories

IOT drives new types of use cases but requires a real-time data platform

Cisco predicts 50 billion

devices will be connected

by 2020.

Forrester estimates 30% usage of IOT in

manufacturing, will double

by 2019.

IOT

© 2015 Forrester Research, Inc.

Page 20: In Memory

© 2014 Forrester Research, Inc. 20

IOT analytical Apps continue to grow rapidly

Fleet management – Monitoring condition, location, and usage of vehicle fleets

Inventory management - Tracking inventory levels and managing operations

Facility management

Customer order and delivery tracking - Enabling customer visibility

Energy management - Monitoring, usage of water, electricity, and other resources

Smart products

Supply chain management - Managing supply chain relationships

Smart home management

Industrial asset management

Security and public safety monitoring or surveillance

20 © 2015 Forrester Research, Inc.

Page 21: In Memory

© 2014 Forrester Research, Inc. 21 © 2014 Forrester Research, Inc. Reproduction Prohibited 21

What does in-memory mean to the business and IT?

• Use in-memory platform to innovate and become a

disruptor, there are endless possibilities

• Intensify customer digital experiences

• Insist on real-time analytics for various use cases to

gain competitive advantage

• Look closely at vendor solutions that can scale

• Focus on tiered memory –Dram, SSD/flash, disk

• Expand in-memory to support more data and new real-

time analytical use cases

IT

Business

Page 22: In Memory

© 2009 Forrester Research, Inc. Reproduction Prohibited

Thank you

Noel Yuhanna

www.forrester.com

Twitter: @nyuhanna

Page 23: In Memory

Teradata Intelligent Memory

Imad Birouty Director, Teradata Product Marketing

Page 24: In Memory

24

Finding The Right Balance

• Memory is 3,000x faster

than disk

• Cost of memory is

decreasing

• Memory per node is

increasing • 96GB -> 256GB ->512GB

-> 768GB -> 1TB

• Memory still 80x more expensive

than disk

• Not all data worth 80x premium • Data Warehouse has wider

variety of data than OLTP systems

• Not all data fits into memory

• Blindly adding memory has

diminishing returns

Page 25: In Memory

25

Teradata’s Approach To In-Memory

• Advanced engineering to use memory intelligently

– Economical use of memory

• Integrated into the Teradata Database

– No separate DBMS to manage

• Automated in-memory data management

• Scale to the largest system sizes (2,048 nodes)

• Database internal code changes to optimize for in-memory processing

• Improve overall system efficiency

Page 26: In Memory

26

In-Memory For Your

Most Valuable Data

Page 27: In Memory

27

Understanding Data Access Patterns Performance of in-memory databases with disk economics

Teradata’s Approach

Hottest data in memory/not all the data

Integrated into Teradata system

No need for separate appliance

Data Temperature Profile – Typical DW

Page 28: In Memory

28

Teradata Intelligent Memory

Sophisticated algorithms to track usage, measure temperature, and rank data

Compliments FSG cache

Dynamically adjusts to new query patterns

Intelligent

Memory

most

recently used

data

most

frequently used data

Hottest data placed and

maintained in memory,

aged out as it cools

cool out very hot in

Memory

Cache

Temporarily store data

required for current queries,

purges least recently used

Page 29: In Memory

29

Advanced Engineering For

Maximum Performance

Page 30: In Memory

30

Smart Engineering To Get Most From Systems Maximize System Throughput By Reducing Bottlenecks

Fast Faster Fastest

Goal of Advanced In-Memory Systems Engineering Goal of In-Memory

Page 31: In Memory

31

It Is More Than The Amount of Memory

• It’s about taking full advantage of modern CPU technology

• It’s about Improving memory bandwidth

• It’s about Improving processor cache effectiveness

• Increasing memory improves Disk I/O

• Improving the bandwidth and cache effectiveness improves cost per instruction, throughput, and response time

Engineering For Performance and System Efficiency

Page 32: In Memory

32

Pipelining & Advanced Use of Memory Query Pipelining & New In-Memory Table Structures

Without Pipelining With Pipelining • Improves Performance with fewer disk I/O’s

• Optimizes Memory Bandwidth • Improves CPU Throughput

Node

Disk

Node

Disk

New in-memory

table structures

hold data as

column partitioned

to reduce size and

store data in the

way the CPU accesses it

Page 33: In Memory

33

SELECT * From Album_Table WHERE Producer = 'Smith';

Album_Table

CPU

SIMD

Vector CPUs

Instr

Instr

Instr

Instr

Instr

Instr

Instr

Instr

Instr

Vector List Pointer

Pointer

Pointer

Pointer

Pointer

Pointer

Pointer

Pointer

Pointer

© 2015 Teradata

Bulk Qualification & Vectorization Memory Bandwidth and Memory Cache Effectiveness

Column Partitioned Table

Page 34: In Memory

34

In-Memory Hash Join

© 2015 Teradata

store_id city state

601 San Diego CA

602 San Diego CA

701 Los Angeles CA

710 Orange CA

712 Tucson AZ

725 Scottsdale AZ

726 Scottsdale AZ

729 Phoenix AZ

store_id Pharmacy Tires

601 Y Y

602 Y N

701 N Y

710 Y Y

712 Y N

725 Y Y

726 N N

729 Y Y

Store Table Store Services

Stores that offer pharmacy services

store_id

601

602

701

710

712

725

726

729

store_id

601

602

701

710

712

725

726

729

Qualified Rows held in In-

Memory Spool

(based on Hash Value bulk

qualification) Row at a Time

Evaluation

Vector-based Evaluation

Page 35: In Memory

35

Teradata Intelligent Memory

Performance of In-Memory

Economics of Disk Storage

Page 36: In Memory

36 © 2015 Teradata

Questions?

Ask via the Chat function now

Page 37: In Memory

37 37