Prototyping Data Intensive Apps: TrendingTopics.org 09/29/09 1 Pete Skomoroch Research Scientist at LinkedIn Consultant at...
Talk Outline <ul><li>TrendingTopics Overview </li></ul><ul><li>Wikipedia Page View Dataset </li></ul><ul><li>Hadoop on Ama...
Data Intensive Web Apps <ul><li>Batch data mining or prediction with Hadoop </li></ul><ul><li>Iterate quickly with high le...
TrendingTopics.org
Daily Pageview Timeline Charts
Detects Rising Trends with Hadoop
TrendingTopics is Open Source <ul><li>Built as a side project at Data Wrangling </li></ul><ul><li>Core code completed over...
Technology Stack
Application Data Flow
Hadoop on Amazon EC2 <ul><li>EC2: Launch servers on demand </li></ul><ul><li>S3: Simple Storage Service </li></ul><ul><li>...
Wikipedia Page View Log Data <ul><li>1 TB of hourly log files from Wikipedia Squid proxy </li></ul><ul><li>Filename contai...
Wikipedia Redirect Data <ul><li>Many articles in the logs redirect to canonical titles </li></ul><ul><li>Amazon Public Dat...
Loading Data into Hadoop on EC2 <ul><li>Hadoop can read directly from Amazon S3 </li></ul><ul><li>Pulling in data from EBS...
Python Streaming for Daily Timelines
Streaming: Filter & Aggregate Logs
Hive Data Warehouse Layer on EC2 <ul><li>Easy for analysts familar with SQL </li></ul><ul><li>Familar syntax (SELECT, GROU...
Fixing Wiki Redirects with Hive JOIN
Trend Detection With Hadoop
Hourly Data: “Java” vs. “Hangover”
Robust Regression with R & Hadoop  <ul><li>Fit R model to hourly data for 2.3M Wiki articles </li></ul><ul><li>“ Trending”...
Call Python Trend Mapper from Hive
Simple Trend Calculation in Python
Use Trend Scores to Rank Articles
Exporting Data from our EC2 Cluster <ul><li>Output files are delimited text for bulk DB load </li></ul><ul><li>Distcp outp...
Hooking It All Together
Ruby on Rails Front End <ul><li>Run a development version of the app locally: </li></ul>
Automate & Deploy with EC2onRails <ul><li>EC2onRails: Ubuntu server images for EC2, runs instances of Ruby on Rails, MySQL...
Cron Triggers Hadoop Jobs on EC2 <ul><li>Modify Cloudera hadoop-ec2-init-remote.sh boot script </li></ul>
Bash Scripts Automate Daily Run <ul><li>run_daily_merge.sh  kicks off Hadoop/Hive jobs </li></ul><ul><li>Queries MySQL for...
Capistrano Tasks <ul><li>Capistrano: Ruby tool for automating tasks on one or more remote servers ( www.capify.org ) </li>...
Visualization APIs <ul><li>Google Viz API: Annotated Timeline for Rails </li></ul><ul><li>Google Charts sparklines: gc4r  ...
Ideas & Next Steps <ul><li>Try out the code on Github </li></ul><ul><li>Show related articles using link graph data </li><...
LinkedIn  A nalytics  Team We’re Hiring
Questions? <ul><li>Contact </li></ul><ul><ul><li>Blog:  datawrangling.com </li></ul></ul><ul><ul><li>Twitter: @peteskomoro...
Backup slides
Wikipedia Views ~ Google Searches “ Dwight Howard” on TrendingTopics “ Dwight Howard” on Google Trends
Wikipedia Log Analysis with Pig <ul><li>Most Read articles Pig Latin example </li></ul><ul><li>Show junk Wikipedia pages t...
Upcoming SlideShare
Loading in...5
×

Hw09 Building Data Intensive Apps A Closer Look At Trending Topics.Org

2,659

Published on

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
2,659
On Slideshare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
89
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Hw09 Building Data Intensive Apps A Closer Look At Trending Topics.Org

