Upload
jonathan-seidman
View
2.020
Download
2
Embed Size (px)
DESCRIPTION
Hadoop provides the ability to extract business intelligence from extremely large, heterogeneous data sets that were previously impractical to store and process in traditional data warehouses. The challenge now is in bridging the gap between the data warehouse and Hadoop. In this talk we’ll discuss some steps that Orbitz has taken to bridge this gap, including examples of how Hadoop and Hive are used to aggregate data from large data sets, and how that data can be combined with relational data to create new reports that provide actionable intelligence to business users.
Citation preview
Extending the Enterprise Data Warehouse with Hadoop
Robert Lancaster and Jonathan Seidman
Hadoop World 2011
November 8 | 2011
Who We Are
• Robert Lancaster
– Solutions Architect, Hotel Supply Team
– @rob1lancaster
– Co-organizer of Chicago Big Data and Chicago Machine Learning Study Group.
• Jonathan Seidman
– Lead Engineer, Business Intelligence/Big Data Team
– Co-founder/organizer of Chicago Hadoop User Group and Chicago Big Data
– @jseidman
page 2
page 3
Launched in 2001
Over 160 million bookings
Some History…
page 4
In 2009…
• The Machine Learning team is formed to improve site performance. For example, improving hotel search results.
• This required access to large volumes of behavioral data for analysis.
– Fortunately, the required data was collected in session data stored in web analytics logs.
page 5
The Problem…
• The only archive of the required data went back about two weeks.
page 6
Transactional data (e.g. bookings) and
aggregated Non-transactional data
Data Warehouse
Non-transactional Data (e.g. searches)
Hadoop Provided a Solution…
page 7
Data Warehouse
Detailed non-transactional data
(what every user sees, clicks, etc.)
Hadoop
Transactional data (e.g. bookings) and
aggregated Non-transactional data
Deploying Hadoop Enabled Multiple Applications…
page 8
2.78%
34.30% 31.87%
71.67%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Queries
Searches
And Useful Analyses…
page 9
But Brought New Challenges…
• Most of these efforts are driven by development teams.
• The challenge now is unlocking the value of this data for non-technical users.
page 10
In Early 2011…
• Big Data team is formed under Business Intelligence team at Orbitz Worldwide.
• Allows the Big Data team to work more closely with the data warehouse and BI teams.
• Reflects the importance of big data to the future of the company.
page 11
A View Shared Beyond Orbitz…
page 12
“We strongly believe that Hadoop is the nucleus of the next-generation cloud EDW…”
*James Kobielus, Forrester Research, “Hadoop, Is It Soup Yet?”
“…but that promise is still three to five years from fruition.”*
Two Primary Ways We Use Hadoop to Complement the EDW
• Extraction and transformation of data for loading into the data warehouse – “ETL”.
• Off-loading of analysis from the data warehouse.
page 13
ETL Example: Proposed Dimensional Model
page 14
Raw logs Hadoop Dimensional model
ETL Example: Click Data Processing
page 15
Web Server
Logs ETL DW
Data Cleansing (Stored procedure)
DW Web Server Web Servers
Several hours of processing ~20% original data size
Current Processing in Data Warehouse
ETL Example: Click Data Processing
• Moving to Hadoop will facilitate:
– Remove load from the data warehouse.
– Adding additional attributes for processing.
– Allow processing to be run more frequently.
page 16
Web Server
Logs Hadoop
Data Cleansing (MapReduce) DW
Web Server Web Servers
Proposed Processing in Hadoop
Analysis Example: Geo-Targeting Ads
• Facilitated analysis that allows for more personalized ad content.
• Allowed marketing team to analyze over a years worth of search data.
• Provided analysis that was difficult to perform in the data warehouse.
page 17
BI Vendors Are Working on Hadoop Integration
page 18
Both big (relatively)…
And small…
page 19
Example Processing Pipeline for Web Analytics Data
page 20
page 21
Example Use Case: Selection Errors
Use Case – Selection Errors: Introduction
• Multiple points of entry.
• Multiple paths through site.
• Goal: tie events together to form picture of customer behavior.
page 22
Use Case – Selection Errors: Processing
page 23
Use Case – Selection Errors: Visualization
page 24
page 25
Example Use Case: Beta Data
Use Case – Beta Data: Introduction
• Hotel Sort Optimization
• Compare A vs. B
• Web Analytics Data – What user saw.
– How user behaved
• Server Log Data – Sorting behavior used.
page 26
Use Case – Beta Data Processing
page 27
page 28
Use Case – Beta Data: Visualization
page 29
Example Use Case: RCDC
Use Case – RCDC: Introduction
• Understand and improve cache behavior.
• Improve “coverage”
– Traditionally search 1 page of hotels at a time.
– Get “just enough” information to present to consumers.
– Increase amount of availability information we have when consumer performs a search.
• Data needed to support needs beyond reporting.
page 30
Use Case – RCDC: Processing
page 31
Use Case – RCDC: Visualization
page 32
Conclusions
• Hadoop market is still immature, but growing quickly. Better tools are on the way.
– Look beyond the usual (enterprise) suspects. Many of the most interesting companies in the big data space are small startups.
• Hadoop will not replace your data warehouse, but any organization with a large data warehouse should at least be exploring Hadoop as a complement to their BI infrastructure.
page 33
Conclusions
• Work closely with your existing data management teams.
– Your idea of what constitutes “big data” might quickly diverge from theirs.
• The flip-side to this is that Hadoop can be an excellent tool to off-load resource-consuming jobs from your data warehouse.
page 34