Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Cloud Computing: Hadoop


Published on

Data Processing in the Cloud with Hadoop from Data Services World conference.

Published in: Technology, Education
  • thanks
    it helps to understand what is hadoop in cloud domain
    Are you sure you want to  Yes  No
    Your message goes here
  • Hope fully interesting topic
    Are you sure you want to  Yes  No
    Your message goes here
  • it clarifies many doubts... thank u
    Are you sure you want to  Yes  No
    Your message goes here
  • Hi dude,
    Thank u, It helped a lot.
    Are you sure you want to  Yes  No
    Your message goes here

Cloud Computing: Hadoop

  1. 1. Data Processing in the Cloud Parand Tony Darugar [email_address]
  2. 2. What is Hadoop <ul><li>Flexible infrastructure for large scale computation and data processing on a network of commodity hardware. </li></ul>
  3. 3. Why? <ul><li>A common infrastructure pattern extracted from building distributed systems </li></ul><ul><li>Scale </li></ul><ul><li>Incremental growth </li></ul><ul><li>Cost </li></ul><ul><li>Flexibility </li></ul>
  4. 4. Built-in Resilience to Failure <ul><li>When dealing with large numbers of commodity servers, failure is a fact of life </li></ul><ul><li>Assume failure, build protections and recovery into your architecture </li></ul><ul><ul><li>Data level redundancy </li></ul></ul><ul><ul><li>Job/Task level monitoring and automated restart and re-allocation </li></ul></ul>
  5. 5. Current State of Hadoop Project <ul><li>Top level Apache Foundation project </li></ul><ul><li>In production use at Yahoo, Facebook, Amazon, IBM, Fox, NY Times, Powerset, … </li></ul><ul><li>Large, active user base, mailing lists, user groups </li></ul><ul><li>Very active development, strong development team </li></ul>
  6. 6. Widely Adopted <ul><li>A valuable and reusable skill set </li></ul><ul><ul><li>Taught at major universities </li></ul></ul><ul><ul><li>Easier to hire for </li></ul></ul><ul><ul><li>Easier to train on </li></ul></ul><ul><ul><li>Portable across projects, groups </li></ul></ul>
  7. 7. Plethora of Related Projects <ul><li>Pig </li></ul><ul><li>Hive </li></ul><ul><li>Hbase </li></ul><ul><li>Cascading </li></ul><ul><li>Hadoop on EC2 </li></ul><ul><li>JAQL , X-Trace, Happy, Mahout </li></ul>
  8. 8. What is Hadoop <ul><li>The Linux of distributed processing. </li></ul>
  9. 9. How Does Hadoop Work?
  10. 10. Hadoop File System <ul><li>A distributed file system for large data </li></ul><ul><ul><li>Your data in triplicate </li></ul></ul><ul><ul><li>Built-in redundancy, resiliency to large scale failures </li></ul></ul><ul><ul><li>Intelligent distribution, striping across racks </li></ul></ul><ul><ul><li>Accommodates very large data sizes </li></ul></ul><ul><ul><li>On commodity hardware </li></ul></ul>
  11. 11. Programming Model: Map/Reduce <ul><li>Very simple programming model: </li></ul><ul><ul><li>Map(anything)->key, value </li></ul></ul><ul><ul><li>Sort, partition on key </li></ul></ul><ul><ul><li>Reduce(key,value)->key, value </li></ul></ul><ul><li>No parallel processing / message passing semantics </li></ul><ul><li>Programmable in Java or any other language (streaming) </li></ul>
  12. 12. Processing Model <ul><li>Create or allocate a cluster </li></ul><ul><li>Put data onto the file system: </li></ul><ul><ul><li>Data is split into blocks, stored in triplicate across your cluster </li></ul></ul><ul><li>Run your job: </li></ul><ul><ul><li>Your Map code is copied to the allocated nodes, preferring nodes that contain copies of your data </li></ul></ul><ul><ul><ul><li>Move computation to data, not data to computation </li></ul></ul></ul>
  13. 13. Processing Model <ul><ul><ul><li>Monitor workers, automatically restarting failed or slow tasks </li></ul></ul></ul><ul><ul><li>Gather output of Map, sort and partition on key </li></ul></ul><ul><ul><li>Run Reduce tasks </li></ul></ul><ul><ul><ul><li>Monitor workers, automatically restarting failed or slow tasks </li></ul></ul></ul><ul><li>Results of your job are now available on the Hadoop file system </li></ul>
  14. 14. Hadoop on the Grid <ul><li>Managed Hadoop clusters </li></ul><ul><li>Shared resources </li></ul><ul><ul><li>improved utilization </li></ul></ul><ul><li>Standard data sets, storage </li></ul><ul><li>Shared, standardized operations management </li></ul><ul><li>Hosted internally or externally (eg. on EC2) </li></ul>
  15. 15. Usage Patterns
  16. 16. ETL <ul><li>Put large data source (eg. Log files) onto the Hadoop File System </li></ul><ul><li>Perform aggregations, transformations, normalizations on the data </li></ul><ul><li>Load into RDBMS / data mart </li></ul>
  17. 17. Reporting and Analytics <ul><li>Run canned and ad-hoc queries over large data </li></ul><ul><li>Run analytics and data mining operations on large data </li></ul><ul><li>Produce reports for end-user consumption or loading into data mart </li></ul>
  18. 18. Data Processing Pipelines <ul><li>Multi-step pipelines for data processing </li></ul><ul><li>Coordination, scheduling, data collection and publishing of feeds </li></ul><ul><li>SLA carrying, regularly scheduled jobs </li></ul>
  19. 19. Machine Learning & Graph Algorithms <ul><li>Traverse large graphs and data sets, building models and classifiers </li></ul><ul><li>Implement machine learning algorithms over massive data sets </li></ul>
  20. 20. General Back-End Processing <ul><li>Implement significant portions of back-end, batch oriented processing on the grid </li></ul><ul><li>General computation framework </li></ul><ul><li>Simplify back-end architecture </li></ul>
  21. 21. What Next? <ul><li>Dowload Hadoop: </li></ul><ul><ul><li> </li></ul></ul><ul><li>Try it on your laptop </li></ul><ul><li>Try Pig </li></ul><ul><ul><li> </li></ul></ul><ul><li>Deploy to multiple boxes </li></ul><ul><li>Try it on EC2 </li></ul>