  1. 1. Prototyping Data Intensive Apps: TrendingTopics.org 09/29/09 1 Pete Skomoroch Research Scientist at LinkedIn Consultant at Data Wrangling @peteskomoroch
  2. 2. Talk Outline <ul><li>TrendingTopics Overview </li></ul><ul><li>Wikipedia Page View Dataset </li></ul><ul><li>Hadoop on Amazon EC2 </li></ul><ul><li>Loading Data on EC2: Amazon EBS & S3 </li></ul><ul><li>Daily Timelines with Hadoop Streaming </li></ul><ul><li>Hive Data Warehouse Layer </li></ul><ul><li>Trend Computation with Hive </li></ul><ul><li>Hooking It All Together </li></ul><ul><li>Front End & Visualizations </li></ul>
  3. 3. Data Intensive Web Apps <ul><li>Batch data mining or prediction with Hadoop </li></ul><ul><li>Iterate quickly with high level languages & tools </li></ul><ul><ul><li>Pig, Hive, Clojure, Cascading, Python, Ruby </li></ul></ul><ul><li>EC2: Get running with limited initial capital </li></ul><ul><li>Use external data and APIs in novel ways </li></ul><ul><li>Recent real world example: FlightCaster </li></ul>
  4. 4. TrendingTopics.org
  5. 5. Daily Pageview Timeline Charts
  6. 6. Detects Rising Trends with Hadoop
  7. 7. TrendingTopics is Open Source <ul><li>Built as a side project at Data Wrangling </li></ul><ul><li>Core code completed over 2 weeks in June </li></ul><ul><li>Code on Github </li></ul><ul><li>Data released on Amazon Public Datasets </li></ul>
  8. 8. Technology Stack
  9. 9. Application Data Flow
  10. 10. Hadoop on Amazon EC2 <ul><li>EC2: Launch servers on demand </li></ul><ul><li>S3: Simple Storage Service </li></ul><ul><li>EBS: Persistent disks for EC2 </li></ul><ul><li>Used Cloudera Hadoop Distribution </li></ul><ul><ul><li>Includes Pig, Hive, HBase, Sqoop, EBS integration </li></ul></ul><ul><ul><li>Allows customization of Hadoop EC2 instances </li></ul></ul><ul><ul><li>See EC2 Docs and Tutorials on the Cloudera site </li></ul></ul>
  11. 11. Wikipedia Page View Log Data <ul><li>1 TB of hourly log files from Wikipedia Squid proxy </li></ul><ul><li>Filename contains timestamp </li></ul><ul><li>Columns: project, pagename, pageviews, & bytes </li></ul>
  12. 12. Wikipedia Redirect Data <ul><li>Many articles in the logs redirect to canonical titles </li></ul><ul><li>Amazon Public Dataset contains a lookup table: </li></ul>
  13. 13. Loading Data into Hadoop on EC2 <ul><li>Hadoop can read directly from Amazon S3 </li></ul><ul><li>Pulling in data from EBS snapshots </li></ul>
  14. 14. Python Streaming for Daily Timelines
  15. 15. Streaming: Filter & Aggregate Logs
  16. 16. Hive Data Warehouse Layer on EC2 <ul><li>Easy for analysts familar with SQL </li></ul><ul><li>Familar syntax (SELECT, GROUP BY, SORT...) </li></ul><ul><li>Hide the MR muck, for JOIN, UNION, etc. </li></ul><ul><li>Supports streaming UDFs </li></ul><ul><li>Sampling: generate datasets for R&D </li></ul><ul><li>Used on Timeline and Redirect data </li></ul>
  17. 17. Fixing Wiki Redirects with Hive JOIN
  18. 18. Trend Detection With Hadoop
  19. 19. Hourly Data: “Java” vs. “Hangover”
  20. 20. Robust Regression with R & Hadoop <ul><li>Fit R model to hourly data for 2.3M Wiki articles </li></ul><ul><li>“ Trending” if article views >> model prediction </li></ul>Page View “Spikes”
  21. 21. Call Python Trend Mapper from Hive
  22. 22. Simple Trend Calculation in Python
  23. 23. Use Trend Scores to Rank Articles
