cs-op analytics

15
1 © Cloudera, Inc. All rights reserved. Smarter Decisions in Less Time Opera?onal Analy?cs with Cloudera

Upload: cloudera-inc

Post on 15-Jul-2015

338 views

Category:

Technology


0 download

TRANSCRIPT

1  ©  Cloudera,  Inc.  All  rights  reserved.  

Smarter  Decisions  in  Less  Time  Opera?onal  Analy?cs  with  Cloudera    

2  ©  Cloudera,  Inc.  All  rights  reserved.  

Opera?onalizing  Reports,  Models,  or  Rules  

Recommenda)on  Engine  

Event  Detec)on  

Model    Scoring  

Point  Solu)ons  

Custom  Development   3rd  Party    

Data  Discovery  &  Analy8cs  

3  ©  Cloudera,  Inc.  All  rights  reserved.  

Custom  Development  Use  Cases  

Recommenda)on  Engine  

Event  Detec)on  

Model    Scoring  

Fraud  Detec?on  Spam  Filter  Marke?ng  Alerts  

Embedded  Analy?cs  Analy?c  Aggregates  Reports  

Next  Best  Offer  Content  Rec  Services  Rec  

4  ©  Cloudera,  Inc.  All  rights  reserved.  

The  Process  of  Opera?onal  Analy?cs  

Data  Discovery    Advanced  Analy8cs  

Data  Volumes  Stream  &  Batch  Processing  

     Data    

Genera?on  

Opera8onal  Analy8cs    Flow  

Op?mize  Analy?c  Func?on  

Processing  

Respond  to  Data  

Feed  Data  Applica?on  

Act  and    Measure  

Model  Flexibility  Scalability  

     

Embedded  Analy8cs  Reports  

5  ©  Cloudera,  Inc.  All  rights  reserved.  

Opera?onal  Analy?c  Needs  

Scale   Embed  Analy8cs  

Enterprise  Data  Warehouse  

Data  Data  Sources  

ETL  

Structured  

Unstructured  

Database  

ELT  

Store  &  Process  

Tradi8onal  Architecture    

Archive  

Serve  

Ac?on  

Model  

Process  f  (D1,  DN)  

Structured  

Unstructured  

Machine  

Drill  Down  

Human  API  

Ingest  

LiHle  Latency  

6  ©  Cloudera,  Inc.  All  rights  reserved.  

Challenges  with  Tradi?onal  Opera?onal  Analy?c  

1)  Limited  Data   3)  Analy8c  Latency  2)  Drill  Down  Performance  

Enterprise  Data  Warehouse  

Data  Data  Sources  

ETL  

Structured  

Unstructured  

Database  

ELT  

Store  &  Process  

Tradi8onal  Architecture    

Archive  

Serve  

Ac?on  

Model  

Process  f  (D1,  DN)  

Structured  

Unstructured  

Machine  

Drill  Down  

Human  API  

Ingest  1  

2  

1  

3  

7  ©  Cloudera,  Inc.  All  rights  reserved.  

A  New  Way  Forward  

1)  Data  Scale     3)  LiHle  Latency  2)  Drill  Down  Speed  

Enterprise  Data  Warehouse  

Data  Data  Sources  

ETL  

Structured  

Unstructured  

Enterprise  Data  Hub  

ELT  

Store  &  Process  

Modern  Architecture    Serve  

Ac?on  

Process  f  (D1,  DN)   Structured  

Unstructured  

Machine  

Drill  Down  

Human  API  

Ingest  

1  

1  

2  3  

8  ©  Cloudera,  Inc.  All  rights  reserved.  

Opower  Customer  Story  

9  ©  Cloudera,  Inc.  All  rights  reserved.  

Opower  Overview  

The  Company  •  Serving  95+  u?li?es  in  9  countries  

•  Over  5TWh  saved  to  date  

•  40%  of  US  household  data  under  management  totaling  300  billion  reads  

 

Our  DNA  •  Behavioral  science  so^ware  

•  Data  analy?cs  

•  Consumer  marke?ng  

•  User-­‐centric  design  

A  So^ware  as  a  Service  Customer  Engagement  Pla`orm  

10  ©  Cloudera,  Inc.  All  rights  reserved.  

Opower’s  Personalized  Insights  

Neighbor  comparisons   Usage  trend  analysis  

11  ©  Cloudera,  Inc.  All  rights  reserved.  

Ini?al  Hadoop  Architecture    

1  

2  

3  

Ingest  performance  

Complex  query  paths  

1  

3  

2  

Challenges  

Mul?ple  workloads  

12  ©  Cloudera,  Inc.  All  rights  reserved.  

Modern  Hadoop  Architecture    

Offline  Analysis  and  Experimenta?on  Product  Analy?cs  

Ingest  Performance  

Workload  separa?on  3  

1   2  

Improvements  

En?ty-­‐centric  HBase  schema  2   1  

3  

13  ©  Cloudera,  Inc.  All  rights  reserved.  

Insight  Crea?on  Environments  

Insight  Delivery  Insight  Calcula?on  

Product  Calcula8on  and  Delivery   Offline  Analysis  and  Experimenta8on  

Meter reads(gas)

Meter reads(electric)

Bill forecast insight

MapReduce

HBase Site Row

Insight Service

Application

Bulkload ETL

Hive BI Raw MR

Batch Tools

HDFS

Reporting

External FeedsHBase Export

Non-product Insights

14  ©  Cloudera,  Inc.  All  rights  reserved.  

What  does  this  mean  to  end  users?  

Batch  Analy8c  Calcula8ons   Individual  Insight  Query  Latency  

Pre-­‐Hadoop   Modern  Hadoop  

Hours  

12  

24  

48  

Hours  

Days  

Pre-­‐Hadoop  

Second

s  

1  

2  

3  

~10ms  

3  secs  

Analy8c  Development  Time  

Pre-­‐Hadoop  

Mon

ths  

1  

3  

5  

Weeks  

Months  

Modern  Hadoop  Modern  Hadoop  

Thank  you.