  24. 24. Exporting Data from our EC2 Cluster <ul><li>Output files are delimited text for bulk DB load </li></ul><ul><li>Distcp outputs to S3: </li></ul><ul><ul><li>hadoop distcp /user/output s3n://trendingtopics/exports </li></ul></ul><ul><li>Hive can create tables directly over S3 filesystem: </li></ul><ul><ul><li>create external table foo ... location 's3n://trendingtopics/hiveoutput’ </li></ul></ul>
  25. 25. Hooking It All Together
  26. 26. Ruby on Rails Front End <ul><li>Run a development version of the app locally: </li></ul>
  27. 27. Automate & Deploy with EC2onRails <ul><li>EC2onRails: Ubuntu server images for EC2, runs instances of Ruby on Rails, MySQL, Apache </li></ul><ul><li>Daily cron job on App server launches Hadoop </li></ul><ul><li>Hadoop EC2 boot script checks out latest trend detection code from Github </li></ul><ul><li>Capistrano tasks deploy new application code, maintain DB, cache, archives </li></ul>
  28. 28. Cron Triggers Hadoop Jobs on EC2 <ul><li>Modify Cloudera hadoop-ec2-init-remote.sh boot script </li></ul>
  29. 29. Bash Scripts Automate Daily Run <ul><li>run_daily_merge.sh kicks off Hadoop/Hive jobs </li></ul><ul><li>Queries MySQL for last run date </li></ul><ul><li>Generates timelines with Hadoop streaming </li></ul><ul><li>Hive operates on Timeline data, computes Trends with Python </li></ul><ul><li>Hive output copied to S3 </li></ul><ul><li>Calls remote script to load MySQL staging tables </li></ul><ul><li>Hadoop cluster self terminates </li></ul><ul><li>Remote script swaps staging and prod after load </li></ul>
  30. 30. Capistrano Tasks <ul><li>Capistrano: Ruby tool for automating tasks on one or more remote servers ( www.capify.org ) </li></ul><ul><li>Deploys Rails code and cron jobs to App server </li></ul><ul><li>Cleanup tasks are triggered at end of load: </li></ul><ul><ul><li>Flags featured Wikipedia articles </li></ul></ul><ul><ul><li>Indexes MySQL tables, Flushes cache </li></ul></ul>
  31. 31. Visualization APIs <ul><li>Google Viz API: Annotated Timeline for Rails </li></ul><ul><li>Google Charts sparklines: gc4r </li></ul><ul><li>Yahoo BOSS image search in Rails </li></ul>
  32. 32. Ideas & Next Steps <ul><li>Try out the code on Github </li></ul><ul><li>Show related articles using link graph data </li></ul><ul><li>Extract trending phrases from article text </li></ul><ul><li>Boost Twitter trending topics with Wikipedia logs </li></ul><ul><li>Update top trends hourly </li></ul><ul><li>Use date partitioned tables, incremental updates </li></ul><ul><li>Add a real search engine (uses MySQL now) </li></ul>
  33. 33. LinkedIn A nalytics Team We’re Hiring
  34. 34. Questions? <ul><li>Contact </li></ul><ul><ul><li>Blog: datawrangling.com </li></ul></ul><ul><ul><li>Twitter: @peteskomoroch </li></ul></ul><ul><li>Code </li></ul><ul><ul><li>http://github.com/datawrangling/trendingtopics </li></ul></ul><ul><ul><li>Hadoop blog posts & tutorials on the Cloudera site </li></ul></ul><ul><ul><li>Amazon EMR tutorial on Finding Similar Artists with Hadoop & Last.FM data http://bit.ly/EMRsimilarity </li></ul></ul>
  35. 35. Backup slides
  36. 36. Wikipedia Views ~ Google Searches “ Dwight Howard” on TrendingTopics “ Dwight Howard” on Google Trends
  37. 37. Wikipedia Log Analysis with Pig <ul><li>Most Read articles Pig Latin example </li></ul><ul><li>Show junk Wikipedia pages that need filtering </li></ul>
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×